Merge remote-tracking branch 'origin/master' into lazy-trees

.github/workflows/backport.yml

@@ -8,7 +8,7 @@ jobs:
     if: github.repository_owner == 'NixOS' && github.event.pull_request.merged == true && (github.event_name != 'labeled' || startsWith('backport', github.event.label.name))
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           ref: ${{ github.event.pull_request.head.sha }}
           # required to find all branches

.github/workflows/ci.yml

@@ -14,10 +14,10 @@ jobs:
     runs-on: ${{ matrix.os }}
     timeout-minutes: 60
     steps:
-    - uses: actions/checkout@v2.4.0
+    - uses: actions/checkout@v3
       with:
         fetch-depth: 0
-    - uses: cachix/install-nix-action@v16
+    - uses: cachix/install-nix-action@v17
     - run: echo CACHIX_NAME="$(echo $GITHUB_REPOSITORY-install-tests | tr "[A-Z]/" "[a-z]-")" >> $GITHUB_ENV
     - uses: cachix/cachix-action@v10
       if: needs.check_cachix.outputs.secret == 'true'
@@ -46,11 +46,11 @@ jobs:
     outputs:
       installerURL: ${{ steps.prepare-installer.outputs.installerURL }}
     steps:
-    - uses: actions/checkout@v2.4.0
+    - uses: actions/checkout@v3
       with:
         fetch-depth: 0
     - run: echo CACHIX_NAME="$(echo $GITHUB_REPOSITORY-install-tests | tr "[A-Z]/" "[a-z]-")" >> $GITHUB_ENV
-    - uses: cachix/install-nix-action@v16
+    - uses: cachix/install-nix-action@v17
     - uses: cachix/cachix-action@v10
       with:
         name: '${{ env.CACHIX_NAME }}'
@@ -67,9 +67,9 @@ jobs:
         os: [ubuntu-latest, macos-latest]
     runs-on: ${{ matrix.os }}
     steps:
-    - uses: actions/checkout@v2.4.0
+    - uses: actions/checkout@v3
     - run: echo CACHIX_NAME="$(echo $GITHUB_REPOSITORY-install-tests | tr "[A-Z]/" "[a-z]-")" >> $GITHUB_ENV
-    - uses: cachix/install-nix-action@v16
+    - uses: cachix/install-nix-action@v17
       with:
         install_url: '${{needs.installer.outputs.installerURL}}'
         install_options: "--tarball-url-prefix https://${{ env.CACHIX_NAME }}.cachix.org/serve"
@@ -83,10 +83,10 @@ jobs:
       needs.check_cachix.outputs.secret == 'true'
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v2.4.0
+    - uses: actions/checkout@v3
       with:
         fetch-depth: 0
-    - uses: cachix/install-nix-action@v16
+    - uses: cachix/install-nix-action@v17
     - run: echo CACHIX_NAME="$(echo $GITHUB_REPOSITORY-install-tests | tr "[A-Z]/" "[a-z]-")" >> $GITHUB_ENV
     - run: echo NIX_VERSION="$(nix-instantiate --eval -E '(import ./default.nix).defaultPackage.${builtins.currentSystem}.version' | tr -d \")" >> $GITHUB_ENV
     - uses: cachix/cachix-action@v10

.github/workflows/hydra_status.yml

@@ -9,7 +9,7 @@ jobs:
     if: github.repository_owner == 'NixOS'
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v2.4.0
+    - uses: actions/checkout@v3
       with:
         fetch-depth: 0
     - run: bash scripts/check-hydra-status.sh

.gitignore

@@ -79,6 +79,7 @@ perl/Makefile.config
 /tests/shell.drv
 /tests/config.nix
 /tests/ca/config.nix
+/tests/repl-result-out
 
 # /tests/lang/
 /tests/lang/*.out
@@ -90,6 +91,7 @@ perl/Makefile.config
 
 /misc/systemd/nix-daemon.service
 /misc/systemd/nix-daemon.socket
+/misc/systemd/nix-daemon.conf
 /misc/upstart/nix-daemon.conf
 
 /src/resolve-system-dependencies/resolve-system-dependencies

.version

@@ -1 +1 @@
-2.8.0
+2.9.0

doc/manual/generate-options.nix

@@ -6,7 +6,8 @@ options:
 concatStrings (map
   (name:
     let option = options.${name}; in
-    "  - `${name}`  \n\n"
+    "  - [`${name}`](#conf-${name})"
+    + "<p id=\"conf-${name}\"></p>\n\n"
     + concatStrings (map (s: "    ${s}\n") (splitLines option.description)) + "\n\n"
     + (if option.documentDefault
        then "    **Default:** " + (

doc/manual/src/SUMMARY.md.in

@@ -72,6 +72,7 @@
   - [CLI guideline](contributing/cli-guideline.md)
 - [Release Notes](release-notes/release-notes.md)
   - [Release X.Y (202?-??-??)](release-notes/rl-next.md)
+  - [Release 2.8 (2022-04-19)](release-notes/rl-2.8.md)
   - [Release 2.7 (2022-03-07)](release-notes/rl-2.7.md)
   - [Release 2.6 (2022-01-24)](release-notes/rl-2.6.md)
   - [Release 2.5 (2021-12-13)](release-notes/rl-2.5.md)

doc/manual/src/release-notes/rl-2.7.md

@@ -1,4 +1,4 @@
-# Release X.Y (2022-03-07)
+# Release 2.7 (2022-03-07)
 
 * Nix will now make some helpful suggestions when you mistype
   something on the command line. For instance, if you type `nix build

doc/manual/src/release-notes/rl-2.8.md

@@ -0,0 +1,53 @@
+# Release 2.8 (2022-04-19)
+
+* New experimental command: `nix fmt`, which applies a formatter
+  defined by the `formatter.<system>` flake output to the Nix
+  expressions in a flake.
+
+* Various Nix commands can now read expressions from standard input
+  using `--file -`.
+
+* New experimental builtin function `builtins.fetchClosure` that
+  copies a closure from a binary cache at evaluation time and rewrites
+  it to content-addressed form (if it isn't already). Like
+  `builtins.storePath`, this allows importing pre-built store paths;
+  the difference is that it doesn't require the user to configure
+  binary caches and trusted public keys.
+
+  This function is only available if you enable the experimental
+  feature `fetch-closure`.
+
+* New experimental feature: *impure derivations*. These are
+  derivations that can produce a different result every time they're
+  built. Here is an example:
+
+  ```nix
+  stdenv.mkDerivation {
+    name = "impure";
+    __impure = true; # marks this derivation as impure
+    buildCommand = "date > $out";
+  }
+  ```
+
+  Running `nix build` twice on this expression will build the
+  derivation twice, producing two different content-addressed store
+  paths. Like fixed-output derivations, impure derivations have access
+  to the network. Only fixed-output derivations and impure derivations
+  can depend on an impure derivation.
+
+* `nix store make-content-addressable` has been renamed to `nix store
+  make-content-addressed`.
+
+* The `nixosModule` flake output attribute has been renamed consistent
+  with the `.default` renames in Nix 2.7.
+
+  * `nixosModule` → `nixosModules.default`
+
+  As before, the old output will continue to work, but `nix flake check` will
+  issue a warning about it.
+
+* `nix run` is now stricter in what it accepts: members of the `apps`
+  flake output are now required to be apps (as defined in [the
+  manual](https://nixos.org/manual/nix/stable/command-ref/new-cli/nix3-run.html#apps)),
+  and members of `packages` or `legacyPackages` must be derivations
+  (not apps).

doc/manual/src/release-notes/rl-next.md

@@ -1,16 +1,26 @@
 # Release X.Y (202?-??-??)
 
-* Various nix commands can now read expressions from stdin with `--file -`.
+* Nix now provides better integration with zsh's run-help feature. It is now
+  included in the Nix installation in the form of an autoloadable shell
+  function, run-help-nix. It picks up Nix subcommands from the currently typed
+  in command and directs the user to the associated man pages.
 
-* `nix store make-content-addressable` has been renamed to `nix store
-  make-content-addressed`.
+* `nix repl` has a new build-'n-link (`:bl`) command that builds a derivation
+  while creating GC root symlinks.
 
-* New experimental builtin function `builtins.fetchClosure` that
-  copies a closure from a binary cache at evaluation time and rewrites
-  it to content-addressed form (if it isn't already). Like
-  `builtins.storePath`, this allows importing pre-built store paths;
-  the difference is that it doesn't require the user to configure
-  binary caches and trusted public keys.
+* The path produced by `builtins.toFile` is now allowed to be imported or read
+  even with restricted evaluation. Note that this will not work with a
+  read-only store.
 
-  This function is only available if you enable the experimental
-  feature `fetch-closure`.
+* `nix build` has a new `--print-out-paths` flag to print the resulting output paths.
+  This matches the default behaviour of `nix-build`.
+
+* You can now specify which outputs of a derivation `nix` should
+  operate on using the syntax `installable^outputs`,
+  e.g. `nixpkgs#glibc^dev,static` or `nixpkgs#glibc^*`. By default,
+  `nix` will use the outputs specified by the derivation's
+  `meta.outputsToInstall` attribute if it exists, or all outputs
+  otherwise.
+
+  Selecting derivation outputs using the attribute selection syntax
+  (e.g. `nixpkgs#glibc.dev`) no longer works.

docker.nix

@@ -22,6 +22,7 @@ let
     findutils
     iana-etc
     git
+    openssh
   ];
 
   users = {

flake.lock

@@ -18,11 +18,11 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1632864508,
-        "narHash": "sha256-d127FIvGR41XbVRDPVvozUPQ/uRHbHwvfyKHwEt5xFM=",
+        "lastModified": 1645296114,
+        "narHash": "sha256-y53N7TyIkXsjMpOG7RhvqJFGDacLs9HlyHeSTBioqYU=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "82891b5e2c2359d7e58d08849e4c89511ab94234",
+        "rev": "530a53dcbc9437363471167a5e4762c5fcfa34a1",
         "type": "github"
       },
       "original": {

flake.nix

@@ -23,7 +23,7 @@
 
       crossSystems = [ "armv6l-linux" "armv7l-linux" ];
 
-      stdenvs = [ "gccStdenv" "clangStdenv" "clang11Stdenv" "stdenv" ];
+      stdenvs = [ "gccStdenv" "clangStdenv" "clang11Stdenv" "stdenv" "libcxxStdenv" ];
 
       forAllSystems = f: nixpkgs.lib.genAttrs systems (system: f system);
       forAllSystemsAndStdenvs = f: forAllSystems (system:

misc/bash/completion.sh

@@ -15,7 +15,7 @@ function _complete_nix {
         else
             COMPREPLY+=("$completion")
         fi
-    done < <(NIX_GET_COMPLETIONS=$cword "${words[@]/#\~/$HOME}" 2>/dev/null)
+    done < <(NIX_GET_COMPLETIONS=$cword "${words[@]}" 2>/dev/null)
     __ltrim_colon_completions "$cur"
 }
 

misc/systemd/nix-daemon.service.in

@@ -3,6 +3,7 @@ Description=Nix Daemon
 Documentation=man:nix-daemon https://nixos.org/manual
 RequiresMountsFor=@storedir@
 RequiresMountsFor=@localstatedir@
+RequiresMountsFor=@localstatedir@/nix/db
 ConditionPathIsReadWrite=@localstatedir@/nix/daemon-socket
 
 [Service]

misc/zsh/local.mk

@@ -1 +1,2 @@
 $(eval $(call install-file-as, $(d)/completion.zsh, $(datarootdir)/zsh/site-functions/_nix, 0644))
+$(eval $(call install-file-as, $(d)/run-help-nix, $(datarootdir)/zsh/site-functions/run-help-nix, 0644))

misc/zsh/run-help-nix

@@ -0,0 +1,42 @@
+emulate -L zsh
+
+# run-help is a zsh widget that can be bound to a key. It mainly looks up the
+# man page for the currently typed in command.
+#
+# Although run-help works for any command without requiring special support,
+# it can only deduce the right man page based solely on the name of the
+# command. Programs like Nix provide better integration with run-help by
+# helping zsh identify Nix subcommands and their corresponding man pages. This
+# is what this function does.
+#
+# To actually use run-help on zsh, place the following lines in your .zshrc:
+#
+#     (( $+aliases[run-help] )) && unalias run-help
+#     autoload -Uz run-help run-help-nix
+#
+# Then also assign run-help to any key of choice:
+#
+#     bindkey '^[h' run-help
+
+while [[ "$#" != 0 && "$1" == -* ]]; do
+  shift
+done
+
+local -a subcommands; subcommands=( nix3 )
+
+local arg
+for arg in "$@"; do
+  if man -w "${(j:-:)subcommands}-$arg" >/dev/null 2>&1; then
+    subcommands+="$arg"
+  else
+    break
+  fi
+done
+
+if (( $#subcommands > 1 )); then
+  man "${(j:-:)subcommands}"
+else
+  man nix
+fi
+
+return $?

scripts/check-hydra-status.sh

@@ -14,7 +14,7 @@ curl -sS -H 'Accept: application/json' https://hydra.nixos.org/jobset/nix/master
 someBuildFailed=0
 
 for buildId in $BUILDS_FOR_LATEST_EVAL; do
-  buildInfo=$(curl -sS -H 'Accept: application/json' "https://hydra.nixos.org/build/$buildId")
+  buildInfo=$(curl --fail -sS -H 'Accept: application/json' "https://hydra.nixos.org/build/$buildId")
 
   finished=$(echo "$buildInfo" | jq -r '.finished')
 

scripts/install-multi-user.sh

@@ -423,6 +423,18 @@ EOF
         fi
     done
 
+    if [ "$(uname -s)" = "Linux" ] && [ ! -e /run/systemd/system ]; then
+        warning <<EOF
+We did not detect systemd on your system. With a multi-user install
+without systemd you will have to manually configure your init system to
+launch the Nix daemon after installation.
+EOF
+        if ! ui_confirm "Do you want to proceed with a multi-user installation?"; then
+            failure <<EOF
+You have aborted the installation.
+EOF
+        fi
+    fi
 }
 
 setup_report() {

scripts/install-systemd-multi-user.sh

@@ -90,7 +90,7 @@ poly_configure_nix_daemon_service() {
               ln -sfn /nix/var/nix/profiles/default/$TMPFILES_SRC $TMPFILES_DEST
 
         _sudo "to run systemd-tmpfiles once to pick that path up" \
-             systemd-tmpfiles create --prefix=/nix/var/nix
+             systemd-tmpfiles --create --prefix=/nix/var/nix
 
         _sudo "to set up the nix-daemon service" \
               systemctl link "/nix/var/nix/profiles/default$SERVICE_SRC"

scripts/install.in

@@ -82,7 +82,7 @@ if [ "$(uname -s)" != "Darwin" ]; then
 fi
 
 if command -v curl > /dev/null 2>&1; then
-    fetch() { curl -L "$1" -o "$2"; }
+    fetch() { curl --fail -L "$1" -o "$2"; }
 elif command -v wget > /dev/null 2>&1; then
     fetch() { wget "$1" -O "$2"; }
 else

src/libcmd/command.cc

@@ -197,17 +197,17 @@ void StorePathCommand::run(ref<Store> store, std::vector<StorePath> && storePath
     run(store, *storePaths.begin());
 }
 
-Strings editorFor(const Pos & pos)
+Strings editorFor(const Path & file, uint32_t line)
 {
     auto editor = getEnv("EDITOR").value_or("cat");
     auto args = tokenizeString<Strings>(editor);
-    if (pos.line > 0 && (
+    if (line > 0 && (
         editor.find("emacs") != std::string::npos ||
         editor.find("nano") != std::string::npos ||
         editor.find("vim") != std::string::npos ||
         editor.find("kak") != std::string::npos))
-        args.push_back(fmt("+%d", pos.line));
-    args.push_back(pos.file);
+        args.push_back(fmt("+%d", line));
+    args.push_back(file);
     return args;
 }
 

src/libcmd/command.hh

@@ -85,11 +85,12 @@ struct SourceExprCommand : virtual Args, MixFlakeOptions
 {
     std::optional<Path> file;
     std::optional<std::string> expr;
+    bool readOnlyMode = false;
 
     // FIXME: move this; not all commands (e.g. 'nix run') use it.
     OperateOn operateOn = OperateOn::Output;
 
-    SourceExprCommand();
+    SourceExprCommand(bool supportReadOnlyMode = false);
 
     std::vector<std::shared_ptr<Installable>> parseInstallables(
         ref<Store> store, std::vector<std::string> ss);
@@ -128,13 +129,13 @@ struct InstallableCommand : virtual Args, SourceExprCommand
 {
     std::shared_ptr<Installable> installable;
 
-    InstallableCommand();
+    InstallableCommand(bool supportReadOnlyMode = false);
 
     void prepare() override;
 
     std::optional<FlakeRef> getFlakeRefForCompletion() override
     {
-        return parseFlakeRef(_installable, absPath("."));
+        return parseFlakeRefWithFragment(_installable, absPath(".")).first;
     }
 
 private:
@@ -218,7 +219,7 @@ static RegisterCommand registerCommand2(std::vector<std::string> && name)
 
 /* Helper function to generate args that invoke $EDITOR on
    filename:lineno. */
-Strings editorFor(const Pos & pos);
+Strings editorFor(const Path & file, uint32_t line);
 
 struct MixProfile : virtual StoreCommand
 {

src/libcmd/common-eval-args.cc

@@ -7,6 +7,7 @@
 #include "registry.hh"
 #include "flake/flakeref.hh"
 #include "store-api.hh"
+#include "command.hh"
 
 namespace nix {
 
@@ -59,6 +60,9 @@ MixEvalArgs::MixEvalArgs()
             fetchers::Attrs extraAttrs;
             if (to.subdir != "") extraAttrs["dir"] = to.subdir;
             fetchers::overrideRegistry(from.input, to.input, extraAttrs);
+        }},
+        .completer = {[&](size_t, std::string_view prefix) {
+            completeFlakeRef(openStore(), prefix);
         }}
     });
 

src/libcmd/installables.cc

@@ -1,4 +1,6 @@
+#include "globals.hh"
 #include "installables.hh"
+#include "util.hh"
 #include "command.hh"
 #include "attr-path.hh"
 #include "common-eval-args.hh"
@@ -99,6 +101,14 @@ MixFlakeOptions::MixFlakeOptions()
             lockFlags.inputOverrides.insert_or_assign(
                 flake::parseInputPath(inputPath),
                 parseFlakeRef(flakeRef, absPath("."), true));
+        }},
+        .completer = {[&](size_t n, std::string_view prefix) {
+            if (n == 0) {
+                if (auto flakeRef = getFlakeRefForCompletion())
+                    completeFlakeInputPath(getEvalState(), *flakeRef, prefix);
+            } else if (n == 1) {
+                completeFlakeRef(getEvalState()->store, prefix);
+            }
         }}
     });
 
@@ -129,7 +139,7 @@ MixFlakeOptions::MixFlakeOptions()
     });
 }
 
-SourceExprCommand::SourceExprCommand()
+SourceExprCommand::SourceExprCommand(bool supportReadOnlyMode)
 {
     addFlag({
         .longName = "file",
@@ -157,6 +167,17 @@ SourceExprCommand::SourceExprCommand()
         .category = installablesCategory,
         .handler = {&operateOn, OperateOn::Derivation},
     });
+
+    if (supportReadOnlyMode) {
+        addFlag({
+            .longName = "read-only",
+            .description =
+                "Do not instantiate each evaluated derivation. "
+                "This improves performance, but can cause errors when accessing "
+                "store paths of derivations during evaluation.",
+            .handler = {&readOnlyMode, true},
+        });
+    }
 }
 
 Strings SourceExprCommand::getDefaultFlakeAttrPaths()
@@ -182,6 +203,8 @@ Strings SourceExprCommand::getDefaultFlakeAttrPathPrefixes()
 void SourceExprCommand::completeInstallable(std::string_view prefix)
 {
     if (file) {
+        completionType = ctAttrs;
+
         evalSettings.pureEval = false;
         auto state = getEvalState();
         Expr *e =
@@ -212,13 +235,14 @@ void SourceExprCommand::completeInstallable(std::string_view prefix)
         Value v2;
         state->autoCallFunction(*autoArgs, v1, v2);
 
-        completionType = ctAttrs;
-
         if (v2.type() == nAttrs) {
             for (auto & i : *v2.attrs) {
-                std::string name = i.name;
+                std::string name = state->symbols[i.name];
                 if (name.find(searchWord) == 0) {
-                    completions->add(i.name);
+                    if (prefix_ == "")
+                        completions->add(name);
+                    else
+                        completions->add(prefix_ + "." + name);
                 }
             }
         }
@@ -246,10 +270,11 @@ void completeFlakeRefWithFragment(
         if (hash == std::string::npos) {
             completeFlakeRef(evalState->store, prefix);
         } else {
+            completionType = ctAttrs;
+
             auto fragment = prefix.substr(hash + 1);
             auto flakeRefS = std::string(prefix.substr(0, hash));
-            // FIXME: do tilde expansion.
-            auto flakeRef = parseFlakeRef(flakeRefS, absPath("."));
+            auto flakeRef = parseFlakeRef(expandTilde(flakeRefS), absPath("."));
 
             auto evalCache = openEvalCache(*evalState,
                 std::make_shared<flake::LockedFlake>(lockFlake(*evalState, flakeRef, lockFlags)));
@@ -261,8 +286,6 @@ void completeFlakeRefWithFragment(
                flake. */
             attrPathPrefixes.push_back("");
 
-            completionType = ctAttrs;
-
             for (auto & attrPathPrefixS : attrPathPrefixes) {
                 auto attrPathPrefix = parseAttrPath(*evalState, attrPathPrefixS);
                 auto attrPathS = attrPathPrefixS + std::string(fragment);
@@ -270,7 +293,7 @@ void completeFlakeRefWithFragment(
 
                 std::string lastAttr;
                 if (!attrPath.empty() && !hasSuffix(attrPathS, ".")) {
-                    lastAttr = attrPath.back();
+                    lastAttr = evalState->symbols[attrPath.back()];
                     attrPath.pop_back();
                 }
 
@@ -278,11 +301,11 @@ void completeFlakeRefWithFragment(
                 if (!attr) continue;
 
                 for (auto & attr2 : (*attr)->getAttrs()) {
-                    if (hasPrefix(attr2, lastAttr)) {
+                    if (hasPrefix(evalState->symbols[attr2], lastAttr)) {
                         auto attrPath2 = (*attr)->getAttrPath(attr2);
                         /* Strip the attrpath prefix. */
                         attrPath2.erase(attrPath2.begin(), attrPath2.begin() + attrPathPrefix.size());
-                        completions->add(flakeRefS + "#" + concatStringsSep(".", attrPath2));
+                        completions->add(flakeRefS + "#" + concatStringsSep(".", evalState->symbols.resolve(attrPath2)));
                     }
                 }
             }
@@ -336,16 +359,16 @@ DerivedPath Installable::toDerivedPath()
     return std::move(buildables[0]);
 }
 
-std::vector<std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>>
+std::vector<ref<eval_cache::AttrCursor>>
 Installable::getCursors(EvalState & state)
 {
     auto evalCache =
         std::make_shared<nix::eval_cache::EvalCache>(std::nullopt, state,
             [&]() { return toValue(state).first; });
-    return {{evalCache->getRoot(), ""}};
+    return {evalCache->getRoot()};
 }
 
-std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>
+ref<eval_cache::AttrCursor>
 Installable::getCursor(EvalState & state)
 {
     auto cursors = getCursors(state);
@@ -419,10 +442,8 @@ DerivedPaths InstallableValue::toDerivedPaths()
 
     // Group by derivation, helps with .all in particular
     for (auto & drv : toDerivations()) {
-        auto outputName = drv.outputName;
-        if (outputName == "")
-            throw Error("derivation '%s' lacks an 'outputName' attribute", state->store->printStorePath(drv.drvPath));
-        drvsToOutputs[drv.drvPath].insert(outputName);
+        for (auto & outputName : drv.outputsToInstall)
+            drvsToOutputs[drv.drvPath].insert(outputName);
         drvsToCopy.insert(drv.drvPath);
     }
 
@@ -445,14 +466,24 @@ struct InstallableAttrPath : InstallableValue
     SourceExprCommand & cmd;
     RootValue v;
     std::string attrPath;
+    OutputsSpec outputsSpec;
 
-    InstallableAttrPath(ref<EvalState> state, SourceExprCommand & cmd, Value * v, const std::string & attrPath)
-        : InstallableValue(state), cmd(cmd), v(allocRootValue(v)), attrPath(attrPath)
+    InstallableAttrPath(
+        ref<EvalState> state,
+        SourceExprCommand & cmd,
+        Value * v,
+        const std::string & attrPath,
+        OutputsSpec outputsSpec)
+        : InstallableValue(state)
+        , cmd(cmd)
+        , v(allocRootValue(v))
+        , attrPath(attrPath)
+        , outputsSpec(std::move(outputsSpec))
     { }
 
     std::string what() const override { return attrPath; }
 
-    std::pair<Value *, Pos> toValue(EvalState & state) override
+    std::pair<Value *, PosIdx> toValue(EvalState & state) override
     {
         auto [vRes, pos] = findAlongAttrPath(state, attrPath, *cmd.getAutoArgs(state), **v);
         state.forceValue(*vRes, pos);
@@ -476,7 +507,19 @@ std::vector<InstallableValue::DerivationInfo> InstallableAttrPath::toDerivations
         auto drvPath = drvInfo.queryDrvPath();
         if (!drvPath)
             throw Error("'%s' is not a derivation", what());
-        res.push_back({ *drvPath, drvInfo.queryOutputName() });
+
+        std::set<std::string> outputsToInstall;
+
+        if (auto outputNames = std::get_if<OutputNames>(&outputsSpec))
+            outputsToInstall = *outputNames;
+        else
+            for (auto & output : drvInfo.queryOutputs(false, std::get_if<DefaultOutputs>(&outputsSpec)))
+                outputsToInstall.insert(output.first);
+
+        res.push_back(DerivationInfo {
+            .drvPath = *drvPath,
+            .outputsToInstall = std::move(outputsToInstall)
+        });
     }
 
     return res;
@@ -553,6 +596,7 @@ InstallableFlake::InstallableFlake(
     ref<EvalState> state,
     FlakeRef && flakeRef,
     std::string_view fragment,
+    OutputsSpec outputsSpec,
     Strings attrPaths,
     Strings prefixes,
     const flake::LockFlags & lockFlags)
@@ -560,6 +604,7 @@ InstallableFlake::InstallableFlake(
       flakeRef(flakeRef),
       attrPaths(fragment == "" ? attrPaths : Strings{(std::string) fragment}),
       prefixes(fragment == "" ? Strings{} : prefixes),
+      outputsSpec(std::move(outputsSpec)),
       lockFlags(lockFlags)
 {
     if (cmd && cmd->getAutoArgs(*state)->size())
@@ -568,43 +613,41 @@ InstallableFlake::InstallableFlake(
 
 std::tuple<std::string, FlakeRef, InstallableValue::DerivationInfo> InstallableFlake::toDerivation()
 {
-    auto lockedFlake = getLockedFlake();
+    auto attr = getCursor(*state);
 
-    auto cache = openEvalCache(*state, lockedFlake);
-    auto root = cache->getRoot();
+    auto attrPath = attr->getAttrPathStr();
 
-    Suggestions suggestions;
+    if (!attr->isDerivation())
+        throw Error("flake output attribute '%s' is not a derivation", attrPath);
 
-    for (auto & attrPath : getActualAttrPaths()) {
-        debug("trying flake output attribute '%s'", attrPath);
+    auto drvPath = attr->forceDerivation();
 
-        auto attrOrSuggestions = root->findAlongAttrPath(
-            parseAttrPath(*state, attrPath),
-            true
-        );
+    std::set<std::string> outputsToInstall;
 
-        if (!attrOrSuggestions) {
-            suggestions += attrOrSuggestions.getSuggestions();
-            continue;
-        }
+    if (auto aMeta = attr->maybeGetAttr(state->sMeta))
+        if (auto aOutputsToInstall = aMeta->maybeGetAttr("outputsToInstall"))
+            for (auto & s : aOutputsToInstall->getListOfStrings())
+                outputsToInstall.insert(s);
 
-        auto attr = *attrOrSuggestions;
-
-        if (!attr->isDerivation())
-            throw Error("flake output attribute '%s' is not a derivation", attrPath);
-
-        auto drvPath = attr->forceDerivation();
-
-        auto drvInfo = DerivationInfo {
-            std::move(drvPath),
-            attr->getAttr(state->sOutputName)->getString()
-        };
-
-        return {attrPath, lockedFlake->flake.lockedRef, std::move(drvInfo)};
+    if (outputsToInstall.empty() || std::get_if<AllOutputs>(&outputsSpec)) {
+        outputsToInstall.clear();
+        if (auto aOutputs = attr->maybeGetAttr(state->sOutputs))
+            for (auto & s : aOutputs->getListOfStrings())
+                outputsToInstall.insert(s);
     }
 
-    throw Error(suggestions, "flake '%s' does not provide attribute %s",
-        flakeRef, showAttrPaths(getActualAttrPaths()));
+    if (outputsToInstall.empty())
+        outputsToInstall.insert("out");
+
+    if (auto outputNames = std::get_if<OutputNames>(&outputsSpec))
+        outputsToInstall = *outputNames;
+
+    auto drvInfo = DerivationInfo {
+        .drvPath = std::move(drvPath),
+        .outputsToInstall = std::move(outputsToInstall),
+    };
+
+    return {attrPath, getLockedFlake()->flake.lockedRef, std::move(drvInfo)};
 }
 
 std::vector<InstallableValue::DerivationInfo> InstallableFlake::toDerivations()
@@ -614,35 +657,12 @@ std::vector<InstallableValue::DerivationInfo> InstallableFlake::toDerivations()
     return res;
 }
 
-std::pair<Value *, Pos> InstallableFlake::toValue(EvalState & state)
+std::pair<Value *, PosIdx> InstallableFlake::toValue(EvalState & state)
 {
-    auto lockedFlake = getLockedFlake();
-
-    auto vOutputs = getFlakeOutputs(state, *lockedFlake);
-
-    auto emptyArgs = state.allocBindings(0);
-
-    Suggestions suggestions;
-
-    for (auto & attrPath : getActualAttrPaths()) {
-        try {
-            auto [v, pos] = findAlongAttrPath(state, attrPath, *emptyArgs, *vOutputs);
-            state.forceValue(*v, pos);
-            return {v, pos};
-        } catch (AttrPathNotFound & e) {
-            suggestions += e.info().suggestions;
-        }
-    }
-
-    throw Error(
-        suggestions,
-        "flake '%s' does not provide attribute %s",
-        flakeRef,
-        showAttrPaths(getActualAttrPaths())
-    );
+    return {&getCursor(state)->forceValue(), noPos};
 }
 
-std::vector<std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>>
+std::vector<ref<eval_cache::AttrCursor>>
 InstallableFlake::getCursors(EvalState & state)
 {
     auto evalCache = openEvalCache(state,
@@ -650,21 +670,55 @@ InstallableFlake::getCursors(EvalState & state)
 
     auto root = evalCache->getRoot();
 
-    std::vector<std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>> res;
+    std::vector<ref<eval_cache::AttrCursor>> res;
 
     for (auto & attrPath : getActualAttrPaths()) {
         auto attr = root->findAlongAttrPath(parseAttrPath(state, attrPath));
-        if (attr) res.push_back({*attr, attrPath});
+        if (attr) res.push_back(ref(*attr));
     }
 
     return res;
 }
 
+ref<eval_cache::AttrCursor> InstallableFlake::getCursor(EvalState & state)
+{
+    auto lockedFlake = getLockedFlake();
+
+    auto cache = openEvalCache(state, lockedFlake);
+    auto root = cache->getRoot();
+
+    Suggestions suggestions;
+
+    auto attrPaths = getActualAttrPaths();
+
+    for (auto & attrPath : attrPaths) {
+        debug("trying flake output attribute '%s'", attrPath);
+
+        auto attrOrSuggestions = root->findAlongAttrPath(
+            parseAttrPath(state, attrPath),
+            true
+        );
+
+        if (!attrOrSuggestions) {
+            suggestions += attrOrSuggestions.getSuggestions();
+            continue;
+        }
+
+        return *attrOrSuggestions;
+    }
+
+    throw Error(
+        suggestions,
+        "flake '%s' does not provide attribute %s",
+        flakeRef,
+        showAttrPaths(attrPaths));
+}
+
 std::shared_ptr<flake::LockedFlake> InstallableFlake::getLockedFlake() const
 {
-    flake::LockFlags lockFlagsApplyConfig = lockFlags;
-    lockFlagsApplyConfig.applyNixConfig = true;
     if (!_lockedFlake) {
+        flake::LockFlags lockFlagsApplyConfig = lockFlags;
+        lockFlagsApplyConfig.applyNixConfig = true;
         _lockedFlake = std::make_shared<flake::LockedFlake>(lockFlake(*state, flakeRef, lockFlagsApplyConfig));
     }
     return _lockedFlake;
@@ -689,6 +743,10 @@ std::vector<std::shared_ptr<Installable>> SourceExprCommand::parseInstallables(
 {
     std::vector<std::shared_ptr<Installable>> result;
 
+    if (readOnlyMode) {
+        settings.readOnlyMode = true;
+    }
+
     if (file || expr) {
         if (file && expr)
             throw UsageError("'--file' and '--expr' are exclusive");
@@ -710,8 +768,14 @@ std::vector<std::shared_ptr<Installable>> SourceExprCommand::parseInstallables(
             state->eval(e, *vFile);
         }
 
-        for (auto & s : ss)
-            result.push_back(std::make_shared<InstallableAttrPath>(state, *this, vFile, s == "." ? "" : s));
+        for (auto & s : ss) {
+            auto [prefix, outputsSpec] = parseOutputsSpec(s);
+            result.push_back(
+                std::make_shared<InstallableAttrPath>(
+                    state, *this, vFile,
+                    prefix == "." ? "" : prefix,
+                    outputsSpec));
+        }
 
     } else {
 
@@ -730,12 +794,13 @@ std::vector<std::shared_ptr<Installable>> SourceExprCommand::parseInstallables(
             }
 
             try {
-                auto [flakeRef, fragment] = parseFlakeRefWithFragment(s, absPath("."));
+                auto [flakeRef, fragment, outputsSpec] = parseFlakeRefWithFragmentAndOutputsSpec(s, absPath("."));
                 result.push_back(std::make_shared<InstallableFlake>(
                         this,
                         getEvalState(),
                         std::move(flakeRef),
                         fragment,
+                        outputsSpec,
                         getDefaultFlakeAttrPaths(),
                         getDefaultFlakeAttrPathPrefixes(),
                         lockFlags));
@@ -809,26 +874,28 @@ std::vector<std::pair<std::shared_ptr<Installable>, BuiltPath>> Installable::bui
                         auto outputHashes = staticOutputHashes(*evalStore, drv); // FIXME: expensive
                         auto drvOutputs = drv.outputsAndOptPaths(*store);
                         for (auto & output : bfd.outputs) {
-                            if (!outputHashes.count(output))
+                            auto outputHash = get(outputHashes, output);
+                            if (!outputHash)
                                 throw Error(
                                     "the derivation '%s' doesn't have an output named '%s'",
                                     store->printStorePath(bfd.drvPath), output);
                             if (settings.isExperimentalFeatureEnabled(Xp::CaDerivations)) {
-                                DrvOutput outputId { outputHashes.at(output), output };
+                                DrvOutput outputId { *outputHash, output };
                                 auto realisation = store->queryRealisation(outputId);
                                 if (!realisation)
                                     throw Error(
-                                        "cannot operate on an output of unbuilt "
-                                        "content-addressed derivation '%s'",
+                                        "cannot operate on an output of the "
+                                        "unbuilt derivation '%s'",
                                         outputId.to_string());
                                 outputs.insert_or_assign(output, realisation->outPath);
                             } else {
                                 // If ca-derivations isn't enabled, assume that
                                 // the output path is statically known.
-                                assert(drvOutputs.count(output));
-                                assert(drvOutputs.at(output).second);
+                                auto drvOutput = get(drvOutputs, output);
+                                assert(drvOutput);
+                                assert(drvOutput->second);
                                 outputs.insert_or_assign(
-                                    output, *drvOutputs.at(output).second);
+                                    output, *drvOutput->second);
                             }
                         }
                         res.push_back({installable, BuiltPath::Built { bfd.drvPath, outputs }});
@@ -957,7 +1024,7 @@ InstallablesCommand::InstallablesCommand()
 void InstallablesCommand::prepare()
 {
     if (_installables.empty() && useDefaultInstallables())
-        // FIXME: commands like "nix install" should not have a
+        // FIXME: commands like "nix profile install" should not have a
         // default, probably.
         _installables.push_back(".");
     installables = parseInstallables(getStore(), _installables);
@@ -967,13 +1034,14 @@ std::optional<FlakeRef> InstallablesCommand::getFlakeRefForCompletion()
 {
     if (_installables.empty()) {
         if (useDefaultInstallables())
-            return parseFlakeRef(".", absPath("."));
+            return parseFlakeRefWithFragment(".", absPath(".")).first;
         return {};
     }
-    return parseFlakeRef(_installables.front(), absPath("."));
+    return parseFlakeRefWithFragment(_installables.front(), absPath(".")).first;
 }
 
-InstallableCommand::InstallableCommand()
+InstallableCommand::InstallableCommand(bool supportReadOnlyMode)
+    : SourceExprCommand(supportReadOnlyMode)
 {
     expectArgs({
         .label = "installable",

src/libcmd/installables.hh

@@ -68,7 +68,7 @@ struct Installable
 
     UnresolvedApp toApp(EvalState & state);
 
-    virtual std::pair<Value *, Pos> toValue(EvalState & state)
+    virtual std::pair<Value *, PosIdx> toValue(EvalState & state)
     {
         throw Error("argument '%s' cannot be evaluated", what());
     }
@@ -80,10 +80,10 @@ struct Installable
         return {};
     }
 
-    virtual std::vector<std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>>
+    virtual std::vector<ref<eval_cache::AttrCursor>>
     getCursors(EvalState & state);
 
-    std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>
+    virtual ref<eval_cache::AttrCursor>
     getCursor(EvalState & state);
 
     virtual FlakeRef nixpkgsFlakeRef() const
@@ -141,7 +141,7 @@ struct InstallableValue : Installable
     struct DerivationInfo
     {
         StorePath drvPath;
-        std::string outputName;
+        std::set<std::string> outputsToInstall;
     };
 
     virtual std::vector<DerivationInfo> toDerivations() = 0;
@@ -156,6 +156,7 @@ struct InstallableFlake : InstallableValue
     FlakeRef flakeRef;
     Strings attrPaths;
     Strings prefixes;
+    OutputsSpec outputsSpec;
     const flake::LockFlags & lockFlags;
     mutable std::shared_ptr<flake::LockedFlake> _lockedFlake;
 
@@ -164,6 +165,7 @@ struct InstallableFlake : InstallableValue
         ref<EvalState> state,
         FlakeRef && flakeRef,
         std::string_view fragment,
+        OutputsSpec outputsSpec,
         Strings attrPaths,
         Strings prefixes,
         const flake::LockFlags & lockFlags);
@@ -178,11 +180,17 @@ struct InstallableFlake : InstallableValue
 
     std::vector<DerivationInfo> toDerivations() override;
 
-    std::pair<Value *, Pos> toValue(EvalState & state) override;
+    std::pair<Value *, PosIdx> toValue(EvalState & state) override;
 
-    std::vector<std::pair<std::shared_ptr<eval_cache::AttrCursor>, std::string>>
+    /* Get a cursor to every attrpath in getActualAttrPaths() that
+       exists. */
+    std::vector<ref<eval_cache::AttrCursor>>
     getCursors(EvalState & state) override;
 
+    /* Get a cursor to the first attrpath in getActualAttrPaths() that
+       exists, or throw an exception with suggestions if none exists. */
+    ref<eval_cache::AttrCursor> getCursor(EvalState & state) override;
+
     std::shared_ptr<flake::LockedFlake> getLockedFlake() const;
 
     FlakeRef nixpkgsFlakeRef() const override;

src/libcmd/markdown.cc

@@ -9,10 +9,12 @@ namespace nix {
 
 std::string renderMarkdownToTerminal(std::string_view markdown)
 {
+    int windowWidth = getWindowSize().second;
+
     struct lowdown_opts opts {
         .type = LOWDOWN_TERM,
         .maxdepth = 20,
-        .cols = std::max(getWindowSize().second, (unsigned short) 80),
+        .cols = (size_t) std::max(windowWidth - 5, 60),
         .hmargin = 0,
         .vmargin = 0,
         .feat = LOWDOWN_COMMONMARK | LOWDOWN_FENCED | LOWDOWN_DEFLIST | LOWDOWN_TABLES,

src/libexpr/attr-path.cc

@@ -41,13 +41,13 @@ std::vector<Symbol> parseAttrPath(EvalState & state, std::string_view s)
 }
 
 
-std::pair<Value *, Pos> findAlongAttrPath(EvalState & state, const std::string & attrPath,
+std::pair<Value *, PosIdx> findAlongAttrPath(EvalState & state, const std::string & attrPath,
     Bindings & autoArgs, Value & vIn)
 {
     Strings tokens = parseAttrPath(attrPath);
 
     Value * v = &vIn;
-    Pos pos = noPos;
+    PosIdx pos = noPos;
 
     for (auto & attr : tokens) {
 
@@ -77,13 +77,13 @@ std::pair<Value *, Pos> findAlongAttrPath(EvalState & state, const std::string &
             if (a == v->attrs->end()) {
                 std::set<std::string> attrNames;
                 for (auto & attr : *v->attrs)
-                    attrNames.insert(attr.name);
+                    attrNames.insert(state.symbols[attr.name]);
 
                 auto suggestions = Suggestions::bestMatches(attrNames, attr);
                 throw AttrPathNotFound(suggestions, "attribute '%1%' in selection path '%2%' not found", attr, attrPath);
             }
             v = &*a->value;
-            pos = *a->pos;
+            pos = a->pos;
         }
 
         else {
@@ -106,7 +106,7 @@ std::pair<Value *, Pos> findAlongAttrPath(EvalState & state, const std::string &
 }
 
 
-Pos findPackageFilename(EvalState & state, Value & v, std::string what)
+std::pair<std::string, uint32_t> findPackageFilename(EvalState & state, Value & v, std::string what)
 {
     Value * v2;
     try {
@@ -132,9 +132,7 @@ Pos findPackageFilename(EvalState & state, Value & v, std::string what)
         throw ParseError("cannot parse line number '%s'", pos);
     }
 
-    Symbol file = state.symbols.create(filename);
-
-    return { foFile, file, lineno, 0 };
+    return { std::move(filename), lineno };
 }
 
 

src/libexpr/attr-path.hh

@@ -10,14 +10,14 @@ namespace nix {
 MakeError(AttrPathNotFound, Error);
 MakeError(NoPositionInfo, Error);
 
-std::pair<Value *, Pos> findAlongAttrPath(
+std::pair<Value *, PosIdx> findAlongAttrPath(
     EvalState & state,
     const std::string & attrPath,
     Bindings & autoArgs,
     Value & vIn);
 
 /* Heuristic to find the filename and lineno or a nix value. */
-Pos findPackageFilename(EvalState & state, Value & v, std::string what);
+std::pair<std::string, uint32_t> findPackageFilename(EvalState & state, Value & v, std::string what);
 
 std::vector<Symbol> parseAttrPath(EvalState & state, std::string_view s);
 

src/libexpr/attr-set.cc

@@ -26,7 +26,7 @@ Bindings * EvalState::allocBindings(size_t capacity)
 /* Create a new attribute named 'name' on an existing attribute set stored
    in 'vAttrs' and return the newly allocated Value which is associated with
    this attribute. */
-Value * EvalState::allocAttr(Value & vAttrs, const Symbol & name)
+Value * EvalState::allocAttr(Value & vAttrs, Symbol name)
 {
     Value * v = allocValue();
     vAttrs.attrs->push_back(Attr(name, v));
@@ -40,7 +40,7 @@ Value * EvalState::allocAttr(Value & vAttrs, std::string_view name)
 }
 
 
-Value & BindingsBuilder::alloc(const Symbol & name, ptr<Pos> pos)
+Value & BindingsBuilder::alloc(Symbol name, PosIdx pos)
 {
     auto value = state.allocValue();
     bindings->push_back(Attr(name, value, pos));
@@ -48,7 +48,7 @@ Value & BindingsBuilder::alloc(const Symbol & name, ptr<Pos> pos)
 }
 
 
-Value & BindingsBuilder::alloc(std::string_view name, ptr<Pos> pos)
+Value & BindingsBuilder::alloc(std::string_view name, PosIdx pos)
 {
     return alloc(state.symbols.create(name), pos);
 }

src/libexpr/attr-set.hh

@@ -15,18 +15,27 @@ struct Value;
 /* Map one attribute name to its value. */
 struct Attr
 {
+    /* the placement of `name` and `pos` in this struct is important.
+       both of them are uint32 wrappers, they are next to each other
+       to make sure that Attr has no padding on 64 bit machines. that
+       way we keep Attr size at two words with no wasted space. */
     Symbol name;
+    PosIdx pos;
     Value * value;
-    ptr<Pos> pos;
-    Attr(Symbol name, Value * value, ptr<Pos> pos = ptr(&noPos))
-        : name(name), value(value), pos(pos) { };
-    Attr() : pos(&noPos) { };
+    Attr(Symbol name, Value * value, PosIdx pos = noPos)
+        : name(name), pos(pos), value(value) { };
+    Attr() { };
     bool operator < (const Attr & a) const
     {
         return name < a.name;
     }
 };
 
+static_assert(sizeof(Attr) == 2 * sizeof(uint32_t) + sizeof(Value *),
+    "performance of the evaluator is highly sensitive to the size of Attr. "
+    "avoid introducing any padding into Attr if at all possible, and do not "
+    "introduce new fields that need not be present for almost every instance.");
+
 /* Bindings contains all the attributes of an attribute set. It is defined
    by its size and its capacity, the capacity being the number of Attr
    elements allocated after this structure, while the size corresponds to
@@ -35,13 +44,13 @@ class Bindings
 {
 public:
     typedef uint32_t size_t;
-    ptr<Pos> pos;
+    PosIdx pos;
 
 private:
     size_t size_, capacity_;
     Attr attrs[0];
 
-    Bindings(size_t capacity) : pos(&noPos), size_(0), capacity_(capacity) { }
+    Bindings(size_t capacity) : size_(0), capacity_(capacity) { }
     Bindings(const Bindings & bindings) = delete;
 
 public:
@@ -57,7 +66,7 @@ public:
         attrs[size_++] = attr;
     }
 
-    iterator find(const Symbol & name)
+    iterator find(Symbol name)
     {
         Attr key(name, 0);
         iterator i = std::lower_bound(begin(), end(), key);
@@ -65,7 +74,7 @@ public:
         return end();
     }
 
-    Attr * get(const Symbol & name)
+    Attr * get(Symbol name)
     {
         Attr key(name, 0);
         iterator i = std::lower_bound(begin(), end(), key);
@@ -73,18 +82,6 @@ public:
         return nullptr;
     }
 
-    Attr & need(const Symbol & name, const Pos & pos = noPos)
-    {
-        auto a = get(name);
-        if (!a)
-            throw Error({
-                .msg = hintfmt("attribute '%s' missing", name),
-                .errPos = pos
-            });
-
-        return *a;
-    }
-
     iterator begin() { return &attrs[0]; }
     iterator end() { return &attrs[size_]; }
 
@@ -98,14 +95,15 @@ public:
     size_t capacity() { return capacity_; }
 
     /* Returns the attributes in lexicographically sorted order. */
-    std::vector<const Attr *> lexicographicOrder() const
+    std::vector<const Attr *> lexicographicOrder(const SymbolTable & symbols) const
     {
         std::vector<const Attr *> res;
         res.reserve(size_);
         for (size_t n = 0; n < size_; n++)
             res.emplace_back(&attrs[n]);
-        std::sort(res.begin(), res.end(), [](const Attr * a, const Attr * b) {
-            return (const std::string &) a->name < (const std::string &) b->name;
+        std::sort(res.begin(), res.end(), [&](const Attr * a, const Attr * b) {
+            std::string_view sa = symbols[a->name], sb = symbols[b->name];
+            return sa < sb;
         });
         return res;
     }
@@ -130,7 +128,7 @@ public:
         : bindings(bindings), state(state)
     { }
 
-    void insert(Symbol name, Value * value, ptr<Pos> pos = ptr(&noPos))
+    void insert(Symbol name, Value * value, PosIdx pos = noPos)
     {
         insert(Attr(name, value, pos));
     }
@@ -145,9 +143,9 @@ public:
         bindings->push_back(attr);
     }
 
-    Value & alloc(const Symbol & name, ptr<Pos> pos = ptr(&noPos));
+    Value & alloc(Symbol name, PosIdx pos = noPos);
 
-    Value & alloc(std::string_view name, ptr<Pos> pos = ptr(&noPos));
+    Value & alloc(std::string_view name, PosIdx pos = noPos);
 
     Bindings * finish()
     {

src/libexpr/eval-cache.cc

@@ -35,13 +35,19 @@ struct AttrDb
 
     std::unique_ptr<Sync<State>> _state;
 
-    AttrDb(const Store & cfg, const Hash & fingerprint)
+    SymbolTable & symbols;
+
+    AttrDb(
+        const Store & cfg,
+        const Hash & fingerprint,
+        SymbolTable & symbols)
         : cfg(cfg)
         , _state(std::make_unique<Sync<State>>())
+        , symbols(symbols)
     {
         auto state(_state->lock());
 
-        Path cacheDir = getCacheDir() + "/nix/eval-cache-v2";
+        Path cacheDir = getCacheDir() + "/nix/eval-cache-v3";
         createDirs(cacheDir);
 
         Path dbPath = cacheDir + "/" + fingerprint.to_string(Base16, false) + ".sqlite";
@@ -100,7 +106,7 @@ struct AttrDb
 
             state->insertAttribute.use()
                 (key.first)
-                (key.second)
+                (symbols[key.second])
                 (AttrType::FullAttrs)
                 (0, false).exec();
 
@@ -110,7 +116,7 @@ struct AttrDb
             for (auto & attr : attrs)
                 state->insertAttribute.use()
                     (rowId)
-                    (attr)
+                    (symbols[attr])
                     (AttrType::Placeholder)
                     (0, false).exec();
 
@@ -135,14 +141,14 @@ struct AttrDb
                 }
                 state->insertAttributeWithContext.use()
                     (key.first)
-                    (key.second)
+                    (symbols[key.second])
                     (AttrType::String)
                     (s)
                     (ctx).exec();
             } else {
                 state->insertAttribute.use()
                     (key.first)
-                    (key.second)
+                    (symbols[key.second])
                     (AttrType::String)
                 (s).exec();
             }
@@ -161,7 +167,7 @@ struct AttrDb
 
             state->insertAttribute.use()
                 (key.first)
-                (key.second)
+                (symbols[key.second])
                 (AttrType::Bool)
                 (b ? 1 : 0).exec();
 
@@ -169,6 +175,24 @@ struct AttrDb
         });
     }
 
+    AttrId setListOfStrings(
+        AttrKey key,
+        const std::vector<std::string> & l)
+    {
+        return doSQLite([&]()
+        {
+            auto state(_state->lock());
+
+            state->insertAttribute.use()
+                (key.first)
+                (symbols[key.second])
+                (AttrType::ListOfStrings)
+                (concatStringsSep("\t", l)).exec();
+
+            return state->db.getLastInsertedRowId();
+        });
+    }
+
     AttrId setPlaceholder(AttrKey key)
     {
         return doSQLite([&]()
@@ -177,7 +201,7 @@ struct AttrDb
 
             state->insertAttribute.use()
                 (key.first)
-                (key.second)
+                (symbols[key.second])
                 (AttrType::Placeholder)
                 (0, false).exec();
 
@@ -193,7 +217,7 @@ struct AttrDb
 
             state->insertAttribute.use()
                 (key.first)
-                (key.second)
+                (symbols[key.second])
                 (AttrType::Missing)
                 (0, false).exec();
 
@@ -209,7 +233,7 @@ struct AttrDb
 
             state->insertAttribute.use()
                 (key.first)
-                (key.second)
+                (symbols[key.second])
                 (AttrType::Misc)
                 (0, false).exec();
 
@@ -225,7 +249,7 @@ struct AttrDb
 
             state->insertAttribute.use()
                 (key.first)
-                (key.second)
+                (symbols[key.second])
                 (AttrType::Failed)
                 (0, false).exec();
 
@@ -233,13 +257,11 @@ struct AttrDb
         });
     }
 
-    std::optional<std::pair<AttrId, AttrValue>> getAttr(
-        AttrKey key,
-        SymbolTable & symbols)
+    std::optional<std::pair<AttrId, AttrValue>> getAttr(AttrKey key)
     {
         auto state(_state->lock());
 
-        auto queryAttribute(state->queryAttribute.use()(key.first)(key.second));
+        auto queryAttribute(state->queryAttribute.use()(key.first)(symbols[key.second]));
         if (!queryAttribute.next()) return {};
 
         auto rowId = (AttrType) queryAttribute.getInt(0);
@@ -253,7 +275,7 @@ struct AttrDb
                 std::vector<Symbol> attrs;
                 auto queryAttributes(state->queryAttributes.use()(rowId));
                 while (queryAttributes.next())
-                    attrs.push_back(symbols.create(queryAttributes.getStr(0)));
+                    attrs.emplace_back(symbols.create(queryAttributes.getStr(0)));
                 return {{rowId, attrs}};
             }
             case AttrType::String: {
@@ -265,6 +287,8 @@ struct AttrDb
             }
             case AttrType::Bool:
                 return {{rowId, queryAttribute.getInt(2) != 0}};
+            case AttrType::ListOfStrings:
+                return {{rowId, tokenizeString<std::vector<std::string>>(queryAttribute.getStr(2), "\t")}};
             case AttrType::Missing:
                 return {{rowId, missing_t()}};
             case AttrType::Misc:
@@ -277,10 +301,13 @@ struct AttrDb
     }
 };
 
-static std::shared_ptr<AttrDb> makeAttrDb(const Store & cfg, const Hash & fingerprint)
+static std::shared_ptr<AttrDb> makeAttrDb(
+    const Store & cfg,
+    const Hash & fingerprint,
+    SymbolTable & symbols)
 {
     try {
-        return std::make_shared<AttrDb>(cfg, fingerprint);
+        return std::make_shared<AttrDb>(cfg, fingerprint, symbols);
     } catch (SQLiteError &) {
         ignoreException();
         return nullptr;
@@ -291,7 +318,7 @@ EvalCache::EvalCache(
     std::optional<std::reference_wrapper<const Hash>> useCache,
     EvalState & state,
     RootLoader rootLoader)
-    : db(useCache ? makeAttrDb(*state.store, *useCache) : nullptr)
+    : db(useCache ? makeAttrDb(*state.store, *useCache, state.symbols) : nullptr)
     , state(state)
     , rootLoader(rootLoader)
 {
@@ -306,9 +333,9 @@ Value * EvalCache::getRootValue()
     return *value;
 }
 
-std::shared_ptr<AttrCursor> EvalCache::getRoot()
+ref<AttrCursor> EvalCache::getRoot()
 {
-    return std::make_shared<AttrCursor>(ref(shared_from_this()), std::nullopt);
+    return make_ref<AttrCursor>(ref(shared_from_this()), std::nullopt);
 }
 
 AttrCursor::AttrCursor(
@@ -327,8 +354,7 @@ AttrKey AttrCursor::getKey()
     if (!parent)
         return {0, root->state.sEpsilon};
     if (!parent->first->cachedValue) {
-        parent->first->cachedValue = root->db->getAttr(
-            parent->first->getKey(), root->state.symbols);
+        parent->first->cachedValue = root->db->getAttr(parent->first->getKey());
         assert(parent->first->cachedValue);
     }
     return {parent->first->cachedValue->first, parent->second};
@@ -369,17 +395,17 @@ std::vector<Symbol> AttrCursor::getAttrPath(Symbol name) const
 
 std::string AttrCursor::getAttrPathStr() const
 {
-    return concatStringsSep(".", getAttrPath());
+    return concatStringsSep(".", root->state.symbols.resolve(getAttrPath()));
 }
 
 std::string AttrCursor::getAttrPathStr(Symbol name) const
 {
-    return concatStringsSep(".", getAttrPath(name));
+    return concatStringsSep(".", root->state.symbols.resolve(getAttrPath(name)));
 }
 
 Value & AttrCursor::forceValue()
 {
-    debug("evaluating uncached attribute %s", getAttrPathStr());
+    debug("evaluating uncached attribute '%s'", getAttrPathStr());
 
     auto & v = getValue();
 
@@ -417,25 +443,25 @@ Suggestions AttrCursor::getSuggestionsForAttr(Symbol name)
     auto attrNames = getAttrs();
     std::set<std::string> strAttrNames;
     for (auto & name : attrNames)
-        strAttrNames.insert(std::string(name));
+        strAttrNames.insert(root->state.symbols[name]);
 
-    return Suggestions::bestMatches(strAttrNames, name);
+    return Suggestions::bestMatches(strAttrNames, root->state.symbols[name]);
 }
 
 std::shared_ptr<AttrCursor> AttrCursor::maybeGetAttr(Symbol name, bool forceErrors)
 {
     if (root->db) {
         if (!cachedValue)
-            cachedValue = root->db->getAttr(getKey(), root->state.symbols);
+            cachedValue = root->db->getAttr(getKey());
 
         if (cachedValue) {
             if (auto attrs = std::get_if<std::vector<Symbol>>(&cachedValue->second)) {
                 for (auto & attr : *attrs)
                     if (attr == name)
-                        return std::make_shared<AttrCursor>(root, std::make_pair(shared_from_this(), name));
+                        return std::make_shared<AttrCursor>(root, std::make_pair(shared_from_this(), attr));
                 return nullptr;
             } else if (std::get_if<placeholder_t>(&cachedValue->second)) {
-                auto attr = root->db->getAttr({cachedValue->first, name}, root->state.symbols);
+                auto attr = root->db->getAttr({cachedValue->first, name});
                 if (attr) {
                     if (std::get_if<missing_t>(&attr->second))
                         return nullptr;
@@ -525,7 +551,7 @@ std::string AttrCursor::getString()
 {
     if (root->db) {
         if (!cachedValue)
-            cachedValue = root->db->getAttr(getKey(), root->state.symbols);
+            cachedValue = root->db->getAttr(getKey());
         if (cachedValue && !std::get_if<placeholder_t>(&cachedValue->second)) {
             if (auto s = std::get_if<string_t>(&cachedValue->second)) {
                 debug("using cached string attribute '%s'", getAttrPathStr());
@@ -547,7 +573,7 @@ string_t AttrCursor::getStringWithContext()
 {
     if (root->db) {
         if (!cachedValue)
-            cachedValue = root->db->getAttr(getKey(), root->state.symbols);
+            cachedValue = root->db->getAttr(getKey());
         if (cachedValue && !std::get_if<placeholder_t>(&cachedValue->second)) {
             if (auto s = std::get_if<string_t>(&cachedValue->second)) {
                 bool valid = true;
@@ -580,7 +606,7 @@ bool AttrCursor::getBool()
 {
     if (root->db) {
         if (!cachedValue)
-            cachedValue = root->db->getAttr(getKey(), root->state.symbols);
+            cachedValue = root->db->getAttr(getKey());
         if (cachedValue && !std::get_if<placeholder_t>(&cachedValue->second)) {
             if (auto b = std::get_if<bool>(&cachedValue->second)) {
                 debug("using cached Boolean attribute '%s'", getAttrPathStr());
@@ -598,11 +624,44 @@ bool AttrCursor::getBool()
     return v.boolean;
 }
 
+std::vector<std::string> AttrCursor::getListOfStrings()
+{
+    if (root->db) {
+        if (!cachedValue)
+            cachedValue = root->db->getAttr(getKey());
+        if (cachedValue && !std::get_if<placeholder_t>(&cachedValue->second)) {
+            if (auto l = std::get_if<std::vector<std::string>>(&cachedValue->second)) {
+                debug("using cached list of strings attribute '%s'", getAttrPathStr());
+                return *l;
+            } else
+                throw TypeError("'%s' is not a list of strings", getAttrPathStr());
+        }
+    }
+
+    debug("evaluating uncached attribute '%s'", getAttrPathStr());
+
+    auto & v = getValue();
+    root->state.forceValue(v, noPos);
+
+    if (v.type() != nList)
+        throw TypeError("'%s' is not a list", getAttrPathStr());
+
+    std::vector<std::string> res;
+
+    for (auto & elem : v.listItems())
+        res.push_back(std::string(root->state.forceStringNoCtx(*elem)));
+
+    if (root->db)
+        cachedValue = {root->db->setListOfStrings(getKey(), res), res};
+
+    return res;
+}
+
 std::vector<Symbol> AttrCursor::getAttrs()
 {
     if (root->db) {
         if (!cachedValue)
-            cachedValue = root->db->getAttr(getKey(), root->state.symbols);
+            cachedValue = root->db->getAttr(getKey());
         if (cachedValue && !std::get_if<placeholder_t>(&cachedValue->second)) {
             if (auto attrs = std::get_if<std::vector<Symbol>>(&cachedValue->second)) {
                 debug("using cached attrset attribute '%s'", getAttrPathStr());
@@ -620,8 +679,9 @@ std::vector<Symbol> AttrCursor::getAttrs()
     std::vector<Symbol> attrs;
     for (auto & attr : *getValue().attrs)
         attrs.push_back(attr.name);
-    std::sort(attrs.begin(), attrs.end(), [](const Symbol & a, const Symbol & b) {
-        return (const std::string &) a < (const std::string &) b;
+    std::sort(attrs.begin(), attrs.end(), [&](Symbol a, Symbol b) {
+        std::string_view sa = root->state.symbols[a], sb = root->state.symbols[b];
+        return sa < sb;
     });
 
     if (root->db)

src/libexpr/eval-cache.hh

@@ -33,7 +33,7 @@ public:
         EvalState & state,
         RootLoader rootLoader);
 
-    std::shared_ptr<AttrCursor> getRoot();
+    ref<AttrCursor> getRoot();
 };
 
 enum AttrType {
@@ -44,6 +44,7 @@ enum AttrType {
     Misc = 4,
     Failed = 5,
     Bool = 6,
+    ListOfStrings = 7,
 };
 
 struct placeholder_t {};
@@ -61,7 +62,8 @@ typedef std::variant<
     missing_t,
     misc_t,
     failed_t,
-    bool
+    bool,
+    std::vector<std::string>
     > AttrValue;
 
 class AttrCursor : public std::enable_shared_from_this<AttrCursor>
@@ -104,6 +106,8 @@ public:
 
     ref<AttrCursor> getAttr(std::string_view name);
 
+    /* Get an attribute along a chain of attrsets. Note that this does
+       not auto-call functors or functions. */
     OrSuggestions<ref<AttrCursor>> findAlongAttrPath(const std::vector<Symbol> & attrPath, bool force = false);
 
     std::string getString();
@@ -112,6 +116,8 @@ public:
 
     bool getBool();
 
+    std::vector<std::string> getListOfStrings();
+
     std::vector<Symbol> getAttrs();
 
     bool isDerivation();

src/libexpr/eval-inline.hh

@@ -2,27 +2,8 @@
 
 #include "eval.hh"
 
-#define LocalNoInline(f) static f __attribute__((noinline)); f
-#define LocalNoInlineNoReturn(f) static f __attribute__((noinline, noreturn)); f
-
 namespace nix {
 
-LocalNoInlineNoReturn(void throwEvalError(const Pos & pos, const char * s))
-{
-    throw EvalError({
-        .msg = hintfmt(s),
-        .errPos = pos
-    });
-}
-
-LocalNoInlineNoReturn(void throwTypeError(const Pos & pos, const char * s, const Value & v))
-{
-    throw TypeError({
-        .msg = hintfmt(s, showType(v)),
-        .errPos = pos
-    });
-}
-
 
 /* Note: Various places expect the allocated memory to be zeroed. */
 [[gnu::always_inline]]
@@ -99,7 +80,7 @@ Env & EvalState::allocEnv(size_t size)
 
 
 [[gnu::always_inline]]
-void EvalState::forceValue(Value & v, const Pos & pos)
+void EvalState::forceValue(Value & v, const PosIdx pos)
 {
     forceValue(v, [&]() { return pos; });
 }
@@ -128,7 +109,7 @@ void EvalState::forceValue(Value & v, Callable getPos)
 
 
 [[gnu::always_inline]]
-inline void EvalState::forceAttrs(Value & v, const Pos & pos)
+inline void EvalState::forceAttrs(Value & v, const PosIdx pos)
 {
     forceAttrs(v, [&]() { return pos; });
 }
@@ -145,7 +126,7 @@ inline void EvalState::forceAttrs(Value & v, Callable getPos)
 
 
 [[gnu::always_inline]]
-inline void EvalState::forceList(Value & v, const Pos & pos)
+inline void EvalState::forceList(Value & v, const PosIdx pos)
 {
     forceValue(v, pos);
     if (!v.isList())

src/libexpr/eval.cc

@@ -96,7 +96,8 @@ RootValue allocRootValue(Value * v)
 }
 
 
-void Value::print(std::ostream & str, std::set<const void *> * seen) const
+void Value::print(const SymbolTable & symbols, std::ostream & str,
+    std::set<const void *> * seen) const
 {
     checkInterrupt();
 
@@ -129,9 +130,9 @@ void Value::print(std::ostream & str, std::set<const void *> * seen) const
             str << "«repeated»";
         else {
             str << "{ ";
-            for (auto & i : attrs->lexicographicOrder()) {
-                str << i->name << " = ";
-                i->value->print(str, seen);
+            for (auto & i : attrs->lexicographicOrder(symbols)) {
+                str << symbols[i->name] << " = ";
+                i->value->print(symbols, str, seen);
                 str << "; ";
             }
             str << "}";
@@ -146,7 +147,7 @@ void Value::print(std::ostream & str, std::set<const void *> * seen) const
         else {
             str << "[ ";
             for (auto v2 : listItems()) {
-                v2->print(str, seen);
+                v2->print(symbols, str, seen);
                 str << " ";
             }
             str << "]";
@@ -177,17 +178,18 @@ void Value::print(std::ostream & str, std::set<const void *> * seen) const
 }
 
 
-void Value::print(std::ostream & str, bool showRepeated) const
+void Value::print(const SymbolTable & symbols, std::ostream & str, bool showRepeated) const
 {
     std::set<const void *> seen;
-    print(str, showRepeated ? nullptr : &seen);
+    print(symbols, str, showRepeated ? nullptr : &seen);
 }
 
 
-std::ostream & operator << (std::ostream & str, const Value & v)
+std::string printValue(const EvalState & state, const Value & v)
 {
-    v.print(str, false);
-    return str;
+    std::ostringstream out;
+    v.print(state.symbols, out);
+    return out.str();
 }
 
 
@@ -236,10 +238,10 @@ std::string showType(const Value & v)
     }
 }
 
-Pos Value::determinePos(const Pos & pos) const
+PosIdx Value::determinePos(const PosIdx pos) const
 {
     switch (internalType) {
-        case tAttrs: return *attrs->pos;
+        case tAttrs: return attrs->pos;
         case tLambda: return lambda.fun->pos;
         case tApp: return app.left->determinePos(pos);
         default: return pos;
@@ -308,7 +310,7 @@ static BoehmGCStackAllocator boehmGCStackAllocator;
 
 static Symbol getName(const AttrName & name, EvalState & state, Env & env)
 {
-    if (name.symbol.set()) {
+    if (name.symbol) {
         return name.symbol;
     } else {
         Value nameValue;
@@ -436,6 +438,7 @@ EvalState::EvalState(
     , sBuilder(symbols.create("builder"))
     , sArgs(symbols.create("args"))
     , sContentAddressed(symbols.create("__contentAddressed"))
+    , sImpure(symbols.create("__impure"))
     , sOutputHash(symbols.create("outputHash"))
     , sOutputHashAlgo(symbols.create("outputHashAlgo"))
     , sOutputHashMode(symbols.create("outputHashMode"))
@@ -646,20 +649,20 @@ Value * EvalState::addPrimOp(const std::string & name,
     size_t arity, PrimOpFun primOp)
 {
     auto name2 = name.substr(0, 2) == "__" ? name.substr(2) : name;
-    Symbol sym = symbols.create(name2);
+    auto sym = symbols.create(name2);
 
     /* Hack to make constants lazy: turn them into a application of
        the primop to a dummy value. */
     if (arity == 0) {
         auto vPrimOp = allocValue();
-        vPrimOp->mkPrimOp(new PrimOp { .fun = primOp, .arity = 1, .name = sym });
+        vPrimOp->mkPrimOp(new PrimOp { .fun = primOp, .arity = 1, .name = name2 });
         Value v;
         v.mkApp(vPrimOp, vPrimOp);
         return addConstant(name, v);
     }
 
     Value * v = allocValue();
-    v->mkPrimOp(new PrimOp { .fun = primOp, .arity = arity, .name = sym });
+    v->mkPrimOp(new PrimOp { .fun = primOp, .arity = arity, .name = name2 });
     staticBaseEnv.vars.emplace_back(symbols.create(name), baseEnvDispl);
     baseEnv.values[baseEnvDispl++] = v;
     baseEnv.values[0]->attrs->push_back(Attr(sym, v));
@@ -674,21 +677,21 @@ Value * EvalState::addPrimOp(PrimOp && primOp)
     if (primOp.arity == 0) {
         primOp.arity = 1;
         auto vPrimOp = allocValue();
-        vPrimOp->mkPrimOp(new PrimOp(std::move(primOp)));
+        vPrimOp->mkPrimOp(new PrimOp(primOp));
         Value v;
         v.mkApp(vPrimOp, vPrimOp);
         return addConstant(primOp.name, v);
     }
 
-    Symbol envName = primOp.name;
+    auto envName = symbols.create(primOp.name);
     if (hasPrefix(primOp.name, "__"))
-        primOp.name = symbols.create(std::string(primOp.name, 2));
+        primOp.name = primOp.name.substr(2);
 
     Value * v = allocValue();
-    v->mkPrimOp(new PrimOp(std::move(primOp)));
+    v->mkPrimOp(new PrimOp(primOp));
     staticBaseEnv.vars.emplace_back(envName, baseEnvDispl);
     baseEnv.values[baseEnvDispl++] = v;
-    baseEnv.values[0]->attrs->push_back(Attr(primOp.name, v));
+    baseEnv.values[0]->attrs->push_back(Attr(symbols.create(primOp.name), v));
     return v;
 }
 
@@ -705,7 +708,7 @@ std::optional<EvalState::Doc> EvalState::getDoc(Value & v)
         auto v2 = &v;
         if (v2->primOp->doc)
             return Doc {
-                .pos = noPos,
+                .pos = {},
                 .name = v2->primOp->name,
                 .arity = v2->primOp->arity,
                 .args = v2->primOp->args,
@@ -721,108 +724,128 @@ std::optional<EvalState::Doc> EvalState::getDoc(Value & v)
    evaluator.  So here are some helper functions for throwing
    exceptions. */
 
-LocalNoInlineNoReturn(void throwEvalError(const Pos & pos, const Suggestions & suggestions, const char * s, const std::string & s2))
+void EvalState::throwEvalError(const PosIdx pos, const char * s) const
+{
+    throw EvalError({
+        .msg = hintfmt(s),
+        .errPos = positions[pos]
+    });
+}
+
+void EvalState::throwTypeError(const PosIdx pos, const char * s, const Value & v) const
+{
+    throw TypeError({
+        .msg = hintfmt(s, showType(v)),
+        .errPos = positions[pos]
+    });
+}
+
+void EvalState::throwEvalError(const PosIdx pos, const Suggestions & suggestions, const char * s,
+    const std::string & s2) const
 {
     throw EvalError(ErrorInfo {
         .msg = hintfmt(s, s2),
-        .errPos = pos,
+        .errPos = positions[pos],
         .suggestions = suggestions,
     });
 }
 
-LocalNoInlineNoReturn(void throwEvalError(const Pos & pos, const char * s, const std::string & s2))
+void EvalState::throwEvalError(const PosIdx pos, const char * s, const std::string & s2) const
 {
     throw EvalError(ErrorInfo {
         .msg = hintfmt(s, s2),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInlineNoReturn(void throwEvalError(const char * s, const std::string & s2, const std::string & s3))
+void EvalState::throwEvalError(const char * s, const std::string & s2, const std::string & s3) const
 {
     throw EvalError(s, s2, s3);
 }
 
-LocalNoInlineNoReturn(void throwEvalError(const Pos & pos, const char * s, const std::string & s2, const std::string & s3))
+void EvalState::throwEvalError(const PosIdx pos, const char * s, const std::string & s2,
+    const std::string & s3) const
 {
     throw EvalError({
         .msg = hintfmt(s, s2, s3),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInlineNoReturn(void throwEvalError(const Pos & p1, const char * s, const Symbol & sym, const Pos & p2))
+void EvalState::throwEvalError(const PosIdx p1, const char * s, const Symbol sym, const PosIdx p2) const
 {
     // p1 is where the error occurred; p2 is a position mentioned in the message.
     throw EvalError({
-        .msg = hintfmt(s, sym, p2),
-        .errPos = p1
+        .msg = hintfmt(s, symbols[sym], positions[p2]),
+        .errPos = positions[p1]
     });
 }
 
-LocalNoInlineNoReturn(void throwTypeError(const Pos & pos, const char * s))
+void EvalState::throwTypeError(const PosIdx pos, const char * s) const
 {
     throw TypeError({
         .msg = hintfmt(s),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInlineNoReturn(void throwTypeError(const Pos & pos, const char * s, const ExprLambda & fun, const Symbol & s2))
+void EvalState::throwTypeError(const PosIdx pos, const char * s, const ExprLambda & fun,
+    const Symbol s2) const
 {
     throw TypeError({
-        .msg = hintfmt(s, fun.showNamePos(), s2),
-        .errPos = pos
+        .msg = hintfmt(s, fun.showNamePos(*this), symbols[s2]),
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInlineNoReturn(void throwTypeError(const Pos & pos, const Suggestions & suggestions, const char * s, const ExprLambda & fun, const Symbol & s2))
+void EvalState::throwTypeError(const PosIdx pos, const Suggestions & suggestions, const char * s,
+    const ExprLambda & fun, const Symbol s2) const
 {
     throw TypeError(ErrorInfo {
-        .msg = hintfmt(s, fun.showNamePos(), s2),
-        .errPos = pos,
+        .msg = hintfmt(s, fun.showNamePos(*this), symbols[s2]),
+        .errPos = positions[pos],
         .suggestions = suggestions,
     });
 }
 
 
-LocalNoInlineNoReturn(void throwTypeError(const char * s, const Value & v))
+void EvalState::throwTypeError(const char * s, const Value & v) const
 {
     throw TypeError(s, showType(v));
 }
 
-LocalNoInlineNoReturn(void throwAssertionError(const Pos & pos, const char * s, const std::string & s1))
+void EvalState::throwAssertionError(const PosIdx pos, const char * s, const std::string & s1) const
 {
     throw AssertionError({
         .msg = hintfmt(s, s1),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInlineNoReturn(void throwUndefinedVarError(const Pos & pos, const char * s, const std::string & s1))
+void EvalState::throwUndefinedVarError(const PosIdx pos, const char * s, const std::string & s1) const
 {
     throw UndefinedVarError({
         .msg = hintfmt(s, s1),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInlineNoReturn(void throwMissingArgumentError(const Pos & pos, const char * s, const std::string & s1))
+void EvalState::throwMissingArgumentError(const PosIdx pos, const char * s, const std::string & s1) const
 {
     throw MissingArgumentError({
         .msg = hintfmt(s, s1),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
-LocalNoInline(void addErrorTrace(Error & e, const char * s, const std::string & s2))
+void EvalState::addErrorTrace(Error & e, const char * s, const std::string & s2) const
 {
     e.addTrace(std::nullopt, s, s2);
 }
 
-LocalNoInline(void addErrorTrace(Error & e, const Pos & pos, const char * s, const std::string & s2))
+void EvalState::addErrorTrace(Error & e, const PosIdx pos, const char * s, const std::string & s2) const
 {
-    e.addTrace(pos, s, s2);
+    e.addTrace(positions[pos], s, s2);
 }
 
 
@@ -882,11 +905,11 @@ inline Value * EvalState::lookupVar(Env * env, const ExprVar & var, bool noEval)
         }
         Bindings::iterator j = env->values[0]->attrs->find(var.name);
         if (j != env->values[0]->attrs->end()) {
-            if (countCalls) attrSelects[*j->pos]++;
+            if (countCalls) attrSelects[j->pos]++;
             return j->value;
         }
         if (!env->prevWith)
-            throwUndefinedVarError(var.pos, "undefined variable '%1%'", var.name);
+            throwUndefinedVarError(var.pos, "undefined variable '%1%'", symbols[var.name]);
         for (size_t l = env->prevWith; l; --l, env = env->up) ;
     }
 }
@@ -916,13 +939,14 @@ void EvalState::mkThunk_(Value & v, Expr * expr)
 }
 
 
-void EvalState::mkPos(Value & v, ptr<Pos> pos)
+void EvalState::mkPos(Value & v, PosIdx p)
 {
-    if (pos->file.set()) {
+    auto pos = positions[p];
+    if (!pos.file.empty()) {
         auto attrs = buildBindings(3);
-        attrs.alloc(sFile).mkString(pos->file);
-        attrs.alloc(sLine).mkInt(pos->line);
-        attrs.alloc(sColumn).mkInt(pos->column);
+        attrs.alloc(sFile).mkString(pos.file);
+        attrs.alloc(sLine).mkInt(pos.line);
+        attrs.alloc(sColumn).mkInt(pos.column);
         v.mkAttrs(attrs);
     } else
         v.mkNull();
@@ -1059,7 +1083,7 @@ inline bool EvalState::evalBool(Env & env, Expr * e)
 }
 
 
-inline bool EvalState::evalBool(Env & env, Expr * e, const Pos & pos)
+inline bool EvalState::evalBool(Env & env, Expr * e, const PosIdx pos)
 {
     Value v;
     e->eval(*this, env, v);
@@ -1133,7 +1157,7 @@ void ExprAttrs::eval(EvalState & state, Env & env, Value & v)
             } else
                 vAttr = i.second.e->maybeThunk(state, i.second.inherited ? env : env2);
             env2.values[displ++] = vAttr;
-            v.attrs->push_back(Attr(i.first, vAttr, ptr(&i.second.pos)));
+            v.attrs->push_back(Attr(i.first, vAttr, i.second.pos));
         }
 
         /* If the rec contains an attribute called `__overrides', then
@@ -1165,7 +1189,7 @@ void ExprAttrs::eval(EvalState & state, Env & env, Value & v)
 
     else
         for (auto & i : attrs)
-            v.attrs->push_back(Attr(i.first, i.second.e->maybeThunk(state, env), ptr(&i.second.pos)));
+            v.attrs->push_back(Attr(i.first, i.second.e->maybeThunk(state, env), i.second.pos));
 
     /* Dynamic attrs apply *after* rec and __overrides. */
     for (auto & i : dynamicAttrs) {
@@ -1175,18 +1199,18 @@ void ExprAttrs::eval(EvalState & state, Env & env, Value & v)
         if (nameVal.type() == nNull)
             continue;
         state.forceStringNoCtx(nameVal);
-        Symbol nameSym = state.symbols.create(nameVal.string.s);
+        auto nameSym = state.symbols.create(nameVal.string.s);
         Bindings::iterator j = v.attrs->find(nameSym);
         if (j != v.attrs->end())
-            throwEvalError(i.pos, "dynamic attribute '%1%' already defined at %2%", nameSym, *j->pos);
+            state.throwEvalError(i.pos, "dynamic attribute '%1%' already defined at %2%", nameSym, j->pos);
 
         i.valueExpr->setName(nameSym);
         /* Keep sorted order so find can catch duplicates */
-        v.attrs->push_back(Attr(nameSym, i.valueExpr->maybeThunk(state, *dynamicEnv), ptr(&i.pos)));
+        v.attrs->push_back(Attr(nameSym, i.valueExpr->maybeThunk(state, *dynamicEnv), i.pos));
         v.attrs->sort(); // FIXME: inefficient
     }
 
-    v.attrs->pos = ptr(&pos);
+    v.attrs->pos = pos;
 }
 
 
@@ -1231,10 +1255,12 @@ static std::string showAttrPath(EvalState & state, Env & env, const AttrPath & a
     for (auto & i : attrPath) {
         if (!first) out << '.'; else first = false;
         try {
-            out << getName(i, state, env);
+            out << state.symbols[getName(i, state, env)];
         } catch (Error & e) {
-            assert(!i.symbol.set());
-            out << "\"${" << *i.expr << "}\"";
+            assert(!i.symbol);
+            out << "\"${";
+            i.expr->show(state.symbols, out);
+            out << "}\"";
         }
     }
     return out.str();
@@ -1244,7 +1270,7 @@ static std::string showAttrPath(EvalState & state, Env & env, const AttrPath & a
 void ExprSelect::eval(EvalState & state, Env & env, Value & v)
 {
     Value vTmp;
-    ptr<Pos> pos2(&noPos);
+    PosIdx pos2;
     Value * vAttrs = &vTmp;
 
     e->eval(state, env, vTmp);
@@ -1254,7 +1280,7 @@ void ExprSelect::eval(EvalState & state, Env & env, Value & v)
         for (auto & i : attrPath) {
             state.nrLookups++;
             Bindings::iterator j;
-            Symbol name = getName(i, state, env);
+            auto name = getName(i, state, env);
             if (def) {
                 state.forceValue(*vAttrs, pos);
                 if (vAttrs->type() != nAttrs ||
@@ -1268,23 +1294,24 @@ void ExprSelect::eval(EvalState & state, Env & env, Value & v)
                 if ((j = vAttrs->attrs->find(name)) == vAttrs->attrs->end()) {
                     std::set<std::string> allAttrNames;
                     for (auto & attr : *vAttrs->attrs)
-                        allAttrNames.insert(attr.name);
-                    throwEvalError(
+                        allAttrNames.insert(state.symbols[attr.name]);
+                    state.throwEvalError(
                         pos,
-                        Suggestions::bestMatches(allAttrNames, name),
-                        "attribute '%1%' missing", name);
+                        Suggestions::bestMatches(allAttrNames, state.symbols[name]),
+                        "attribute '%1%' missing", state.symbols[name]);
                 }
             }
             vAttrs = j->value;
             pos2 = j->pos;
-            if (state.countCalls) state.attrSelects[*pos2]++;
+            if (state.countCalls) state.attrSelects[pos2]++;
         }
 
-        state.forceValue(*vAttrs, (*pos2 != noPos ? *pos2 : this->pos ) );
+        state.forceValue(*vAttrs, (pos2 ? pos2 : this->pos ) );
 
     } catch (Error & e) {
-        if (*pos2 != noPos && pos2->file != state.sDerivationNix)
-            addErrorTrace(e, *pos2, "while evaluating the attribute '%1%'",
+        auto pos2r = state.positions[pos2];
+        if (pos2 && pos2r.file != state.derivationNixPath)
+            state.addErrorTrace(e, pos2, "while evaluating the attribute '%1%'",
                 showAttrPath(state, env, attrPath));
         throw;
     }
@@ -1303,7 +1330,7 @@ void ExprOpHasAttr::eval(EvalState & state, Env & env, Value & v)
     for (auto & i : attrPath) {
         state.forceValue(*vAttrs, noPos);
         Bindings::iterator j;
-        Symbol name = getName(i, state, env);
+        auto name = getName(i, state, env);
         if (vAttrs->type() != nAttrs ||
             (j = vAttrs->attrs->find(name)) == vAttrs->attrs->end())
         {
@@ -1324,9 +1351,11 @@ void ExprLambda::eval(EvalState & state, Env & env, Value & v)
 }
 
 
-void EvalState::callFunction(Value & fun, size_t nrArgs, Value * * args, Value & vRes, const Pos & pos)
+void EvalState::callFunction(Value & fun, size_t nrArgs, Value * * args, Value & vRes, const PosIdx pos)
 {
-    auto trace = evalSettings.traceFunctionCalls ? std::make_unique<FunctionCallTrace>(pos) : nullptr;
+    auto trace = evalSettings.traceFunctionCalls
+        ? std::make_unique<FunctionCallTrace>(positions[pos])
+        : nullptr;
 
     forceValue(fun, pos);
 
@@ -1351,7 +1380,7 @@ void EvalState::callFunction(Value & fun, size_t nrArgs, Value * * args, Value &
             ExprLambda & lambda(*vCur.lambda.fun);
 
             auto size =
-                (lambda.arg.empty() ? 0 : 1) +
+                (!lambda.arg ? 0 : 1) +
                 (lambda.hasFormals() ? lambda.formals->formals.size() : 0);
             Env & env2(allocEnv(size));
             env2.up = vCur.lambda.env;
@@ -1364,7 +1393,7 @@ void EvalState::callFunction(Value & fun, size_t nrArgs, Value * * args, Value &
             else {
                 forceAttrs(*args[0], pos);
 
-                if (!lambda.arg.empty())
+                if (lambda.arg)
                     env2.values[displ++] = args[0];
 
                 /* For each formal argument, get the actual argument.  If
@@ -1392,10 +1421,10 @@ void EvalState::callFunction(Value & fun, size_t nrArgs, Value * * args, Value &
                         if (!lambda.formals->has(i.name)) {
                             std::set<std::string> formalNames;
                             for (auto & formal : lambda.formals->formals)
-                                formalNames.insert(formal.name);
+                                formalNames.insert(symbols[formal.name]);
                             throwTypeError(
                                 pos,
-                                Suggestions::bestMatches(formalNames, i.name),
+                                Suggestions::bestMatches(formalNames, symbols[i.name]),
                                 "%1% called with unexpected argument '%2%'",
                                 lambda,
                                 i.name);
@@ -1413,8 +1442,8 @@ void EvalState::callFunction(Value & fun, size_t nrArgs, Value * * args, Value &
             } catch (Error & e) {
                 if (loggerSettings.showTrace.get()) {
                     addErrorTrace(e, lambda.pos, "while evaluating %s",
-                        (lambda.name.set()
-                            ? "'" + (const std::string &) lambda.name + "'"
+                        (lambda.name
+                            ? concatStrings("'", symbols[lambda.name], "'")
                             : "anonymous lambda"));
                     addErrorTrace(e, pos, "from call site%s", "");
                 }
@@ -1563,7 +1592,7 @@ void EvalState::autoCallFunction(Bindings & args, Value & fun, Value & res)
 Nix attempted to evaluate a function as a top level expression; in
 this case it must have its arguments supplied either by default
 values, or passed explicitly with '--arg' or '--argstr'. See
-https://nixos.org/manual/nix/stable/#ss-functions.)", i.name);
+https://nixos.org/manual/nix/stable/#ss-functions.)", symbols[i.name]);
 
             }
         }
@@ -1595,8 +1624,8 @@ void ExprAssert::eval(EvalState & state, Env & env, Value & v)
 {
     if (!state.evalBool(env, cond, pos)) {
         std::ostringstream out;
-        cond->show(out);
-        throwAssertionError(pos, "assertion '%1%' failed", out.str());
+        cond->show(state.symbols, out);
+        state.throwAssertionError(pos, "assertion '%1%' failed", out.str());
     }
     body->eval(state, env, v);
 }
@@ -1689,7 +1718,7 @@ void ExprOpConcatLists::eval(EvalState & state, Env & env, Value & v)
 }
 
 
-void EvalState::concatLists(Value & v, size_t nrLists, Value * * lists, const Pos & pos)
+void EvalState::concatLists(Value & v, size_t nrLists, Value * * lists, const PosIdx pos)
 {
     nrListConcats++;
 
@@ -1773,14 +1802,14 @@ void ExprConcatStrings::eval(EvalState & state, Env & env, Value & v)
                 nf = n;
                 nf += vTmp.fpoint;
             } else
-                throwEvalError(i_pos, "cannot add %1% to an integer", showType(vTmp));
+                state.throwEvalError(i_pos, "cannot add %1% to an integer", showType(vTmp));
         } else if (firstType == nFloat) {
             if (vTmp.type() == nInt) {
                 nf += vTmp.integer;
             } else if (vTmp.type() == nFloat) {
                 nf += vTmp.fpoint;
             } else
-                throwEvalError(i_pos, "cannot add %1% to a float", showType(vTmp));
+                state.throwEvalError(i_pos, "cannot add %1% to a float", showType(vTmp));
         } else {
             if (s.empty()) s.reserve(es->size());
             /* Skip canonization of first path, which would only be
@@ -1800,7 +1829,7 @@ void ExprConcatStrings::eval(EvalState & state, Env & env, Value & v)
         v.mkFloat(nf);
     else if (firstType == nPath) {
         if (!context.empty())
-            throwEvalError(pos, "a string that refers to a store path cannot be appended to a path");
+            state.throwEvalError(pos, "a string that refers to a store path cannot be appended to a path");
         v.mkPath({.accessor = *values[0]._path.accessor, .path = canonPath(str())});
     } else
         v.mkStringMove(c_str(), context);
@@ -1809,7 +1838,7 @@ void ExprConcatStrings::eval(EvalState & state, Env & env, Value & v)
 
 void ExprPos::eval(EvalState & state, Env & env, Value & v)
 {
-    state.mkPos(v, ptr(&pos));
+    state.mkPos(v, pos);
 }
 
 
@@ -1829,7 +1858,7 @@ void EvalState::forceValueDeep(Value & v)
                 try {
                     recurse(*i.value);
                 } catch (Error & e) {
-                    addErrorTrace(e, *i.pos, "while evaluating the attribute '%1%'", i.name);
+                    addErrorTrace(e, i.pos, "while evaluating the attribute '%1%'", symbols[i.name]);
                     throw;
                 }
         }
@@ -1844,7 +1873,7 @@ void EvalState::forceValueDeep(Value & v)
 }
 
 
-NixInt EvalState::forceInt(Value & v, const Pos & pos)
+NixInt EvalState::forceInt(Value & v, const PosIdx pos)
 {
     forceValue(v, pos);
     if (v.type() != nInt)
@@ -1853,7 +1882,7 @@ NixInt EvalState::forceInt(Value & v, const Pos & pos)
 }
 
 
-NixFloat EvalState::forceFloat(Value & v, const Pos & pos)
+NixFloat EvalState::forceFloat(Value & v, const PosIdx pos)
 {
     forceValue(v, pos);
     if (v.type() == nInt)
@@ -1864,7 +1893,7 @@ NixFloat EvalState::forceFloat(Value & v, const Pos & pos)
 }
 
 
-bool EvalState::forceBool(Value & v, const Pos & pos)
+bool EvalState::forceBool(Value & v, const PosIdx pos)
 {
     forceValue(v, pos);
     if (v.type() != nBool)
@@ -1879,7 +1908,7 @@ bool EvalState::isFunctor(Value & fun)
 }
 
 
-void EvalState::forceFunction(Value & v, const Pos & pos)
+void EvalState::forceFunction(Value & v, const PosIdx pos)
 {
     forceValue(v, pos);
     if (v.type() != nFunction && !isFunctor(v))
@@ -1887,7 +1916,7 @@ void EvalState::forceFunction(Value & v, const Pos & pos)
 }
 
 
-std::string_view EvalState::forceString(Value & v, const Pos & pos)
+std::string_view EvalState::forceString(Value & v, const PosIdx pos)
 {
     forceValue(v, pos);
     if (v.type() != nString) {
@@ -1940,7 +1969,7 @@ NixStringContext Value::getContext(const Store & store)
 }
 
 
-std::string_view EvalState::forceString(Value & v, PathSet & context, const Pos & pos)
+std::string_view EvalState::forceString(Value & v, PathSet & context, const PosIdx pos)
 {
     auto s = forceString(v, pos);
     copyContext(v, context);
@@ -1948,7 +1977,7 @@ std::string_view EvalState::forceString(Value & v, PathSet & context, const Pos
 }
 
 
-std::string_view EvalState::forceStringNoCtx(Value & v, const Pos & pos)
+std::string_view EvalState::forceStringNoCtx(Value & v, const PosIdx pos)
 {
     auto s = forceString(v, pos);
     if (v.string.context) {
@@ -1968,13 +1997,13 @@ bool EvalState::isDerivation(Value & v)
     if (v.type() != nAttrs) return false;
     Bindings::iterator i = v.attrs->find(sType);
     if (i == v.attrs->end()) return false;
-    forceValue(*i->value, *i->pos);
+    forceValue(*i->value, i->pos);
     if (i->value->type() != nString) return false;
     return strcmp(i->value->string.s, "derivation") == 0;
 }
 
 
-std::optional<std::string> EvalState::tryAttrsToString(const Pos & pos, Value & v,
+std::optional<std::string> EvalState::tryAttrsToString(const PosIdx pos, Value & v,
     PathSet & context, bool coerceMore, bool copyToStore)
 {
     auto i = v.attrs->find(sToString);
@@ -1987,7 +2016,7 @@ std::optional<std::string> EvalState::tryAttrsToString(const Pos & pos, Value &
     return {};
 }
 
-BackedStringView EvalState::coerceToString(const Pos & pos, Value & v, PathSet & context,
+BackedStringView EvalState::coerceToString(const PosIdx pos, Value & v, PathSet & context,
     bool coerceMore, bool copyToStore, bool canonicalizePath)
 {
     forceValue(v, pos);
@@ -2017,7 +2046,7 @@ BackedStringView EvalState::coerceToString(const Pos & pos, Value & v, PathSet &
     }
 
     if (v.type() == nExternal)
-        return v.external->coerceToString(pos, context, coerceMore, copyToStore);
+        return v.external->coerceToString(positions[pos], context, coerceMore, copyToStore);
 
     if (coerceMore) {
 
@@ -2082,7 +2111,7 @@ std::string EvalState::copyPathToStore(PathSet & context, const Path & path)
 }
 
 
-SourcePath EvalState::coerceToPath(const Pos & pos, Value & v, PathSet & context)
+SourcePath EvalState::coerceToPath(const PosIdx pos, Value & v, PathSet & context)
 {
     auto path = coerceToString(pos, v, context, false, false).toOwned();
     if (path == "" || path[0] != '/')
@@ -2092,14 +2121,14 @@ SourcePath EvalState::coerceToPath(const Pos & pos, Value & v, PathSet & context
 }
 
 
-StorePath EvalState::coerceToStorePath(const Pos & pos, Value & v, PathSet & context)
+StorePath EvalState::coerceToStorePath(const PosIdx pos, Value & v, PathSet & context)
 {
     auto path = coerceToString(pos, v, context, false, false).toOwned();
     if (auto storePath = store->maybeParseStorePath(path))
         return *storePath;
     throw EvalError({
         .msg = hintfmt("path '%1%' is not in the Nix store", path),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 
@@ -2268,14 +2297,14 @@ void EvalState::printStats()
                 auto list = topObj.list("functions");
                 for (auto & i : functionCalls) {
                     auto obj = list.object();
-                    if (i.first->name.set())
+                    if (i.first->name)
                         obj.attr("name", (const std::string &) i.first->name);
                     else
                         obj.attr("name", nullptr);
-                    if (i.first->pos) {
-                        obj.attr("file", (const std::string &) i.first->pos.file);
-                        obj.attr("line", i.first->pos.line);
-                        obj.attr("column", i.first->pos.column);
+                    if (auto pos = positions[i.first->pos]) {
+                        obj.attr("file", (const std::string &) pos.file);
+                        obj.attr("line", pos.line);
+                        obj.attr("column", pos.column);
                     }
                     obj.attr("count", i.second);
                 }
@@ -2284,10 +2313,10 @@ void EvalState::printStats()
                 auto list = topObj.list("attributes");
                 for (auto & i : attrSelects) {
                     auto obj = list.object();
-                    if (i.first) {
-                        obj.attr("file", (const std::string &) i.first.file);
-                        obj.attr("line", i.first.line);
-                        obj.attr("column", i.first.column);
+                    if (auto pos = positions[i.first]) {
+                        obj.attr("file", (const std::string &) pos.file);
+                        obj.attr("line", pos.line);
+                        obj.attr("column", pos.column);
                     }
                     obj.attr("count", i.second);
                 }

src/libexpr/eval.hh

@@ -25,14 +25,14 @@ struct SourcePath;
 enum RepairFlag : bool;
 
 
-typedef void (* PrimOpFun) (EvalState & state, const Pos & pos, Value * * args, Value & v);
+typedef void (* PrimOpFun) (EvalState & state, const PosIdx pos, Value * * args, Value & v);
 
 
 struct PrimOp
 {
     PrimOpFun fun;
     size_t arity;
-    Symbol name;
+    std::string name;
     std::vector<std::string> args;
     const char * doc = nullptr;
 };
@@ -55,7 +55,8 @@ void copyContext(const Value & v, PathSet & context);
 typedef std::map<Path, StorePath> SrcToStore;
 
 
-std::ostream & operator << (std::ostream & str, const Value & v);
+std::ostream & printValue(const EvalState & state, std::ostream & str, const Value & v);
+std::string printValue(const EvalState & state, const Value & v);
 
 
 typedef std::pair<std::string, std::string> SearchPathElem;
@@ -75,12 +76,15 @@ class EvalState
 {
 public:
     SymbolTable symbols;
+    PosTable positions;
+
+    static inline std::string derivationNixPath = "//builtin/derivation.nix";
 
     const Symbol sWith, sOutPath, sDrvPath, sType, sMeta, sName, sValue,
         sSystem, sOverrides, sOutputs, sOutputName, sIgnoreNulls,
         sFile, sLine, sColumn, sFunctor, sToString,
         sRight, sWrong, sStructuredAttrs, sBuilder, sArgs,
-        sContentAddressed,
+        sContentAddressed, sImpure,
         sOutputHash, sOutputHashAlgo, sOutputHashMode,
         sRecurseForDerivations,
         sDescription, sSelf, sEpsilon, sStartSet, sOperator, sKey, sPath,
@@ -208,7 +212,7 @@ public:
 
     /* Look up a file in the search path. */
     Path findFile(const std::string_view path);
-    Path findFile(SearchPath & searchPath, const std::string_view path, const Pos & pos = noPos);
+    Path findFile(SearchPath & searchPath, const std::string_view path, const PosIdx pos = noPos);
 
     /* If the specified search path element is a URI, download it. */
     std::pair<bool, std::string> resolveSearchPathElem(const SearchPathElem & elem);
@@ -220,14 +224,14 @@ public:
     /* Evaluation the expression, then verify that it has the expected
        type. */
     inline bool evalBool(Env & env, Expr * e);
-    inline bool evalBool(Env & env, Expr * e, const Pos & pos);
+    inline bool evalBool(Env & env, Expr * e, const PosIdx pos);
     inline void evalAttrs(Env & env, Expr * e, Value & v);
 
     /* If `v' is a thunk, enter it and overwrite `v' with the result
        of the evaluation of the thunk.  If `v' is a delayed function
        application, call the function and overwrite `v' with the
        result.  Otherwise, this is a no-op. */
-    inline void forceValue(Value & v, const Pos & pos);
+    inline void forceValue(Value & v, const PosIdx pos);
 
     template <typename Callable>
     inline void forceValue(Value & v, Callable getPos);
@@ -237,33 +241,70 @@ public:
     void forceValueDeep(Value & v);
 
     /* Force `v', and then verify that it has the expected type. */
-    NixInt forceInt(Value & v, const Pos & pos);
-    NixFloat forceFloat(Value & v, const Pos & pos);
-    bool forceBool(Value & v, const Pos & pos);
+    NixInt forceInt(Value & v, const PosIdx pos);
+    NixFloat forceFloat(Value & v, const PosIdx pos);
+    bool forceBool(Value & v, const PosIdx pos);
 
-    void forceAttrs(Value & v, const Pos & pos);
+    void forceAttrs(Value & v, const PosIdx pos);
 
     template <typename Callable>
     inline void forceAttrs(Value & v, Callable getPos);
 
-    inline void forceList(Value & v, const Pos & pos);
-    void forceFunction(Value & v, const Pos & pos); // either lambda or primop
-    std::string_view forceString(Value & v, const Pos & pos = noPos);
-    std::string_view forceString(Value & v, PathSet & context, const Pos & pos = noPos);
-    std::string_view forceStringNoCtx(Value & v, const Pos & pos = noPos);
+    inline void forceList(Value & v, const PosIdx pos);
+    void forceFunction(Value & v, const PosIdx pos); // either lambda or primop
+    std::string_view forceString(Value & v, const PosIdx pos = noPos);
+    std::string_view forceString(Value & v, PathSet & context, const PosIdx pos = noPos);
+    std::string_view forceStringNoCtx(Value & v, const PosIdx pos = noPos);
 
+    [[gnu::noinline, gnu::noreturn]]
+    void throwEvalError(const PosIdx pos, const char * s) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwTypeError(const PosIdx pos, const char * s, const Value & v) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwEvalError(const PosIdx pos, const Suggestions & suggestions, const char * s,
+        const std::string & s2) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwEvalError(const PosIdx pos, const char * s, const std::string & s2) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwEvalError(const char * s, const std::string & s2, const std::string & s3) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwEvalError(const PosIdx pos, const char * s, const std::string & s2, const std::string & s3) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwEvalError(const PosIdx p1, const char * s, const Symbol sym, const PosIdx p2) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwTypeError(const PosIdx pos, const char * s) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwTypeError(const PosIdx pos, const char * s, const ExprLambda & fun, const Symbol s2) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwTypeError(const PosIdx pos, const Suggestions & suggestions, const char * s,
+        const ExprLambda & fun, const Symbol s2) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwTypeError(const char * s, const Value & v) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwAssertionError(const PosIdx pos, const char * s, const std::string & s1) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwUndefinedVarError(const PosIdx pos, const char * s, const std::string & s1) const;
+    [[gnu::noinline, gnu::noreturn]]
+    void throwMissingArgumentError(const PosIdx pos, const char * s, const std::string & s1) const;
+
+    [[gnu::noinline]]
+    void addErrorTrace(Error & e, const char * s, const std::string & s2) const;
+    [[gnu::noinline]]
+    void addErrorTrace(Error & e, const PosIdx pos, const char * s, const std::string & s2) const;
+
+public:
     /* Return true iff the value `v' denotes a derivation (i.e. a
        set with attribute `type = "derivation"'). */
     bool isDerivation(Value & v);
 
-    std::optional<std::string> tryAttrsToString(const Pos & pos, Value & v,
+    std::optional<std::string> tryAttrsToString(const PosIdx pos, Value & v,
         PathSet & context, bool coerceMore = false, bool copyToStore = true);
 
     /* String coercion.  Converts strings, paths and derivations to a
        string.  If `coerceMore' is set, also converts nulls, integers,
        booleans and lists to a string.  If `copyToStore' is set,
        referenced paths are copied to the Nix store as a side effect. */
-    BackedStringView coerceToString(const Pos & pos, Value & v, PathSet & context,
+    BackedStringView coerceToString(const PosIdx pos, Value & v, PathSet & context,
         bool coerceMore = false, bool copyToStore = true,
         bool canonicalizePath = true);
 
@@ -272,10 +313,10 @@ public:
     /* Path coercion.  Converts strings, paths and derivations to a
        path.  The result is guaranteed to be a canonicalised, absolute
        path.  Nothing is copied to the store. */
-    SourcePath coerceToPath(const Pos & pos, Value & v, PathSet & context);
+    SourcePath coerceToPath(const PosIdx pos, Value & v, PathSet & context);
 
     /* Like coerceToPath, but the result must be a store path. */
-    StorePath coerceToStorePath(const Pos & pos, Value & v, PathSet & context);
+    StorePath coerceToStorePath(const PosIdx pos, Value & v, PathSet & context);
 
 public:
 
@@ -308,7 +349,7 @@ public:
     struct Doc
     {
         Pos pos;
-        std::optional<Symbol> name;
+        std::optional<std::string> name;
         size_t arity;
         std::vector<std::string> args;
         const char * doc;
@@ -336,9 +377,9 @@ public:
     bool isFunctor(Value & fun);
 
     // FIXME: use std::span
-    void callFunction(Value & fun, size_t nrArgs, Value * * args, Value & vRes, const Pos & pos);
+    void callFunction(Value & fun, size_t nrArgs, Value * * args, Value & vRes, const PosIdx pos);
 
-    void callFunction(Value & fun, Value & arg, Value & vRes, const Pos & pos)
+    void callFunction(Value & fun, Value & arg, Value & vRes, const PosIdx pos)
     {
         Value * args[] = {&arg};
         callFunction(fun, 1, args, vRes, pos);
@@ -352,7 +393,7 @@ public:
     inline Value * allocValue();
     inline Env & allocEnv(size_t size);
 
-    Value * allocAttr(Value & vAttrs, const Symbol & name);
+    Value * allocAttr(Value & vAttrs, Symbol name);
     Value * allocAttr(Value & vAttrs, std::string_view name);
 
     Bindings * allocBindings(size_t capacity);
@@ -364,9 +405,9 @@ public:
 
     void mkList(Value & v, size_t length);
     void mkThunk_(Value & v, Expr * expr);
-    void mkPos(Value & v, ptr<Pos> pos);
+    void mkPos(Value & v, PosIdx pos);
 
-    void concatLists(Value & v, size_t nrLists, Value * * lists, const Pos & pos);
+    void concatLists(Value & v, size_t nrLists, Value * * lists, const PosIdx pos);
 
     /* Print statistics. */
     void printStats();
@@ -394,7 +435,7 @@ private:
 
     bool countCalls;
 
-    typedef std::map<Symbol, size_t> PrimOpCalls;
+    typedef std::map<std::string, size_t> PrimOpCalls;
     PrimOpCalls primOpCalls;
 
     typedef std::map<ExprLambda *, size_t> FunctionCalls;
@@ -402,7 +443,7 @@ private:
 
     void incrFunctionCall(ExprLambda * fun);
 
-    typedef std::map<Pos, size_t> AttrSelects;
+    typedef std::map<PosIdx, size_t> AttrSelects;
     AttrSelects attrSelects;
 
     friend struct ExprOpUpdate;
@@ -413,9 +454,9 @@ private:
     friend struct ExprFloat;
     friend struct ExprPath;
     friend struct ExprSelect;
-    friend void prim_getAttr(EvalState & state, const Pos & pos, Value * * args, Value & v);
-    friend void prim_match(EvalState & state, const Pos & pos, Value * * args, Value & v);
-    friend void prim_split(EvalState & state, const Pos & pos, Value * * args, Value & v);
+    friend void prim_getAttr(EvalState & state, const PosIdx pos, Value * * args, Value & v);
+    friend void prim_match(EvalState & state, const PosIdx pos, Value * * args, Value & v);
+    friend void prim_split(EvalState & state, const PosIdx pos, Value * * args, Value & v);
 
     friend struct Value;
 };

src/libexpr/flake/config.cc

@@ -50,13 +50,11 @@ void ConfigFile::apply()
         else
             assert(false);
 
-        if (!whitelist.count(baseName)) {
-            auto trustedList = readTrustedList();
-
+        if (!whitelist.count(baseName) && !nix::fetchSettings.acceptFlakeConfig) {
             bool trusted = false;
-            if (nix::fetchSettings.acceptFlakeConfig){
-                trusted = true;
-            } else if (auto saved = get(get(trustedList, name).value_or(std::map<std::string, bool>()), valueS)) {
+            auto trustedList = readTrustedList();
+            auto tlname = get(trustedList, name);
+            if (auto saved = tlname ? get(*tlname, valueS) : nullptr) {
                 trusted = *saved;
                 warn("Using saved setting for '%s = %s' from ~/.local/share/nix/trusted-settings.json.", name,valueS);
             } else {
@@ -69,7 +67,6 @@ void ConfigFile::apply()
                     writeTrustedList(trustedList);
                 }
             }
-
             if (!trusted) {
                 warn("ignoring untrusted flake configuration setting '%s'", name);
                 continue;

src/libexpr/flake/flake.cc

@@ -72,7 +72,7 @@ static std::tuple<fetchers::Tree, FlakeRef, FlakeRef> fetchOrSubstituteTree(
     return {std::move(tree), resolvedRef, lockedRef};
 }
 
-static void forceTrivialValue(EvalState & state, Value & value, const Pos & pos)
+static void forceTrivialValue(EvalState & state, Value & value, const PosIdx pos)
 {
     if (value.isThunk() && value.isTrivial())
         state.forceValue(value, pos);
@@ -80,18 +80,18 @@ static void forceTrivialValue(EvalState & state, Value & value, const Pos & pos)
 
 
 static void expectType(EvalState & state, ValueType type,
-    Value & value, const Pos & pos)
+    Value & value, const PosIdx pos)
 {
     forceTrivialValue(state, value, pos);
     if (value.type() != type)
         throw Error("expected %s but got %s at %s",
-            showType(type), showType(value.type()), pos);
+            showType(type), showType(value.type()), state.positions[pos]);
 }
 
 static std::map<FlakeId, FlakeInput> parseFlakeInputs(
     EvalState & state,
     Value * value,
-    const Pos & pos,
+    const PosIdx pos,
     const std::optional<Path> & baseDir,
     const InputPath & lockRootPath);
 
@@ -99,7 +99,7 @@ static FlakeInput parseFlakeInput(
     EvalState & state,
     const std::string & inputName,
     Value * value,
-    const Pos & pos,
+    const PosIdx pos,
     const std::optional<Path> & baseDir,
     const InputPath & lockRootPath)
 {
@@ -118,37 +118,39 @@ static FlakeInput parseFlakeInput(
     for (nix::Attr attr : *(value->attrs)) {
         try {
             if (attr.name == sUrl) {
-                expectType(state, nString, *attr.value, *attr.pos);
+                expectType(state, nString, *attr.value, attr.pos);
                 url = attr.value->string.s;
                 attrs.emplace("url", *url);
             } else if (attr.name == sFlake) {
-                expectType(state, nBool, *attr.value, *attr.pos);
+                expectType(state, nBool, *attr.value, attr.pos);
                 input.isFlake = attr.value->boolean;
             } else if (attr.name == sInputs) {
-                input.overrides = parseFlakeInputs(state, attr.value, *attr.pos, baseDir, lockRootPath);
+                input.overrides = parseFlakeInputs(state, attr.value, attr.pos, baseDir, lockRootPath);
             } else if (attr.name == sFollows) {
-                expectType(state, nString, *attr.value, *attr.pos);
+                expectType(state, nString, *attr.value, attr.pos);
                 auto follows(parseInputPath(attr.value->string.s));
                 follows.insert(follows.begin(), lockRootPath.begin(), lockRootPath.end());
                 input.follows = follows;
             } else {
                 switch (attr.value->type()) {
                     case nString:
-                        attrs.emplace(attr.name, attr.value->string.s);
+                        attrs.emplace(state.symbols[attr.name], attr.value->string.s);
                         break;
                     case nBool:
-                        attrs.emplace(attr.name, Explicit<bool> { attr.value->boolean });
+                        attrs.emplace(state.symbols[attr.name], Explicit<bool> { attr.value->boolean });
                         break;
                     case nInt:
-                        attrs.emplace(attr.name, (long unsigned int)attr.value->integer);
+                        attrs.emplace(state.symbols[attr.name], (long unsigned int)attr.value->integer);
                         break;
                     default:
                         throw TypeError("flake input attribute '%s' is %s while a string, Boolean, or integer is expected",
-                            attr.name, showType(*attr.value));
+                            state.symbols[attr.name], showType(*attr.value));
                 }
             }
         } catch (Error & e) {
-            e.addTrace(*attr.pos, hintfmt("in flake attribute '%s'", attr.name));
+            e.addTrace(
+                state.positions[attr.pos],
+                hintfmt("in flake attribute '%s'", state.symbols[attr.name]));
             throw;
         }
     }
@@ -157,13 +159,13 @@ static FlakeInput parseFlakeInput(
         try {
             input.ref = FlakeRef::fromAttrs(attrs);
         } catch (Error & e) {
-            e.addTrace(pos, hintfmt("in flake input"));
+            e.addTrace(state.positions[pos], hintfmt("in flake input"));
             throw;
         }
     else {
         attrs.erase("url");
         if (!attrs.empty())
-            throw Error("unexpected flake input attribute '%s', at %s", attrs.begin()->first, pos);
+            throw Error("unexpected flake input attribute '%s', at %s", attrs.begin()->first, state.positions[pos]);
         if (url)
             input.ref = parseFlakeRef(*url, baseDir, true, input.isFlake);
     }
@@ -177,7 +179,7 @@ static FlakeInput parseFlakeInput(
 static std::map<FlakeId, FlakeInput> parseFlakeInputs(
     EvalState & state,
     Value * value,
-    const Pos & pos,
+    const PosIdx pos,
     const std::optional<Path> & baseDir,
     const InputPath & lockRootPath)
 {
@@ -186,11 +188,11 @@ static std::map<FlakeId, FlakeInput> parseFlakeInputs(
     expectType(state, nAttrs, *value, pos);
 
     for (nix::Attr & inputAttr : *(*value).attrs) {
-        inputs.emplace(inputAttr.name,
+        inputs.emplace(state.symbols[inputAttr.name],
             parseFlakeInput(state,
-                inputAttr.name,
+                state.symbols[inputAttr.name],
                 inputAttr.value,
-                *inputAttr.pos,
+                inputAttr.pos,
                 baseDir,
                 lockRootPath));
     }
@@ -213,37 +215,37 @@ static Flake readFlake(
     Value vInfo;
     state.evalFile(flakePath, vInfo, true);
 
-    expectType(state, nAttrs, vInfo, Pos(foFile, state.symbols.create(flakePath.to_string()), 0, 0));
+    expectType(state, nAttrs, vInfo, state.positions.add({flakePath.to_string(), foFile}, 0, 0));
 
     Flake flake {
         // FIXME
         .originalRef = lockedRef,
         .resolvedRef = lockedRef,
         .lockedRef = lockedRef,
-        .accessor = ptr(&accessor),
+        .accessor = &accessor,
         .flakePath = dirOf(flakePath.path),
     };
 
     if (auto description = vInfo.attrs->get(state.sDescription)) {
-        expectType(state, nString, *description->value, *description->pos);
+        expectType(state, nString, *description->value, description->pos);
         flake.description = description->value->string.s;
     }
 
     auto sInputs = state.symbols.create("inputs");
 
     if (auto inputs = vInfo.attrs->get(sInputs))
-        flake.inputs = parseFlakeInputs(state, inputs->value, *inputs->pos, flakeDir, lockRootPath);
+        flake.inputs = parseFlakeInputs(state, inputs->value, inputs->pos, flakeDir, lockRootPath);
 
     auto sOutputs = state.symbols.create("outputs");
 
     if (auto outputs = vInfo.attrs->get(sOutputs)) {
-        expectType(state, nFunction, *outputs->value, *outputs->pos);
+        expectType(state, nFunction, *outputs->value, outputs->pos);
 
         if (outputs->value->isLambda() && outputs->value->lambda.fun->hasFormals()) {
             for (auto & formal : outputs->value->lambda.fun->formals->formals) {
                 if (formal.name != state.sSelf)
-                    flake.inputs.emplace(formal.name, FlakeInput {
-                        .ref = parseFlakeRef(formal.name)
+                    flake.inputs.emplace(state.symbols[formal.name], FlakeInput {
+                        .ref = parseFlakeRef(state.symbols[formal.name])
                     });
             }
         }
@@ -254,35 +256,41 @@ static Flake readFlake(
     auto sNixConfig = state.symbols.create("nixConfig");
 
     if (auto nixConfig = vInfo.attrs->get(sNixConfig)) {
-        expectType(state, nAttrs, *nixConfig->value, *nixConfig->pos);
+        expectType(state, nAttrs, *nixConfig->value, nixConfig->pos);
 
         for (auto & setting : *nixConfig->value->attrs) {
-            forceTrivialValue(state, *setting.value, *setting.pos);
+            forceTrivialValue(state, *setting.value, setting.pos);
             if (setting.value->type() == nString)
-                flake.config.settings.insert({setting.name, std::string(state.forceStringNoCtx(*setting.value, *setting.pos))});
+                flake.config.settings.emplace(
+                    state.symbols[setting.name],
+                    std::string(state.forceStringNoCtx(*setting.value, setting.pos)));
             else if (setting.value->type() == nPath) {
                 PathSet emptyContext = {};
                 flake.config.settings.emplace(
-                    setting.name,
-                    state.coerceToString(*setting.pos, *setting.value, emptyContext, false, true, true) .toOwned());
+                    state.symbols[setting.name],
+                    state.coerceToString(setting.pos, *setting.value, emptyContext, false, true, true) .toOwned());
             }
             else if (setting.value->type() == nInt)
-                flake.config.settings.insert({setting.name, state.forceInt(*setting.value, *setting.pos)});
+                flake.config.settings.emplace(
+                    state.symbols[setting.name],
+                    state.forceInt(*setting.value, setting.pos));
             else if (setting.value->type() == nBool)
-                flake.config.settings.insert({setting.name, Explicit<bool> { state.forceBool(*setting.value, *setting.pos) }});
+                flake.config.settings.emplace(
+                    state.symbols[setting.name],
+                    Explicit<bool> { state.forceBool(*setting.value, setting.pos) });
             else if (setting.value->type() == nList) {
                 std::vector<std::string> ss;
                 for (auto elem : setting.value->listItems()) {
                     if (elem->type() != nString)
                         throw TypeError("list element in flake configuration setting '%s' is %s while a string is expected",
-                            setting.name, showType(*setting.value));
-                    ss.emplace_back(state.forceStringNoCtx(*elem, *setting.pos));
+                            state.symbols[setting.name], showType(*setting.value));
+                    ss.emplace_back(state.forceStringNoCtx(*elem, setting.pos));
                 }
-                flake.config.settings.insert({setting.name, ss});
+                flake.config.settings.emplace(state.symbols[setting.name], ss);
             }
             else
                 throw TypeError("flake configuration setting '%s' is %s",
-                    setting.name, showType(*setting.value));
+                    state.symbols[setting.name], showType(*setting.value));
         }
     }
 
@@ -292,7 +300,7 @@ static Flake readFlake(
             attr.name != sOutputs &&
             attr.name != sNixConfig)
             throw Error("flake '%s' has an unsupported attribute '%s', at %s",
-                lockedRef, attr.name, *attr.pos);
+                lockedRef, state.symbols[attr.name], state.positions[attr.pos]);
     }
 
     return flake;
@@ -723,12 +731,12 @@ void callFlake(EvalState & state,
     state.callFunction(*vTmp2, *vRootSubdir, vRes, noPos);
 }
 
-static void prim_getFlake(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_getFlake(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     std::string flakeRefS(state.forceStringNoCtx(*args[0], pos));
     auto flakeRef = parseFlakeRef(flakeRefS, {}, true);
     if (evalSettings.pureEval && !flakeRef.input.isLocked())
-        throw Error("cannot call 'getFlake' on unlocked flake reference '%s', at %s (use --impure to override)", flakeRefS, pos);
+        throw Error("cannot call 'getFlake' on unlocked flake reference '%s', at %s (use --impure to override)", flakeRefS, state.positions[pos]);
 
     callFlake(state,
         lockFlake(state, flakeRef,

src/libexpr/flake/flake.hh

@@ -61,7 +61,7 @@ struct Flake
     FlakeRef originalRef; // the original flake specification (by the user)
     FlakeRef resolvedRef; // registry references and caching resolved to the specific underlying flake
     FlakeRef lockedRef; // the specific local store result of invoking the fetcher
-    ptr<InputAccessor> accessor;
+    InputAccessor * accessor; // FIXME: must be non-null
     Path flakePath;
     bool forceDirty = false; // pretend that 'lockedRef' is dirty
     std::optional<std::string> description;

src/libexpr/flake/flakeref.cc

@@ -176,7 +176,7 @@ std::pair<FlakeRef, std::string> parseFlakeRefWithFragment(
                             parsedURL.query.insert_or_assign("shallow", "1");
 
                         return std::make_pair(
-                            FlakeRef(Input::fromURL(parsedURL), get(parsedURL.query, "dir").value_or("")),
+                            FlakeRef(Input::fromURL(parsedURL), getOr(parsedURL.query, "dir", "")),
                             fragment);
                     }
 
@@ -189,7 +189,7 @@ std::pair<FlakeRef, std::string> parseFlakeRefWithFragment(
             if (!hasPrefix(path, "/"))
                 throw BadURL("flake reference '%s' is not an absolute path", url);
             auto query = decodeQuery(match[2]);
-            path = canonPath(path + "/" + get(query, "dir").value_or(""));
+            path = canonPath(path + "/" + getOr(query, "dir", ""));
         }
 
         fetchers::Attrs attrs;
@@ -208,7 +208,7 @@ std::pair<FlakeRef, std::string> parseFlakeRefWithFragment(
         input.parent = baseDir;
 
         return std::make_pair(
-            FlakeRef(std::move(input), get(parsedURL.query, "dir").value_or("")),
+            FlakeRef(std::move(input), getOr(parsedURL.query, "dir", "")),
             fragment);
     }
 }
@@ -238,4 +238,15 @@ std::pair<fetchers::Tree, FlakeRef> FlakeRef::fetchTree(ref<Store> store) const
     return {std::move(tree), FlakeRef(std::move(lockedInput), subdir)};
 }
 
+std::tuple<FlakeRef, std::string, OutputsSpec> parseFlakeRefWithFragmentAndOutputsSpec(
+    const std::string & url,
+    const std::optional<Path> & baseDir,
+    bool allowMissing,
+    bool isFlake)
+{
+    auto [prefix, outputsSpec] = parseOutputsSpec(url);
+    auto [flakeRef, fragment] = parseFlakeRefWithFragment(prefix, baseDir, allowMissing, isFlake);
+    return {std::move(flakeRef), fragment, outputsSpec};
+}
+
 }

src/libexpr/flake/flakeref.hh

@@ -3,6 +3,7 @@
 #include "types.hh"
 #include "hash.hh"
 #include "fetchers.hh"
+#include "path-with-outputs.hh"
 
 #include <variant>
 
@@ -79,4 +80,11 @@ std::pair<FlakeRef, std::string> parseFlakeRefWithFragment(
 std::optional<std::pair<FlakeRef, std::string>> maybeParseFlakeRefWithFragment(
     const std::string & url, const std::optional<Path> & baseDir = {});
 
+std::tuple<FlakeRef, std::string, OutputsSpec> parseFlakeRefWithFragmentAndOutputsSpec(
+    const std::string & url,
+    const std::optional<Path> & baseDir = {},
+    bool allowMissing = false,
+    bool isFlake = true);
+
+
 }

src/libexpr/function-trace.hh

@@ -8,7 +8,7 @@ namespace nix {
 
 struct FunctionCallTrace
 {
-    const Pos & pos;
+    const Pos pos;
     FunctionCallTrace(const Pos & pos);
     ~FunctionCallTrace();
 };

src/libexpr/get-drvs.cc

@@ -34,7 +34,7 @@ DrvInfo::DrvInfo(EvalState & state, ref<Store> store, const std::string & drvPat
 
     outputName =
         selectedOutputs.empty()
-        ? get(drv.env, "outputName").value_or("out")
+        ? getOr(drv.env, "outputName", "out")
         : *selectedOutputs.begin();
 
     auto i = drv.outputs.find(outputName);
@@ -61,7 +61,7 @@ std::string DrvInfo::querySystem() const
 {
     if (system == "" && attrs) {
         auto i = attrs->find(state->sSystem);
-        system = i == attrs->end() ? "unknown" : state->forceStringNoCtx(*i->value, *i->pos);
+        system = i == attrs->end() ? "unknown" : state->forceStringNoCtx(*i->value, i->pos);
     }
     return system;
 }
@@ -75,7 +75,7 @@ std::optional<StorePath> DrvInfo::queryDrvPath() const
         if (i == attrs->end())
             drvPath = {std::nullopt};
         else
-            drvPath = {state->coerceToStorePath(*i->pos, *i->value, context)};
+            drvPath = {state->coerceToStorePath(i->pos, *i->value, context)};
     }
     return drvPath.value_or(std::nullopt);
 }
@@ -95,7 +95,7 @@ StorePath DrvInfo::queryOutPath() const
         Bindings::iterator i = attrs->find(state->sOutPath);
         PathSet context;
         if (i != attrs->end())
-            outPath = state->coerceToStorePath(*i->pos, *i->value, context);
+            outPath = state->coerceToStorePath(i->pos, *i->value, context);
     }
     if (!outPath)
         throw UnimplementedError("CA derivations are not yet supported");
@@ -109,23 +109,23 @@ DrvInfo::Outputs DrvInfo::queryOutputs(bool withPaths, bool onlyOutputsToInstall
         /* Get the ‘outputs’ list. */
         Bindings::iterator i;
         if (attrs && (i = attrs->find(state->sOutputs)) != attrs->end()) {
-            state->forceList(*i->value, *i->pos);
+            state->forceList(*i->value, i->pos);
 
             /* For each output... */
             for (auto elem : i->value->listItems()) {
-                std::string output(state->forceStringNoCtx(*elem, *i->pos));
+                std::string output(state->forceStringNoCtx(*elem, i->pos));
 
                 if (withPaths) {
                     /* Evaluate the corresponding set. */
                     Bindings::iterator out = attrs->find(state->symbols.create(output));
                     if (out == attrs->end()) continue; // FIXME: throw error?
-                    state->forceAttrs(*out->value, *i->pos);
+                    state->forceAttrs(*out->value, i->pos);
 
                     /* And evaluate its ‘outPath’ attribute. */
                     Bindings::iterator outPath = out->value->attrs->find(state->sOutPath);
                     if (outPath == out->value->attrs->end()) continue; // FIXME: throw error?
                     PathSet context;
-                    outputs.emplace(output, state->coerceToStorePath(*outPath->pos, *outPath->value, context));
+                    outputs.emplace(output, state->coerceToStorePath(outPath->pos, *outPath->value, context));
                 } else
                     outputs.emplace(output, std::nullopt);
             }
@@ -168,7 +168,7 @@ Bindings * DrvInfo::getMeta()
     if (!attrs) return 0;
     Bindings::iterator a = attrs->find(state->sMeta);
     if (a == attrs->end()) return 0;
-    state->forceAttrs(*a->value, *a->pos);
+    state->forceAttrs(*a->value, a->pos);
     meta = a->value->attrs;
     return meta;
 }
@@ -179,7 +179,7 @@ StringSet DrvInfo::queryMetaNames()
     StringSet res;
     if (!getMeta()) return res;
     for (auto & i : *meta)
-        res.insert(i.name);
+        res.emplace(state->symbols[i.name]);
     return res;
 }
 
@@ -269,7 +269,7 @@ void DrvInfo::setMeta(const std::string & name, Value * v)
 {
     getMeta();
     auto attrs = state->buildBindings(1 + (meta ? meta->size() : 0));
-    Symbol sym = state->symbols.create(name);
+    auto sym = state->symbols.create(name);
     if (meta)
         for (auto i : *meta)
             if (i.name != sym)
@@ -356,11 +356,11 @@ static void getDerivations(EvalState & state, Value & vIn,
            there are names clashes between derivations, the derivation
            bound to the attribute with the "lower" name should take
            precedence). */
-        for (auto & i : v.attrs->lexicographicOrder()) {
-            debug("evaluating attribute '%1%'", i->name);
-            if (!std::regex_match(std::string(i->name), attrRegex))
+        for (auto & i : v.attrs->lexicographicOrder(state.symbols)) {
+            debug("evaluating attribute '%1%'", state.symbols[i->name]);
+            if (!std::regex_match(std::string(state.symbols[i->name]), attrRegex))
                 continue;
-            std::string pathPrefix2 = addToPath(pathPrefix, i->name);
+            std::string pathPrefix2 = addToPath(pathPrefix, state.symbols[i->name]);
             if (combineChannels)
                 getDerivations(state, *i->value, pathPrefix2, autoArgs, drvs, done, ignoreAssertionFailures);
             else if (getDerivation(state, *i->value, pathPrefix2, drvs, done, ignoreAssertionFailures)) {
@@ -369,7 +369,7 @@ static void getDerivations(EvalState & state, Value & vIn,
                    `recurseForDerivations = true' attribute. */
                 if (i->value->type() == nAttrs) {
                     Bindings::iterator j = i->value->attrs->find(state.sRecurseForDerivations);
-                    if (j != i->value->attrs->end() && state.forceBool(*j->value, *j->pos))
+                    if (j != i->value->attrs->end() && state.forceBool(*j->value, j->pos))
                         getDerivations(state, *i->value, pathPrefix2, autoArgs, drvs, done, ignoreAssertionFailures);
                 }
             }

src/libexpr/lexer.l

@@ -28,9 +28,9 @@ using namespace nix;
 
 namespace nix {
 
-static inline Pos makeCurPos(const YYLTYPE & loc, ParseData * data)
+static inline PosIdx makeCurPos(const YYLTYPE & loc, ParseData * data)
 {
-    return Pos(data->origin, data->file, loc.first_line, loc.first_column);
+    return data->state.positions.add(data->origin, loc.first_line, loc.first_column);
 }
 
 #define CUR_POS makeCurPos(*yylloc, data)
@@ -155,7 +155,7 @@ or          { return OR_KW; }
               } catch (const boost::bad_lexical_cast &) {
                   throw ParseError({
                       .msg = hintfmt("invalid integer '%1%'", yytext),
-                      .errPos = CUR_POS,
+                      .errPos = data->state.positions[CUR_POS],
                   });
               }
               return INT;
@@ -165,7 +165,7 @@ or          { return OR_KW; }
               if (errno != 0)
                   throw ParseError({
                       .msg = hintfmt("invalid float '%1%'", yytext),
-                      .errPos = CUR_POS,
+                      .errPos = data->state.positions[CUR_POS],
                   });
               return FLOAT;
             }
@@ -294,7 +294,7 @@ or          { return OR_KW; }
 <INPATH_SLASH><<EOF>> {
   throw ParseError({
       .msg = hintfmt("path has a trailing slash"),
-      .errPos = CUR_POS,
+      .errPos = data->state.positions[CUR_POS],
   });
 }
 

src/libexpr/nixexpr.cc

@@ -1,5 +1,7 @@
 #include "nixexpr.hh"
 #include "derivations.hh"
+#include "eval.hh"
+#include "symbol-table.hh"
 #include "util.hh"
 
 #include <cstdlib>
@@ -10,12 +12,6 @@ namespace nix {
 
 /* Displaying abstract syntax trees. */
 
-std::ostream & operator << (std::ostream & str, const Expr & e)
-{
-    e.show(str);
-    return str;
-}
-
 static void showString(std::ostream & str, std::string_view s)
 {
     str << '"';
@@ -28,8 +24,10 @@ static void showString(std::ostream & str, std::string_view s)
     str << '"';
 }
 
-static void showId(std::ostream & str, std::string_view s)
+std::ostream & operator <<(std::ostream & str, const SymbolStr & symbol)
 {
+    std::string_view s = symbol;
+
     if (s.empty())
         str << "\"\"";
     else if (s == "if") // FIXME: handle other keywords
@@ -38,7 +36,7 @@ static void showId(std::ostream & str, std::string_view s)
         char c = s[0];
         if (!((c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || c == '_')) {
             showString(str, s);
-            return;
+            return str;
         }
         for (auto c : s)
             if (!((c >= 'a' && c <= 'z') ||
@@ -46,89 +44,104 @@ static void showId(std::ostream & str, std::string_view s)
                   (c >= '0' && c <= '9') ||
                   c == '_' || c == '\'' || c == '-')) {
                 showString(str, s);
-                return;
+                return str;
             }
         str << s;
     }
-}
-
-std::ostream & operator << (std::ostream & str, const Symbol & sym)
-{
-    showId(str, *sym.s);
     return str;
 }
 
-void Expr::show(std::ostream & str) const
+void Expr::show(const SymbolTable & symbols, std::ostream & str) const
 {
     abort();
 }
 
-void ExprInt::show(std::ostream & str) const
+void ExprInt::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << n;
 }
 
-void ExprFloat::show(std::ostream & str) const
+void ExprFloat::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << nf;
 }
 
-void ExprString::show(std::ostream & str) const
+void ExprString::show(const SymbolTable & symbols, std::ostream & str) const
 {
     showString(str, s);
 }
 
-void ExprPath::show(std::ostream & str) const
+void ExprPath::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << s;
 }
 
-void ExprVar::show(std::ostream & str) const
+void ExprVar::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << name;
+    str << symbols[name];
 }
 
-void ExprSelect::show(std::ostream & str) const
+void ExprSelect::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << "(" << *e << ")." << showAttrPath(attrPath);
-    if (def) str << " or (" << *def << ")";
+    str << "(";
+    e->show(symbols, str);
+    str << ")." << showAttrPath(symbols, attrPath);
+    if (def) {
+        str << " or (";
+        def->show(symbols, str);
+        str << ")";
+    }
 }
 
-void ExprOpHasAttr::show(std::ostream & str) const
+void ExprOpHasAttr::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << "((" << *e << ") ? " << showAttrPath(attrPath) << ")";
+    str << "((";
+    e->show(symbols, str);
+    str << ") ? " << showAttrPath(symbols, attrPath) << ")";
 }
 
-void ExprAttrs::show(std::ostream & str) const
+void ExprAttrs::show(const SymbolTable & symbols, std::ostream & str) const
 {
     if (recursive) str << "rec ";
     str << "{ ";
     typedef const decltype(attrs)::value_type * Attr;
     std::vector<Attr> sorted;
     for (auto & i : attrs) sorted.push_back(&i);
-        std::sort(sorted.begin(), sorted.end(), [](Attr a, Attr b) {
-            return (const std::string &) a->first < (const std::string &) b->first;
-        });
+    std::sort(sorted.begin(), sorted.end(), [&](Attr a, Attr b) {
+        std::string_view sa = symbols[a->first], sb = symbols[b->first];
+        return sa < sb;
+    });
     for (auto & i : sorted) {
         if (i->second.inherited)
-            str << "inherit " << i->first << " " << "; ";
-        else
-            str << i->first << " = " << *i->second.e << "; ";
+            str << "inherit " << symbols[i->first] << " " << "; ";
+        else {
+            str << symbols[i->first] << " = ";
+            i->second.e->show(symbols, str);
+            str << "; ";
+        }
+    }
+    for (auto & i : dynamicAttrs) {
+        str << "\"${";
+        i.nameExpr->show(symbols, str);
+        str << "}\" = ";
+        i.valueExpr->show(symbols, str);
+        str << "; ";
     }
-    for (auto & i : dynamicAttrs)
-        str << "\"${" << *i.nameExpr << "}\" = " << *i.valueExpr << "; ";
     str << "}";
 }
 
-void ExprList::show(std::ostream & str) const
+void ExprList::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << "[ ";
-    for (auto & i : elems)
-        str << "(" << *i << ") ";
+    for (auto & i : elems) {
+        str << "(";
+        i->show(symbols, str);
+        str << ") ";
+    }
     str << "]";
 }
 
-void ExprLambda::show(std::ostream & str) const
+void ExprLambda::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << "(";
     if (hasFormals()) {
@@ -136,74 +149,100 @@ void ExprLambda::show(std::ostream & str) const
         bool first = true;
         for (auto & i : formals->formals) {
             if (first) first = false; else str << ", ";
-            str << i.name;
-            if (i.def) str << " ? " << *i.def;
+            str << symbols[i.name];
+            if (i.def) {
+                str << " ? ";
+                i.def->show(symbols, str);
+            }
         }
         if (formals->ellipsis) {
             if (!first) str << ", ";
             str << "...";
         }
         str << " }";
-        if (!arg.empty()) str << " @ ";
+        if (arg) str << " @ ";
     }
-    if (!arg.empty()) str << arg;
-    str << ": " << *body << ")";
+    if (arg) str << symbols[arg];
+    str << ": ";
+    body->show(symbols, str);
+    str << ")";
 }
 
-void ExprCall::show(std::ostream & str) const
+void ExprCall::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << '(' << *fun;
+    str << '(';
+    fun->show(symbols, str);
     for (auto e : args) {
         str <<  ' ';
-        str << *e;
+        e->show(symbols, str);
     }
     str << ')';
 }
 
-void ExprLet::show(std::ostream & str) const
+void ExprLet::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << "(let ";
     for (auto & i : attrs->attrs)
         if (i.second.inherited) {
-            str << "inherit " << i.first << "; ";
+            str << "inherit " << symbols[i.first] << "; ";
         }
-        else
-            str << i.first << " = " << *i.second.e << "; ";
-    str << "in " << *body << ")";
+        else {
+            str << symbols[i.first] << " = ";
+            i.second.e->show(symbols, str);
+            str << "; ";
+        }
+    str << "in ";
+    body->show(symbols, str);
+    str << ")";
 }
 
-void ExprWith::show(std::ostream & str) const
+void ExprWith::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << "(with " << *attrs << "; " << *body << ")";
+    str << "(with ";
+    attrs->show(symbols, str);
+    str << "; ";
+    body->show(symbols, str);
+    str << ")";
 }
 
-void ExprIf::show(std::ostream & str) const
+void ExprIf::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << "(if " << *cond << " then " << *then << " else " << *else_ << ")";
+    str << "(if ";
+    cond->show(symbols, str);
+    str << " then ";
+    then->show(symbols, str);
+    str << " else ";
+    else_->show(symbols, str);
+    str << ")";
 }
 
-void ExprAssert::show(std::ostream & str) const
+void ExprAssert::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << "assert " << *cond << "; " << *body;
+    str << "assert ";
+    cond->show(symbols, str);
+    str << "; ";
+    body->show(symbols, str);
 }
 
-void ExprOpNot::show(std::ostream & str) const
+void ExprOpNot::show(const SymbolTable & symbols, std::ostream & str) const
 {
-    str << "(! " << *e << ")";
+    str << "(! ";
+    e->show(symbols, str);
+    str << ")";
 }
 
-void ExprConcatStrings::show(std::ostream & str) const
+void ExprConcatStrings::show(const SymbolTable & symbols, std::ostream & str) const
 {
     bool first = true;
     str << "(";
     for (auto & i : *es) {
         if (first) first = false; else str << " + ";
-        str << *i.second;
+        i.second->show(symbols, str);
     }
     str << ")";
 }
 
-void ExprPos::show(std::ostream & str) const
+void ExprPos::show(const SymbolTable & symbols, std::ostream & str) const
 {
     str << "__curPos";
 }
@@ -234,48 +273,49 @@ std::ostream & operator << (std::ostream & str, const Pos & pos)
 }
 
 
-std::string showAttrPath(const AttrPath & attrPath)
+std::string showAttrPath(const SymbolTable & symbols, const AttrPath & attrPath)
 {
     std::ostringstream out;
     bool first = true;
     for (auto & i : attrPath) {
         if (!first) out << '.'; else first = false;
-        if (i.symbol.set())
-            out << i.symbol;
-        else
-            out << "\"${" << *i.expr << "}\"";
+        if (i.symbol)
+            out << symbols[i.symbol];
+        else {
+            out << "\"${";
+            i.expr->show(symbols, out);
+            out << "}\"";
+        }
     }
     return out.str();
 }
 
 
-Pos noPos;
-
 
 /* Computing levels/displacements for variables. */
 
-void Expr::bindVars(const StaticEnv & env)
+void Expr::bindVars(const EvalState & es, const StaticEnv & env)
 {
     abort();
 }
 
-void ExprInt::bindVars(const StaticEnv & env)
+void ExprInt::bindVars(const EvalState & es, const StaticEnv & env)
 {
 }
 
-void ExprFloat::bindVars(const StaticEnv & env)
+void ExprFloat::bindVars(const EvalState & es, const StaticEnv & env)
 {
 }
 
-void ExprString::bindVars(const StaticEnv & env)
+void ExprString::bindVars(const EvalState & es, const StaticEnv & env)
 {
 }
 
-void ExprPath::bindVars(const StaticEnv & env)
+void ExprPath::bindVars(const EvalState & es, const StaticEnv & env)
 {
 }
 
-void ExprVar::bindVars(const StaticEnv & env)
+void ExprVar::bindVars(const EvalState & es, const StaticEnv & env)
 {
     /* Check whether the variable appears in the environment.  If so,
        set its level and displacement. */
@@ -301,31 +341,31 @@ void ExprVar::bindVars(const StaticEnv & env)
        "undefined variable" error now. */
     if (withLevel == -1)
         throw UndefinedVarError({
-            .msg = hintfmt("undefined variable '%1%'", name),
-            .errPos = pos
+            .msg = hintfmt("undefined variable '%1%'", es.symbols[name]),
+            .errPos = es.positions[pos]
         });
     fromWith = true;
     this->level = withLevel;
 }
 
-void ExprSelect::bindVars(const StaticEnv & env)
+void ExprSelect::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    e->bindVars(env);
-    if (def) def->bindVars(env);
+    e->bindVars(es, env);
+    if (def) def->bindVars(es, env);
     for (auto & i : attrPath)
-        if (!i.symbol.set())
-            i.expr->bindVars(env);
+        if (!i.symbol)
+            i.expr->bindVars(es, env);
 }
 
-void ExprOpHasAttr::bindVars(const StaticEnv & env)
+void ExprOpHasAttr::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    e->bindVars(env);
+    e->bindVars(es, env);
     for (auto & i : attrPath)
-        if (!i.symbol.set())
-            i.expr->bindVars(env);
+        if (!i.symbol)
+            i.expr->bindVars(es, env);
 }
 
-void ExprAttrs::bindVars(const StaticEnv & env)
+void ExprAttrs::bindVars(const EvalState & es, const StaticEnv & env)
 {
     const StaticEnv * dynamicEnv = &env;
     StaticEnv newEnv(false, &env, recursive ? attrs.size() : 0);
@@ -340,35 +380,35 @@ void ExprAttrs::bindVars(const StaticEnv & env)
         // No need to sort newEnv since attrs is in sorted order.
 
         for (auto & i : attrs)
-            i.second.e->bindVars(i.second.inherited ? env : newEnv);
+            i.second.e->bindVars(es, i.second.inherited ? env : newEnv);
     }
 
     else
         for (auto & i : attrs)
-            i.second.e->bindVars(env);
+            i.second.e->bindVars(es, env);
 
     for (auto & i : dynamicAttrs) {
-        i.nameExpr->bindVars(*dynamicEnv);
-        i.valueExpr->bindVars(*dynamicEnv);
+        i.nameExpr->bindVars(es, *dynamicEnv);
+        i.valueExpr->bindVars(es, *dynamicEnv);
     }
 }
 
-void ExprList::bindVars(const StaticEnv & env)
+void ExprList::bindVars(const EvalState & es, const StaticEnv & env)
 {
     for (auto & i : elems)
-        i->bindVars(env);
+        i->bindVars(es, env);
 }
 
-void ExprLambda::bindVars(const StaticEnv & env)
+void ExprLambda::bindVars(const EvalState & es, const StaticEnv & env)
 {
     StaticEnv newEnv(
         false, &env,
         (hasFormals() ? formals->formals.size() : 0) +
-        (arg.empty() ? 0 : 1));
+        (!arg ? 0 : 1));
 
     Displacement displ = 0;
 
-    if (!arg.empty()) newEnv.vars.emplace_back(arg, displ++);
+    if (arg) newEnv.vars.emplace_back(arg, displ++);
 
     if (hasFormals()) {
         for (auto & i : formals->formals)
@@ -377,20 +417,20 @@ void ExprLambda::bindVars(const StaticEnv & env)
         newEnv.sort();
 
         for (auto & i : formals->formals)
-            if (i.def) i.def->bindVars(newEnv);
+            if (i.def) i.def->bindVars(es, newEnv);
     }
 
-    body->bindVars(newEnv);
+    body->bindVars(es, newEnv);
 }
 
-void ExprCall::bindVars(const StaticEnv & env)
+void ExprCall::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    fun->bindVars(env);
+    fun->bindVars(es, env);
     for (auto e : args)
-        e->bindVars(env);
+        e->bindVars(es, env);
 }
 
-void ExprLet::bindVars(const StaticEnv & env)
+void ExprLet::bindVars(const EvalState & es, const StaticEnv & env)
 {
     StaticEnv newEnv(false, &env, attrs->attrs.size());
 
@@ -401,12 +441,12 @@ void ExprLet::bindVars(const StaticEnv & env)
     // No need to sort newEnv since attrs->attrs is in sorted order.
 
     for (auto & i : attrs->attrs)
-        i.second.e->bindVars(i.second.inherited ? env : newEnv);
+        i.second.e->bindVars(es, i.second.inherited ? env : newEnv);
 
-    body->bindVars(newEnv);
+    body->bindVars(es, newEnv);
 }
 
-void ExprWith::bindVars(const StaticEnv & env)
+void ExprWith::bindVars(const EvalState & es, const StaticEnv & env)
 {
     /* Does this `with' have an enclosing `with'?  If so, record its
        level so that `lookupVar' can look up variables in the previous
@@ -420,57 +460,60 @@ void ExprWith::bindVars(const StaticEnv & env)
             break;
         }
 
-    attrs->bindVars(env);
+    attrs->bindVars(es, env);
     StaticEnv newEnv(true, &env);
-    body->bindVars(newEnv);
+    body->bindVars(es, newEnv);
 }
 
-void ExprIf::bindVars(const StaticEnv & env)
+void ExprIf::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    cond->bindVars(env);
-    then->bindVars(env);
-    else_->bindVars(env);
+    cond->bindVars(es, env);
+    then->bindVars(es, env);
+    else_->bindVars(es, env);
 }
 
-void ExprAssert::bindVars(const StaticEnv & env)
+void ExprAssert::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    cond->bindVars(env);
-    body->bindVars(env);
+    cond->bindVars(es, env);
+    body->bindVars(es, env);
 }
 
-void ExprOpNot::bindVars(const StaticEnv & env)
+void ExprOpNot::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    e->bindVars(env);
+    e->bindVars(es, env);
 }
 
-void ExprConcatStrings::bindVars(const StaticEnv & env)
+void ExprConcatStrings::bindVars(const EvalState & es, const StaticEnv & env)
 {
-    for (auto & i : *es)
-        i.second->bindVars(env);
+    for (auto & i : *this->es)
+        i.second->bindVars(es, env);
 }
 
-void ExprPos::bindVars(const StaticEnv & env)
+void ExprPos::bindVars(const EvalState & es, const StaticEnv & env)
 {
 }
 
 
 /* Storing function names. */
 
-void Expr::setName(Symbol & name)
+void Expr::setName(Symbol name)
 {
 }
 
 
-void ExprLambda::setName(Symbol & name)
+void ExprLambda::setName(Symbol name)
 {
     this->name = name;
     body->setName(name);
 }
 
 
-std::string ExprLambda::showNamePos() const
+std::string ExprLambda::showNamePos(const EvalState & state) const
 {
-    return fmt("%1% at %2%", name.set() ? "'" + (std::string) name + "'" : "anonymous function", pos);
+    std::string id(name
+        ? concatStrings("'", state.symbols[name], "'")
+        : "anonymous function");
+    return fmt("%1% at %2%", id, state.positions[pos]);
 }
 
 
@@ -480,8 +523,7 @@ std::string ExprLambda::showNamePos() const
 size_t SymbolTable::totalSize() const
 {
     size_t n = 0;
-    for (auto & i : store)
-        n += i.size();
+    dump([&] (const std::string & s) { n += s.size(); });
     return n;
 }
 

src/libexpr/nixexpr.hh

@@ -1,8 +1,12 @@
 #pragma once
 
+#include <map>
+#include <vector>
+
 #include "value.hh"
 #include "symbol-table.hh"
 #include "error.hh"
+#include "chunked-vector.hh"
 
 
 namespace nix {
@@ -23,32 +27,92 @@ MakeError(RestrictedPathError, Error);
 
 struct Pos
 {
-    Symbol file;
+    std::string file;
+    FileOrigin origin;
     uint32_t line;
-    FileOrigin origin:2;
-    uint32_t column:30;
-    Pos() : line(0), origin(foString), column(0) { };
-    Pos(FileOrigin origin, const Symbol & file, uint32_t line, uint32_t column)
-        : file(file), line(line), origin(origin), column(column) { };
-    operator bool() const
+    uint32_t column;
+
+    explicit operator bool() const { return line > 0; }
+};
+
+class PosIdx {
+    friend class PosTable;
+
+private:
+    uint32_t id;
+
+    explicit PosIdx(uint32_t id): id(id) {}
+
+public:
+    PosIdx() : id(0) {}
+
+    explicit operator bool() const { return id > 0; }
+
+    bool operator<(const PosIdx other) const { return id < other.id; }
+};
+
+class PosTable
+{
+public:
+    class Origin {
+        friend PosTable;
+    private:
+        // must always be invalid by default, add() replaces this with the actual value.
+        // subsequent add() calls use this index as a token to quickly check whether the
+        // current origins.back() can be reused or not.
+        mutable uint32_t idx = std::numeric_limits<uint32_t>::max();
+
+        explicit Origin(uint32_t idx): idx(idx), file{}, origin{} {}
+
+    public:
+        const std::string file;
+        const FileOrigin origin;
+
+        Origin(std::string file, FileOrigin origin): file(std::move(file)), origin(origin) {}
+    };
+
+    struct Offset {
+        uint32_t line, column;
+    };
+
+private:
+    std::vector<Origin> origins;
+    ChunkedVector<Offset, 8192> offsets;
+
+public:
+    PosTable(): offsets(1024)
     {
-        return line != 0;
+        origins.reserve(1024);
     }
 
-    bool operator < (const Pos & p2) const
+    PosIdx add(const Origin & origin, uint32_t line, uint32_t column)
     {
-        if (!line) return p2.line;
-        if (!p2.line) return false;
-        int d = ((const std::string &) file).compare((const std::string &) p2.file);
-        if (d < 0) return true;
-        if (d > 0) return false;
-        if (line < p2.line) return true;
-        if (line > p2.line) return false;
-        return column < p2.column;
+        const auto idx = offsets.add({line, column}).second;
+        if (origins.empty() || origins.back().idx != origin.idx) {
+            origin.idx = idx;
+            origins.push_back(origin);
+        }
+        return PosIdx(idx + 1);
+    }
+
+    Pos operator[](PosIdx p) const
+    {
+        if (p.id == 0 || p.id > offsets.size())
+            return {};
+        const auto idx = p.id - 1;
+        /* we want the last key <= idx, so we'll take prev(first key > idx).
+           this is guaranteed to never rewind origin.begin because the first
+           key is always 0. */
+        const auto pastOrigin = std::upper_bound(
+            origins.begin(), origins.end(), Origin(idx),
+            [] (const auto & a, const auto & b) { return a.idx < b.idx; });
+        const auto origin = *std::prev(pastOrigin);
+        const auto offset = offsets[idx];
+        return {origin.file, origin.origin, offset.line, offset.column};
     }
 };
 
-extern Pos noPos;
+inline PosIdx noPos = {};
 
 std::ostream & operator << (std::ostream & str, const Pos & pos);
 
@@ -64,13 +128,13 @@ struct AttrName
 {
     Symbol symbol;
     Expr * expr;
-    AttrName(const Symbol & s) : symbol(s) {};
+    AttrName(Symbol s) : symbol(s) {};
     AttrName(Expr * e) : expr(e) {};
 };
 
 typedef std::vector<AttrName> AttrPath;
 
-std::string showAttrPath(const AttrPath & attrPath);
+std::string showAttrPath(const SymbolTable & symbols, const AttrPath & attrPath);
 
 
 /* Abstract syntax of Nix expressions. */
@@ -78,19 +142,17 @@ std::string showAttrPath(const AttrPath & attrPath);
 struct Expr
 {
     virtual ~Expr() { };
-    virtual void show(std::ostream & str) const;
-    virtual void bindVars(const StaticEnv & env);
+    virtual void show(const SymbolTable & symbols, std::ostream & str) const;
+    virtual void bindVars(const EvalState & es, const StaticEnv & env);
     virtual void eval(EvalState & state, Env & env, Value & v);
     virtual Value * maybeThunk(EvalState & state, Env & env);
-    virtual void setName(Symbol & name);
+    virtual void setName(Symbol name);
 };
 
-std::ostream & operator << (std::ostream & str, const Expr & e);
-
 #define COMMON_METHODS \
-    void show(std::ostream & str) const; \
+    void show(const SymbolTable & symbols, std::ostream & str) const;    \
     void eval(EvalState & state, Env & env, Value & v); \
-    void bindVars(const StaticEnv & env);
+    void bindVars(const EvalState & es, const StaticEnv & env);
 
 struct ExprInt : Expr
 {
@@ -137,7 +199,7 @@ typedef uint32_t Displacement;
 
 struct ExprVar : Expr
 {
-    Pos pos;
+    PosIdx pos;
     Symbol name;
 
     /* Whether the variable comes from an environment (e.g. a rec, let
@@ -153,19 +215,19 @@ struct ExprVar : Expr
     Level level;
     Displacement displ;
 
-    ExprVar(const Symbol & name) : name(name) { };
-    ExprVar(const Pos & pos, const Symbol & name) : pos(pos), name(name) { };
+    ExprVar(Symbol name) : name(name) { };
+    ExprVar(const PosIdx & pos, Symbol name) : pos(pos), name(name) { };
     COMMON_METHODS
     Value * maybeThunk(EvalState & state, Env & env);
 };
 
 struct ExprSelect : Expr
 {
-    Pos pos;
+    PosIdx pos;
     Expr * e, * def;
     AttrPath attrPath;
-    ExprSelect(const Pos & pos, Expr * e, const AttrPath & attrPath, Expr * def) : pos(pos), e(e), def(def), attrPath(attrPath) { };
-    ExprSelect(const Pos & pos, Expr * e, const Symbol & name) : pos(pos), e(e), def(0) { attrPath.push_back(AttrName(name)); };
+    ExprSelect(const PosIdx & pos, Expr * e, const AttrPath & attrPath, Expr * def) : pos(pos), e(e), def(def), attrPath(attrPath) { };
+    ExprSelect(const PosIdx & pos, Expr * e, Symbol name) : pos(pos), e(e), def(0) { attrPath.push_back(AttrName(name)); };
     COMMON_METHODS
 };
 
@@ -180,13 +242,13 @@ struct ExprOpHasAttr : Expr
 struct ExprAttrs : Expr
 {
     bool recursive;
-    Pos pos;
+    PosIdx pos;
     struct AttrDef {
         bool inherited;
         Expr * e;
-        Pos pos;
+        PosIdx pos;
         Displacement displ; // displacement
-        AttrDef(Expr * e, const Pos & pos, bool inherited=false)
+        AttrDef(Expr * e, const PosIdx & pos, bool inherited=false)
             : inherited(inherited), e(e), pos(pos) { };
         AttrDef() { };
     };
@@ -194,14 +256,14 @@ struct ExprAttrs : Expr
     AttrDefs attrs;
     struct DynamicAttrDef {
         Expr * nameExpr, * valueExpr;
-        Pos pos;
-        DynamicAttrDef(Expr * nameExpr, Expr * valueExpr, const Pos & pos)
+        PosIdx pos;
+        DynamicAttrDef(Expr * nameExpr, Expr * valueExpr, const PosIdx & pos)
             : nameExpr(nameExpr), valueExpr(valueExpr), pos(pos) { };
     };
     typedef std::vector<DynamicAttrDef> DynamicAttrDefs;
     DynamicAttrDefs dynamicAttrs;
-    ExprAttrs(const Pos &pos) : recursive(false), pos(pos) { };
-    ExprAttrs() : recursive(false), pos(noPos) { };
+    ExprAttrs(const PosIdx &pos) : recursive(false), pos(pos) { };
+    ExprAttrs() : recursive(false) { };
     COMMON_METHODS
 };
 
@@ -214,10 +276,9 @@ struct ExprList : Expr
 
 struct Formal
 {
-    Pos pos;
+    PosIdx pos;
     Symbol name;
     Expr * def;
-    Formal(const Pos & pos, const Symbol & name, Expr * def) : pos(pos), name(name), def(def) { };
 };
 
 struct Formals
@@ -226,18 +287,20 @@ struct Formals
     Formals_ formals;
     bool ellipsis;
 
-    bool has(Symbol arg) const {
+    bool has(Symbol arg) const
+    {
         auto it = std::lower_bound(formals.begin(), formals.end(), arg,
             [] (const Formal & f, const Symbol & sym) { return f.name < sym; });
         return it != formals.end() && it->name == arg;
     }
 
-    std::vector<Formal> lexicographicOrder() const
+    std::vector<Formal> lexicographicOrder(const SymbolTable & symbols) const
     {
         std::vector<Formal> result(formals.begin(), formals.end());
         std::sort(result.begin(), result.end(),
-            [] (const Formal & a, const Formal & b) {
-                return std::string_view(a.name) < std::string_view(b.name);
+            [&] (const Formal & a, const Formal & b) {
+                std::string_view sa = symbols[a.name], sb = symbols[b.name];
+                return sa < sb;
             });
         return result;
     }
@@ -245,17 +308,21 @@ struct Formals
 
 struct ExprLambda : Expr
 {
-    Pos pos;
+    PosIdx pos;
     Symbol name;
     Symbol arg;
     Formals * formals;
     Expr * body;
-    ExprLambda(const Pos & pos, const Symbol & arg, Formals * formals, Expr * body)
+    ExprLambda(PosIdx pos, Symbol arg, Formals * formals, Expr * body)
         : pos(pos), arg(arg), formals(formals), body(body)
     {
     };
-    void setName(Symbol & name);
-    std::string showNamePos() const;
+    ExprLambda(PosIdx pos, Formals * formals, Expr * body)
+        : pos(pos), formals(formals), body(body)
+    {
+    }
+    void setName(Symbol name);
+    std::string showNamePos(const EvalState & state) const;
     inline bool hasFormals() const { return formals != nullptr; }
     COMMON_METHODS
 };
@@ -264,8 +331,8 @@ struct ExprCall : Expr
 {
     Expr * fun;
     std::vector<Expr *> args;
-    Pos pos;
-    ExprCall(const Pos & pos, Expr * fun, std::vector<Expr *> && args)
+    PosIdx pos;
+    ExprCall(const PosIdx & pos, Expr * fun, std::vector<Expr *> && args)
         : fun(fun), args(args), pos(pos)
     { }
     COMMON_METHODS
@@ -281,26 +348,26 @@ struct ExprLet : Expr
 
 struct ExprWith : Expr
 {
-    Pos pos;
+    PosIdx pos;
     Expr * attrs, * body;
     size_t prevWith;
-    ExprWith(const Pos & pos, Expr * attrs, Expr * body) : pos(pos), attrs(attrs), body(body) { };
+    ExprWith(const PosIdx & pos, Expr * attrs, Expr * body) : pos(pos), attrs(attrs), body(body) { };
     COMMON_METHODS
 };
 
 struct ExprIf : Expr
 {
-    Pos pos;
+    PosIdx pos;
     Expr * cond, * then, * else_;
-    ExprIf(const Pos & pos, Expr * cond, Expr * then, Expr * else_) : pos(pos), cond(cond), then(then), else_(else_) { };
+    ExprIf(const PosIdx & pos, Expr * cond, Expr * then, Expr * else_) : pos(pos), cond(cond), then(then), else_(else_) { };
     COMMON_METHODS
 };
 
 struct ExprAssert : Expr
 {
-    Pos pos;
+    PosIdx pos;
     Expr * cond, * body;
-    ExprAssert(const Pos & pos, Expr * cond, Expr * body) : pos(pos), cond(cond), body(body) { };
+    ExprAssert(const PosIdx & pos, Expr * cond, Expr * body) : pos(pos), cond(cond), body(body) { };
     COMMON_METHODS
 };
 
@@ -314,17 +381,17 @@ struct ExprOpNot : Expr
 #define MakeBinOp(name, s) \
     struct name : Expr \
     { \
-        Pos pos; \
+        PosIdx pos; \
         Expr * e1, * e2; \
         name(Expr * e1, Expr * e2) : e1(e1), e2(e2) { }; \
-        name(const Pos & pos, Expr * e1, Expr * e2) : pos(pos), e1(e1), e2(e2) { }; \
-        void show(std::ostream & str) const \
+        name(const PosIdx & pos, Expr * e1, Expr * e2) : pos(pos), e1(e1), e2(e2) { }; \
+        void show(const SymbolTable & symbols, std::ostream & str) const \
         { \
-            str << "(" << *e1 << " " s " " << *e2 << ")";   \
+            str << "("; e1->show(symbols, str); str << " " s " "; e2->show(symbols, str); str << ")"; \
         } \
-        void bindVars(const StaticEnv & env) \
+        void bindVars(const EvalState & es, const StaticEnv & env)    \
         { \
-            e1->bindVars(env); e2->bindVars(env); \
+            e1->bindVars(es, env); e2->bindVars(es, env);    \
         } \
         void eval(EvalState & state, Env & env, Value & v); \
     };
@@ -339,18 +406,18 @@ MakeBinOp(ExprOpConcatLists, "++")
 
 struct ExprConcatStrings : Expr
 {
-    Pos pos;
+    PosIdx pos;
     bool forceString;
-    std::vector<std::pair<Pos, Expr *> > * es;
-    ExprConcatStrings(const Pos & pos, bool forceString, std::vector<std::pair<Pos, Expr *> > * es)
+    std::vector<std::pair<PosIdx, Expr *> > * es;
+    ExprConcatStrings(const PosIdx & pos, bool forceString, std::vector<std::pair<PosIdx, Expr *> > * es)
         : pos(pos), forceString(forceString), es(es) { };
     COMMON_METHODS
 };
 
 struct ExprPos : Expr
 {
-    Pos pos;
-    ExprPos(const Pos & pos) : pos(pos) { };
+    PosIdx pos;
+    ExprPos(const PosIdx & pos) : pos(pos) { };
     COMMON_METHODS
 };
 
@@ -388,7 +455,7 @@ struct StaticEnv
         vars.erase(it, end);
     }
 
-    Vars::const_iterator find(const Symbol & name) const
+    Vars::const_iterator find(Symbol name) const
     {
         Vars::value_type key(name, 0);
         auto i = std::lower_bound(vars.begin(), vars.end(), key);

src/libexpr/parser.y

@@ -32,12 +32,12 @@ namespace nix {
         SymbolTable & symbols;
         Expr * result;
         Path basePath;
-        Symbol file;
-        FileOrigin origin;
+        PosTable::Origin origin;
         std::optional<ErrorInfo> error;
-        ParseData(EvalState & state)
+        ParseData(EvalState & state, PosTable::Origin origin)
             : state(state)
             , symbols(state.symbols)
+            , origin(std::move(origin))
             { };
     };
 
@@ -77,26 +77,26 @@ using namespace nix;
 namespace nix {
 
 
-static void dupAttr(const AttrPath & attrPath, const Pos & pos, const Pos & prevPos)
+static void dupAttr(const EvalState & state, const AttrPath & attrPath, const PosIdx pos, const PosIdx prevPos)
 {
     throw ParseError({
          .msg = hintfmt("attribute '%1%' already defined at %2%",
-             showAttrPath(attrPath), prevPos),
-         .errPos = pos
+             showAttrPath(state.symbols, attrPath), state.positions[prevPos]),
+         .errPos = state.positions[pos]
     });
 }
 
-static void dupAttr(Symbol attr, const Pos & pos, const Pos & prevPos)
+static void dupAttr(const EvalState & state, Symbol attr, const PosIdx pos, const PosIdx prevPos)
 {
     throw ParseError({
-        .msg = hintfmt("attribute '%1%' already defined at %2%", attr, prevPos),
-        .errPos = pos
+        .msg = hintfmt("attribute '%1%' already defined at %2%", state.symbols[attr], state.positions[prevPos]),
+        .errPos = state.positions[pos]
     });
 }
 
 
 static void addAttr(ExprAttrs * attrs, AttrPath & attrPath,
-    Expr * e, const Pos & pos)
+    Expr * e, const PosIdx pos, const nix::EvalState & state)
 {
     AttrPath::iterator i;
     // All attrpaths have at least one attr
@@ -104,15 +104,15 @@ static void addAttr(ExprAttrs * attrs, AttrPath & attrPath,
     // Checking attrPath validity.
     // ===========================
     for (i = attrPath.begin(); i + 1 < attrPath.end(); i++) {
-        if (i->symbol.set()) {
+        if (i->symbol) {
             ExprAttrs::AttrDefs::iterator j = attrs->attrs.find(i->symbol);
             if (j != attrs->attrs.end()) {
                 if (!j->second.inherited) {
                     ExprAttrs * attrs2 = dynamic_cast<ExprAttrs *>(j->second.e);
-                    if (!attrs2) dupAttr(attrPath, pos, j->second.pos);
+                    if (!attrs2) dupAttr(state, attrPath, pos, j->second.pos);
                     attrs = attrs2;
                 } else
-                    dupAttr(attrPath, pos, j->second.pos);
+                    dupAttr(state, attrPath, pos, j->second.pos);
             } else {
                 ExprAttrs * nested = new ExprAttrs;
                 attrs->attrs[i->symbol] = ExprAttrs::AttrDef(nested, pos);
@@ -126,7 +126,7 @@ static void addAttr(ExprAttrs * attrs, AttrPath & attrPath,
     }
     // Expr insertion.
     // ==========================
-    if (i->symbol.set()) {
+    if (i->symbol) {
         ExprAttrs::AttrDefs::iterator j = attrs->attrs.find(i->symbol);
         if (j != attrs->attrs.end()) {
             // This attr path is already defined. However, if both
@@ -139,11 +139,11 @@ static void addAttr(ExprAttrs * attrs, AttrPath & attrPath,
                 for (auto & ad : ae->attrs) {
                     auto j2 = jAttrs->attrs.find(ad.first);
                     if (j2 != jAttrs->attrs.end()) // Attr already defined in iAttrs, error.
-                        dupAttr(ad.first, j2->second.pos, ad.second.pos);
+                        dupAttr(state, ad.first, j2->second.pos, ad.second.pos);
                     jAttrs->attrs.emplace(ad.first, ad.second);
                 }
             } else {
-                dupAttr(attrPath, pos, j->second.pos);
+                dupAttr(state, attrPath, pos, j->second.pos);
             }
         } else {
             // This attr path is not defined. Let's create it.
@@ -157,14 +157,14 @@ static void addAttr(ExprAttrs * attrs, AttrPath & attrPath,
 
 
 static Formals * toFormals(ParseData & data, ParserFormals * formals,
-    Pos pos = noPos, Symbol arg = {})
+    PosIdx pos = noPos, Symbol arg = {})
 {
     std::sort(formals->formals.begin(), formals->formals.end(),
         [] (const auto & a, const auto & b) {
             return std::tie(a.name, a.pos) < std::tie(b.name, b.pos);
         });
 
-    std::optional<std::pair<Symbol, Pos>> duplicate;
+    std::optional<std::pair<Symbol, PosIdx>> duplicate;
     for (size_t i = 0; i + 1 < formals->formals.size(); i++) {
         if (formals->formals[i].name != formals->formals[i + 1].name)
             continue;
@@ -173,18 +173,18 @@ static Formals * toFormals(ParseData & data, ParserFormals * formals,
     }
     if (duplicate)
         throw ParseError({
-            .msg = hintfmt("duplicate formal function argument '%1%'", duplicate->first),
-            .errPos = duplicate->second
+            .msg = hintfmt("duplicate formal function argument '%1%'", data.symbols[duplicate->first]),
+            .errPos = data.state.positions[duplicate->second]
         });
 
     Formals result;
     result.ellipsis = formals->ellipsis;
     result.formals = std::move(formals->formals);
 
-    if (arg.set() && result.has(arg))
+    if (arg && result.has(arg))
         throw ParseError({
-            .msg = hintfmt("duplicate formal function argument '%1%'", arg),
-            .errPos = pos
+            .msg = hintfmt("duplicate formal function argument '%1%'", data.symbols[arg]),
+            .errPos = data.state.positions[pos]
         });
 
     delete formals;
@@ -192,8 +192,8 @@ static Formals * toFormals(ParseData & data, ParserFormals * formals,
 }
 
 
-static Expr * stripIndentation(const Pos & pos, SymbolTable & symbols,
-    std::vector<std::pair<Pos, std::variant<Expr *, StringToken> > > & es)
+static Expr * stripIndentation(const PosIdx pos, SymbolTable & symbols,
+    std::vector<std::pair<PosIdx, std::variant<Expr *, StringToken> > > & es)
 {
     if (es.empty()) return new ExprString("");
 
@@ -233,7 +233,7 @@ static Expr * stripIndentation(const Pos & pos, SymbolTable & symbols,
     }
 
     /* Strip spaces from each line. */
-    std::vector<std::pair<Pos, Expr *> > * es2 = new std::vector<std::pair<Pos, Expr *> >;
+    auto * es2 = new std::vector<std::pair<PosIdx, Expr *> >;
     atStartOfLine = true;
     size_t curDropped = 0;
     size_t n = es.size();
@@ -284,9 +284,9 @@ static Expr * stripIndentation(const Pos & pos, SymbolTable & symbols,
 }
 
 
-static inline Pos makeCurPos(const YYLTYPE & loc, ParseData * data)
+static inline PosIdx makeCurPos(const YYLTYPE & loc, ParseData * data)
 {
-    return Pos(data->origin, data->file, loc.first_line, loc.first_column);
+    return data->state.positions.add(data->origin, loc.first_line, loc.first_column);
 }
 
 #define CUR_POS makeCurPos(*yylocp, data)
@@ -299,7 +299,7 @@ void yyerror(YYLTYPE * loc, yyscan_t scanner, ParseData * data, const char * err
 {
     data->error = {
         .msg = hintfmt(error),
-        .errPos = makeCurPos(*loc, data)
+        .errPos = data->state.positions[makeCurPos(*loc, data)]
     };
 }
 
@@ -320,8 +320,8 @@ void yyerror(YYLTYPE * loc, yyscan_t scanner, ParseData * data, const char * err
   StringToken uri;
   StringToken str;
   std::vector<nix::AttrName> * attrNames;
-  std::vector<std::pair<nix::Pos, nix::Expr *> > * string_parts;
-  std::vector<std::pair<nix::Pos, std::variant<nix::Expr *, StringToken> > > * ind_string_parts;
+  std::vector<std::pair<nix::PosIdx, nix::Expr *> > * string_parts;
+  std::vector<std::pair<nix::PosIdx, std::variant<nix::Expr *, StringToken> > > * ind_string_parts;
 }
 
 %type <e> start expr expr_function expr_if expr_op
@@ -369,15 +369,15 @@ expr_function
   : ID ':' expr_function
     { $$ = new ExprLambda(CUR_POS, data->symbols.create($1), 0, $3); }
   | '{' formals '}' ':' expr_function
-    { $$ = new ExprLambda(CUR_POS, data->symbols.create(""), toFormals(*data, $2), $5); }
+    { $$ = new ExprLambda(CUR_POS, toFormals(*data, $2), $5); }
   | '{' formals '}' '@' ID ':' expr_function
     {
-      Symbol arg = data->symbols.create($5);
+      auto arg = data->symbols.create($5);
       $$ = new ExprLambda(CUR_POS, arg, toFormals(*data, $2, CUR_POS, arg), $7);
     }
   | ID '@' '{' formals '}' ':' expr_function
     {
-      Symbol arg = data->symbols.create($1);
+      auto arg = data->symbols.create($1);
       $$ = new ExprLambda(CUR_POS, arg, toFormals(*data, $4, CUR_POS, arg), $7);
     }
   | ASSERT expr ';' expr_function
@@ -388,7 +388,7 @@ expr_function
     { if (!$2->dynamicAttrs.empty())
         throw ParseError({
             .msg = hintfmt("dynamic attributes not allowed in let"),
-            .errPos = CUR_POS
+            .errPos = data->state.positions[CUR_POS]
         });
       $$ = new ExprLet($2, $4);
     }
@@ -415,7 +415,7 @@ expr_op
   | expr_op UPDATE expr_op { $$ = new ExprOpUpdate(CUR_POS, $1, $3); }
   | expr_op '?' attrpath { $$ = new ExprOpHasAttr($1, *$3); }
   | expr_op '+' expr_op
-    { $$ = new ExprConcatStrings(CUR_POS, false, new std::vector<std::pair<Pos, Expr *> >({{makeCurPos(@1, data), $1}, {makeCurPos(@3, data), $3}})); }
+    { $$ = new ExprConcatStrings(CUR_POS, false, new std::vector<std::pair<PosIdx, Expr *> >({{makeCurPos(@1, data), $1}, {makeCurPos(@3, data), $3}})); }
   | expr_op '-' expr_op { $$ = new ExprCall(CUR_POS, new ExprVar(data->symbols.create("__sub")), {$1, $3}); }
   | expr_op '*' expr_op { $$ = new ExprCall(CUR_POS, new ExprVar(data->symbols.create("__mul")), {$1, $3}); }
   | expr_op '/' expr_op { $$ = new ExprCall(CUR_POS, new ExprVar(data->symbols.create("__div")), {$1, $3}); }
@@ -477,7 +477,7 @@ expr_simple
       if (noURLLiterals)
           throw ParseError({
               .msg = hintfmt("URL literals are disabled"),
-              .errPos = CUR_POS
+              .errPos = data->state.positions[CUR_POS]
           });
       $$ = new ExprString(std::string($1));
   }
@@ -503,9 +503,9 @@ string_parts_interpolated
   : string_parts_interpolated STR
   { $$ = $1; $1->emplace_back(makeCurPos(@2, data), new ExprString(std::string($2))); }
   | string_parts_interpolated DOLLAR_CURLY expr '}' { $$ = $1; $1->emplace_back(makeCurPos(@2, data), $3); }
-  | DOLLAR_CURLY expr '}' { $$ = new std::vector<std::pair<Pos, Expr *> >; $$->emplace_back(makeCurPos(@1, data), $2); }
+  | DOLLAR_CURLY expr '}' { $$ = new std::vector<std::pair<PosIdx, Expr *> >; $$->emplace_back(makeCurPos(@1, data), $2); }
   | STR DOLLAR_CURLY expr '}' {
-      $$ = new std::vector<std::pair<Pos, Expr *> >;
+      $$ = new std::vector<std::pair<PosIdx, Expr *> >;
       $$->emplace_back(makeCurPos(@1, data), new ExprString(std::string($1)));
       $$->emplace_back(makeCurPos(@2, data), $3);
     }
@@ -528,17 +528,17 @@ path_start
 ind_string_parts
   : ind_string_parts IND_STR { $$ = $1; $1->emplace_back(makeCurPos(@2, data), $2); }
   | ind_string_parts DOLLAR_CURLY expr '}' { $$ = $1; $1->emplace_back(makeCurPos(@2, data), $3); }
-  | { $$ = new std::vector<std::pair<Pos, std::variant<Expr *, StringToken> > >; }
+  | { $$ = new std::vector<std::pair<PosIdx, std::variant<Expr *, StringToken> > >; }
   ;
 
 binds
-  : binds attrpath '=' expr ';' { $$ = $1; addAttr($$, *$2, $4, makeCurPos(@2, data)); }
+  : binds attrpath '=' expr ';' { $$ = $1; addAttr($$, *$2, $4, makeCurPos(@2, data), data->state); }
   | binds INHERIT attrs ';'
     { $$ = $1;
       for (auto & i : *$3) {
           if ($$->attrs.find(i.symbol) != $$->attrs.end())
-              dupAttr(i.symbol, makeCurPos(@3, data), $$->attrs[i.symbol].pos);
-          Pos pos = makeCurPos(@3, data);
+              dupAttr(data->state, i.symbol, makeCurPos(@3, data), $$->attrs[i.symbol].pos);
+          auto pos = makeCurPos(@3, data);
           $$->attrs.emplace(i.symbol, ExprAttrs::AttrDef(new ExprVar(CUR_POS, i.symbol), pos, true));
       }
     }
@@ -547,7 +547,7 @@ binds
       /* !!! Should ensure sharing of the expression in $4. */
       for (auto & i : *$6) {
           if ($$->attrs.find(i.symbol) != $$->attrs.end())
-              dupAttr(i.symbol, makeCurPos(@6, data), $$->attrs[i.symbol].pos);
+              dupAttr(data->state, i.symbol, makeCurPos(@6, data), $$->attrs[i.symbol].pos);
           $$->attrs.emplace(i.symbol, ExprAttrs::AttrDef(new ExprSelect(CUR_POS, $4, i.symbol), makeCurPos(@6, data)));
       }
     }
@@ -565,7 +565,7 @@ attrs
       } else
           throw ParseError({
               .msg = hintfmt("dynamic attributes not allowed in inherit"),
-              .errPos = makeCurPos(@2, data)
+              .errPos = data->state.positions[makeCurPos(@2, data)]
           });
     }
   | { $$ = new AttrPath; }
@@ -621,8 +621,8 @@ formals
   ;
 
 formal
-  : ID { $$ = new Formal(CUR_POS, data->symbols.create($1), 0); }
-  | ID '?' expr { $$ = new Formal(CUR_POS, data->symbols.create($1), $3); }
+  : ID { $$ = new Formal{CUR_POS, data->symbols.create($1), 0}; }
+  | ID '?' expr { $$ = new Formal{CUR_POS, data->symbols.create($1), $3}; }
   ;
 
 %%
@@ -646,19 +646,19 @@ Expr * EvalState::parse(char * text, size_t length, FileOrigin origin,
     const PathView path, const PathView basePath, StaticEnv & staticEnv)
 {
     yyscan_t scanner;
-    ParseData data(*this);
-    data.origin = origin;
+    std::string file;
     switch (origin) {
         case foFile:
-            data.file = data.symbols.create(path);
+            file = path;
             break;
         case foStdin:
         case foString:
-            data.file = data.symbols.create(text);
+            file = text;
             break;
         default:
             assert(false);
     }
+    ParseData data(*this, {file, origin});
     data.basePath = basePath;
 
     yylex_init(&scanner);
@@ -668,7 +668,7 @@ Expr * EvalState::parse(char * text, size_t length, FileOrigin origin,
 
     if (res) throw ParseError(data.error.value());
 
-    data.result->bindVars(staticEnv);
+    data.result->bindVars(*this, staticEnv);
 
     return data.result;
 }
@@ -768,7 +768,7 @@ Path EvalState::findFile(const std::string_view path)
 }
 
 
-Path EvalState::findFile(SearchPath & searchPath, const std::string_view path, const Pos & pos)
+Path EvalState::findFile(SearchPath & searchPath, const std::string_view path, const PosIdx pos)
 {
     for (auto & i : searchPath) {
         std::string suffix;
@@ -796,7 +796,7 @@ Path EvalState::findFile(SearchPath & searchPath, const std::string_view path, c
             ? "cannot look up '<%s>' in pure evaluation mode (use '--impure' to override)"
             : "file '%s' was not found in the Nix search path (add it using $NIX_PATH or -I)",
             path),
-        .errPos = pos
+        .errPos = positions[pos]
     });
 }
 

src/libexpr/primops.hh

@@ -38,9 +38,9 @@ struct RegisterPrimOp
    them. */
 
 /* Load a ValueInitializer from a DSO and return whatever it initializes */
-void prim_importNative(EvalState & state, const Pos & pos, Value * * args, Value & v);
+void prim_importNative(EvalState & state, const PosIdx pos, Value * * args, Value & v);
 
 /* Execute a program and parse its output */
-void prim_exec(EvalState & state, const Pos & pos, Value * * args, Value & v);
+void prim_exec(EvalState & state, const PosIdx pos, Value * * args, Value & v);
 
 }

src/libexpr/primops/context.cc

@@ -5,7 +5,7 @@
 
 namespace nix {
 
-static void prim_unsafeDiscardStringContext(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_unsafeDiscardStringContext(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     PathSet context;
     auto s = state.coerceToString(pos, *args[0], context);
@@ -15,7 +15,7 @@ static void prim_unsafeDiscardStringContext(EvalState & state, const Pos & pos,
 static RegisterPrimOp primop_unsafeDiscardStringContext("__unsafeDiscardStringContext", 1, prim_unsafeDiscardStringContext);
 
 
-static void prim_hasContext(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_hasContext(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     PathSet context;
     state.forceString(*args[0], context, pos);
@@ -31,7 +31,7 @@ static RegisterPrimOp primop_hasContext("__hasContext", 1, prim_hasContext);
    source-only deployment).  This primop marks the string context so
    that builtins.derivation adds the path to drv.inputSrcs rather than
    drv.inputDrvs. */
-static void prim_unsafeDiscardOutputDependency(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_unsafeDiscardOutputDependency(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     PathSet context;
     auto s = state.coerceToString(pos, *args[0], context);
@@ -65,7 +65,7 @@ static RegisterPrimOp primop_unsafeDiscardOutputDependency("__unsafeDiscardOutpu
    Note that for a given path any combination of the above attributes
    may be present.
 */
-static void prim_getContext(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_getContext(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     struct ContextInfo {
         bool path = false;
@@ -134,7 +134,7 @@ static RegisterPrimOp primop_getContext("__getContext", 1, prim_getContext);
    See the commentary above unsafeGetContext for details of the
    context representation.
 */
-static void prim_appendContext(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_appendContext(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     PathSet context;
     auto orig = state.forceString(*args[0], context, pos);
@@ -144,45 +144,46 @@ static void prim_appendContext(EvalState & state, const Pos & pos, Value * * arg
     auto sPath = state.symbols.create("path");
     auto sAllOutputs = state.symbols.create("allOutputs");
     for (auto & i : *args[1]->attrs) {
-        if (!state.store->isStorePath(i.name))
+        const auto & name = state.symbols[i.name];
+        if (!state.store->isStorePath(name))
             throw EvalError({
-                .msg = hintfmt("Context key '%s' is not a store path", i.name),
-                .errPos = *i.pos
+                .msg = hintfmt("Context key '%s' is not a store path", name),
+                .errPos = state.positions[i.pos]
             });
         if (!settings.readOnlyMode)
-            state.store->ensurePath(state.store->parseStorePath(i.name));
-        state.forceAttrs(*i.value, *i.pos);
+            state.store->ensurePath(state.store->parseStorePath(name));
+        state.forceAttrs(*i.value, i.pos);
         auto iter = i.value->attrs->find(sPath);
         if (iter != i.value->attrs->end()) {
-            if (state.forceBool(*iter->value, *iter->pos))
-                context.insert(i.name);
+            if (state.forceBool(*iter->value, iter->pos))
+                context.emplace(name);
         }
 
         iter = i.value->attrs->find(sAllOutputs);
         if (iter != i.value->attrs->end()) {
-            if (state.forceBool(*iter->value, *iter->pos)) {
-                if (!isDerivation(i.name)) {
+            if (state.forceBool(*iter->value, iter->pos)) {
+                if (!isDerivation(name)) {
                     throw EvalError({
-                        .msg = hintfmt("Tried to add all-outputs context of %s, which is not a derivation, to a string", i.name),
-                        .errPos = *i.pos
+                        .msg = hintfmt("Tried to add all-outputs context of %s, which is not a derivation, to a string", name),
+                        .errPos = state.positions[i.pos]
                     });
                 }
-                context.insert("=" + std::string(i.name));
+                context.insert(concatStrings("=", name));
             }
         }
 
         iter = i.value->attrs->find(state.sOutputs);
         if (iter != i.value->attrs->end()) {
-            state.forceList(*iter->value, *iter->pos);
-            if (iter->value->listSize() && !isDerivation(i.name)) {
+            state.forceList(*iter->value, iter->pos);
+            if (iter->value->listSize() && !isDerivation(name)) {
                 throw EvalError({
-                    .msg = hintfmt("Tried to add derivation output context of %s, which is not a derivation, to a string", i.name),
-                    .errPos = *i.pos
+                    .msg = hintfmt("Tried to add derivation output context of %s, which is not a derivation, to a string", name),
+                    .errPos = state.positions[i.pos]
                 });
             }
             for (auto elem : iter->value->listItems()) {
-                auto name = state.forceStringNoCtx(*elem, *iter->pos);
-                context.insert(concatStrings("!", name, "!", i.name));
+                auto outputName = state.forceStringNoCtx(*elem, iter->pos);
+                context.insert(concatStrings("!", outputName, "!", name));
             }
         }
     }

src/libexpr/primops/fetchClosure.cc

@@ -5,7 +5,7 @@
 
 namespace nix {
 
-static void prim_fetchClosure(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_fetchClosure(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     state.forceAttrs(*args[0], pos);
 
@@ -15,40 +15,42 @@ static void prim_fetchClosure(EvalState & state, const Pos & pos, Value * * args
     std::optional<StorePath> toPath;
 
     for (auto & attr : *args[0]->attrs) {
-        if (attr.name == "fromPath") {
+        const auto & attrName = state.symbols[attr.name];
+
+        if (attrName == "fromPath") {
             PathSet context;
-            fromPath = state.coerceToStorePath(*attr.pos, *attr.value, context);
+            fromPath = state.coerceToStorePath(attr.pos, *attr.value, context);
         }
 
-        else if (attr.name == "toPath") {
-            state.forceValue(*attr.value, *attr.pos);
+        else if (attrName == "toPath") {
+            state.forceValue(*attr.value, attr.pos);
             toCA = true;
             if (attr.value->type() != nString || attr.value->string.s != std::string("")) {
                 PathSet context;
-                toPath = state.coerceToStorePath(*attr.pos, *attr.value, context);
+                toPath = state.coerceToStorePath(attr.pos, *attr.value, context);
             }
         }
 
-        else if (attr.name == "fromStore")
-            fromStoreUrl = state.forceStringNoCtx(*attr.value, *attr.pos);
+        else if (attrName == "fromStore")
+            fromStoreUrl = state.forceStringNoCtx(*attr.value, attr.pos);
 
         else
             throw Error({
-                .msg = hintfmt("attribute '%s' isn't supported in call to 'fetchClosure'", attr.name),
-                .errPos = pos
+                .msg = hintfmt("attribute '%s' isn't supported in call to 'fetchClosure'", attrName),
+                .errPos = state.positions[pos]
             });
     }
 
     if (!fromPath)
         throw Error({
             .msg = hintfmt("attribute '%s' is missing in call to 'fetchClosure'", "fromPath"),
-            .errPos = pos
+            .errPos = state.positions[pos]
         });
 
     if (!fromStoreUrl)
         throw Error({
             .msg = hintfmt("attribute '%s' is missing in call to 'fetchClosure'", "fromStore"),
-            .errPos = pos
+            .errPos = state.positions[pos]
         });
 
     auto parsedURL = parseURL(*fromStoreUrl);
@@ -58,7 +60,13 @@ static void prim_fetchClosure(EvalState & state, const Pos & pos, Value * * args
         !(getEnv("_NIX_IN_TEST").has_value() && parsedURL.scheme == "file"))
         throw Error({
             .msg = hintfmt("'fetchClosure' only supports http:// and https:// stores"),
-            .errPos = pos
+            .errPos = state.positions[pos]
+        });
+
+    if (!parsedURL.query.empty())
+        throw Error({
+            .msg = hintfmt("'fetchClosure' does not support URL query parameters (in '%s')", *fromStoreUrl),
+            .errPos = state.positions[pos]
         });
 
     auto fromStore = openStore(parsedURL.to_string());
@@ -74,7 +82,7 @@ static void prim_fetchClosure(EvalState & state, const Pos & pos, Value * * args
                         state.store->printStorePath(*fromPath),
                         state.store->printStorePath(i->second),
                         state.store->printStorePath(*toPath)),
-                    .errPos = pos
+                    .errPos = state.positions[pos]
                 });
             if (!toPath)
                 throw Error({
@@ -83,11 +91,12 @@ static void prim_fetchClosure(EvalState & state, const Pos & pos, Value * * args
                         "please set this in the 'toPath' attribute passed to 'fetchClosure'",
                         state.store->printStorePath(*fromPath),
                         state.store->printStorePath(i->second)),
-                    .errPos = pos
+                    .errPos = state.positions[pos]
                 });
         }
     } else {
-        copyClosure(*fromStore, *state.store, RealisedPath::Set { *fromPath });
+        if (!state.store->isValidPath(*fromPath))
+            copyClosure(*fromStore, *state.store, RealisedPath::Set { *fromPath });
         toPath = fromPath;
     }
 
@@ -98,7 +107,7 @@ static void prim_fetchClosure(EvalState & state, const Pos & pos, Value * * args
             throw Error({
                 .msg = hintfmt("in pure mode, 'fetchClosure' requires a content-addressed path, which '%s' isn't",
                     state.store->printStorePath(*toPath)),
-                .errPos = pos
+                .errPos = state.positions[pos]
             });
     }
 

src/libexpr/primops/fetchMercurial.cc

@@ -7,7 +7,7 @@
 
 namespace nix {
 
-static void prim_fetchMercurial(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_fetchMercurial(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     std::string url;
     std::optional<Hash> rev;
@@ -22,31 +22,31 @@ static void prim_fetchMercurial(EvalState & state, const Pos & pos, Value * * ar
         state.forceAttrs(*args[0], pos);
 
         for (auto & attr : *args[0]->attrs) {
-            std::string_view n(attr.name);
+            std::string_view n(state.symbols[attr.name]);
             if (n == "url")
-                url = state.coerceToString(*attr.pos, *attr.value, context, false, false).toOwned();
+                url = state.coerceToString(attr.pos, *attr.value, context, false, false).toOwned();
             else if (n == "rev") {
                 // Ugly: unlike fetchGit, here the "rev" attribute can
                 // be both a revision or a branch/tag name.
-                auto value = state.forceStringNoCtx(*attr.value, *attr.pos);
+                auto value = state.forceStringNoCtx(*attr.value, attr.pos);
                 if (std::regex_match(value.begin(), value.end(), revRegex))
                     rev = Hash::parseAny(value, htSHA1);
                 else
                     ref = value;
             }
             else if (n == "name")
-                name = state.forceStringNoCtx(*attr.value, *attr.pos);
+                name = state.forceStringNoCtx(*attr.value, attr.pos);
             else
                 throw EvalError({
-                    .msg = hintfmt("unsupported argument '%s' to 'fetchMercurial'", attr.name),
-                    .errPos = *attr.pos
+                    .msg = hintfmt("unsupported argument '%s' to 'fetchMercurial'", state.symbols[attr.name]),
+                    .errPos = state.positions[attr.pos]
                 });
         }
 
         if (url.empty())
             throw EvalError({
                 .msg = hintfmt("'url' argument required"),
-                .errPos = pos
+                .errPos = state.positions[pos]
             });
 
     } else

src/libexpr/primops/fetchTree.cc

@@ -97,7 +97,7 @@ struct FetchTreeParams {
 
 static void fetchTree(
     EvalState & state,
-    const Pos & pos,
+    const PosIdx pos,
     Value * * args,
     Value & v,
     std::optional<std::string> type,
@@ -118,13 +118,13 @@ static void fetchTree(
             if (type)
                 throw Error({
                     .msg = hintfmt("unexpected attribute 'type'"),
-                    .errPos = pos
+                    .errPos = state.positions[pos]
                 });
-            type = state.forceStringNoCtx(*aType->value, *aType->pos);
+            type = state.forceStringNoCtx(*aType->value, aType->pos);
         } else if (!type)
             throw Error({
                 .msg = hintfmt("attribute 'type' is missing in call to 'fetchTree'"),
-                .errPos = pos
+                .errPos = state.positions[pos]
             });
 
         attrs.emplace("type", type.value());
@@ -132,8 +132,8 @@ static void fetchTree(
         for (auto & attr : *args[0]->attrs) {
             if (attr.name == state.sType) continue;
 
-            if (attr.name == "patches") {
-                state.forceList(*attr.value, *attr.pos);
+            if (state.symbols[attr.name] == "patches") {
+                state.forceList(*attr.value, attr.pos);
 
                 for (auto elem : attr.value->listItems()) {
                     // FIXME: use realisePath
@@ -145,31 +145,31 @@ static void fetchTree(
                 continue;
             }
 
-            state.forceValue(*attr.value, *attr.pos);
+            state.forceValue(*attr.value, attr.pos);
 
             if (attr.value->type() == nPath || attr.value->type() == nString) {
-                auto s = state.coerceToString(*attr.pos, *attr.value, context, false, false).toOwned();
-                attrs.emplace(attr.name,
-                    attr.name == "url"
+                auto s = state.coerceToString(attr.pos, *attr.value, context, false, false).toOwned();
+                attrs.emplace(state.symbols[attr.name],
+                    state.symbols[attr.name] == "url"
                     ? type == "git"
                       ? fixURIForGit(s, state)
                       : fixURI(s, state)
                     : s);
             }
             else if (attr.value->type() == nBool)
-                attrs.emplace(attr.name, Explicit<bool>{attr.value->boolean});
+                attrs.emplace(state.symbols[attr.name], Explicit<bool>{attr.value->boolean});
             else if (attr.value->type() == nInt)
-                attrs.emplace(attr.name, uint64_t(attr.value->integer));
+                attrs.emplace(state.symbols[attr.name], uint64_t(attr.value->integer));
             else
                 throw TypeError("fetchTree argument '%s' is %s while a string, Boolean or integer is expected",
-                    attr.name, showType(*attr.value));
+                    state.symbols[attr.name], showType(*attr.value));
         }
 
         if (!params.allowNameArgument)
             if (auto nameIter = attrs.find("name"); nameIter != attrs.end())
                 throw Error({
                     .msg = hintfmt("attribute 'name' isn't supported in call to 'fetchTree'"),
-                    .errPos = pos
+                    .errPos = state.positions[pos]
                 });
 
         input = fetchers::Input::fromAttrs(std::move(attrs));
@@ -190,7 +190,7 @@ static void fetchTree(
         input = lookupInRegistries(state.store, input).first;
 
     if (evalSettings.pureEval && !input.isLocked())
-        throw Error("in pure evaluation mode, 'fetchTree' requires a locked input, at %s", pos);
+        throw Error("in pure evaluation mode, 'fetchTree' requires a locked input, at %s", state.positions[pos]);
 
     auto [accessor, input2] = input.lazyFetch(state.store);
 
@@ -208,7 +208,7 @@ static void fetchTree(
         false);
 }
 
-static void prim_fetchTree(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_fetchTree(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     settings.requireExperimentalFeature(Xp::Flakes);
     fetchTree(state, pos, args, v, std::nullopt, FetchTreeParams { .allowNameArgument = false });
@@ -217,7 +217,7 @@ static void prim_fetchTree(EvalState & state, const Pos & pos, Value * * args, V
 // FIXME: document
 static RegisterPrimOp primop_fetchTree("fetchTree", 1, prim_fetchTree);
 
-static void fetch(EvalState & state, const Pos & pos, Value * * args, Value & v,
+static void fetch(EvalState & state, const PosIdx pos, Value * * args, Value & v,
     const std::string & who, bool unpack, std::string name)
 {
     std::optional<std::string> url;
@@ -230,24 +230,24 @@ static void fetch(EvalState & state, const Pos & pos, Value * * args, Value & v,
         state.forceAttrs(*args[0], pos);
 
         for (auto & attr : *args[0]->attrs) {
-            std::string n(attr.name);
+            std::string_view n(state.symbols[attr.name]);
             if (n == "url")
-                url = state.forceStringNoCtx(*attr.value, *attr.pos);
+                url = state.forceStringNoCtx(*attr.value, attr.pos);
             else if (n == "sha256")
-                expectedHash = newHashAllowEmpty(state.forceStringNoCtx(*attr.value, *attr.pos), htSHA256);
+                expectedHash = newHashAllowEmpty(state.forceStringNoCtx(*attr.value, attr.pos), htSHA256);
             else if (n == "name")
-                name = state.forceStringNoCtx(*attr.value, *attr.pos);
+                name = state.forceStringNoCtx(*attr.value, attr.pos);
             else
                 throw EvalError({
-                    .msg = hintfmt("unsupported argument '%s' to '%s'", attr.name, who),
-                    .errPos = *attr.pos
+                    .msg = hintfmt("unsupported argument '%s' to '%s'", n, who),
+                    .errPos = state.positions[attr.pos]
                 });
             }
 
         if (!url)
             throw EvalError({
                 .msg = hintfmt("'url' argument required"),
-                .errPos = pos
+                .errPos = state.positions[pos]
             });
     } else
         url = state.forceStringNoCtx(*args[0], pos);
@@ -294,7 +294,7 @@ static void fetch(EvalState & state, const Pos & pos, Value * * args, Value & v,
     state.allowAndSetStorePathString(storePath, v);
 }
 
-static void prim_fetchurl(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_fetchurl(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     fetch(state, pos, args, v, "fetchurl", false, "");
 }
@@ -310,7 +310,7 @@ static RegisterPrimOp primop_fetchurl({
     .fun = prim_fetchurl,
 });
 
-static void prim_fetchTarball(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_fetchTarball(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     fetch(state, pos, args, v, "fetchTarball", true, "source");
 }
@@ -361,7 +361,7 @@ static RegisterPrimOp primop_fetchTarball({
     .fun = prim_fetchTarball,
 });
 
-static void prim_fetchGit(EvalState & state, const Pos & pos, Value * * args, Value & v)
+static void prim_fetchGit(EvalState & state, const PosIdx pos, Value * * args, Value & v)
 {
     fetchTree(state, pos, args, v, "git", FetchTreeParams { .emptyRevFallback = true, .allowNameArgument = true });
 }

src/libexpr/primops/fromTOML.cc

@@ -5,7 +5,7 @@
 
 namespace nix {
 
-static void prim_fromTOML(EvalState & state, const Pos & pos, Value * * args, Value & val)
+static void prim_fromTOML(EvalState & state, const PosIdx pos, Value * * args, Value & val)
 {
     auto toml = state.forceStringNoCtx(*args[0], pos);
 
@@ -73,7 +73,7 @@ static void prim_fromTOML(EvalState & state, const Pos & pos, Value * * args, Va
     } catch (std::exception & e) { // TODO: toml::syntax_error
         throw EvalError({
             .msg = hintfmt("while parsing a TOML string: %s", e.what()),
-            .errPos = pos
+            .errPos = state.positions[pos]
         });
     }
 }

src/libexpr/symbol-table.hh

@@ -5,44 +5,32 @@
 #include <unordered_map>
 
 #include "types.hh"
+#include "chunked-vector.hh"
 
 namespace nix {
 
 /* Symbol table used by the parser and evaluator to represent and look
    up identifiers and attributes efficiently.  SymbolTable::create()
    converts a string into a symbol.  Symbols have the property that
-   they can be compared efficiently (using a pointer equality test),
+   they can be compared efficiently (using an equality test),
    because the symbol table stores only one copy of each string. */
 
-class Symbol
+/* This class mainly exists to give us an operator<< for ostreams. We could also
+   return plain strings from SymbolTable, but then we'd have to wrap every
+   instance of a symbol that is fmt()ed, which is inconvenient and error-prone. */
+class SymbolStr
 {
-private:
-    const std::string * s; // pointer into SymbolTable
-    Symbol(const std::string * s) : s(s) { };
     friend class SymbolTable;
 
+private:
+    const std::string * s;
+
+    explicit SymbolStr(const std::string & symbol): s(&symbol) {}
+
 public:
-    Symbol() : s(0) { };
-
-    bool operator == (const Symbol & s2) const
-    {
-        return s == s2.s;
-    }
-
-    // FIXME: remove
     bool operator == (std::string_view s2) const
     {
-        return s->compare(s2) == 0;
-    }
-
-    bool operator != (const Symbol & s2) const
-    {
-        return s != s2.s;
-    }
-
-    bool operator < (const Symbol & s2) const
-    {
-        return s < s2.s;
+        return *s == s2;
     }
 
     operator const std::string & () const
@@ -55,51 +43,78 @@ public:
         return *s;
     }
 
-    bool set() const
-    {
-        return s;
-    }
+    friend std::ostream & operator <<(std::ostream & os, const SymbolStr & symbol);
+};
 
-    bool empty() const
-    {
-        return s->empty();
-    }
+class Symbol
+{
+    friend class SymbolTable;
 
-    friend std::ostream & operator << (std::ostream & str, const Symbol & sym);
+private:
+    uint32_t id;
+
+    explicit Symbol(uint32_t id): id(id) {}
+
+public:
+    Symbol() : id(0) {}
+
+    explicit operator bool() const { return id > 0; }
+
+    bool operator<(const Symbol other) const { return id < other.id; }
+    bool operator==(const Symbol other) const { return id == other.id; }
+    bool operator!=(const Symbol other) const { return id != other.id; }
 };
 
 class SymbolTable
 {
 private:
-    std::unordered_map<std::string_view, Symbol> symbols;
-    std::list<std::string> store;
+    std::unordered_map<std::string_view, std::pair<const std::string *, uint32_t>> symbols;
+    ChunkedVector<std::string, 8192> store{16};
 
 public:
+
     Symbol create(std::string_view s)
     {
         // Most symbols are looked up more than once, so we trade off insertion performance
         // for lookup performance.
         // TODO: could probably be done more efficiently with transparent Hash and Equals
         // on the original implementation using unordered_set
+        // FIXME: make this thread-safe.
         auto it = symbols.find(s);
-        if (it != symbols.end()) return it->second;
+        if (it != symbols.end()) return Symbol(it->second.second + 1);
 
-        auto & rawSym = store.emplace_back(s);
-        return symbols.emplace(rawSym, Symbol(&rawSym)).first->second;
+        const auto & [rawSym, idx] = store.add(std::string(s));
+        symbols.emplace(rawSym, std::make_pair(&rawSym, idx));
+        return Symbol(idx + 1);
+    }
+
+    std::vector<SymbolStr> resolve(const std::vector<Symbol> & symbols) const
+    {
+        std::vector<SymbolStr> result;
+        result.reserve(symbols.size());
+        for (auto sym : symbols)
+            result.push_back((*this)[sym]);
+        return result;
+    }
+
+    SymbolStr operator[](Symbol s) const
+    {
+        if (s.id == 0 || s.id > store.size())
+            abort();
+        return SymbolStr(store[s.id - 1]);
     }
 
     size_t size() const
     {
-        return symbols.size();
+        return store.size();
     }
 
     size_t totalSize() const;
 
     template<typename T>
-    void dump(T callback)
+    void dump(T callback) const
     {
-        for (auto & s : store)
-            callback(s);
+        store.forEach(callback);
     }
 };
 

src/libexpr/value-to-json.cc

@@ -10,7 +10,7 @@
 namespace nix {
 
 void printValueAsJSON(EvalState & state, bool strict,
-    Value & v, const Pos & pos, JSONPlaceholder & out, PathSet & context)
+    Value & v, const PosIdx pos, JSONPlaceholder & out, PathSet & context)
 {
     checkInterrupt();
 
@@ -51,14 +51,14 @@ void printValueAsJSON(EvalState & state, bool strict,
                 auto obj(out.object());
                 StringSet names;
                 for (auto & j : *v.attrs)
-                    names.insert(j.name);
+                    names.emplace(state.symbols[j.name]);
                 for (auto & j : names) {
                     Attr & a(*v.attrs->find(state.symbols.create(j)));
                     auto placeholder(obj.placeholder(j));
-                    printValueAsJSON(state, strict, *a.value, *a.pos, placeholder, context);
+                    printValueAsJSON(state, strict, *a.value, a.pos, placeholder, context);
                 }
             } else
-                printValueAsJSON(state, strict, *i->value, *i->pos, out, context);
+                printValueAsJSON(state, strict, *i->value, i->pos, out, context);
             break;
         }
 
@@ -83,14 +83,15 @@ void printValueAsJSON(EvalState & state, bool strict,
         case nFunction:
             auto e = TypeError({
                 .msg = hintfmt("cannot convert %1% to JSON", showType(v)),
-                .errPos = v.determinePos(pos)
+                .errPos = state.positions[v.determinePos(pos)]
             });
-            throw e.addTrace(pos, hintfmt("message for the trace"));
+            e.addTrace(state.positions[pos], hintfmt("message for the trace"));
+            throw e;
     }
 }
 
 void printValueAsJSON(EvalState & state, bool strict,
-    Value & v, const Pos & pos, std::ostream & str, PathSet & context)
+    Value & v, const PosIdx pos, std::ostream & str, PathSet & context)
 {
     JSONPlaceholder out(str);
     printValueAsJSON(state, strict, v, pos, out, context);

src/libexpr/value-to-json.hh

@@ -11,9 +11,9 @@ namespace nix {
 class JSONPlaceholder;
 
 void printValueAsJSON(EvalState & state, bool strict,
-    Value & v, const Pos & pos, JSONPlaceholder & out, PathSet & context);
+    Value & v, const PosIdx pos, JSONPlaceholder & out, PathSet & context);
 
 void printValueAsJSON(EvalState & state, bool strict,
-    Value & v, const Pos & pos, std::ostream & str, PathSet & context);
+    Value & v, const PosIdx pos, std::ostream & str, PathSet & context);
 
 }

src/libexpr/value-to-xml.cc

@@ -19,10 +19,10 @@ static XMLAttrs singletonAttrs(const std::string & name, const std::string & val
 
 static void printValueAsXML(EvalState & state, bool strict, bool location,
     Value & v, XMLWriter & doc, PathSet & context, PathSet & drvsSeen,
-    const Pos & pos);
+    const PosIdx pos);
 
 
-static void posToXML(XMLAttrs & xmlAttrs, const Pos & pos)
+static void posToXML(EvalState & state, XMLAttrs & xmlAttrs, const Pos & pos)
 {
     xmlAttrs["path"] = pos.file;
     xmlAttrs["line"] = (format("%1%") % pos.line).str();
@@ -36,25 +36,25 @@ static void showAttrs(EvalState & state, bool strict, bool location,
     StringSet names;
 
     for (auto & i : attrs)
-        names.insert(i.name);
+        names.emplace(state.symbols[i.name]);
 
     for (auto & i : names) {
         Attr & a(*attrs.find(state.symbols.create(i)));
 
         XMLAttrs xmlAttrs;
         xmlAttrs["name"] = i;
-        if (location && a.pos != ptr(&noPos)) posToXML(xmlAttrs, *a.pos);
+        if (location && a.pos) posToXML(state, xmlAttrs, state.positions[a.pos]);
 
         XMLOpenElement _(doc, "attr", xmlAttrs);
         printValueAsXML(state, strict, location,
-            *a.value, doc, context, drvsSeen, *a.pos);
+            *a.value, doc, context, drvsSeen, a.pos);
     }
 }
 
 
 static void printValueAsXML(EvalState & state, bool strict, bool location,
     Value & v, XMLWriter & doc, PathSet & context, PathSet & drvsSeen,
-    const Pos & pos)
+    const PosIdx pos)
 {
     checkInterrupt();
 
@@ -93,14 +93,14 @@ static void printValueAsXML(EvalState & state, bool strict, bool location,
                 Path drvPath;
                 a = v.attrs->find(state.sDrvPath);
                 if (a != v.attrs->end()) {
-                    if (strict) state.forceValue(*a->value, *a->pos);
+                    if (strict) state.forceValue(*a->value, a->pos);
                     if (a->value->type() == nString)
                         xmlAttrs["drvPath"] = drvPath = a->value->string.s;
                 }
 
                 a = v.attrs->find(state.sOutPath);
                 if (a != v.attrs->end()) {
-                    if (strict) state.forceValue(*a->value, *a->pos);
+                    if (strict) state.forceValue(*a->value, a->pos);
                     if (a->value->type() == nString)
                         xmlAttrs["outPath"] = a->value->string.s;
                 }
@@ -134,18 +134,18 @@ static void printValueAsXML(EvalState & state, bool strict, bool location,
                 break;
             }
             XMLAttrs xmlAttrs;
-            if (location) posToXML(xmlAttrs, v.lambda.fun->pos);
+            if (location) posToXML(state, xmlAttrs, state.positions[v.lambda.fun->pos]);
             XMLOpenElement _(doc, "function", xmlAttrs);
 
             if (v.lambda.fun->hasFormals()) {
                 XMLAttrs attrs;
-                if (!v.lambda.fun->arg.empty()) attrs["name"] = v.lambda.fun->arg;
+                if (v.lambda.fun->arg) attrs["name"] = state.symbols[v.lambda.fun->arg];
                 if (v.lambda.fun->formals->ellipsis) attrs["ellipsis"] = "1";
                 XMLOpenElement _(doc, "attrspat", attrs);
-                for (auto & i : v.lambda.fun->formals->lexicographicOrder())
-                    doc.writeEmptyElement("attr", singletonAttrs("name", i.name));
+                for (auto & i : v.lambda.fun->formals->lexicographicOrder(state.symbols))
+                    doc.writeEmptyElement("attr", singletonAttrs("name", state.symbols[i.name]));
             } else
-                doc.writeEmptyElement("varpat", singletonAttrs("name", v.lambda.fun->arg));
+                doc.writeEmptyElement("varpat", singletonAttrs("name", state.symbols[v.lambda.fun->arg]));
 
             break;
         }
@@ -166,14 +166,14 @@ static void printValueAsXML(EvalState & state, bool strict, bool location,
 
 void ExternalValueBase::printValueAsXML(EvalState & state, bool strict,
     bool location, XMLWriter & doc, PathSet & context, PathSet & drvsSeen,
-    const Pos & pos) const
+    const PosIdx pos) const
 {
     doc.writeEmptyElement("unevaluated");
 }
 
 
 void printValueAsXML(EvalState & state, bool strict, bool location,
-    Value & v, std::ostream & out, PathSet & context, const Pos & pos)
+    Value & v, std::ostream & out, PathSet & context, const PosIdx pos)
 {
     XMLWriter doc(true, out);
     XMLOpenElement root(doc, "expr");

src/libexpr/value-to-xml.hh

@@ -9,6 +9,6 @@
 namespace nix {
 
 void printValueAsXML(EvalState & state, bool strict, bool location,
-    Value & v, std::ostream & out, PathSet & context, const Pos & pos);
+    Value & v, std::ostream & out, PathSet & context, const PosIdx pos);
 
 }

src/libexpr/value.hh

@@ -57,6 +57,7 @@ struct Expr;
 struct ExprLambda;
 struct PrimOp;
 class Symbol;
+class PosIdx;
 struct Pos;
 class StorePath;
 class Store;
@@ -104,7 +105,7 @@ class ExternalValueBase
     /* Print the value as XML. Defaults to unevaluated */
     virtual void printValueAsXML(EvalState & state, bool strict, bool location,
         XMLWriter & doc, PathSet & context, PathSet & drvsSeen,
-        const Pos & pos) const;
+        const PosIdx pos) const;
 
     virtual ~ExternalValueBase()
     {
@@ -121,11 +122,11 @@ private:
 
     friend std::string showType(const Value & v);
 
-    void print(std::ostream & str, std::set<const void *> * seen) const;
+    void print(const SymbolTable & symbols, std::ostream & str, std::set<const void *> * seen) const;
 
 public:
 
-    void print(std::ostream & str, bool showRepeated = false) const;
+    void print(const SymbolTable & symbols, std::ostream & str, bool showRepeated = false) const;
 
     // Functions needed to distinguish the type
     // These should be removed eventually, by putting the functionality that's
@@ -366,7 +367,7 @@ public:
         return internalType == tList1 ? 1 : internalType == tList2 ? 2 : bigList.size;
     }
 
-    Pos determinePos(const Pos & pos) const;
+    PosIdx determinePos(const PosIdx pos) const;
 
     /* Check whether forcing this value requires a trivial amount of
        computation. In particular, function applications are

src/libfetchers/fetchers.cc

@@ -251,9 +251,18 @@ std::optional<std::string> Input::getRef() const
 
 std::optional<Hash> Input::getRev() const
 {
-    if (auto s = maybeGetStrAttr(attrs, "rev"))
-        return Hash::parseAny(*s, htSHA1);
-    return {};
+    std::optional<Hash> hash = {};
+
+    if (auto s = maybeGetStrAttr(attrs, "rev")) {
+        try {
+            hash = Hash::parseAnyPrefixed(*s);
+        } catch (BadHash &e) {
+            // Default to sha1 for backwards compatibility with existing flakes
+            hash = Hash::parseAny(*s, htSHA1);
+        }
+    }
+
+    return hash;
 }
 
 std::optional<uint64_t> Input::getRevCount() const

src/libfetchers/git.cc

@@ -5,9 +5,13 @@
 #include "store-api.hh"
 #include "url-parts.hh"
 #include "pathlocks.hh"
+#include "util.hh"
+#include "git.hh"
 
 #include "fetch-settings.hh"
 
+#include <regex>
+#include <string.h>
 #include <sys/time.h>
 #include <sys/wait.h>
 
@@ -15,24 +19,240 @@ using namespace std::string_literals;
 
 namespace nix::fetchers {
 
+namespace {
+
 // Explicit initial branch of our bare repo to suppress warnings from new version of git.
 // The value itself does not matter, since we always fetch a specific revision or branch.
 // It is set with `-c init.defaultBranch=` instead of `--initial-branch=` to stay compatible with
 // old version of git, which will ignore unrecognized `-c` options.
 const std::string gitInitialBranch = "__nix_dummy_branch";
 
-static std::string readHead(const Path & path)
+std::string getGitDir()
 {
-    return chomp(runProgram("git", true, { "-C", path, "rev-parse", "--abbrev-ref", "HEAD" }));
+    return getEnv("GIT_DIR").value_or(".git");
 }
 
-static bool isNotDotGitDirectory(const Path & path)
+bool isCacheFileWithinTtl(time_t now, const struct stat & st)
 {
-    static const std::regex gitDirRegex("^(?:.*/)?\\.git$");
-
-    return not std::regex_match(path, gitDirRegex);
+    return st.st_mtime + settings.tarballTtl > now;
 }
 
+bool touchCacheFile(const Path & path, time_t touch_time)
+{
+    struct timeval times[2];
+    times[0].tv_sec = touch_time;
+    times[0].tv_usec = 0;
+    times[1].tv_sec = touch_time;
+    times[1].tv_usec = 0;
+
+    return lutimes(path.c_str(), times) == 0;
+}
+
+Path getCachePath(std::string_view key)
+{
+    return getCacheDir() + "/nix/gitv3/" +
+        hashString(htSHA256, key).to_string(Base32, false);
+}
+
+// Returns the name of the HEAD branch.
+//
+// Returns the head branch name as reported by git ls-remote --symref, e.g., if
+// ls-remote returns the output below, "main" is returned based on the ref line.
+//
+//   ref: refs/heads/main       HEAD
+//   ...
+std::optional<std::string> readHead(const Path & path)
+{
+    auto [exit_code, output] = runProgram(RunOptions {
+        .program = "git",
+        // FIXME: use 'HEAD'
+        .args = {"ls-remote", "--symref", path},
+    });
+    if (exit_code != 0) {
+        return std::nullopt;
+    }
+
+    std::string_view line = output;
+    line = line.substr(0, line.find("\n"));
+    if (const auto parseResult = git::parseLsRemoteLine(line)) {
+        switch (parseResult->kind) {
+            case git::LsRemoteRefLine::Kind::Symbolic:
+                debug("resolved HEAD ref '%s' for repo '%s'", parseResult->target, path);
+                break;
+            case git::LsRemoteRefLine::Kind::Object:
+                debug("resolved HEAD rev '%s' for repo '%s'", parseResult->target, path);
+                break;
+        }
+        return parseResult->target;
+    }
+    return std::nullopt;
+}
+
+// Persist the HEAD ref from the remote repo in the local cached repo.
+bool storeCachedHead(std::string_view actualUrl, std::string headRef)
+{
+    Path cacheDir = getCachePath(actualUrl);
+    try {
+        runProgram("git", true, { "-C", cacheDir, "symbolic-ref", "--", "HEAD", headRef });
+    } catch (ExecError & e) {
+        if (!WIFEXITED(e.status)) throw;
+        return false;
+    }
+    /* No need to touch refs/HEAD, because `git symbolic-ref` updates the mtime. */
+    return true;
+}
+
+std::optional<std::string> readHeadCached(std::string_view actualUrl)
+{
+    // Create a cache path to store the branch of the HEAD ref. Append something
+    // in front of the URL to prevent collision with the repository itself.
+    Path cacheDir = getCachePath(actualUrl);
+    Path headRefFile = cacheDir + "/HEAD";
+
+    time_t now = time(0);
+    struct stat st;
+    std::optional<std::string> cachedRef;
+    if (stat(headRefFile.c_str(), &st) == 0) {
+        cachedRef = readHead(cacheDir);
+        if (cachedRef != std::nullopt &&
+            *cachedRef != gitInitialBranch &&
+            isCacheFileWithinTtl(now, st))
+        {
+            debug("using cached HEAD ref '%s' for repo '%s'", *cachedRef, actualUrl);
+            return cachedRef;
+        }
+    }
+
+    auto ref = readHead(actualUrl);
+    if (ref) return ref;
+
+    if (cachedRef) {
+        // If the cached git ref is expired in fetch() below, and the 'git fetch'
+        // fails, it falls back to continuing with the most recent version.
+        // This function must behave the same way, so we return the expired
+        // cached ref here.
+        warn("could not get HEAD ref for repository '%s'; using expired cached ref '%s'", actualUrl, *cachedRef);
+        return *cachedRef;
+    }
+
+    return std::nullopt;
+}
+
+bool isNotDotGitDirectory(const Path & path)
+{
+    return baseNameOf(path) != ".git";
+}
+
+struct WorkdirInfo
+{
+    bool clean = false;
+    bool hasHead = false;
+};
+
+// Returns whether a git workdir is clean and has commits.
+WorkdirInfo getWorkdirInfo(const Input & input, const Path & workdir)
+{
+    const bool submodules = maybeGetBoolAttr(input.attrs, "submodules").value_or(false);
+    auto gitDir = getGitDir();
+
+    auto env = getEnv();
+    // Set LC_ALL to C: because we rely on the error messages from git rev-parse to determine what went wrong
+    // that way unknown errors can lead to a failure instead of continuing through the wrong code path
+    env["LC_ALL"] = "C";
+
+    /* Check whether HEAD points to something that looks like a commit,
+       since that is the refrence we want to use later on. */
+    auto result = runProgram(RunOptions {
+        .program = "git",
+        .args = { "-C", workdir, "--git-dir", gitDir, "rev-parse", "--verify", "--no-revs", "HEAD^{commit}" },
+        .environment = env,
+        .mergeStderrToStdout = true
+    });
+    auto exitCode = WEXITSTATUS(result.first);
+    auto errorMessage = result.second;
+
+    if (errorMessage.find("fatal: not a git repository") != std::string::npos) {
+        throw Error("'%s' is not a Git repository", workdir);
+    } else if (errorMessage.find("fatal: Needed a single revision") != std::string::npos) {
+        // indicates that the repo does not have any commits
+        // we want to proceed and will consider it dirty later
+    } else if (exitCode != 0) {
+        // any other errors should lead to a failure
+        throw Error("getting the HEAD of the Git tree '%s' failed with exit code %d:\n%s", workdir, exitCode, errorMessage);
+    }
+
+    bool clean = false;
+    bool hasHead = exitCode == 0;
+
+    try {
+        if (hasHead) {
+            // Using git diff is preferrable over lower-level operations here,
+            // because its conceptually simpler and we only need the exit code anyways.
+            auto gitDiffOpts = Strings({ "-C", workdir, "diff", "HEAD", "--quiet"});
+            if (!submodules) {
+                // Changes in submodules should only make the tree dirty
+                // when those submodules will be copied as well.
+                gitDiffOpts.emplace_back("--ignore-submodules");
+            }
+            gitDiffOpts.emplace_back("--");
+            runProgram("git", true, gitDiffOpts);
+
+            clean = true;
+        }
+    } catch (ExecError & e) {
+        if (!WIFEXITED(e.status) || WEXITSTATUS(e.status) != 1) throw;
+    }
+
+    return WorkdirInfo { .clean = clean, .hasHead = hasHead };
+}
+
+std::pair<StorePath, Input> fetchFromWorkdir(ref<Store> store, Input & input, const Path & workdir, const WorkdirInfo & workdirInfo)
+{
+    const bool submodules = maybeGetBoolAttr(input.attrs, "submodules").value_or(false);
+
+    if (!fetchSettings.allowDirty)
+        throw Error("Git tree '%s' is dirty", workdir);
+
+    if (fetchSettings.warnDirty)
+        warn("Git tree '%s' is dirty", workdir);
+
+    auto gitOpts = Strings({ "-C", workdir, "ls-files", "-z" });
+    if (submodules)
+        gitOpts.emplace_back("--recurse-submodules");
+
+    auto files = tokenizeString<std::set<std::string>>(
+        runProgram("git", true, gitOpts), "\0"s);
+
+    Path actualPath(absPath(workdir));
+
+    PathFilter filter = [&](const Path & p) -> bool {
+        assert(hasPrefix(p, actualPath));
+        std::string file(p, actualPath.size() + 1);
+
+        auto st = lstat(p);
+
+        if (S_ISDIR(st.st_mode)) {
+            auto prefix = file + "/";
+            auto i = files.lower_bound(prefix);
+            return i != files.end() && hasPrefix(*i, prefix);
+        }
+
+        return files.count(file);
+    };
+
+    auto storePath = store->addToStore(input.getName(), actualPath, FileIngestionMethod::Recursive, htSHA256, filter);
+
+    // FIXME: maybe we should use the timestamp of the last
+    // modified dirty file?
+    input.attrs.insert_or_assign(
+        "lastModified",
+        workdirInfo.hasHead ? std::stoull(runProgram("git", true, { "-C", actualPath, "log", "-1", "--format=%ct", "--no-show-signature", "HEAD" })) : 0);
+
+    return {std::move(storePath), input};
+}
+
+}  // end namespace
+
 struct GitInputScheme : InputScheme
 {
     std::optional<Input> inputFromURL(const ParsedURL & url) override
@@ -50,7 +270,7 @@ struct GitInputScheme : InputScheme
         Attrs attrs;
         attrs.emplace("type", "git");
 
-        for (auto &[name, value] : url.query) {
+        for (auto & [name, value] : url.query) {
             if (name == "rev" || name == "ref")
                 attrs.emplace(name, value);
             else if (name == "shallow" || name == "submodules")
@@ -152,13 +372,14 @@ struct GitInputScheme : InputScheme
     {
         auto sourcePath = getSourcePath(input);
         assert(sourcePath);
+        auto gitDir = getGitDir();
 
         runProgram("git", true,
-            { "-C", *sourcePath, "add", "--force", "--intent-to-add", "--", std::string(file) });
+            { "-C", *sourcePath, "--git-dir", gitDir, "add", "--force", "--intent-to-add", "--", std::string(file) });
 
         if (commitMsg)
             runProgram("git", true,
-                { "-C", *sourcePath, "commit", std::string(file), "-m", *commitMsg });
+                { "-C", *sourcePath, "--git-dir", gitDir, "commit", std::string(file), "-m", *commitMsg });
     }
 
     struct RepoInfo
@@ -371,6 +592,15 @@ struct GitInputScheme : InputScheme
             repoDir = repoInfo.url;
 
         } else {
+            const bool useHeadRef = !input.getRef();
+            if (useHeadRef) {
+                auto head = readHeadCached(actualUrl);
+                if (!head) {
+                    warn("could not read HEAD ref from repo at '%s', using 'master'", actualUrl);
+                    head = "master";
+                }
+                input.attrs.insert_or_assign("ref", *head);
+            }
 
             if (auto res = getCache()->lookup(store, unlockedAttrs)) {
                 auto rev2 = Hash::parseAny(getStrAttr(res->first, "rev"), htSHA1);
@@ -382,6 +612,7 @@ struct GitInputScheme : InputScheme
 
             Path cacheDir = getCacheDir() + "/nix/gitv3/" + hashString(htSHA256, repoInfo.url).to_string(Base32, false);
             repoDir = cacheDir;
+            gitDir = ".";
 
             createDirs(dirOf(cacheDir));
             PathLocks cacheDirLock({cacheDir + ".lock"});
@@ -402,7 +633,7 @@ struct GitInputScheme : InputScheme
                repo. */
             if (input.getRev()) {
                 try {
-                    runProgram("git", true, { "-C", repoDir, "cat-file", "-e", input.getRev()->gitRev() });
+                    runProgram("git", true, { "-C", repoDir, "--git-dir", gitDir, "cat-file", "-e", input.getRev()->gitRev() });
                     doFetch = false;
                 } catch (ExecError & e) {
                     if (WIFEXITED(e.status)) {
@@ -419,7 +650,7 @@ struct GitInputScheme : InputScheme
                        git fetch to update the local ref to the remote ref. */
                     struct stat st;
                     doFetch = stat(localRefFile.c_str(), &st) != 0 ||
-                        (uint64_t) st.st_mtime + settings.tarballTtl <= (uint64_t) now;
+                        !isCacheFileWithinTtl(now, st);
                 }
             }
 
@@ -443,13 +674,10 @@ struct GitInputScheme : InputScheme
                     warn("could not update local clone of Git repository '%s'; continuing with the most recent version", repoInfo.url);
                 }
 
-                struct timeval times[2];
-                times[0].tv_sec = now;
-                times[0].tv_usec = 0;
-                times[1].tv_sec = now;
-                times[1].tv_usec = 0;
-
-                utimes(localRefFile.c_str(), times);
+                if (!touchCacheFile(localRefFile, now))
+                    warn("could not update mtime for file '%s': %s", localRefFile, strerror(errno));
+                if (useHeadRef && !storeCachedHead(actualUrl, *input.getRef()))
+                    warn("could not update cached head '%s' for '%s'", *input.getRef(), actualUrl);
             }
 
             if (!input.getRev())
@@ -458,7 +686,7 @@ struct GitInputScheme : InputScheme
             // cache dir lock is removed at scope end; we will only use read-only operations on specific revisions in the remainder
         }
 
-        bool isShallow = chomp(runProgram("git", true, { "-C", repoDir, "rev-parse", "--is-shallow-repository" })) == "true";
+        bool isShallow = chomp(runProgram("git", true, { "-C", repoDir, "--git-dir", gitDir, "rev-parse", "--is-shallow-repository" })) == "true";
 
         if (isShallow && !repoInfo.shallow)
             throw Error("'%s' is a shallow Git repository, but a non-shallow repository is needed", repoInfo.url);
@@ -478,7 +706,7 @@ struct GitInputScheme : InputScheme
 
         auto result = runProgram(RunOptions {
             .program = "git",
-            .args = { "-C", repoDir, "cat-file", "commit", input.getRev()->gitRev() },
+            .args = { "-C", repoDir, "--git-dir", gitDir, "cat-file", "commit", input.getRev()->gitRev() },
             .mergeStderrToStdout = true
         });
         if (WEXITSTATUS(result.first) == 128
@@ -517,7 +745,7 @@ struct GitInputScheme : InputScheme
             auto source = sinkToSource([&](Sink & sink) {
                 runProgram2({
                     .program = "git",
-                    .args = { "-C", repoDir, "archive", input.getRev()->gitRev() },
+                    .args = { "-C", repoDir, "--git-dir", gitDir, "archive", input.getRev()->gitRev() },
                     .standardOut = &sink
                 });
             });
@@ -527,7 +755,7 @@ struct GitInputScheme : InputScheme
 
         auto storePath = store->addToStore(name, tmpDir, FileIngestionMethod::Recursive, htSHA256, filter);
 
-        auto lastModified = std::stoull(runProgram("git", true, { "-C", repoDir, "log", "-1", "--format=%ct", "--no-show-signature", input.getRev()->gitRev() }));
+        auto lastModified = std::stoull(runProgram("git", true, { "-C", repoDir, "--git-dir", gitDir, "log", "-1", "--format=%ct", "--no-show-signature", input.getRev()->gitRev() }));
 
         Attrs infoAttrs({
             {"rev", input.getRev()->gitRev()},
@@ -536,7 +764,7 @@ struct GitInputScheme : InputScheme
 
         if (!repoInfo.shallow)
             infoAttrs.insert_or_assign("revCount",
-                std::stoull(runProgram("git", true, { "-C", repoDir, "rev-list", "--count", input.getRev()->gitRev() })));
+                std::stoull(runProgram("git", true, { "-C", repoDir, "--git-dir", gitDir, "rev-list", "--count", input.getRev()->gitRev() })));
 
         if (!_input.getRev())
             getCache()->add(

src/libfetchers/github.cc

@@ -4,7 +4,7 @@
 #include "store-api.hh"
 #include "types.hh"
 #include "url-parts.hh"
-
+#include "git.hh"
 #include "fetchers.hh"
 #include "fetch-settings.hh"
 
@@ -395,35 +395,31 @@ struct SourceHutInputScheme : GitArchiveInputScheme
             std::string line;
             getline(is, line);
 
-            auto ref_index = line.find("ref: ");
-            if (ref_index == std::string::npos) {
+            auto remoteLine = git::parseLsRemoteLine(line);
+            if (!remoteLine) {
                 throw BadURL("in '%d', couldn't resolve HEAD ref '%d'", input.to_string(), ref);
             }
-
-            ref_uri = line.substr(ref_index+5, line.length()-1);
-        } else
+            ref_uri = remoteLine->target;
+        } else {
             ref_uri = fmt("refs/(heads|tags)/%s", ref);
+        }
 
         auto file = store->toRealPath(
             downloadFile(store, fmt("%s/info/refs", base_url), "source", false, headers).storePath);
         std::ifstream is(file);
 
         std::string line;
-        std::string id;
-        while(getline(is, line)) {
-            // Append $ to avoid partial name matches
-            std::regex pattern(fmt("%s$", ref_uri));
-
-            if (std::regex_search(line, pattern)) {
-                id = line.substr(0, line.find('\t'));
-                break;
-            }
+        std::optional<std::string> id;
+        while(!id && getline(is, line)) {
+            auto parsedLine = git::parseLsRemoteLine(line);
+            if (parsedLine && parsedLine->reference == ref_uri)
+                id = parsedLine->target;
         }
 
-        if(id.empty())
+        if(!id)
             throw BadURL("in '%d', couldn't find ref '%d'", input.to_string(), ref);
 
-        auto rev = Hash::parseAny(id, htSHA1);
+        auto rev = Hash::parseAny(*id, htSHA1);
         debug("HEAD revision for '%s' is %s", fmt("%s/%s", base_url, ref), rev.gitRev());
         return rev;
     }

src/libfetchers/mercurial.cc

@@ -36,7 +36,7 @@ static std::string runHg(const Strings & args, const std::optional<std::string>
     auto res = runProgram(std::move(opts));
 
     if (!statusOk(res.first))
-        throw ExecError(res.first, fmt("hg %1%", statusToString(res.first)));
+        throw ExecError(res.first, "hg %1%", statusToString(res.first));
 
     return res.second;
 }
@@ -178,9 +178,11 @@ struct MercurialInputScheme : InputScheme
                 auto files = tokenizeString<std::set<std::string>>(
                     runHg({ "status", "-R", actualUrl, "--clean", "--modified", "--added", "--no-status", "--print0" }), "\0"s);
 
+                Path actualPath(absPath(actualUrl));
+
                 PathFilter filter = [&](const Path & p) -> bool {
-                    assert(hasPrefix(p, actualUrl));
-                    std::string file(p, actualUrl.size() + 1);
+                    assert(hasPrefix(p, actualPath));
+                    std::string file(p, actualPath.size() + 1);
 
                     auto st = lstat(p);
 
@@ -193,7 +195,7 @@ struct MercurialInputScheme : InputScheme
                     return files.count(file);
                 };
 
-                auto storePath = store->addToStore(input.getName(), actualUrl, FileIngestionMethod::Recursive, htSHA256, filter);
+                auto storePath = store->addToStore(input.getName(), actualPath, FileIngestionMethod::Recursive, htSHA256, filter);
 
                 return {std::move(storePath), input};
             }
@@ -201,8 +203,17 @@ struct MercurialInputScheme : InputScheme
 
         if (!input.getRef()) input.attrs.insert_or_assign("ref", "default");
 
+        auto checkHashType = [&](const std::optional<Hash> & hash)
+        {
+            if (hash.has_value() && hash->type != htSHA1)
+                throw Error("Hash '%s' is not supported by Mercurial. Only sha1 is supported.", hash->to_string(Base16, true));
+        };
+
+
         auto getLockedAttrs = [&]()
         {
+            checkHashType(input.getRev());
+
             return Attrs({
                 {"type", "hg"},
                 {"name", name},
@@ -262,7 +273,7 @@ struct MercurialInputScheme : InputScheme
                         runHg({ "recover", "-R", cacheDir });
                         runHg({ "pull", "-R", cacheDir, "--", actualUrl });
                     } else {
-                        throw ExecError(e.status, fmt("'hg pull' %s", statusToString(e.status)));
+                        throw ExecError(e.status, "'hg pull' %s", statusToString(e.status));
                     }
                 }
             } else {

src/libmain/shared.cc

@@ -60,37 +60,37 @@ void printMissing(ref<Store> store, const StorePathSet & willBuild,
 {
     if (!willBuild.empty()) {
         if (willBuild.size() == 1)
-            printMsg(lvl, fmt("this derivation will be built:"));
+            printMsg(lvl, "this derivation will be built:");
         else
-            printMsg(lvl, fmt("these %d derivations will be built:", willBuild.size()));
+            printMsg(lvl, "these %d derivations will be built:", willBuild.size());
         auto sorted = store->topoSortPaths(willBuild);
         reverse(sorted.begin(), sorted.end());
         for (auto & i : sorted)
-            printMsg(lvl, fmt("  %s", store->printStorePath(i)));
+            printMsg(lvl, "  %s", store->printStorePath(i));
     }
 
     if (!willSubstitute.empty()) {
         const float downloadSizeMiB = downloadSize / (1024.f * 1024.f);
         const float narSizeMiB = narSize / (1024.f * 1024.f);
         if (willSubstitute.size() == 1) {
-            printMsg(lvl, fmt("this path will be fetched (%.2f MiB download, %.2f MiB unpacked):",
+            printMsg(lvl, "this path will be fetched (%.2f MiB download, %.2f MiB unpacked):",
                 downloadSizeMiB,
-                narSizeMiB));
+                narSizeMiB);
         } else {
-            printMsg(lvl, fmt("these %d paths will be fetched (%.2f MiB download, %.2f MiB unpacked):",
+            printMsg(lvl, "these %d paths will be fetched (%.2f MiB download, %.2f MiB unpacked):",
                 willSubstitute.size(),
                 downloadSizeMiB,
-                narSizeMiB));
+                narSizeMiB);
         }
         for (auto & i : willSubstitute)
-            printMsg(lvl, fmt("  %s", store->printStorePath(i)));
+            printMsg(lvl, "  %s", store->printStorePath(i));
     }
 
     if (!unknown.empty()) {
-        printMsg(lvl, fmt("don't know how to build these paths%s:",
-                (settings.readOnlyMode ? " (may be caused by read-only store access)" : "")));
+        printMsg(lvl, "don't know how to build these paths%s:",
+                (settings.readOnlyMode ? " (may be caused by read-only store access)" : ""));
         for (auto & i : unknown)
-            printMsg(lvl, fmt("  %s", store->printStorePath(i)));
+            printMsg(lvl, "  %s", store->printStorePath(i));
     }
 }
 

src/libstore/build-result.hh

@@ -1,6 +1,7 @@
 #pragma once
 
 #include "realisation.hh"
+#include "derived-path.hh"
 
 #include <string>
 #include <chrono>
@@ -30,6 +31,8 @@ struct BuildResult
         ResolvesToAlreadyValid,
         NoSubstituters,
     } status = MiscFailure;
+
+    // FIXME: include entire ErrorInfo object.
     std::string errorMsg;
 
     std::string toString() const {

src/libstore/build/derivation-goal.cc

@@ -204,9 +204,34 @@ void DerivationGoal::haveDerivation()
 {
     trace("have derivation");
 
+    parsedDrv = std::make_unique<ParsedDerivation>(drvPath, *drv);
+
     if (!drv->type().hasKnownOutputPaths())
         settings.requireExperimentalFeature(Xp::CaDerivations);
 
+    if (!drv->type().isPure()) {
+        settings.requireExperimentalFeature(Xp::ImpureDerivations);
+
+        for (auto & [outputName, output] : drv->outputs) {
+            auto randomPath = StorePath::random(outputPathName(drv->name, outputName));
+            assert(!worker.store.isValidPath(randomPath));
+            initialOutputs.insert({
+                outputName,
+                InitialOutput {
+                    .wanted = true,
+                    .outputHash = impureOutputHash,
+                    .known = InitialOutputStatus {
+                        .path = randomPath,
+                        .status = PathStatus::Absent
+                    }
+                }
+            });
+        }
+
+        gaveUpOnSubstitution();
+        return;
+    }
+
     for (auto & i : drv->outputsAndOptPaths(worker.store))
         if (i.second.second)
             worker.store.addTempRoot(*i.second.second);
@@ -230,9 +255,6 @@ void DerivationGoal::haveDerivation()
         return;
     }
 
-    parsedDrv = std::make_unique<ParsedDerivation>(drvPath, *drv);
-
-
     /* We are first going to try to create the invalid output paths
        through substitutes.  If that doesn't work, we'll build
        them. */
@@ -266,6 +288,8 @@ void DerivationGoal::outputsSubstitutionTried()
 {
     trace("all outputs substituted (maybe)");
 
+    assert(drv->type().isPure());
+
     if (nrFailed > 0 && nrFailed > nrNoSubstituters + nrIncompleteClosure && !settings.tryFallback) {
         done(BuildResult::TransientFailure, {},
             Error("some substitutes for the outputs of derivation '%s' failed (usually happens due to networking issues); try '--fallback' to build derivation from source ",
@@ -315,9 +339,21 @@ void DerivationGoal::outputsSubstitutionTried()
 void DerivationGoal::gaveUpOnSubstitution()
 {
     /* The inputs must be built before we can build this goal. */
+    inputDrvOutputs.clear();
     if (useDerivation)
-        for (auto & i : dynamic_cast<Derivation *>(drv.get())->inputDrvs)
+        for (auto & i : dynamic_cast<Derivation *>(drv.get())->inputDrvs) {
+            /* Ensure that pure, non-fixed-output derivations don't
+               depend on impure derivations. */
+            if (drv->type().isPure() && !drv->type().isFixed()) {
+                auto inputDrv = worker.evalStore.readDerivation(i.first);
+                if (!inputDrv.type().isPure())
+                    throw Error("pure derivation '%s' depends on impure derivation '%s'",
+                        worker.store.printStorePath(drvPath),
+                        worker.store.printStorePath(i.first));
+            }
+
             addWaitee(worker.makeDerivationGoal(i.first, i.second, buildMode == bmRepair ? bmRepair : bmNormal));
+        }
 
     /* Copy the input sources from the eval store to the build
        store. */
@@ -345,6 +381,8 @@ void DerivationGoal::gaveUpOnSubstitution()
 
 void DerivationGoal::repairClosure()
 {
+    assert(drv->type().isPure());
+
     /* If we're repairing, we now know that our own outputs are valid.
        Now check whether the other paths in the outputs closure are
        good.  If not, then start derivation goals for the derivations
@@ -452,22 +490,24 @@ void DerivationGoal::inputsRealised()
                        drvs. */
                     : true);
             },
+            [&](const DerivationType::Impure &) {
+                return true;
+            }
         }, drvType.raw());
 
-        if (resolveDrv)
-        {
+        if (resolveDrv && !fullDrv.inputDrvs.empty()) {
             settings.requireExperimentalFeature(Xp::CaDerivations);
 
             /* We are be able to resolve this derivation based on the
-               now-known results of dependencies. If so, we become a stub goal
-               aliasing that resolved derivation goal */
-            std::optional attempt = fullDrv.tryResolve(worker.store);
+               now-known results of dependencies. If so, we become a
+               stub goal aliasing that resolved derivation goal. */
+            std::optional attempt = fullDrv.tryResolve(worker.store, inputDrvOutputs);
             assert(attempt);
             Derivation drvResolved { *std::move(attempt) };
 
             auto pathResolved = writeDerivation(worker.store, drvResolved);
 
-            auto msg = fmt("Resolved derivation: '%s' -> '%s'",
+            auto msg = fmt("resolved derivation: '%s' -> '%s'",
                 worker.store.printStorePath(drvPath),
                 worker.store.printStorePath(pathResolved));
             act = std::make_unique<Activity>(*logger, lvlInfo, actBuildWaiting, msg,
@@ -488,21 +528,13 @@ void DerivationGoal::inputsRealised()
             /* Add the relevant output closures of the input derivation
                `i' as input paths.  Only add the closures of output paths
                that are specified as inputs. */
-            assert(worker.evalStore.isValidPath(drvPath));
-            auto outputs = worker.evalStore.queryPartialDerivationOutputMap(depDrvPath);
-            for (auto & j : wantedDepOutputs) {
-                if (outputs.count(j) > 0) {
-                    auto optRealizedInput = outputs.at(j);
-                    if (!optRealizedInput)
-                        throw Error(
-                            "derivation '%s' requires output '%s' from input derivation '%s', which is supposedly realized already, yet we still don't know what path corresponds to that output",
-                            worker.store.printStorePath(drvPath), j, worker.store.printStorePath(depDrvPath));
-                    worker.store.computeFSClosure(*optRealizedInput, inputPaths);
-                } else
+            for (auto & j : wantedDepOutputs)
+                if (auto outPath = get(inputDrvOutputs, { depDrvPath, j }))
+                    worker.store.computeFSClosure(*outPath, inputPaths);
+                else
                     throw Error(
                         "derivation '%s' requires non-existent output '%s' from input derivation '%s'",
                         worker.store.printStorePath(drvPath), j, worker.store.printStorePath(depDrvPath));
-            }
         }
     }
 
@@ -754,8 +786,7 @@ void runPostBuildHook(
     Store & store,
     Logger & logger,
     const StorePath & drvPath,
-    StorePathSet outputPaths
-)
+    const StorePathSet & outputPaths)
 {
     auto hook = settings.postBuildHook;
     if (hook == "")
@@ -874,7 +905,7 @@ void DerivationGoal::buildDone()
         auto builtOutputs = registerOutputs();
 
         StorePathSet outputPaths;
-        for (auto & [_, output] : buildResult.builtOutputs)
+        for (auto & [_, output] : builtOutputs)
             outputPaths.insert(output.outPath);
         runPostBuildHook(
             worker.store,
@@ -923,7 +954,7 @@ void DerivationGoal::buildDone()
             st =
                 dynamic_cast<NotDeterministic*>(&e) ? BuildResult::NotDeterministic :
                 statusOk(status) ? BuildResult::OutputRejected :
-                derivationType.isImpure() || diskFull ? BuildResult::TransientFailure :
+                !derivationType.isSandboxed() || diskFull ? BuildResult::TransientFailure :
                 BuildResult::PermanentFailure;
         }
 
@@ -934,60 +965,60 @@ void DerivationGoal::buildDone()
 
 void DerivationGoal::resolvedFinished()
 {
+    trace("resolved derivation finished");
+
     assert(resolvedDrvGoal);
     auto resolvedDrv = *resolvedDrvGoal->drv;
-
-    auto resolvedHashes = staticOutputHashes(worker.store, resolvedDrv);
-
-    StorePathSet outputPaths;
-
-    // `wantedOutputs` might be empty, which means “all the outputs”
-    auto realWantedOutputs = wantedOutputs;
-    if (realWantedOutputs.empty())
-        realWantedOutputs = resolvedDrv.outputNames();
+    auto & resolvedResult = resolvedDrvGoal->buildResult;
 
     DrvOutputs builtOutputs;
 
-    for (auto & wantedOutput : realWantedOutputs) {
-        assert(initialOutputs.count(wantedOutput) != 0);
-        assert(resolvedHashes.count(wantedOutput) != 0);
-        auto realisation = worker.store.queryRealisation(
-            DrvOutput{resolvedHashes.at(wantedOutput), wantedOutput}
-        );
-        // We've just built it, but maybe the build failed, in which case the
-        // realisation won't be there
-        if (realisation) {
-            auto newRealisation = *realisation;
-            newRealisation.id = DrvOutput{initialOutputs.at(wantedOutput).outputHash, wantedOutput};
-            newRealisation.signatures.clear();
-            newRealisation.dependentRealisations = drvOutputReferences(worker.store, *drv, realisation->outPath);
-            signRealisation(newRealisation);
-            worker.store.registerDrvOutput(newRealisation);
+    if (resolvedResult.success()) {
+        auto resolvedHashes = staticOutputHashes(worker.store, resolvedDrv);
+
+        StorePathSet outputPaths;
+
+        // `wantedOutputs` might be empty, which means “all the outputs”
+        auto realWantedOutputs = wantedOutputs;
+        if (realWantedOutputs.empty())
+            realWantedOutputs = resolvedDrv.outputNames();
+
+        for (auto & wantedOutput : realWantedOutputs) {
+            auto initialOutput = get(initialOutputs, wantedOutput);
+            auto resolvedHash = get(resolvedHashes, wantedOutput);
+            if ((!initialOutput) || (!resolvedHash))
+                throw Error(
+                    "derivation '%s' doesn't have expected output '%s' (derivation-goal.cc/resolvedFinished,resolve)",
+                    worker.store.printStorePath(drvPath), wantedOutput);
+            auto realisation = get(resolvedResult.builtOutputs, DrvOutput { *resolvedHash, wantedOutput });
+            if (!realisation)
+                throw Error(
+                    "derivation '%s' doesn't have expected output '%s' (derivation-goal.cc/resolvedFinished,realisation)",
+                    worker.store.printStorePath(resolvedDrvGoal->drvPath), wantedOutput);
+            if (drv->type().isPure()) {
+                auto newRealisation = *realisation;
+                newRealisation.id = DrvOutput { initialOutput->outputHash, wantedOutput };
+                newRealisation.signatures.clear();
+                if (!drv->type().isFixed())
+                    newRealisation.dependentRealisations = drvOutputReferences(worker.store, *drv, realisation->outPath);
+                signRealisation(newRealisation);
+                worker.store.registerDrvOutput(newRealisation);
+            }
             outputPaths.insert(realisation->outPath);
             builtOutputs.emplace(realisation->id, *realisation);
-        } else {
-            // If we don't have a realisation, then it must mean that something
-            // failed when building the resolved drv
-            assert(!buildResult.success());
         }
+
+        runPostBuildHook(
+            worker.store,
+            *logger,
+            drvPath,
+            outputPaths
+        );
     }
 
-    runPostBuildHook(
-        worker.store,
-        *logger,
-        drvPath,
-        outputPaths
-    );
-
-    auto status = [&]() {
-        auto & resolvedResult = resolvedDrvGoal->buildResult;
-        switch (resolvedResult.status) {
-            case BuildResult::AlreadyValid:
-                return BuildResult::ResolvesToAlreadyValid;
-            default:
-                return resolvedResult.status;
-        }
-    }();
+    auto status = resolvedResult.status;
+    if (status == BuildResult::AlreadyValid)
+        status = BuildResult::ResolvesToAlreadyValid;
 
     done(status, std::move(builtOutputs));
 }
@@ -1236,6 +1267,7 @@ void DerivationGoal::flushLine()
 
 std::map<std::string, std::optional<StorePath>> DerivationGoal::queryPartialDerivationOutputMap()
 {
+    assert(drv->type().isPure());
     if (!useDerivation || drv->type().hasKnownOutputPaths()) {
         std::map<std::string, std::optional<StorePath>> res;
         for (auto & [name, output] : drv->outputs)
@@ -1248,6 +1280,7 @@ std::map<std::string, std::optional<StorePath>> DerivationGoal::queryPartialDeri
 
 OutputPathMap DerivationGoal::queryDerivationOutputMap()
 {
+    assert(drv->type().isPure());
     if (!useDerivation || drv->type().hasKnownOutputPaths()) {
         OutputPathMap res;
         for (auto & [name, output] : drv->outputsAndOptPaths(worker.store))
@@ -1261,12 +1294,18 @@ OutputPathMap DerivationGoal::queryDerivationOutputMap()
 
 std::pair<bool, DrvOutputs> DerivationGoal::checkPathValidity()
 {
+    if (!drv->type().isPure()) return { false, {} };
+
     bool checkHash = buildMode == bmRepair;
     auto wantedOutputsLeft = wantedOutputs;
     DrvOutputs validOutputs;
 
     for (auto & i : queryPartialDerivationOutputMap()) {
-        InitialOutput & info = initialOutputs.at(i.first);
+        auto initialOutput = get(initialOutputs, i.first);
+        if (!initialOutput)
+            // this is an invalid output, gets catched with (!wantedOutputsLeft.empty())
+            continue;
+        auto & info = *initialOutput;
         info.wanted = wantOutput(i.first, wantedOutputs);
         if (info.wanted)
             wantedOutputsLeft.erase(i.first);
@@ -1281,7 +1320,7 @@ std::pair<bool, DrvOutputs> DerivationGoal::checkPathValidity()
                     : PathStatus::Corrupt,
             };
         }
-        auto drvOutput = DrvOutput{initialOutputs.at(i.first).outputHash, i.first};
+        auto drvOutput = DrvOutput{info.outputHash, i.first};
         if (settings.isExperimentalFeatureEnabled(Xp::CaDerivations)) {
             if (auto real = worker.store.queryRealisation(drvOutput)) {
                 info.known = {
@@ -1304,6 +1343,7 @@ std::pair<bool, DrvOutputs> DerivationGoal::checkPathValidity()
         if (info.wanted && info.known && info.known->isValid())
             validOutputs.emplace(drvOutput, Realisation { drvOutput, info.known->path });
     }
+
     // If we requested all the outputs via the empty set, we are always fine.
     // If we requested specific elements, the loop above removes all the valid
     // ones, so any that are left must be invalid.
@@ -1341,9 +1381,7 @@ void DerivationGoal::done(
 {
     buildResult.status = status;
     if (ex)
-        // FIXME: strip: "error: "
-        buildResult.errorMsg = ex->what();
-    amDone(buildResult.success() ? ecSuccess : ecFailed, ex);
+        buildResult.errorMsg = fmt("%s", normaltxt(ex->info().msg));
     if (buildResult.status == BuildResult::TimedOut)
         worker.timedOut = true;
     if (buildResult.status == BuildResult::PermanentFailure)
@@ -1370,7 +1408,21 @@ void DerivationGoal::done(
         fs.open(traceBuiltOutputsFile, std::fstream::out);
         fs << worker.store.printStorePath(drvPath) << "\t" << buildResult.toString() << std::endl;
     }
+
+    amDone(buildResult.success() ? ecSuccess : ecFailed, ex);
 }
 
 
+void DerivationGoal::waiteeDone(GoalPtr waitee, ExitCode result)
+{
+    Goal::waiteeDone(waitee, result);
+
+    if (waitee->buildResult.success())
+        if (auto bfd = std::get_if<DerivedPath::Built>(&waitee->buildResult.path))
+            for (auto & [output, realisation] : waitee->buildResult.builtOutputs)
+                inputDrvOutputs.insert_or_assign(
+                    { bfd->drvPath, output.outputName },
+                    realisation.outPath);
+}
+
 }

src/libstore/build/derivation-goal.hh

@@ -57,6 +57,11 @@ struct DerivationGoal : public Goal
        them. */
     StringSet wantedOutputs;
 
+    /* Mapping from input derivations + output names to actual store
+       paths. This is filled in by waiteeDone() as each dependency
+       finishes, before inputsRealised() is reached, */
+    std::map<std::pair<StorePath, std::string>, StorePath> inputDrvOutputs;
+
     /* Whether additional wanted outputs have been added. */
     bool needRestart = false;
 
@@ -224,6 +229,8 @@ struct DerivationGoal : public Goal
         DrvOutputs builtOutputs = {},
         std::optional<Error> ex = {});
 
+    void waiteeDone(GoalPtr waitee, ExitCode result) override;
+
     StorePathSet exportReferences(const StorePathSet & storePaths);
 };
 

src/libstore/build/drv-output-substitution-goal.cc

@@ -41,7 +41,7 @@ void DrvOutputSubstitutionGoal::tryNext()
     if (subs.size() == 0) {
         /* None left.  Terminate this goal and let someone else deal
            with it. */
-        debug("drv output '%s' is required, but there is no substituter that can provide it", id.to_string());
+        debug("derivation output '%s' is required, but there is no substituter that can provide it", id.to_string());
 
         /* Hack: don't indicate failure if there were no substituters.
            In that case the calling derivation should just do a

src/libstore/build/local-derivation-goal.cc

@@ -14,6 +14,7 @@
 #include "worker-protocol.hh"
 #include "topo-sort.hh"
 #include "callback.hh"
+#include "json-utils.hh"
 
 #include <regex>
 #include <queue>
@@ -56,8 +57,6 @@
 #include <pwd.h>
 #include <grp.h>
 
-#include <nlohmann/json.hpp>
-
 namespace nix {
 
 void handleDiffHook(
@@ -395,7 +394,7 @@ void LocalDerivationGoal::startBuilder()
         else if (settings.sandboxMode == smDisabled)
             useChroot = false;
         else if (settings.sandboxMode == smRelaxed)
-            useChroot = !(derivationType.isImpure()) && !noChroot;
+            useChroot = derivationType.isSandboxed() && !noChroot;
     }
 
     auto & localStore = getLocalStore();
@@ -482,7 +481,7 @@ void LocalDerivationGoal::startBuilder()
            temporary build directory.  The text files have the format used
            by `nix-store --register-validity'.  However, the deriver
            fields are left empty. */
-        auto s = get(drv->env, "exportReferencesGraph").value_or("");
+        auto s = getOr(drv->env, "exportReferencesGraph", "");
         Strings ss = tokenizeString<Strings>(s);
         if (ss.size() % 2 != 0)
             throw BuildError("odd number of tokens in 'exportReferencesGraph': '%1%'", s);
@@ -608,7 +607,7 @@ void LocalDerivationGoal::startBuilder()
                 "nogroup:x:65534:\n", sandboxGid()));
 
         /* Create /etc/hosts with localhost entry. */
-        if (!(derivationType.isImpure()))
+        if (derivationType.isSandboxed())
             writeFile(chrootRootDir + "/etc/hosts", "127.0.0.1 localhost\n::1 localhost\n");
 
         /* Make the closure of the inputs available in the chroot,
@@ -704,6 +703,9 @@ void LocalDerivationGoal::startBuilder()
 
     /* Run the builder. */
     printMsg(lvlChatty, "executing builder '%1%'", drv->builder);
+    printMsg(lvlChatty, "using builder args '%1%'", concatStringsSep(" ", drv->args));
+    for (auto & i : drv->env)
+        printMsg(lvlVomit, "setting builder env variable '%1%'='%2%'", i.first, i.second);
 
     /* Create the log file. */
     Path logFile = openLogFile();
@@ -796,7 +798,7 @@ void LocalDerivationGoal::startBuilder()
            us.
         */
 
-        if (!(derivationType.isImpure()))
+        if (derivationType.isSandboxed())
             privateNetwork = true;
 
         userNamespaceSync.create();
@@ -986,7 +988,7 @@ void LocalDerivationGoal::initTmpDir() {
        there is no size constraint). */
     if (!parsedDrv->getStructuredAttrs()) {
 
-        StringSet passAsFile = tokenizeString<StringSet>(get(drv->env, "passAsFile").value_or(""));
+        StringSet passAsFile = tokenizeString<StringSet>(getOr(drv->env, "passAsFile", ""));
         for (auto & i : drv->env) {
             if (passAsFile.find(i.first) == passAsFile.end()) {
                 env[i.first] = i.second;
@@ -1060,7 +1062,7 @@ void LocalDerivationGoal::initEnv()
        to the builder is generally impure, but the output of
        fixed-output derivations is by definition pure (since we
        already know the cryptographic hash of the output). */
-    if (derivationType.isImpure()) {
+    if (!derivationType.isSandboxed()) {
         for (auto & i : parsedDrv->getStringsAttr("impureEnvVars").value_or(Strings()))
             env[i] = getEnv(i).value_or("");
     }
@@ -1674,7 +1676,7 @@ void LocalDerivationGoal::runChild()
             /* Fixed-output derivations typically need to access the
                network, so give them access to /etc/resolv.conf and so
                on. */
-            if (derivationType.isImpure()) {
+            if (!derivationType.isSandboxed()) {
                 // Only use nss functions to resolve hosts and
                 // services. Don’t use it for anything else that may
                 // be configured for this system. This limits the
@@ -1918,7 +1920,7 @@ void LocalDerivationGoal::runChild()
 
                 sandboxProfile += "(import \"sandbox-defaults.sb\")\n";
 
-                if (derivationType.isImpure())
+                if (!derivationType.isSandboxed())
                     sandboxProfile += "(import \"sandbox-network.sb\")\n";
 
                 /* Add the output paths we'll use at build-time to the chroot */
@@ -2125,12 +2127,22 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
     std::map<std::string, std::variant<AlreadyRegistered, PerhapsNeedToRegister>> outputReferencesIfUnregistered;
     std::map<std::string, struct stat> outputStats;
     for (auto & [outputName, _] : drv->outputs) {
-        auto actualPath = toRealPathChroot(worker.store.printStorePath(scratchOutputs.at(outputName)));
+        auto scratchOutput = get(scratchOutputs, outputName);
+        if (!scratchOutput)
+            throw BuildError(
+                "builder for '%s' has no scratch output for '%s'",
+                worker.store.printStorePath(drvPath), outputName);
+        auto actualPath = toRealPathChroot(worker.store.printStorePath(*scratchOutput));
 
         outputsToSort.insert(outputName);
 
         /* Updated wanted info to remove the outputs we definitely don't need to register */
-        auto & initialInfo = initialOutputs.at(outputName);
+        auto initialOutput = get(initialOutputs, outputName);
+        if (!initialOutput)
+            throw BuildError(
+                "builder for '%s' has no initial output for '%s'",
+                worker.store.printStorePath(drvPath), outputName);
+        auto & initialInfo = *initialOutput;
 
         /* Don't register if already valid, and not checking */
         initialInfo.wanted = buildMode == bmCheck
@@ -2182,6 +2194,11 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
 
     auto sortedOutputNames = topoSort(outputsToSort,
         {[&](const std::string & name) {
+            auto orifu = get(outputReferencesIfUnregistered, name);
+            if (!orifu)
+                throw BuildError(
+                    "no output reference for '%s' in build of '%s'",
+                    name, worker.store.printStorePath(drvPath));
             return std::visit(overloaded {
                 /* Since we'll use the already installed versions of these, we
                    can treat them as leaves and ignore any references they
@@ -2196,7 +2213,7 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
                                 referencedOutputs.insert(o);
                     return referencedOutputs;
                 },
-            }, outputReferencesIfUnregistered.at(name));
+            }, *orifu);
         }},
         {[&](const std::string & path, const std::string & parent) {
             // TODO with more -vvvv also show the temporary paths for manual inspection.
@@ -2210,9 +2227,10 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
     OutputPathMap finalOutputs;
 
     for (auto & outputName : sortedOutputNames) {
-        auto output = drv->outputs.at(outputName);
-        auto & scratchPath = scratchOutputs.at(outputName);
-        auto actualPath = toRealPathChroot(worker.store.printStorePath(scratchPath));
+        auto output = get(drv->outputs, outputName);
+        auto scratchPath = get(scratchOutputs, outputName);
+        assert(output && scratchPath);
+        auto actualPath = toRealPathChroot(worker.store.printStorePath(*scratchPath));
 
         auto finish = [&](StorePath finalStorePath) {
             /* Store the final path */
@@ -2220,10 +2238,13 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
             /* The rewrite rule will be used in downstream outputs that refer to
                use. This is why the topological sort is essential to do first
                before this for loop. */
-            if (scratchPath != finalStorePath)
-                outputRewrites[std::string { scratchPath.hashPart() }] = std::string { finalStorePath.hashPart() };
+            if (*scratchPath != finalStorePath)
+                outputRewrites[std::string { scratchPath->hashPart() }] = std::string { finalStorePath.hashPart() };
         };
 
+        auto orifu = get(outputReferencesIfUnregistered, outputName);
+        assert(orifu);
+
         std::optional<StorePathSet> referencesOpt = std::visit(overloaded {
             [&](const AlreadyRegistered & skippedFinalPath) -> std::optional<StorePathSet> {
                 finish(skippedFinalPath.path);
@@ -2232,7 +2253,7 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
             [&](const PerhapsNeedToRegister & r) -> std::optional<StorePathSet> {
                 return r.refs;
             },
-        }, outputReferencesIfUnregistered.at(outputName));
+        }, *orifu);
 
         if (!referencesOpt)
             continue;
@@ -2265,25 +2286,29 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
             for (auto & r : references) {
                 auto name = r.name();
                 auto origHash = std::string { r.hashPart() };
-                if (r == scratchPath)
+                if (r == *scratchPath) {
                     res.first = true;
-                else if (outputRewrites.count(origHash) == 0)
-                    res.second.insert(r);
-                else {
-                    std::string newRef = outputRewrites.at(origHash);
+                } else if (auto outputRewrite = get(outputRewrites, origHash)) {
+                    std::string newRef = *outputRewrite;
                     newRef += '-';
                     newRef += name;
                     res.second.insert(StorePath { newRef });
+                } else {
+                    res.second.insert(r);
                 }
             }
             return res;
         };
 
         auto newInfoFromCA = [&](const DerivationOutput::CAFloating outputHash) -> ValidPathInfo {
-            auto & st = outputStats.at(outputName);
+            auto st = get(outputStats, outputName);
+            if (!st)
+                throw BuildError(
+                    "output path %1% without valid stats info",
+                    actualPath);
             if (outputHash.method == FileIngestionMethod::Flat) {
                 /* The output path should be a regular file without execute permission. */
-                if (!S_ISREG(st.st_mode) || (st.st_mode & S_IXUSR) != 0)
+                if (!S_ISREG(st->st_mode) || (st->st_mode & S_IXUSR) != 0)
                     throw BuildError(
                         "output path '%1%' should be a non-executable regular file "
                         "since recursive hashing is not enabled (outputHashMode=flat)",
@@ -2291,7 +2316,7 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
             }
             rewriteOutput();
             /* FIXME optimize and deduplicate with addToStore */
-            std::string oldHashPart { scratchPath.hashPart() };
+            std::string oldHashPart { scratchPath->hashPart() };
             HashModuloSink caSink { outputHash.hashType, oldHashPart };
             switch (outputHash.method) {
             case FileIngestionMethod::Recursive:
@@ -2310,7 +2335,7 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
                     outputPathName(drv->name, outputName),
                     refs.second,
                     refs.first);
-            if (scratchPath != finalPath) {
+            if (*scratchPath != finalPath) {
                 // Also rewrite the output path
                 auto source = sinkToSource([&](Sink & nextSink) {
                     StringSink sink;
@@ -2351,9 +2376,9 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
                 auto requiredFinalPath = output.path;
                 /* Preemptively add rewrite rule for final hash, as that is
                    what the NAR hash will use rather than normalized-self references */
-                if (scratchPath != requiredFinalPath)
+                if (*scratchPath != requiredFinalPath)
                     outputRewrites.insert_or_assign(
-                        std::string { scratchPath.hashPart() },
+                        std::string { scratchPath->hashPart() },
                         std::string { requiredFinalPath.hashPart() });
                 rewriteOutput();
                 auto narHashAndSize = hashPath(htSHA256, actualPath);
@@ -2399,7 +2424,14 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
                 assert(false);
             },
 
-        }, output.raw());
+            [&](const DerivationOutput::Impure & doi) {
+                return newInfoFromCA(DerivationOutput::CAFloating {
+                    .method = doi.method,
+                    .hashType = doi.hashType,
+                });
+            },
+
+        }, output->raw());
 
         /* FIXME: set proper permissions in restorePath() so
             we don't have to do another traversal. */
@@ -2415,7 +2447,7 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
            derivations. */
         PathLocks dynamicOutputLock;
         dynamicOutputLock.setDeletion(true);
-        auto optFixedPath = output.path(worker.store, drv->name, outputName);
+        auto optFixedPath = output->path(worker.store, drv->name, outputName);
         if (!optFixedPath ||
             worker.store.printStorePath(*optFixedPath) != finalDestPath)
         {
@@ -2481,11 +2513,10 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
 
         /* For debugging, print out the referenced and unreferenced paths. */
         for (auto & i : inputPaths) {
-            auto j = references.find(i);
-            if (j == references.end())
-                debug("unreferenced input: '%1%'", worker.store.printStorePath(i));
-            else
+            if (references.count(i))
                 debug("referenced input: '%1%'", worker.store.printStorePath(i));
+            else
+                debug("unreferenced input: '%1%'", worker.store.printStorePath(i));
         }
 
         if (curRound == nrRounds) {
@@ -2602,14 +2633,18 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
     DrvOutputs builtOutputs;
 
     for (auto & [outputName, newInfo] : infos) {
+        auto oldinfo = get(initialOutputs, outputName);
+        assert(oldinfo);
         auto thisRealisation = Realisation {
             .id = DrvOutput {
-                initialOutputs.at(outputName).outputHash,
+                oldinfo->outputHash,
                 outputName
             },
             .outPath = newInfo.path
         };
-        if (settings.isExperimentalFeatureEnabled(Xp::CaDerivations)) {
+        if (settings.isExperimentalFeatureEnabled(Xp::CaDerivations)
+            && drv->type().isPure())
+        {
             signRealisation(thisRealisation);
             worker.store.registerDrvOutput(thisRealisation);
         }
@@ -2698,9 +2733,10 @@ void LocalDerivationGoal::checkOutputs(const std::map<std::string, ValidPathInfo
                 for (auto & i : *value) {
                     if (worker.store.isStorePath(i))
                         spec.insert(worker.store.parseStorePath(i));
-                    else if (outputs.count(i))
-                        spec.insert(outputs.at(i).path);
-                    else throw BuildError("derivation contains an illegal reference specifier '%s'", i);
+                    else if (auto output = get(outputs, i))
+                        spec.insert(output->path);
+                    else
+                        throw BuildError("derivation contains an illegal reference specifier '%s'", i);
                 }
 
                 auto used = recursive
@@ -2739,24 +2775,18 @@ void LocalDerivationGoal::checkOutputs(const std::map<std::string, ValidPathInfo
         };
 
         if (auto structuredAttrs = parsedDrv->getStructuredAttrs()) {
-            auto outputChecks = structuredAttrs->find("outputChecks");
-            if (outputChecks != structuredAttrs->end()) {
-                auto output = outputChecks->find(outputName);
-
-                if (output != outputChecks->end()) {
+            if (auto outputChecks = get(*structuredAttrs, "outputChecks")) {
+                if (auto output = get(*outputChecks, outputName)) {
                     Checks checks;
 
-                    auto maxSize = output->find("maxSize");
-                    if (maxSize != output->end())
+                    if (auto maxSize = get(*output, "maxSize"))
                         checks.maxSize = maxSize->get<uint64_t>();
 
-                    auto maxClosureSize = output->find("maxClosureSize");
-                    if (maxClosureSize != output->end())
+                    if (auto maxClosureSize = get(*output, "maxClosureSize"))
                         checks.maxClosureSize = maxClosureSize->get<uint64_t>();
 
-                    auto get = [&](const std::string & name) -> std::optional<Strings> {
-                        auto i = output->find(name);
-                        if (i != output->end()) {
+                    auto get_ = [&](const std::string & name) -> std::optional<Strings> {
+                        if (auto i = get(*output, name)) {
                             Strings res;
                             for (auto j = i->begin(); j != i->end(); ++j) {
                                 if (!j->is_string())
@@ -2769,10 +2799,10 @@ void LocalDerivationGoal::checkOutputs(const std::map<std::string, ValidPathInfo
                         return {};
                     };
 
-                    checks.allowedReferences = get("allowedReferences");
-                    checks.allowedRequisites = get("allowedRequisites");
-                    checks.disallowedReferences = get("disallowedReferences");
-                    checks.disallowedRequisites = get("disallowedRequisites");
+                    checks.allowedReferences = get_("allowedReferences");
+                    checks.allowedRequisites = get_("allowedRequisites");
+                    checks.disallowedReferences = get_("disallowedReferences");
+                    checks.disallowedRequisites = get_("disallowedRequisites");
 
                     applyChecks(checks);
                 }

src/libstore/build/substitution-goal.cc

@@ -24,9 +24,16 @@ PathSubstitutionGoal::~PathSubstitutionGoal()
 }
 
 
-void PathSubstitutionGoal::done(ExitCode result, BuildResult::Status status)
+void PathSubstitutionGoal::done(
+    ExitCode result,
+    BuildResult::Status status,
+    std::optional<std::string> errorMsg)
 {
     buildResult.status = status;
+    if (errorMsg) {
+        debug(*errorMsg);
+        buildResult.errorMsg = *errorMsg;
+    }
     amDone(result);
 }
 
@@ -67,12 +74,14 @@ void PathSubstitutionGoal::tryNext()
     if (subs.size() == 0) {
         /* None left.  Terminate this goal and let someone else deal
            with it. */
-        debug("path '%s' is required, but there is no substituter that can build it", worker.store.printStorePath(storePath));
 
         /* Hack: don't indicate failure if there were no substituters.
            In that case the calling derivation should just do a
            build. */
-        done(substituterFailed ? ecFailed : ecNoSubstituters, BuildResult::NoSubstituters);
+        done(
+            substituterFailed ? ecFailed : ecNoSubstituters,
+            BuildResult::NoSubstituters,
+            fmt("path '%s' is required, but there is no substituter that can build it", worker.store.printStorePath(storePath)));
 
         if (substituterFailed) {
             worker.failedSubstitutions++;
@@ -169,10 +178,10 @@ void PathSubstitutionGoal::referencesValid()
     trace("all references realised");
 
     if (nrFailed > 0) {
-        debug("some references of path '%s' could not be realised", worker.store.printStorePath(storePath));
         done(
             nrNoSubstituters > 0 || nrIncompleteClosure > 0 ? ecIncompleteClosure : ecFailed,
-            BuildResult::DependencyFailed);
+            BuildResult::DependencyFailed,
+            fmt("some references of path '%s' could not be realised", worker.store.printStorePath(storePath)));
         return;
     }
 

src/libstore/build/substitution-goal.hh

@@ -53,7 +53,10 @@ struct PathSubstitutionGoal : public Goal
     /* Content address for recomputing store path */
     std::optional<ContentAddress> ca;
 
-    void done(ExitCode result, BuildResult::Status status);
+    void done(
+        ExitCode result,
+        BuildResult::Status status,
+        std::optional<std::string> errorMsg = {});
 
 public:
     PathSubstitutionGoal(const StorePath & storePath, Worker & worker, RepairFlag repair = NoRepair, std::optional<ContentAddress> ca = std::nullopt);

src/libstore/build/worker.cc

@@ -350,7 +350,7 @@ void Worker::waitForInput()
        become `available'.  Note that `available' (i.e., non-blocking)
        includes EOF. */
     std::vector<struct pollfd> pollStatus;
-    std::map <int, int> fdToPollStatus;
+    std::map<int, size_t> fdToPollStatus;
     for (auto & i : children) {
         for (auto & j : i.fds) {
             pollStatus.push_back((struct pollfd) { .fd = j, .events = POLLIN });
@@ -380,7 +380,10 @@ void Worker::waitForInput()
         std::set<int> fds2(j->fds);
         std::vector<unsigned char> buffer(4096);
         for (auto & k : fds2) {
-            if (pollStatus.at(fdToPollStatus.at(k)).revents) {
+            const auto fdPollStatusId = get(fdToPollStatus, k);
+            assert(fdPollStatusId);
+            assert(*fdPollStatusId < pollStatus.size());
+            if (pollStatus.at(*fdPollStatusId).revents) {
                 ssize_t rd = ::read(k, buffer.data(), buffer.size());
                 // FIXME: is there a cleaner way to handle pt close
                 // than EIO? Is this even standard?

src/libstore/builtins/fetchurl.cc

@@ -24,7 +24,7 @@ void builtinFetchurl(const BasicDerivation & drv, const std::string & netrcData)
 
     Path storePath = getAttr("out");
     auto mainUrl = getAttr("url");
-    bool unpack = get(drv.env, "unpack").value_or("") == "1";
+    bool unpack = getOr(drv.env, "unpack", "") == "1";
 
     /* Note: have to use a fresh fileTransfer here because we're in
        a forked process. */

src/libstore/ca-specific-schema.sql

@@ -13,12 +13,27 @@ create table if not exists Realisations (
 
 create index if not exists IndexRealisations on Realisations(drvPath, outputName);
 
+-- We can end-up in a weird edge-case where a path depends on itself because
+-- it’s an output of a CA derivation, that happens to be the same as one of its
+-- dependencies.
+-- In that case we have a dependency loop (path -> realisation1 -> realisation2
+-- -> path) that we need to break by removing the dependencies between the
+-- realisations
+create trigger if not exists DeleteSelfRefsViaRealisations before delete on ValidPaths
+  begin
+    delete from RealisationsRefs where realisationReference in (
+      select id from Realisations where outputPath = old.id
+    );
+  end;
+
 create table if not exists RealisationsRefs (
     referrer integer not null,
     realisationReference integer,
     foreign key (referrer) references Realisations(id) on delete cascade,
     foreign key (realisationReference) references Realisations(id) on delete restrict
 );
+-- used by deletion trigger
+create index if not exists IndexRealisationsRefsRealisationReference on RealisationsRefs(realisationReference);
 
 -- used by QueryRealisationReferences
 create index if not exists IndexRealisationsRefs on RealisationsRefs(referrer);

src/libstore/derivations.cc

@@ -25,26 +25,42 @@ std::optional<StorePath> DerivationOutput::path(const Store & store, std::string
         [](const DerivationOutput::Deferred &) -> std::optional<StorePath> {
             return std::nullopt;
         },
+        [](const DerivationOutput::Impure &) -> std::optional<StorePath> {
+            return std::nullopt;
+        },
     }, raw());
 }
 
 
-StorePath DerivationOutput::CAFixed::path(const Store & store, std::string_view drvName, std::string_view outputName) const {
+StorePath DerivationOutput::CAFixed::path(const Store & store, std::string_view drvName, std::string_view outputName) const
+{
     return store.makeFixedOutputPath(
         hash.method, hash.hash,
         outputPathName(drvName, outputName));
 }
 
 
-bool DerivationType::isCA() const {
+bool DerivationType::isCA() const
+{
     /* Normally we do the full `std::visit` to make sure we have
        exhaustively handled all variants, but so long as there is a
        variant called `ContentAddressed`, it must be the only one for
        which `isCA` is true for this to make sense!. */
-    return std::holds_alternative<ContentAddressed>(raw());
+    return std::visit(overloaded {
+        [](const InputAddressed & ia) {
+            return false;
+        },
+        [](const ContentAddressed & ca) {
+            return true;
+        },
+        [](const Impure &) {
+            return true;
+        },
+    }, raw());
 }
 
-bool DerivationType::isFixed() const {
+bool DerivationType::isFixed() const
+{
     return std::visit(overloaded {
         [](const InputAddressed & ia) {
             return false;
@@ -52,10 +68,14 @@ bool DerivationType::isFixed() const {
         [](const ContentAddressed & ca) {
             return ca.fixed;
         },
+        [](const Impure &) {
+            return false;
+        },
     }, raw());
 }
 
-bool DerivationType::hasKnownOutputPaths() const {
+bool DerivationType::hasKnownOutputPaths() const
+{
     return std::visit(overloaded {
         [](const InputAddressed & ia) {
             return !ia.deferred;
@@ -63,17 +83,40 @@ bool DerivationType::hasKnownOutputPaths() const {
         [](const ContentAddressed & ca) {
             return ca.fixed;
         },
+        [](const Impure &) {
+            return false;
+        },
     }, raw());
 }
 
 
-bool DerivationType::isImpure() const {
+bool DerivationType::isSandboxed() const
+{
     return std::visit(overloaded {
         [](const InputAddressed & ia) {
-            return false;
+            return true;
         },
         [](const ContentAddressed & ca) {
-            return !ca.pure;
+            return ca.sandboxed;
+        },
+        [](const Impure &) {
+            return false;
+        },
+    }, raw());
+}
+
+
+bool DerivationType::isPure() const
+{
+    return std::visit(overloaded {
+        [](const InputAddressed & ia) {
+            return true;
+        },
+        [](const ContentAddressed & ca) {
+            return true;
+        },
+        [](const Impure &) {
+            return false;
         },
     }, raw());
 }
@@ -176,7 +219,14 @@ static DerivationOutput parseDerivationOutput(const Store & store,
             hashAlgo = hashAlgo.substr(2);
         }
         const auto hashType = parseHashType(hashAlgo);
-        if (hash != "") {
+        if (hash == "impure") {
+            settings.requireExperimentalFeature(Xp::ImpureDerivations);
+            assert(pathS == "");
+            return DerivationOutput::Impure {
+                .method = std::move(method),
+                .hashType = std::move(hashType),
+            };
+        } else if (hash != "") {
             validatePath(pathS);
             return DerivationOutput::CAFixed {
                 .hash = FixedOutputHash {
@@ -345,6 +395,12 @@ std::string Derivation::unparse(const Store & store, bool maskOutputs,
                 s += ','; printUnquotedString(s, "");
                 s += ','; printUnquotedString(s, "");
                 s += ','; printUnquotedString(s, "");
+            },
+            [&](const DerivationOutputImpure & doi) {
+                // FIXME
+                s += ','; printUnquotedString(s, "");
+                s += ','; printUnquotedString(s, makeFileIngestionPrefix(doi.method) + printHashType(doi.hashType));
+                s += ','; printUnquotedString(s, "impure");
             }
         }, i.second.raw());
         s += ')';
@@ -410,8 +466,14 @@ std::string outputPathName(std::string_view drvName, std::string_view outputName
 
 DerivationType BasicDerivation::type() const
 {
-    std::set<std::string_view> inputAddressedOutputs, fixedCAOutputs, floatingCAOutputs, deferredIAOutputs;
+    std::set<std::string_view>
+        inputAddressedOutputs,
+        fixedCAOutputs,
+        floatingCAOutputs,
+        deferredIAOutputs,
+        impureOutputs;
     std::optional<HashType> floatingHashType;
+
     for (auto & i : outputs) {
         std::visit(overloaded {
             [&](const DerivationOutput::InputAddressed &) {
@@ -426,43 +488,78 @@ DerivationType BasicDerivation::type() const
                     floatingHashType = dof.hashType;
                 } else {
                     if (*floatingHashType != dof.hashType)
-                        throw Error("All floating outputs must use the same hash type");
+                        throw Error("all floating outputs must use the same hash type");
                 }
             },
             [&](const DerivationOutput::Deferred &) {
-               deferredIAOutputs.insert(i.first);
+                deferredIAOutputs.insert(i.first);
+            },
+            [&](const DerivationOutput::Impure &) {
+                impureOutputs.insert(i.first);
             },
         }, i.second.raw());
     }
 
-    if (inputAddressedOutputs.empty() && fixedCAOutputs.empty() && floatingCAOutputs.empty() && deferredIAOutputs.empty()) {
-        throw Error("Must have at least one output");
-    } else if (! inputAddressedOutputs.empty() && fixedCAOutputs.empty() && floatingCAOutputs.empty() && deferredIAOutputs.empty()) {
+    if (inputAddressedOutputs.empty()
+        && fixedCAOutputs.empty()
+        && floatingCAOutputs.empty()
+        && deferredIAOutputs.empty()
+        && impureOutputs.empty())
+        throw Error("must have at least one output");
+
+    if (!inputAddressedOutputs.empty()
+        && fixedCAOutputs.empty()
+        && floatingCAOutputs.empty()
+        && deferredIAOutputs.empty()
+        && impureOutputs.empty())
         return DerivationType::InputAddressed {
             .deferred = false,
         };
-    } else if (inputAddressedOutputs.empty() && ! fixedCAOutputs.empty() && floatingCAOutputs.empty() && deferredIAOutputs.empty()) {
+
+    if (inputAddressedOutputs.empty()
+        && !fixedCAOutputs.empty()
+        && floatingCAOutputs.empty()
+        && deferredIAOutputs.empty()
+        && impureOutputs.empty())
+    {
         if (fixedCAOutputs.size() > 1)
             // FIXME: Experimental feature?
-            throw Error("Only one fixed output is allowed for now");
+            throw Error("only one fixed output is allowed for now");
         if (*fixedCAOutputs.begin() != "out")
-            throw Error("Single fixed output must be named \"out\"");
+            throw Error("single fixed output must be named \"out\"");
         return DerivationType::ContentAddressed {
-            .pure = false,
+            .sandboxed = false,
             .fixed = true,
         };
-    } else if (inputAddressedOutputs.empty() && fixedCAOutputs.empty() && ! floatingCAOutputs.empty() && deferredIAOutputs.empty()) {
+    }
+
+    if (inputAddressedOutputs.empty()
+        && fixedCAOutputs.empty()
+        && !floatingCAOutputs.empty()
+        && deferredIAOutputs.empty()
+        && impureOutputs.empty())
         return DerivationType::ContentAddressed {
-            .pure = true,
+            .sandboxed = true,
             .fixed = false,
         };
-    } else if (inputAddressedOutputs.empty() && fixedCAOutputs.empty() && floatingCAOutputs.empty() && !deferredIAOutputs.empty()) {
+
+    if (inputAddressedOutputs.empty()
+        && fixedCAOutputs.empty()
+        && floatingCAOutputs.empty()
+        && !deferredIAOutputs.empty()
+        && impureOutputs.empty())
         return DerivationType::InputAddressed {
             .deferred = true,
         };
-    } else {
-        throw Error("Can't mix derivation output types");
-    }
+
+    if (inputAddressedOutputs.empty()
+        && fixedCAOutputs.empty()
+        && floatingCAOutputs.empty()
+        && deferredIAOutputs.empty()
+        && !impureOutputs.empty())
+        return DerivationType::Impure { };
+
+    throw Error("can't mix derivation output types");
 }
 
 
@@ -474,7 +571,7 @@ Sync<DrvHashes> drvHashes;
 /* Look up the derivation by value and memoize the
    `hashDerivationModulo` call.
  */
-static const DrvHashModulo pathDerivationModulo(Store & store, const StorePath & drvPath)
+static const DrvHash pathDerivationModulo(Store & store, const StorePath & drvPath)
 {
     {
         auto hashes = drvHashes.lock();
@@ -509,7 +606,7 @@ static const DrvHashModulo pathDerivationModulo(Store & store, const StorePath &
    don't leak the provenance of fixed outputs, reducing pointless cache
    misses as the build itself won't know this.
  */
-DrvHashModulo hashDerivationModulo(Store & store, const Derivation & drv, bool maskOutputs)
+DrvHash hashDerivationModulo(Store & store, const Derivation & drv, bool maskOutputs)
 {
     auto type = drv.type();
 
@@ -524,7 +621,20 @@ DrvHashModulo hashDerivationModulo(Store & store, const Derivation & drv, bool m
                 + store.printStorePath(dof.path(store, drv.name, i.first)));
             outputHashes.insert_or_assign(i.first, std::move(hash));
         }
-        return outputHashes;
+        return DrvHash {
+            .hashes = outputHashes,
+            .kind = DrvHash::Kind::Regular,
+        };
+    }
+
+    if (!type.isPure()) {
+        std::map<std::string, Hash> outputHashes;
+        for (const auto & [outputName, _] : drv.outputs)
+            outputHashes.insert_or_assign(outputName, impureOutputHash);
+        return DrvHash {
+            .hashes = outputHashes,
+            .kind = DrvHash::Kind::Deferred,
+        };
     }
 
     auto kind = std::visit(overloaded {
@@ -538,67 +648,43 @@ DrvHashModulo hashDerivationModulo(Store & store, const Derivation & drv, bool m
                 ? DrvHash::Kind::Regular
                 : DrvHash::Kind::Deferred;
         },
+        [](const DerivationType::Impure &) -> DrvHash::Kind {
+            assert(false);
+        }
     }, drv.type().raw());
 
-    /* For other derivations, replace the inputs paths with recursive
-       calls to this function. */
     std::map<std::string, StringSet> inputs2;
     for (auto & [drvPath, inputOutputs0] : drv.inputDrvs) {
         // Avoid lambda capture restriction with standard / Clang
         auto & inputOutputs = inputOutputs0;
         const auto & res = pathDerivationModulo(store, drvPath);
-        std::visit(overloaded {
-            // Regular non-CA derivation, replace derivation
-            [&](const DrvHash & drvHash) {
-                kind |= drvHash.kind;
-                inputs2.insert_or_assign(drvHash.hash.to_string(Base16, false), inputOutputs);
-            },
-            // CA derivation's output hashes
-            [&](const CaOutputHashes & outputHashes) {
-                std::set<std::string> justOut = { "out" };
-                for (auto & output : inputOutputs) {
-                    /* Put each one in with a single "out" output.. */
-                    const auto h = outputHashes.at(output);
-                    inputs2.insert_or_assign(
-                        h.to_string(Base16, false),
-                        justOut);
-                }
-            },
-        }, res.raw());
+        if (res.kind == DrvHash::Kind::Deferred)
+            kind = DrvHash::Kind::Deferred;
+        for (auto & outputName : inputOutputs) {
+            const auto h = get(res.hashes, outputName);
+            if (!h)
+                throw Error("no hash for output '%s' of derivation '%s'", outputName, drv.name);
+            inputs2[h->to_string(Base16, false)].insert(outputName);
+        }
     }
 
     auto hash = hashString(htSHA256, drv.unparse(store, maskOutputs, &inputs2));
 
-    return DrvHash { .hash = hash, .kind = kind };
-}
-
-
-void operator |= (DrvHash::Kind & self, const DrvHash::Kind & other) noexcept
-{
-    switch (other) {
-    case DrvHash::Kind::Regular:
-        break;
-    case DrvHash::Kind::Deferred:
-        self = other;
-        break;
+    std::map<std::string, Hash> outputHashes;
+    for (const auto & [outputName, _] : drv.outputs) {
+        outputHashes.insert_or_assign(outputName, hash);
     }
+
+    return DrvHash {
+        .hashes = outputHashes,
+        .kind = kind,
+    };
 }
 
 
 std::map<std::string, Hash> staticOutputHashes(Store & store, const Derivation & drv)
 {
-    std::map<std::string, Hash> res;
-    std::visit(overloaded {
-        [&](const DrvHash & drvHash) {
-            for (auto & outputName : drv.outputNames()) {
-                res.insert({outputName, drvHash.hash});
-            }
-        },
-        [&](const CaOutputHashes & outputHashes) {
-            res = outputHashes;
-        },
-    }, hashDerivationModulo(store, drv, true).raw());
-    return res;
+    return hashDerivationModulo(store, drv, true).hashes;
 }
 
 
@@ -625,7 +711,8 @@ StringSet BasicDerivation::outputNames() const
     return names;
 }
 
-DerivationOutputsAndOptPaths BasicDerivation::outputsAndOptPaths(const Store & store) const {
+DerivationOutputsAndOptPaths BasicDerivation::outputsAndOptPaths(const Store & store) const
+{
     DerivationOutputsAndOptPaths outsAndOptPaths;
     for (auto output : outputs)
         outsAndOptPaths.insert(std::make_pair(
@@ -636,7 +723,8 @@ DerivationOutputsAndOptPaths BasicDerivation::outputsAndOptPaths(const Store & s
     return outsAndOptPaths;
 }
 
-std::string_view BasicDerivation::nameFromPath(const StorePath & drvPath) {
+std::string_view BasicDerivation::nameFromPath(const StorePath & drvPath)
+{
     auto nameWithSuffix = drvPath.name();
     constexpr std::string_view extension = ".drv";
     assert(hasSuffix(nameWithSuffix, extension));
@@ -698,6 +786,11 @@ void writeDerivation(Sink & out, const Store & store, const BasicDerivation & dr
                     << ""
                     << "";
             },
+            [&](const DerivationOutput::Impure & doi) {
+                out << ""
+                    << (makeFileIngestionPrefix(doi.method) + printHashType(doi.hashType))
+                    << "impure";
+            },
         }, i.second.raw());
     }
     worker_proto::write(store, out, drv.inputSrcs);
@@ -723,21 +816,19 @@ std::string downstreamPlaceholder(const Store & store, const StorePath & drvPath
 }
 
 
-static void rewriteDerivation(Store & store, BasicDerivation & drv, const StringMap & rewrites) {
-
-    debug("Rewriting the derivation");
-
-    for (auto &rewrite: rewrites) {
+static void rewriteDerivation(Store & store, BasicDerivation & drv, const StringMap & rewrites)
+{
+    for (auto & rewrite : rewrites) {
         debug("rewriting %s as %s", rewrite.first, rewrite.second);
     }
 
     drv.builder = rewriteStrings(drv.builder, rewrites);
-    for (auto & arg: drv.args) {
+    for (auto & arg : drv.args) {
         arg = rewriteStrings(arg, rewrites);
     }
 
     StringPairs newEnv;
-    for (auto & envVar: drv.env) {
+    for (auto & envVar : drv.env) {
         auto envName = rewriteStrings(envVar.first, rewrites);
         auto envValue = rewriteStrings(envVar.second, rewrites);
         newEnv.emplace(envName, envValue);
@@ -747,8 +838,11 @@ static void rewriteDerivation(Store & store, BasicDerivation & drv, const String
     auto hashModulo = hashDerivationModulo(store, Derivation(drv), true);
     for (auto & [outputName, output] : drv.outputs) {
         if (std::holds_alternative<DerivationOutput::Deferred>(output.raw())) {
-            auto & h = hashModulo.requireNoFixedNonDeferred();
-            auto outPath = store.makeOutputPath(outputName, h, drv.name);
+            auto h = get(hashModulo.hashes, outputName);
+            if (!h)
+                throw Error("derivation '%s' output '%s' has no hash (derivations.cc/rewriteDerivation)",
+                    drv.name, outputName);
+            auto outPath = store.makeOutputPath(outputName, *h, drv.name);
             drv.env[outputName] = store.printStorePath(outPath);
             output = DerivationOutput::InputAddressed {
                 .path = std::move(outPath),
@@ -758,55 +852,48 @@ static void rewriteDerivation(Store & store, BasicDerivation & drv, const String
 
 }
 
-const Hash & DrvHashModulo::requireNoFixedNonDeferred() const {
-    auto * drvHashOpt = std::get_if<DrvHash>(&raw());
-    assert(drvHashOpt);
-    assert(drvHashOpt->kind == DrvHash::Kind::Regular);
-    return drvHashOpt->hash;
-}
-
-static bool tryResolveInput(
-    Store & store, StorePathSet & inputSrcs, StringMap & inputRewrites,
-    const StorePath & inputDrv, const StringSet & inputOutputs)
+std::optional<BasicDerivation> Derivation::tryResolve(Store & store) const
 {
-    auto inputDrvOutputs = store.queryPartialDerivationOutputMap(inputDrv);
+    std::map<std::pair<StorePath, std::string>, StorePath> inputDrvOutputs;
 
-    auto getOutput = [&](const std::string & outputName) {
-        auto & actualPathOpt = inputDrvOutputs.at(outputName);
-        if (!actualPathOpt)
-            warn("output %s of input %s missing, aborting the resolving",
-                outputName,
-                store.printStorePath(inputDrv)
-            );
-        return actualPathOpt;
-    };
+    for (auto & input : inputDrvs)
+        for (auto & [outputName, outputPath] : store.queryPartialDerivationOutputMap(input.first))
+            if (outputPath)
+                inputDrvOutputs.insert_or_assign({input.first, outputName}, *outputPath);
 
-    for (auto & outputName : inputOutputs) {
-        auto actualPathOpt = getOutput(outputName);
-        if (!actualPathOpt) return false;
-        auto actualPath = *actualPathOpt;
-        inputRewrites.emplace(
-            downstreamPlaceholder(store, inputDrv, outputName),
-            store.printStorePath(actualPath));
-        inputSrcs.insert(std::move(actualPath));
-    }
-
-    return true;
+    return tryResolve(store, inputDrvOutputs);
 }
 
-std::optional<BasicDerivation> Derivation::tryResolve(Store & store) {
+std::optional<BasicDerivation> Derivation::tryResolve(
+    Store & store,
+    const std::map<std::pair<StorePath, std::string>, StorePath> & inputDrvOutputs) const
+{
     BasicDerivation resolved { *this };
 
     // Input paths that we'll want to rewrite in the derivation
     StringMap inputRewrites;
 
-    for (auto & [inputDrv, inputOutputs] : inputDrvs)
-        if (!tryResolveInput(store, resolved.inputSrcs, inputRewrites, inputDrv, inputOutputs))
-            return std::nullopt;
+    for (auto & [inputDrv, inputOutputs] : inputDrvs) {
+        for (auto & outputName : inputOutputs) {
+            if (auto actualPath = get(inputDrvOutputs, { inputDrv, outputName })) {
+                inputRewrites.emplace(
+                    downstreamPlaceholder(store, inputDrv, outputName),
+                    store.printStorePath(*actualPath));
+                resolved.inputSrcs.insert(*actualPath);
+            } else {
+                warn("output '%s' of input '%s' missing, aborting the resolving",
+                    outputName,
+                    store.printStorePath(inputDrv));
+                return {};
+            }
+        }
+    }
 
     rewriteDerivation(store, resolved, inputRewrites);
 
     return resolved;
 }
 
+const Hash impureOutputHash = hashString(htSHA256, "impure");
+
 }

src/libstore/derivations.hh

@@ -41,15 +41,26 @@ struct DerivationOutputCAFloating
 };
 
 /* Input-addressed output which depends on a (CA) derivation whose hash isn't
- * known atm
+ * known yet.
  */
 struct DerivationOutputDeferred {};
 
+/* Impure output which is moved to a content-addressed location (like
+   CAFloating) but isn't registered as a realization.
+ */
+struct DerivationOutputImpure
+{
+    /* information used for expected hash computation */
+    FileIngestionMethod method;
+    HashType hashType;
+};
+
 typedef std::variant<
     DerivationOutputInputAddressed,
     DerivationOutputCAFixed,
     DerivationOutputCAFloating,
-    DerivationOutputDeferred
+    DerivationOutputDeferred,
+    DerivationOutputImpure
 > _DerivationOutputRaw;
 
 struct DerivationOutput : _DerivationOutputRaw
@@ -61,6 +72,7 @@ struct DerivationOutput : _DerivationOutputRaw
     using CAFixed = DerivationOutputCAFixed;
     using CAFloating = DerivationOutputCAFloating;
     using Deferred = DerivationOutputDeferred;
+    using Impure = DerivationOutputImpure;
 
     /* Note, when you use this function you should make sure that you're passing
        the right derivation name. When in doubt, you should use the safer
@@ -90,13 +102,17 @@ struct DerivationType_InputAddressed {
 };
 
 struct DerivationType_ContentAddressed {
-    bool pure;
+    bool sandboxed;
     bool fixed;
 };
 
+struct DerivationType_Impure {
+};
+
 typedef std::variant<
     DerivationType_InputAddressed,
-    DerivationType_ContentAddressed
+    DerivationType_ContentAddressed,
+    DerivationType_Impure
 > _DerivationTypeRaw;
 
 struct DerivationType : _DerivationTypeRaw {
@@ -104,7 +120,7 @@ struct DerivationType : _DerivationTypeRaw {
     using Raw::Raw;
     using InputAddressed = DerivationType_InputAddressed;
     using ContentAddressed = DerivationType_ContentAddressed;
-
+    using Impure = DerivationType_Impure;
 
     /* Do the outputs of the derivation have paths calculated from their content,
        or from the derivation itself? */
@@ -114,10 +130,18 @@ struct DerivationType : _DerivationTypeRaw {
        non-CA derivations. */
     bool isFixed() const;
 
-    /* Is the derivation impure and needs to access non-deterministic resources, or
-       pure and can be sandboxed? Note that whether or not we actually sandbox the
-       derivation is controlled separately. Never true for non-CA derivations. */
-    bool isImpure() const;
+    /* Whether the derivation is fully sandboxed. If false, the
+       sandbox is opened up, e.g. the derivation has access to the
+       network. Note that whether or not we actually sandbox the
+       derivation is controlled separately. Always true for non-CA
+       derivations. */
+    bool isSandboxed() const;
+
+    /* Whether the derivation is expected to produce the same result
+       every time, and therefore it only needs to be built once. This
+       is only false for derivations that have the attribute '__impure
+       = true'. */
+    bool isPure() const;
 
     /* Does the derivation knows its own output paths?
        Only true when there's no floating-ca derivation involved in the
@@ -173,7 +197,14 @@ struct Derivation : BasicDerivation
           added directly to input sources.
 
        2. Input placeholders are replaced with realized input store paths. */
-    std::optional<BasicDerivation> tryResolve(Store & store);
+    std::optional<BasicDerivation> tryResolve(Store & store) const;
+
+    /* Like the above, but instead of querying the Nix database for
+       realisations, uses a given mapping from input derivation paths
+       + output names to actual output store paths. */
+    std::optional<BasicDerivation> tryResolve(
+        Store & store,
+        const std::map<std::pair<StorePath, std::string>, StorePath> & inputDrvOutputs) const;
 
     Derivation() = default;
     Derivation(const BasicDerivation & bd) : BasicDerivation(bd) { }
@@ -202,14 +233,16 @@ bool isDerivation(const std::string & fileName);
    the output name is "out". */
 std::string outputPathName(std::string_view drvName, std::string_view outputName);
 
-// known CA drv's output hashes, current just for fixed-output derivations
-// whose output hashes are always known since they are fixed up-front.
-typedef std::map<std::string, Hash> CaOutputHashes;
 
+// The hashes modulo of a derivation.
+//
+// Each output is given a hash, although in practice only the content-addressed
+// derivations (fixed-output or not) will have a different hash for each
+// output.
 struct DrvHash {
-    Hash hash;
+    std::map<std::string, Hash> hashes;
 
-    enum struct Kind: bool {
+    enum struct Kind : bool {
         // Statically determined derivations.
         // This hash will be directly used to compute the output paths
         Regular,
@@ -222,28 +255,6 @@ struct DrvHash {
 
 void operator |= (DrvHash::Kind & self, const DrvHash::Kind & other) noexcept;
 
-typedef std::variant<
-    // Regular normalized derivation hash, and whether it was deferred (because
-    // an ancestor derivation is a floating content addressed derivation).
-    DrvHash,
-    // Fixed-output derivation hashes
-    CaOutputHashes
-> _DrvHashModuloRaw;
-
-struct DrvHashModulo : _DrvHashModuloRaw {
-    using Raw = _DrvHashModuloRaw;
-    using Raw::Raw;
-
-    /* Get hash, throwing if it is per-output CA hashes or a
-       deferred Drv hash.
-     */
-    const Hash & requireNoFixedNonDeferred() const;
-
-    inline const Raw & raw() const {
-        return static_cast<const Raw &>(*this);
-    }
-};
-
 /* Returns hashes with the details of fixed-output subderivations
    expunged.
 
@@ -267,16 +278,18 @@ struct DrvHashModulo : _DrvHashModuloRaw {
    ATerm, after subderivations have been likewise expunged from that
    derivation.
  */
-DrvHashModulo hashDerivationModulo(Store & store, const Derivation & drv, bool maskOutputs);
+DrvHash hashDerivationModulo(Store & store, const Derivation & drv, bool maskOutputs);
 
 /*
    Return a map associating each output to a hash that uniquely identifies its
    derivation (modulo the self-references).
+
+   FIXME: what is the Hash in this map?
  */
-std::map<std::string, Hash> staticOutputHashes(Store& store, const Derivation& drv);
+std::map<std::string, Hash> staticOutputHashes(Store & store, const Derivation & drv);
 
 /* Memoisation of hashDerivationModulo(). */
-typedef std::map<StorePath, DrvHashModulo> DrvHashes;
+typedef std::map<StorePath, DrvHash> DrvHashes;
 
 // FIXME: global, though at least thread-safe.
 extern Sync<DrvHashes> drvHashes;
@@ -306,4 +319,6 @@ std::string hashPlaceholder(const std::string_view outputName);
    dependency which is a CA derivation. */
 std::string downstreamPlaceholder(const Store & store, const StorePath & drvPath, std::string_view outputName);
 
+extern const Hash impureOutputHash;
+
 }

src/libstore/derived-path.cc

@@ -4,6 +4,8 @@
 
 #include <nlohmann/json.hpp>
 
+#include <optional>
+
 namespace nix {
 
 nlohmann::json DerivedPath::Opaque::toJSON(ref<Store> store) const {
@@ -17,12 +19,12 @@ nlohmann::json DerivedPath::Built::toJSON(ref<Store> store) const {
     res["drvPath"] = store->printStorePath(drvPath);
     // Fallback for the input-addressed derivation case: We expect to always be
     // able to print the output paths, so let’s do it
-    auto knownOutputs = store->queryPartialDerivationOutputMap(drvPath);
+    const auto knownOutputs = store->queryPartialDerivationOutputMap(drvPath);
     for (const auto& output : outputs) {
-        if (knownOutputs.at(output))
-            res["outputs"][output] = store->printStorePath(knownOutputs.at(output).value());
-        else
-            res["outputs"][output] = nullptr;
+        auto knownOutput = get(knownOutputs, output);
+        res["outputs"][output] = (knownOutput && *knownOutput)
+          ? store->printStorePath(**knownOutput)
+          : nullptr;
     }
     return res;
 }
@@ -123,10 +125,15 @@ RealisedPath::Set BuiltPath::toRealisedPaths(Store & store) const
                 for (auto& [outputName, outputPath] : p.outputs) {
                     if (settings.isExperimentalFeatureEnabled(
                                 Xp::CaDerivations)) {
+                        auto drvOutput = get(drvHashes, outputName);
+                        if (!drvOutput)
+                            throw Error(
+                                "the derivation '%s' has unrealised output '%s' (derived-path.cc/toRealisedPaths)",
+                                store.printStorePath(p.drvPath), outputName);
                         auto thisRealisation = store.queryRealisation(
-                            DrvOutput{drvHashes.at(outputName), outputName});
-                        assert(thisRealisation);  // We’ve built it, so we must h
-                                                  // ve the realisation
+                            DrvOutput{*drvOutput, outputName});
+                        assert(thisRealisation);  // We’ve built it, so we must
+                                                  // have the realisation
                         res.insert(*thisRealisation);
                     } else {
                         res.insert(outputPath);

src/libstore/filetransfer.cc

@@ -443,14 +443,13 @@ struct curlFileTransfer : public FileTransfer
                     : httpStatus != 0
                     ? FileTransferError(err,
                         std::move(response),
-                        fmt("unable to %s '%s': HTTP error %d ('%s')",
-                            request.verb(), request.uri, httpStatus, statusMsg)
-                        + (code == CURLE_OK ? "" : fmt(" (curl error: %s)", curl_easy_strerror(code)))
-                        )
+                        "unable to %s '%s': HTTP error %d%s",
+                        request.verb(), request.uri, httpStatus,
+                        code == CURLE_OK ? "" : fmt(" (curl error: %s)", curl_easy_strerror(code)))
                     : FileTransferError(err,
                         std::move(response),
-                        fmt("unable to %s '%s': %s (%d)",
-                            request.verb(), request.uri, curl_easy_strerror(code), code));
+                        "unable to %s '%s': %s (%d)",
+                        request.verb(), request.uri, curl_easy_strerror(code), code);
 
                 /* If this is a transient error, then maybe retry the
                    download after a while. If we're writing to a
@@ -693,10 +692,10 @@ struct curlFileTransfer : public FileTransfer
 #if ENABLE_S3
                 auto [bucketName, key, params] = parseS3Uri(request.uri);
 
-                std::string profile = get(params, "profile").value_or("");
-                std::string region = get(params, "region").value_or(Aws::Region::US_EAST_1);
-                std::string scheme = get(params, "scheme").value_or("");
-                std::string endpoint = get(params, "endpoint").value_or("");
+                std::string profile = getOr(params, "profile", "");
+                std::string region = getOr(params, "region", Aws::Region::US_EAST_1);
+                std::string scheme = getOr(params, "scheme", "");
+                std::string endpoint = getOr(params, "endpoint", "");
 
                 S3Helper s3Helper(profile, region, scheme, endpoint);
 
@@ -704,7 +703,7 @@ struct curlFileTransfer : public FileTransfer
                 auto s3Res = s3Helper.getObject(bucketName, key);
                 FileTransferResult res;
                 if (!s3Res.data)
-                    throw FileTransferError(NotFound, {}, "S3 object '%s' does not exist", request.uri);
+                    throw FileTransferError(NotFound, "S3 object '%s' does not exist", request.uri);
                 res.data = std::move(*s3Res.data);
                 callback(std::move(res));
 #else

src/libstore/filetransfer.hh

@@ -31,7 +31,7 @@ struct FileTransferSettings : Config
         R"(
           The timeout (in seconds) for establishing connections in the
           binary cache substituter. It corresponds to `curl`’s
-          `--connect-timeout` option.
+          `--connect-timeout` option. A value of 0 means no limit.
         )"};
 
     Setting<unsigned long> stalledDownloadTimeout{
@@ -123,8 +123,6 @@ public:
 
     template<typename... Args>
     FileTransferError(FileTransfer::Error error, std::optional<std::string> response, const Args & ... args);
-
-    virtual const char* sname() const override { return "FileTransferError"; }
 };
 
 bool isUri(std::string_view s);

src/libstore/local-store.cc

@@ -81,7 +81,7 @@ int getSchema(Path schemaPath)
 
 void migrateCASchema(SQLite& db, Path schemaPath, AutoCloseFD& lockFd)
 {
-    const int nixCASchemaVersion = 3;
+    const int nixCASchemaVersion = 4;
     int curCASchema = getSchema(schemaPath);
     if (curCASchema != nixCASchemaVersion) {
         if (curCASchema > nixCASchemaVersion) {
@@ -143,6 +143,21 @@ void migrateCASchema(SQLite& db, Path schemaPath, AutoCloseFD& lockFd)
             )");
             txn.commit();
         }
+        if (curCASchema < 4) {
+            SQLiteTxn txn(db);
+            db.exec(R"(
+                create trigger if not exists DeleteSelfRefsViaRealisations before delete on ValidPaths
+                begin
+                    delete from RealisationsRefs where realisationReference in (
+                    select id from Realisations where outputPath = old.id
+                    );
+                end;
+                -- used by deletion trigger
+                create index if not exists IndexRealisationsRefsRealisationReference on RealisationsRefs(realisationReference);
+            )");
+            txn.commit();
+        }
+
         writeFile(schemaPath, fmt("%d", nixCASchemaVersion));
         lockFile(lockFd.get(), ltRead, true);
     }
@@ -482,18 +497,18 @@ void LocalStore::openDB(State & state, bool create)
         SQLiteStmt stmt;
         stmt.create(db, "pragma main.journal_mode;");
         if (sqlite3_step(stmt) != SQLITE_ROW)
-            throwSQLiteError(db, "querying journal mode");
+            SQLiteError::throw_(db, "querying journal mode");
         prevMode = std::string((const char *) sqlite3_column_text(stmt, 0));
     }
     if (prevMode != mode &&
         sqlite3_exec(db, ("pragma main.journal_mode = " + mode + ";").c_str(), 0, 0, 0) != SQLITE_OK)
-        throwSQLiteError(db, "setting journal mode");
+        SQLiteError::throw_(db, "setting journal mode");
 
     /* Increase the auto-checkpoint interval to 40000 pages.  This
        seems enough to ensure that instantiating the NixOS system
        derivation is done in a single fsync(). */
     if (mode == "wal" && sqlite3_exec(db, "pragma wal_autocheckpoint = 40000;", 0, 0, 0) != SQLITE_OK)
-        throwSQLiteError(db, "setting autocheckpoint interval");
+        SQLiteError::throw_(db, "setting autocheckpoint interval");
 
     /* Initialise the database schema, if necessary. */
     if (create) {
@@ -695,16 +710,19 @@ void LocalStore::checkDerivationOutputs(const StorePath & drvPath, const Derivat
     // combinations that are currently prohibited.
     drv.type();
 
-    std::optional<Hash> h;
+    std::optional<DrvHash> hashesModulo;
     for (auto & i : drv.outputs) {
         std::visit(overloaded {
             [&](const DerivationOutput::InputAddressed & doia) {
-                if (!h) {
+                if (!hashesModulo) {
                     // somewhat expensive so we do lazily
-                    auto h0 = hashDerivationModulo(*this, drv, true);
-                    h = h0.requireNoFixedNonDeferred();
+                    hashesModulo = hashDerivationModulo(*this, drv, true);
                 }
-                StorePath recomputed = makeOutputPath(i.first, *h, drvName);
+                auto currentOutputHash = get(hashesModulo->hashes, i.first);
+                if (!currentOutputHash)
+                    throw Error("derivation '%s' has unexpected output '%s' (local-store / hashesModulo) named '%s'",
+                        printStorePath(drvPath), printStorePath(doia.path), i.first);
+                StorePath recomputed = makeOutputPath(i.first, *currentOutputHash, drvName);
                 if (doia.path != recomputed)
                     throw Error("derivation '%s' has incorrect output '%s', should be '%s'",
                         printStorePath(drvPath), printStorePath(doia.path), printStorePath(recomputed));
@@ -720,6 +738,9 @@ void LocalStore::checkDerivationOutputs(const StorePath & drvPath, const Derivat
             [&](const DerivationOutput::Deferred &) {
                 /* Nothing to check */
             },
+            [&](const DerivationOutput::Impure &) {
+                /* Nothing to check */
+            },
         }, i.second.raw());
     }
 }

src/libstore/misc.cc

@@ -277,15 +277,20 @@ std::map<DrvOutput, StorePath> drvOutputReferences(
 {
     std::set<Realisation> inputRealisations;
 
-    for (const auto& [inputDrv, outputNames] : drv.inputDrvs) {
-        auto outputHashes =
+    for (const auto & [inputDrv, outputNames] : drv.inputDrvs) {
+        const auto outputHashes =
             staticOutputHashes(store, store.readDerivation(inputDrv));
-        for (const auto& outputName : outputNames) {
+        for (const auto & outputName : outputNames) {
+            auto outputHash = get(outputHashes, outputName);
+            if (!outputHash)
+                throw Error(
+                    "output '%s' of derivation '%s' isn't realised", outputName,
+                    store.printStorePath(inputDrv));
             auto thisRealisation = store.queryRealisation(
-                DrvOutput{outputHashes.at(outputName), outputName});
+                DrvOutput{*outputHash, outputName});
             if (!thisRealisation)
                 throw Error(
-                    "output '%s' of derivation '%s' isn’t built", outputName,
+                    "output '%s' of derivation '%s' isn't built", outputName,
                     store.printStorePath(inputDrv));
             inputRealisations.insert(*thisRealisation);
         }
@@ -295,4 +300,5 @@ std::map<DrvOutput, StorePath> drvOutputReferences(
 
     return drvOutputReferences(Realisation::closure(store, inputRealisations), info->references);
 }
+
 }

src/libstore/path-with-outputs.cc

@@ -1,5 +1,8 @@
 #include "path-with-outputs.hh"
 #include "store-api.hh"
+#include "nlohmann/json.hpp"
+
+#include <regex>
 
 namespace nix {
 
@@ -68,4 +71,57 @@ StorePathWithOutputs followLinksToStorePathWithOutputs(const Store & store, std:
     return StorePathWithOutputs { store.followLinksToStorePath(path), std::move(outputs) };
 }
 
+std::pair<std::string, OutputsSpec> parseOutputsSpec(const std::string & s)
+{
+    static std::regex regex(R"((.*)\^((\*)|([a-z]+(,[a-z]+)*)))");
+
+    std::smatch match;
+    if (!std::regex_match(s, match, regex))
+        return {s, DefaultOutputs()};
+
+    if (match[3].matched)
+        return {match[1], AllOutputs()};
+
+    return {match[1], tokenizeString<OutputNames>(match[4].str(), ",")};
+}
+
+std::string printOutputsSpec(const OutputsSpec & outputsSpec)
+{
+    if (std::get_if<DefaultOutputs>(&outputsSpec))
+        return "";
+
+    if (std::get_if<AllOutputs>(&outputsSpec))
+        return "^*";
+
+    if (auto outputNames = std::get_if<OutputNames>(&outputsSpec))
+        return "^" + concatStringsSep(",", *outputNames);
+
+    assert(false);
+}
+
+void to_json(nlohmann::json & json, const OutputsSpec & outputsSpec)
+{
+    if (std::get_if<DefaultOutputs>(&outputsSpec))
+        json = nullptr;
+
+    else if (std::get_if<AllOutputs>(&outputsSpec))
+        json = std::vector<std::string>({"*"});
+
+    else if (auto outputNames = std::get_if<OutputNames>(&outputsSpec))
+        json = *outputNames;
+}
+
+void from_json(const nlohmann::json & json, OutputsSpec & outputsSpec)
+{
+    if (json.is_null())
+        outputsSpec = DefaultOutputs();
+    else {
+        auto names = json.get<OutputNames>();
+        if (names == OutputNames({"*"}))
+            outputsSpec = AllOutputs();
+        else
+            outputsSpec = names;
+    }
+}
+
 }

src/libstore/path-with-outputs.hh

@@ -4,6 +4,7 @@
 
 #include "path.hh"
 #include "derived-path.hh"
+#include "nlohmann/json_fwd.hpp"
 
 namespace nix {
 
@@ -32,4 +33,25 @@ StorePathWithOutputs parsePathWithOutputs(const Store & store, std::string_view
 
 StorePathWithOutputs followLinksToStorePathWithOutputs(const Store & store, std::string_view pathWithOutputs);
 
+typedef std::set<std::string> OutputNames;
+
+struct AllOutputs {
+    bool operator < (const AllOutputs & _) const { return false; }
+};
+
+struct DefaultOutputs {
+    bool operator < (const DefaultOutputs & _) const { return false; }
+};
+
+typedef std::variant<DefaultOutputs, AllOutputs, OutputNames> OutputsSpec;
+
+/* Parse a string of the form 'prefix^output1,...outputN' or
+   'prefix^*', returning the prefix and the outputs spec. */
+std::pair<std::string, OutputsSpec> parseOutputsSpec(const std::string & s);
+
+std::string printOutputsSpec(const OutputsSpec & outputsSpec);
+
+void to_json(nlohmann::json &, const OutputsSpec &);
+void from_json(const nlohmann::json &, OutputsSpec &);
+
 }

src/libstore/path.cc

@@ -1,5 +1,7 @@
 #include "store-api.hh"
 
+#include <sodium.h>
+
 namespace nix {
 
 static void checkName(std::string_view path, std::string_view name)
@@ -41,6 +43,13 @@ bool StorePath::isDerivation() const
 
 StorePath StorePath::dummy("ffffffffffffffffffffffffffffffff-x");
 
+StorePath StorePath::random(std::string_view name)
+{
+    Hash hash(htSHA1);
+    randombytes_buf(hash.hash, hash.hashSize);
+    return StorePath(hash, name);
+}
+
 StorePath Store::parseStorePath(std::string_view path) const
 {
     auto p = canonPath(std::string(path));

src/libstore/path.hh

@@ -58,6 +58,8 @@ public:
     }
 
     static StorePath dummy;
+
+    static StorePath random(std::string_view name);
 };
 
 typedef std::set<StorePath> StorePathSet;

src/libstore/remote-store.cc

@@ -853,15 +853,15 @@ std::vector<BuildResult> RemoteStore::buildPathsWithResults(
 
                         OutputPathMap outputs;
                         auto drv = evalStore->readDerivation(bfd.drvPath);
-                        auto outputHashes = staticOutputHashes(*evalStore, drv); // FIXME: expensive
-                        auto drvOutputs = drv.outputsAndOptPaths(*this);
+                        const auto outputHashes = staticOutputHashes(*evalStore, drv); // FIXME: expensive
+                        const auto drvOutputs = drv.outputsAndOptPaths(*this);
                         for (auto & output : bfd.outputs) {
-                            if (!outputHashes.count(output))
+                            auto outputHash = get(outputHashes, output);
+                            if (!outputHash)
                                 throw Error(
                                     "the derivation '%s' doesn't have an output named '%s'",
                                     printStorePath(bfd.drvPath), output);
-                            auto outputId =
-                                DrvOutput{outputHashes.at(output), output};
+                            auto outputId = DrvOutput{ *outputHash, output };
                             if (settings.isExperimentalFeatureEnabled(Xp::CaDerivations)) {
                                 auto realisation =
                                     queryRealisation(outputId);
@@ -874,13 +874,14 @@ std::vector<BuildResult> RemoteStore::buildPathsWithResults(
                             } else {
                                 // If ca-derivations isn't enabled, assume that
                                 // the output path is statically known.
-                                assert(drvOutputs.count(output));
-                                assert(drvOutputs.at(output).second);
+                                const auto drvOutput = get(drvOutputs, output);
+                                assert(drvOutput);
+                                assert(drvOutput->second);
                                 res.builtOutputs.emplace(
                                     outputId,
                                     Realisation {
                                         .id = outputId,
-                                        .outPath = *drvOutputs.at(output).second
+                                        .outPath = *drvOutput->second,
                                     });
                             }
                         }

src/libstore/s3.hh

@@ -5,6 +5,7 @@
 #include "ref.hh"
 
 #include <optional>
+#include <string>
 
 namespace Aws { namespace Client { class ClientConfiguration; } }
 namespace Aws { namespace S3 { class S3Client; } }

src/libstore/sqlite.cc

@@ -8,22 +8,32 @@
 
 namespace nix {
 
-[[noreturn]] void throwSQLiteError(sqlite3 * db, const FormatOrString & fs)
+SQLiteError::SQLiteError(const char *path, int errNo, int extendedErrNo, hintformat && hf)
+  : Error(""), path(path), errNo(errNo), extendedErrNo(extendedErrNo)
+{
+    err.msg = hintfmt("%s: %s (in '%s')",
+        normaltxt(hf.str()),
+        sqlite3_errstr(extendedErrNo),
+        path ? path : "(in-memory)");
+}
+
+[[noreturn]] void SQLiteError::throw_(sqlite3 * db, hintformat && hf)
 {
     int err = sqlite3_errcode(db);
     int exterr = sqlite3_extended_errcode(db);
 
     auto path = sqlite3_db_filename(db, nullptr);
-    if (!path) path = "(in-memory)";
 
     if (err == SQLITE_BUSY || err == SQLITE_PROTOCOL) {
-        throw SQLiteBusy(
+        auto exp = SQLiteBusy(path, err, exterr, std::move(hf));
+        exp.err.msg = hintfmt(
             err == SQLITE_PROTOCOL
-            ? fmt("SQLite database '%s' is busy (SQLITE_PROTOCOL)", path)
-            : fmt("SQLite database '%s' is busy", path));
-    }
-    else
-        throw SQLiteError("%s: %s (in '%s')", fs.s, sqlite3_errstr(exterr), path);
+                ? "SQLite database '%s' is busy (SQLITE_PROTOCOL)"
+                : "SQLite database '%s' is busy",
+            path ? path : "(in-memory)");
+        throw exp;
+    } else
+        throw SQLiteError(path, err, exterr, std::move(hf));
 }
 
 SQLite::SQLite(const Path & path, bool create)
@@ -37,7 +47,7 @@ SQLite::SQLite(const Path & path, bool create)
         throw Error("cannot open SQLite database '%s'", path);
 
     if (sqlite3_busy_timeout(db, 60 * 60 * 1000) != SQLITE_OK)
-        throwSQLiteError(db, "setting timeout");
+        SQLiteError::throw_(db, "setting timeout");
 
     exec("pragma foreign_keys = 1");
 }
@@ -46,7 +56,7 @@ SQLite::~SQLite()
 {
     try {
         if (db && sqlite3_close(db) != SQLITE_OK)
-            throwSQLiteError(db, "closing database");
+            SQLiteError::throw_(db, "closing database");
     } catch (...) {
         ignoreException();
     }
@@ -62,7 +72,7 @@ void SQLite::exec(const std::string & stmt)
 {
     retrySQLite<void>([&]() {
         if (sqlite3_exec(db, stmt.c_str(), 0, 0, 0) != SQLITE_OK)
-            throwSQLiteError(db, format("executing SQLite statement '%s'") % stmt);
+            SQLiteError::throw_(db, "executing SQLite statement '%s'", stmt);
     });
 }
 
@@ -76,7 +86,7 @@ void SQLiteStmt::create(sqlite3 * db, const std::string & sql)
     checkInterrupt();
     assert(!stmt);
     if (sqlite3_prepare_v2(db, sql.c_str(), -1, &stmt, 0) != SQLITE_OK)
-        throwSQLiteError(db, fmt("creating statement '%s'", sql));
+        SQLiteError::throw_(db, "creating statement '%s'", sql);
     this->db = db;
     this->sql = sql;
 }
@@ -85,7 +95,7 @@ SQLiteStmt::~SQLiteStmt()
 {
     try {
         if (stmt && sqlite3_finalize(stmt) != SQLITE_OK)
-            throwSQLiteError(db, fmt("finalizing statement '%s'", sql));
+            SQLiteError::throw_(db, "finalizing statement '%s'", sql);
     } catch (...) {
         ignoreException();
     }
@@ -109,7 +119,7 @@ SQLiteStmt::Use & SQLiteStmt::Use::operator () (std::string_view value, bool not
 {
     if (notNull) {
         if (sqlite3_bind_text(stmt, curArg++, value.data(), -1, SQLITE_TRANSIENT) != SQLITE_OK)
-            throwSQLiteError(stmt.db, "binding argument");
+            SQLiteError::throw_(stmt.db, "binding argument");
     } else
         bind();
     return *this;
@@ -119,7 +129,7 @@ SQLiteStmt::Use & SQLiteStmt::Use::operator () (const unsigned char * data, size
 {
     if (notNull) {
         if (sqlite3_bind_blob(stmt, curArg++, data, len, SQLITE_TRANSIENT) != SQLITE_OK)
-            throwSQLiteError(stmt.db, "binding argument");
+            SQLiteError::throw_(stmt.db, "binding argument");
     } else
         bind();
     return *this;
@@ -129,7 +139,7 @@ SQLiteStmt::Use & SQLiteStmt::Use::operator () (int64_t value, bool notNull)
 {
     if (notNull) {
         if (sqlite3_bind_int64(stmt, curArg++, value) != SQLITE_OK)
-            throwSQLiteError(stmt.db, "binding argument");
+            SQLiteError::throw_(stmt.db, "binding argument");
     } else
         bind();
     return *this;
@@ -138,7 +148,7 @@ SQLiteStmt::Use & SQLiteStmt::Use::operator () (int64_t value, bool notNull)
 SQLiteStmt::Use & SQLiteStmt::Use::bind()
 {
     if (sqlite3_bind_null(stmt, curArg++) != SQLITE_OK)
-        throwSQLiteError(stmt.db, "binding argument");
+        SQLiteError::throw_(stmt.db, "binding argument");
     return *this;
 }
 
@@ -152,14 +162,14 @@ void SQLiteStmt::Use::exec()
     int r = step();
     assert(r != SQLITE_ROW);
     if (r != SQLITE_DONE)
-        throwSQLiteError(stmt.db, fmt("executing SQLite statement '%s'", sqlite3_expanded_sql(stmt.stmt)));
+        SQLiteError::throw_(stmt.db, fmt("executing SQLite statement '%s'", sqlite3_expanded_sql(stmt.stmt)));
 }
 
 bool SQLiteStmt::Use::next()
 {
     int r = step();
     if (r != SQLITE_DONE && r != SQLITE_ROW)
-        throwSQLiteError(stmt.db, fmt("executing SQLite query '%s'", sqlite3_expanded_sql(stmt.stmt)));
+        SQLiteError::throw_(stmt.db, fmt("executing SQLite query '%s'", sqlite3_expanded_sql(stmt.stmt)));
     return r == SQLITE_ROW;
 }
 
@@ -185,14 +195,14 @@ SQLiteTxn::SQLiteTxn(sqlite3 * db)
 {
     this->db = db;
     if (sqlite3_exec(db, "begin;", 0, 0, 0) != SQLITE_OK)
-        throwSQLiteError(db, "starting transaction");
+        SQLiteError::throw_(db, "starting transaction");
     active = true;
 }
 
 void SQLiteTxn::commit()
 {
     if (sqlite3_exec(db, "commit;", 0, 0, 0) != SQLITE_OK)
-        throwSQLiteError(db, "committing transaction");
+        SQLiteError::throw_(db, "committing transaction");
     active = false;
 }
 
@@ -200,7 +210,7 @@ SQLiteTxn::~SQLiteTxn()
 {
     try {
         if (active && sqlite3_exec(db, "rollback;", 0, 0, 0) != SQLITE_OK)
-            throwSQLiteError(db, "aborting transaction");
+            SQLiteError::throw_(db, "aborting transaction");
     } catch (...) {
         ignoreException();
     }
@@ -215,7 +225,6 @@ void handleSQLiteBusy(const SQLiteBusy & e)
     if (now > lastWarned + 10) {
         lastWarned = now;
         logWarning({
-            .name = "Sqlite busy",
             .msg = hintfmt(e.what())
         });
     }

src/libstore/sqlite.hh

@@ -96,10 +96,30 @@ struct SQLiteTxn
 };
 
 
-MakeError(SQLiteError, Error);
-MakeError(SQLiteBusy, SQLiteError);
+struct SQLiteError : Error
+{
+    const char *path;
+    int errNo, extendedErrNo;
 
-[[noreturn]] void throwSQLiteError(sqlite3 * db, const FormatOrString & fs);
+    template<typename... Args>
+    [[noreturn]] static void throw_(sqlite3 * db, const std::string & fs, const Args & ... args) {
+        throw_(db, hintfmt(fs, args...));
+    }
+
+    SQLiteError(const char *path, int errNo, int extendedErrNo, hintformat && hf);
+
+protected:
+
+    template<typename... Args>
+    SQLiteError(const char *path, int errNo, int extendedErrNo, const std::string & fs, const Args & ... args)
+      : SQLiteError(path, errNo, extendedErrNo, hintfmt(fs, args...))
+    { }
+
+    [[noreturn]] static void throw_(sqlite3 * db, hintformat && hf);
+
+};
+
+MakeError(SQLiteBusy, SQLiteError);
 
 void handleSQLiteBusy(const SQLiteBusy & e);
 

src/libstore/store-api.cc

@@ -1314,7 +1314,7 @@ static bool isNonUriPath(const std::string & spec) {
 std::shared_ptr<Store> openFromNonUri(const std::string & uri, const Store::Params & params)
 {
     if (uri == "" || uri == "auto") {
-        auto stateDir = get(params, "state").value_or(settings.nixStateDir);
+        auto stateDir = getOr(params, "state", settings.nixStateDir);
         if (access(stateDir.c_str(), R_OK | W_OK) == 0)
             return std::make_shared<LocalStore>(params);
         else if (pathExists(settings.nixDaemonSocketFile))

src/libstore/tests/path-with-outputs.cc

@@ -0,0 +1,46 @@
+#include "path-with-outputs.hh"
+
+#include <gtest/gtest.h>
+
+namespace nix {
+
+TEST(parseOutputsSpec, basic)
+{
+    {
+        auto [prefix, outputsSpec] = parseOutputsSpec("foo");
+        ASSERT_EQ(prefix, "foo");
+        ASSERT_TRUE(std::get_if<DefaultOutputs>(&outputsSpec));
+    }
+
+    {
+        auto [prefix, outputsSpec] = parseOutputsSpec("foo^*");
+        ASSERT_EQ(prefix, "foo");
+        ASSERT_TRUE(std::get_if<AllOutputs>(&outputsSpec));
+    }
+
+    {
+        auto [prefix, outputsSpec] = parseOutputsSpec("foo^out");
+        ASSERT_EQ(prefix, "foo");
+        ASSERT_TRUE(std::get<OutputNames>(outputsSpec) == OutputNames({"out"}));
+    }
+
+    {
+        auto [prefix, outputsSpec] = parseOutputsSpec("foo^out,bin");
+        ASSERT_EQ(prefix, "foo");
+        ASSERT_TRUE(std::get<OutputNames>(outputsSpec) == OutputNames({"out", "bin"}));
+    }
+
+    {
+        auto [prefix, outputsSpec] = parseOutputsSpec("foo^bar^out,bin");
+        ASSERT_EQ(prefix, "foo^bar");
+        ASSERT_TRUE(std::get<OutputNames>(outputsSpec) == OutputNames({"out", "bin"}));
+    }
+
+    {
+        auto [prefix, outputsSpec] = parseOutputsSpec("foo^&*()");
+        ASSERT_EQ(prefix, "foo^&*()");
+        ASSERT_TRUE(std::get_if<DefaultOutputs>(&outputsSpec));
+    }
+}
+
+}

src/libutil/args.cc

@@ -127,11 +127,11 @@ bool Args::processFlag(Strings::iterator & pos, Strings::iterator end)
                 if (flag.handler.arity == ArityAny) break;
                 throw UsageError("flag '%s' requires %d argument(s)", name, flag.handler.arity);
             }
-            if (flag.completer)
-                if (auto prefix = needsCompletion(*pos)) {
-                    anyCompleted = true;
+            if (auto prefix = needsCompletion(*pos)) {
+                anyCompleted = true;
+                if (flag.completer)
                     flag.completer(n, *prefix);
-                }
+            }
             args.push_back(*pos++);
         }
         if (!anyCompleted)
@@ -146,6 +146,7 @@ bool Args::processFlag(Strings::iterator & pos, Strings::iterator end)
                     && hasPrefix(name, std::string(*prefix, 2)))
                     completions->add("--" + name, flag->description);
             }
+            return false;
         }
         auto i = longFlags.find(std::string(*pos, 2));
         if (i == longFlags.end()) return false;
@@ -187,10 +188,12 @@ bool Args::processArgs(const Strings & args, bool finish)
     {
         std::vector<std::string> ss;
         for (const auto &[n, s] : enumerate(args)) {
-            ss.push_back(s);
-            if (exp.completer)
-                if (auto prefix = needsCompletion(s))
+            if (auto prefix = needsCompletion(s)) {
+                ss.push_back(*prefix);
+                if (exp.completer)
                     exp.completer(n, *prefix);
+            } else
+                ss.push_back(s);
         }
         exp.handler.fun(ss);
         expectedArgs.pop_front();
@@ -279,21 +282,22 @@ static void _completePath(std::string_view prefix, bool onlyDirs)
 {
     completionType = ctFilenames;
     glob_t globbuf;
-    int flags = GLOB_NOESCAPE | GLOB_TILDE;
+    int flags = GLOB_NOESCAPE;
     #ifdef GLOB_ONLYDIR
     if (onlyDirs)
         flags |= GLOB_ONLYDIR;
     #endif
-    if (glob((std::string(prefix) + "*").c_str(), flags, nullptr, &globbuf) == 0) {
+    // using expandTilde here instead of GLOB_TILDE(_CHECK) so that ~<Tab> expands to /home/user/
+    if (glob((expandTilde(prefix) + "*").c_str(), flags, nullptr, &globbuf) == 0) {
         for (size_t i = 0; i < globbuf.gl_pathc; ++i) {
             if (onlyDirs) {
-                auto st = lstat(globbuf.gl_pathv[i]);
+                auto st = stat(globbuf.gl_pathv[i]);
                 if (!S_ISDIR(st.st_mode)) continue;
             }
             completions->add(globbuf.gl_pathv[i]);
         }
-        globfree(&globbuf);
     }
+    globfree(&globbuf);
 }
 
 void completePath(size_t, std::string_view prefix)
@@ -322,11 +326,6 @@ MultiCommand::MultiCommand(const Commands & commands_)
         .optional = true,
         .handler = {[=](std::string s) {
             assert(!command);
-            if (auto prefix = needsCompletion(s)) {
-                for (auto & [name, command] : commands)
-                    if (hasPrefix(name, *prefix))
-                        completions->add(name);
-            }
             auto i = commands.find(s);
             if (i == commands.end()) {
                 std::set<std::string> commandNames;
@@ -337,6 +336,11 @@ MultiCommand::MultiCommand(const Commands & commands_)
             }
             command = {s, i->second()};
             command->second->parent = this;
+        }},
+        .completer = {[&](size_t, std::string_view prefix) {
+            for (auto & [name, command] : commands)
+                if (hasPrefix(name, prefix))
+                    completions->add(name);
         }}
     });
 

src/libutil/chunked-vector.hh

@@ -0,0 +1,68 @@
+#pragma once
+
+#include <cstdint>
+#include <cstdlib>
+#include <vector>
+#include <limits>
+
+namespace nix {
+
+/* Provides an indexable container like vector<> with memory overhead
+   guarantees like list<> by allocating storage in chunks of ChunkSize
+   elements instead of using a contiguous memory allocation like vector<>
+   does. Not using a single vector that is resized reduces memory overhead
+   on large data sets by on average (growth factor)/2, mostly
+   eliminates copies within the vector during resizing, and provides stable
+   references to its elements. */
+template<typename T, size_t ChunkSize>
+class ChunkedVector {
+private:
+    uint32_t size_ = 0;
+    std::vector<std::vector<T>> chunks;
+
+    /* keep this out of the ::add hot path */
+    [[gnu::noinline]]
+    auto & addChunk()
+    {
+        if (size_ >= std::numeric_limits<uint32_t>::max() - ChunkSize)
+            abort();
+        chunks.emplace_back();
+        chunks.back().reserve(ChunkSize);
+        return chunks.back();
+    }
+
+public:
+    ChunkedVector(uint32_t reserve)
+    {
+        chunks.reserve(reserve);
+        addChunk();
+    }
+
+    uint32_t size() const { return size_; }
+
+    std::pair<T &, uint32_t> add(T value)
+    {
+        const auto idx = size_++;
+        auto & chunk = [&] () -> auto & {
+            if (auto & back = chunks.back(); back.size() < ChunkSize)
+                return back;
+            return addChunk();
+        }();
+        auto & result = chunk.emplace_back(std::move(value));
+        return {result, idx};
+    }
+
+    const T & operator[](uint32_t idx) const
+    {
+        return chunks[idx / ChunkSize][idx % ChunkSize];
+    }
+
+    template<typename Fn>
+    void forEach(Fn fn) const
+    {
+        for (const auto & c : chunks)
+            for (const auto & e : c)
+                fn(e);
+    }
+};
+}

src/libutil/error.cc

@@ -9,10 +9,9 @@ namespace nix {
 
 const std::string nativeSystem = SYSTEM;
 
-BaseError & BaseError::addTrace(std::optional<ErrPos> e, hintformat hint)
+void BaseError::addTrace(std::optional<ErrPos> e, hintformat hint)
 {
     err.traces.push_front(Trace { .pos = e, .hint = hint });
-    return *this;
 }
 
 // c++ std::exception descendants must have a 'const char* what()' function.
@@ -22,12 +21,9 @@ const std::string & BaseError::calcWhat() const
     if (what_.has_value())
         return *what_;
     else {
-        err.name = sname();
-
         std::ostringstream oss;
         showErrorInfo(oss, err, loggerSettings.showTrace);
         what_ = oss.str();
-
         return *what_;
     }
 }

src/libutil/error.hh

@@ -87,11 +87,7 @@ struct ErrPos {
         origin = pos.origin;
         line = pos.line;
         column = pos.column;
-        // is file symbol null?
-        if (pos.file.set())
-            file = pos.file;
-        else
-            file = "";
+        file = pos.file;
         return *this;
     }
 
@@ -109,7 +105,6 @@ struct Trace {
 
 struct ErrorInfo {
     Verbosity level;
-    std::string name; // FIXME: rename
     hintformat msg;
     std::optional<ErrPos> errPos;
     std::list<Trace> traces;
@@ -162,8 +157,6 @@ public:
         : err(e)
     { }
 
-    virtual const char* sname() const { return "BaseError"; }
-
 #ifdef EXCEPTION_NEEDS_THROW_SPEC
     ~BaseError() throw () { };
     const char * what() const throw () { return calcWhat().c_str(); }
@@ -175,12 +168,12 @@ public:
     const ErrorInfo & info() const { calcWhat(); return err; }
 
     template<typename... Args>
-    BaseError & addTrace(std::optional<ErrPos> e, const std::string & fs, const Args & ... args)
+    void addTrace(std::optional<ErrPos> e, const std::string & fs, const Args & ... args)
     {
-        return addTrace(e, hintfmt(fs, args...));
+        addTrace(e, hintfmt(fs, args...));
     }
 
-    BaseError & addTrace(std::optional<ErrPos> e, hintformat hint);
+    void addTrace(std::optional<ErrPos> e, hintformat hint);
 
     bool hasTrace() const { return !err.traces.empty(); }
 };
@@ -190,7 +183,6 @@ public:
     {                                                   \
     public:                                             \
         using superClass::superClass;                   \
-        virtual const char* sname() const override { return #newClass; } \
     }
 
 MakeError(Error, BaseError);
@@ -210,8 +202,6 @@ public:
         auto hf = hintfmt(args...);
         err.msg = hintfmt("%1%: %2%", normaltxt(hf.str()), strerror(errNo));
     }
-
-    virtual const char* sname() const override { return "SysError"; }
 };
 
 }

src/libutil/experimental-features.cc

@@ -7,6 +7,7 @@ namespace nix {
 
 std::map<ExperimentalFeature, std::string> stringifiedXpFeatures = {
     { Xp::CaDerivations, "ca-derivations" },
+    { Xp::ImpureDerivations, "impure-derivations" },
     { Xp::Flakes, "flakes" },
     { Xp::NixCommand, "nix-command" },
     { Xp::RecursiveNix, "recursive-nix" },
@@ -34,7 +35,9 @@ const std::optional<ExperimentalFeature> parseExperimentalFeature(const std::str
 
 std::string_view showExperimentalFeature(const ExperimentalFeature feature)
 {
-    return stringifiedXpFeatures.at(feature);
+    const auto ret = get(stringifiedXpFeatures, feature);
+    assert(ret);
+    return *ret;
 }
 
 std::set<ExperimentalFeature> parseFeatures(const std::set<std::string> & rawFeatures)
@@ -57,4 +60,20 @@ std::ostream & operator <<(std::ostream & str, const ExperimentalFeature & featu
     return str << showExperimentalFeature(feature);
 }
 
+void to_json(nlohmann::json & j, const ExperimentalFeature & feature)
+{
+    j = showExperimentalFeature(feature);
+}
+
+void from_json(const nlohmann::json & j, ExperimentalFeature & feature)
+{
+    const std::string input = j;
+    const auto parsed = parseExperimentalFeature(input);
+
+    if (parsed.has_value())
+        feature = *parsed;
+    else
+        throw Error("Unknown experimental feature '%s' in JSON input", input);
+}
+
 }

src/libutil/experimental-features.hh

@@ -16,6 +16,7 @@ namespace nix {
 enum struct ExperimentalFeature
 {
     CaDerivations,
+    ImpureDerivations,
     Flakes,
     NixCommand,
     RecursiveNix,
@@ -48,10 +49,13 @@ public:
     ExperimentalFeature missingFeature;
 
     MissingExperimentalFeature(ExperimentalFeature);
-    virtual const char * sname() const override
-    {
-        return "MissingExperimentalFeature";
-    }
 };
 
+/**
+ * Semi-magic conversion to and from json.
+ * See the nlohmann/json readme for more details.
+ */
+void to_json(nlohmann::json &, const ExperimentalFeature &);
+void from_json(const nlohmann::json &, ExperimentalFeature &);
+
 }

src/libutil/fmt.hh

@@ -2,7 +2,6 @@
 
 #include <boost/format.hpp>
 #include <string>
-#include <regex>
 #include "ansicolor.hh"
 
 
@@ -155,15 +154,4 @@ inline hintformat hintfmt(std::string plain_string)
     return hintfmt("%s", normaltxt(plain_string));
 }
 
-/* Highlight all the given matches in the given string `s` by wrapping
-   them between `prefix` and `postfix`.
-
-   If some matches overlap, then their union will be wrapped rather
-   than the individual matches. */
-std::string hiliteMatches(
-    std::string_view s,
-    std::vector<std::smatch> matches,
-    std::string_view prefix,
-    std::string_view postfix);
-
 }