When working on some more complex Nix code, there are sometimes rather
unhelpful or misleading error messages, especially if coerce-errors are
thrown.
This patch is a first steps towards improving that. I'm happy to file
more changes after that, but I'd like to gather some feedback first.
To summarize, this patch does the following things:
* Attrsets (a.k.a. `Bindings` in `libexpr`) now have a `Pos`. This is
helpful e.g. to identify which attribute-set in `listToAttrs` is
invalid.
* The `Value`-struct has a new method named `determinePos` which tries
to guess the position of a value and falls back to a default if that's
not possible.
This can be used to provide better messages if a coercion fails.
* The new `determinePos`-API is used by `builtins.concatMap` now. With
that change, Nix shows the exact position in the error where a wrong
value was returned by the lambda.
To make sure it's still obvious that `concatMap` is the problem,
another stack-frame was added.
* The changes described above can be added to every other `primop`, but
first I'd like to get some feedback about the overall approach.
Changes:
* The divider lines are gone. These were in practice a bit confusing,
in particular with --show-trace or --keep-going, since then there
were multiple lines, suggesting a start/end which wasn't the case.
* Instead, multi-line error messages are now indented to align with
the prefix (e.g. "error: ").
* The 'description' field is gone since we weren't really using it.
* 'hint' is renamed to 'msg' since it really wasn't a hint.
* The error is now printed *before* the location info.
* The 'name' field is no longer printed since most of the time it
wasn't very useful since it was just the name of the exception (like
EvalError). Ideally in the future this would be a unique, easily
googleable error ID (like rustc).
* "trace:" is now just "…". This assumes error contexts start with
something like "while doing X".
Example before:
error: --- AssertionError ---------------------------------------------------------------------------------------- nix
at: (7:7) in file: /home/eelco/Dev/nixpkgs/pkgs/applications/misc/hello/default.nix
6|
7| x = assert false; 1;
| ^
8|
assertion 'false' failed
----------------------------------------------------- show-trace -----------------------------------------------------
trace: while evaluating the attribute 'x' of the derivation 'hello-2.10'
at: (192:11) in file: /home/eelco/Dev/nixpkgs/pkgs/stdenv/generic/make-derivation.nix
191| // (lib.optionalAttrs (!(attrs ? name) && attrs ? pname && attrs ? version)) {
192| name = "${attrs.pname}-${attrs.version}";
| ^
193| } // (lib.optionalAttrs (stdenv.hostPlatform != stdenv.buildPlatform && !dontAddHostSuffix && (attrs ? name || (attrs ? pname && attrs ? version)))) {
Example after:
error: assertion 'false' failed
at: (7:7) in file: /home/eelco/Dev/nixpkgs/pkgs/applications/misc/hello/default.nix
6|
7| x = assert false; 1;
| ^
8|
… while evaluating the attribute 'x' of the derivation 'hello-2.10'
at: (192:11) in file: /home/eelco/Dev/nixpkgs/pkgs/stdenv/generic/make-derivation.nix
191| // (lib.optionalAttrs (!(attrs ? name) && attrs ? pname && attrs ? version)) {
192| name = "${attrs.pname}-${attrs.version}";
| ^
193| } // (lib.optionalAttrs (stdenv.hostPlatform != stdenv.buildPlatform && !dontAddHostSuffix && (attrs ? name || (attrs ? pname && attrs ? version)))) {
Move clearValue inside Value
mkInt instead of setInt
mkBool instead of setBool
mkString instead of setString
mkPath instead of setPath
mkNull instead of setNull
mkAttrs instead of setAttrs
mkList instead of setList*
mkThunk instead of setThunk
mkApp instead of setApp
mkLambda instead of setLambda
mkBlackhole instead of setBlackhole
mkPrimOp instead of setPrimOp
mkPrimOpApp instead of setPrimOpApp
mkExternal instead of setExternal
mkFloat instead of setFloat
Add note that the static mk* function should be removed eventually
Crucially this introduces BoehmGCStackAllocator, but it also
adds a bunch of wiring to avoid making libutil depend on bdw-gc.
Part of the solutions for #4178, #4200
Otherwise the result of the printing can't be parsed back correctly by
Nix (because the unescaped `${` will be parsed as the begining of an
anti-quotation).
Fix#3989
The change in 626200713b didn't account
for when the number of auto arguments is bigger than the number of
formal arguments. This causes the following:
$ nix-instantiate --eval -E '{ ... }@args: args.foo' --argstr foo foo
nix-instantiate: src/libexpr/attr-set.hh:55: void nix::Bindings::push_back(const nix::Attr&): Assertion `size_ < capacity_' failed.
Aborted (core dumped)
The command line options --arg and --argstr that are used by a bunch of
CLI commands to pass arguments to top-level functions in files go
through the same code-path as auto-calling top-level functions with
their default arguments - this, however, was only passing the arguments
that were *explicitly* mentioned in the formals of the function - in the
case of an as-pattern with an ellipsis (eg args @ { ... }) extra passed
arguments would get omitted. This fixes that to instead pass *all*
specified auto args in the case that our function has an ellipsis.
Fixes#598
This allows interactively inspecting the state of the evaluator at the
point of failure.
Example:
$ nix eval path:///home/eelco/Dev/nix/flake2#modules.hello-closure._final --start-repl-on-eval-errors
error: --- TypeError -------------------------------------------------------------------------------------------------------------------------------------------------------------------- nix
at: (20:53) in file: /nix/store/4264z41dxfdiqr95svmpnxxxwhfplhy0-source/flake.nix
19|
20| _final = builtins.foldl' (xs: mod: xs // (mod._module.config { config = _final; })) _defaults _allModules;
| ^
21| };
attempt to call something which is not a function but a set
Starting REPL to allow you to inspect the current state of the evaluator.
The following extra variables are in scope: arg, fun
Welcome to Nix version 2.4. Type :? for help.
nix-repl> fun
error: --- EvalError -------------------------------------------------------------------------------------------------------------------------------------------------------------------- nix
at: (150:28) in file: /nix/store/4264z41dxfdiqr95svmpnxxxwhfplhy0-source/flake.nix
149|
150| tarballClosure = (module {
| ^
151| extends = [ self.modules.derivation ];
attribute 'derivation' missing
nix-repl> :t fun
a set
nix-repl> builtins.attrNames fun
[ "tarballClosure" ]
nix-repl>
If you do a fetchTree on a Git repository, whether the result contains
a revCount attribute should not depend on whether that repository
happens to be a shallow clone or not. That would complicate caching a
lot and would be semantically messy. So applying fetchTree/fetchGit to
a shallow repository is now an error unless you pass the attribute
'shallow = true'. If 'shallow = true', we don't return revCount, even
if the repository is not actually shallow.
Note that Nix itself is not doing shallow clones at the moment. But it
could do so as an optimisation if the user specifies 'shallow = true'.
Issue #2988.
Includes the expression of the condition in the assertion message if
the assertion failed, making assertions much easier to debug. eg.
error: assertion (withPython -> (python2Packages != null)) failed at pkgs/tools/security/nmap/default.nix:11:1
This prevents them from being inlined. On gcc 9, this reduces the
stack size needed for
nix-instantiate '<nixpkgs>' -A texlive.combined.scheme-full --dry-run
from 12.9 MiB to 4.8 MiB.
Most functions now take a StorePath argument rather than a Path (which
is just an alias for std::string). The StorePath constructor ensures
that the path is syntactically correct (i.e. it looks like
<store-dir>/<base32-hash>-<name>). Similarly, functions like
buildPaths() now take a StorePathWithOutputs, rather than abusing Path
by adding a '!<outputs>' suffix.
Note that the StorePath type is implemented in Rust. This involves
some hackery to allow Rust values to be used directly in C++, via a
helper type whose destructor calls the Rust type's drop()
function. The main issue is the dynamic nature of C++ move semantics:
after we have moved a Rust value, we should not call the drop function
on the original value. So when we move a value, we set the original
value to bitwise zero, and the destructor only calls drop() if the
value is not bitwise zero. This should be sufficient for most types.
Also lots of minor cleanups to the C++ API to make it more modern
(e.g. using std::optional and std::string_view in some places).
The FunctionCallTrace object consumes a few hundred bytes of stack
space, even when tracing is disabled. This was causing stack overflows:
$ nix-instantiate '<nixpkgs> -A texlive.combined.scheme-full --dry-run
error: stack overflow (possible infinite recursion)
This is with the default stack size of 8 MiB.
Putting the object on the heap reduces stack usage to < 5 MiB.
With this patch, and this file I called `log.py`:
#!/usr/bin/env nix-shell
#!nix-shell -i python3 -p python3 --pure
import sys
from pprint import pprint
stack = []
timestack = []
for line in open(sys.argv[1]):
components = line.strip().split(" ", 2)
if components[0] != "function-trace":
continue
direction = components[1]
components = components[2].rsplit(" ", 2)
loc = components[0]
_at = components[1]
time = int(components[2])
if direction == "entered":
stack.append(loc)
timestack.append(time)
elif direction == "exited":
dur = time - timestack.pop()
vst = ";".join(stack)
print(f"{vst} {dur}")
stack.pop()
and:
nix-instantiate --trace-function-calls -vvvv ../nixpkgs/pkgs/top-level/release.nix -A unstable > log.matthewbauer 2>&1
./log.py ./log.matthewbauer > log.matthewbauer.folded
flamegraph.pl --title matthewbauer-post-pr log.matthewbauer.folded > log.matthewbauer.folded.svg
I can make flame graphs like: http://gsc.io/log.matthewbauer.folded.svg
---
Includes test cases around function call failures and tryEval. Uses
RAII so the finish is always called at the end of the function.
In EvalState::checkSourcePath, the path is checked against the list of
allowed paths first and later it's checked again *after* resolving
symlinks.
The resolving of the symlinks is done via canonPath, which also strips
out "../" and "./". However after the canonicalisation the error message
pointing out that the path is not allowed prints the symlink target in
the error message.
Even if we'd suppress the message, symlink targets could still be leaked
if the symlink target doesn't exist (in this case the error is thrown in
canonPath).
So instead, we now do canonPath() without symlink resolving first before
even checking against the list of allowed paths and then later do the
symlink resolving and checking the allowed paths again.
The first call to canonPath() should get rid of all the "../" and "./",
so in theory the only way to leak a symlink if the attacker is able to
put a symlink in one of the paths allowed by restricted evaluation mode.
For the latter I don't think this is part of the threat model, because
if the attacker can write to that path, the attack vector is even
larger.
Signed-off-by: aszlig <aszlig@nix.build>
