this slightly increases the amount of memory used for any given symbol, but this
increase is more than made up for if the symbol is referenced more than once in
the EvalState that holds it. on average every symbol should be referenced at
least twice (once to introduce a binding, once to use it), so we expect no
increase in memory on average.
symbol tables are limited to 2³² entries like position tables, and similar
arguments apply to why overflow is not likely: 2³² symbols would require as many
string instances (at 24 bytes each) and map entries (at 24 bytes or more each,
assuming that the map holds on average at most one item per bucket as the docs
say). a full symbol table would require at least 192GB of memory just for
symbols, which is well out of reach. (an ofborg eval of nixpks today creates
less than a million symbols!)
Pos objects are somewhat wasteful as they duplicate the origin file name and
input type for each object. on files that produce more than one Pos when parsed
this a sizeable waste of memory (one pointer per Pos). the same goes for
ptr<Pos> on 64 bit machines: parsing enough source to require 8 bytes to locate
a position would need at least 8GB of input and 64GB of expression memory. it's
not likely that we'll hit that any time soon, so we can use a uint32_t index to
locate positions instead.
We now memoize on Bindings / list element vectors rather than Values,
so that e.g. two Values that point to the same Bindings will be
printed only once.
constructing an ostringstream for non-string concats (like integer addition) is
a small constant cost that we can avoid. for string concats we can keep all the
string temporaries we get from coerceToString and concatenate them in one go,
which saves a lot of intermediate temporaries and copies in ostringstream. we
can also avoid copying the concatenated string again by directly allocating it
in GC memory and moving ownership of the concatenated string into the target
value.
saves about 2% on system eval.
before:
Benchmark 1: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.837 s ± 0.031 s [User: 2.562 s, System: 0.191 s]
Range (min … max): 2.796 s … 2.892 s 20 runs
after:
Benchmark 1: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.790 s ± 0.035 s [User: 2.532 s, System: 0.187 s]
Range (min … max): 2.722 s … 2.836 s 20 runs
Previously you had to remember to call value->attrs->sort() after
populating value->attrs. Now there is a BindingsBuilder helper that
wraps Bindings and ensures that sort() is called before you can use
it.
This function is very useful in nixpkgs, but its implementation in Nix
itself is rather slow due to it requiring a lot of attribute set and
list appends.
- This change applies to builtins.toXML and inner workings
- Proof of concept:
```nix
let e = builtins.toXML e; in e
```
- Before:
```
$ nix-instantiate --eval poc.nix
error: infinite recursion encountered
```
- After:
```
$ nix-instantiate --eval poc.nix
error: infinite recursion encountered
at /data/github/kamadorueda/nix/poc.nix:1:9:
1| let e = builtins.toXML e; in e
|
```
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.
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
Using a 64bit integer on 32bit systems will come with a bit of a
performance overhead, but given that Nix doesn't use a lot of integers
compared to other types, I think the overhead is negligible also
considering that 32bit systems are in decline.
The biggest advantage however is that when we use a consistent integer
size across all platforms it's less likely that we miss things that we
break due to that. One example would be:
https://github.com/NixOS/nixpkgs/pull/44233
On Hydra it will evaluate, because the evaluator runs on a 64bit
machine, but when evaluating the same on a 32bit machine it will fail,
so using 64bit integers should make that consistent.
While the change of the type in value.hh is rather easy to do, we have a
few more options available for doing the conversion in the lexer:
* Via an #ifdef on the architecture and using strtol() or strtoll()
accordingly depending on which architecture we are. For the #ifdef
we would need another AX_COMPILE_CHECK_SIZEOF in configure.ac.
* Using istringstream, which would involve copying the value.
* As we're already using boost, lexical_cast might be a good idea.
Spoiler: I went for the latter, first of all because lexical_cast does
have an overload for const char* and second of all, because it doesn't
involve copying around the input string. Also, because istringstream
seems to come with a bigger overhead than boost::lexical_cast:
https://www.boost.org/doc/libs/release/doc/html/boost_lexical_cast/performance.html
The first method (still using strtol/strtoll) also wasn't something I
pursued further, because it is also locale-aware which I doubt is what
we want, given that the regex for int is [0-9]+.
Signed-off-by: aszlig <aszlig@nix.build>
Fixes: #2339
Because config.h can #define things like _FILE_OFFSET_BITS=64 and not
every compilation unit includes config.h, we currently compile half of
Nix with _FILE_OFFSET_BITS=64 and other half with _FILE_OFFSET_BITS
unset. This causes major havoc with the Settings class on e.g. 32-bit ARM,
where different compilation units disagree with the struct layout.
E.g.:
diff --git a/src/libstore/globals.cc b/src/libstore/globals.cc
@@ -166,6 +166,8 @@ void Settings::update()
_get(useSubstitutes, "build-use-substitutes");
+ fprintf(stderr, "at Settings::update(): &useSubstitutes = %p\n", &nix::settings.useSubstitutes);
_get(buildUsersGroup, "build-users-group");
diff --git a/src/libstore/remote-store.cc b/src/libstore/remote-store.cc
+++ b/src/libstore/remote-store.cc
@@ -138,6 +138,8 @@ void RemoteStore::initConnection(Connection & conn)
void RemoteStore::setOptions(Connection & conn)
{
+ fprintf(stderr, "at RemoteStore::setOptions(): &useSubstitutes = %p\n", &nix::settings.useSubstitutes);
conn.to << wopSetOptions
Gave me:
at Settings::update(): &useSubstitutes = 0xb6e5c5cb
at RemoteStore::setOptions(): &useSubstitutes = 0xb6e5c5c7
That was not a fun one to debug!
The implementation of "partition" in Nixpkgs is O(n^2) (because of the
use of ++), and for some reason was causing stack overflows in
multi-threaded evaluation (not sure why).
This reduces "nix-env -qa --drv-path" runtime by 0.197s and memory
usage by 298 MiB (in non-Boehm mode).
The value pointers of lists with 1 or 2 elements are now stored in the
list value itself. In particular, this makes the "concatMap (x: if
cond then [(f x)] else [])" idiom cheaper.
Code that links to libnixexpr (e.g. plugins loaded with importNative, or
nix-exec) may want to provide custom value types and operations on
values of those types. For example, nix-exec is currently using sets
where a custom IO value type would be more appropriate. This commit
provides a generic hook for such types in the form of tExternal and the
ExternalBase virtual class, which contains all functions necessary for
libnixexpr's type-polymorphic functions (e.g. `showType`) to be
implemented.
Clearing v.app.right was not enough, because the length field of a
list only takes 32 bits, so the most significant 32 bits of v.app.left
(a.k.a. v.thunk.env) would remain. This could cause Boehm GC to
interpret it as a valid pointer.
This change reduces maximum RSS for evaluating the ‘tested’ job in
nixos/release-small.nix from 1.33 GiB to 0.80 GiB, and runtime by
about 8%.
