calling GC_malloc for each value is significantly more expensive than
allocating a bunch of values at once with GC_malloc_many. "a bunch" here
is a GC block size, ie 16KiB or less.
this gives a 1.5% performance boost when evaluating our nixos system.
tested with
nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
# on master
Time (mean ± σ): 3.335 s ± 0.007 s [User: 2.774 s, System: 0.293 s]
Range (min … max): 3.315 s … 3.347 s 50 runs
# with this change
Time (mean ± σ): 3.288 s ± 0.006 s [User: 2.728 s, System: 0.292 s]
Range (min … max): 3.274 s … 3.307 s 50 runs
We now parse function applications as a vector of arguments rather
than as a chain of binary applications, e.g. 'substring 1 2 "foo"' is
parsed as
ExprCall { .fun = <substring>, .args = [ <1>, <2>, <"foo"> ] }
rather than
ExprApp (ExprApp (ExprApp <substring> <1>) <2>) <"foo">
This allows primops to be called immediately (if enough arguments are
supplied) without having to allocate intermediate tPrimOpApp values.
On
$ nix-instantiate --dry-run '<nixpkgs/nixos/release-combined.nix>' -A nixos.tests.simple.x86_64-linux
this gives a substantial performance improvement:
user CPU time: median = 0.9209 mean = 0.9218 stddev = 0.0073 min = 0.9086 max = 0.9340 [rejected, p=0.00000, Δ=-0.21433±0.00677]
elapsed time: median = 1.0585 mean = 1.0584 stddev = 0.0024 min = 1.0523 max = 1.0623 [rejected, p=0.00000, Δ=-0.20594±0.00236]
because it reduces the number of tPrimOpApp allocations from 551990 to
42534 (i.e. only small minority of primop calls are partially
applied) which in turn reduces time spent in the garbage collector.
Rather than having them plain strings scattered through the whole
codebase, create an enum containing all the known experimental features.
This means that
- Nix can now `warn` when an unkwown experimental feature is passed
(making it much nicer to spot typos and spot deprecated features)
- It’s now easy to remove a feature altogether (once the feature isn’t
experimental anymore or is dropped) by just removing the field for the
enum and letting the compiler point us to all the now invalid usages
of it.
I found it somewhat confusing to have an error like
error: attribute 'getFlake' missing
if the required experimental-feature (`flakes`) is not enabled. Instead,
I'd expect Nix to throw an error just like it's the case when using e.g. `nix
flake` without `flakes` being enabled.
With this change, the error looks like this:
$ nix-instantiate -E 'builtins.getFlake "nixpkgs"'
error: Cannot call 'builtins.getFlake' because experimental Nix feature 'flakes' is disabled. You can enable it via '--extra-experimental-features flakes'.
at «string»:1:1:
1| builtins.getFlake "nixpkgs"
| ^
I didn't use `settings.requireExperimentalFeature` here on purpose
because this doesn't contain a position. Also, it doesn't seem as if we
need to catch the error and check for the missing feature here since
this already happens at evaluation time.
This fixes a use-after-free bug:
1. s = new EvalState();
2. callFlake()
3. static vCallFlake now references s
4. delete s;
5. s2 = new EvalState();
6. callFlake()
7. static vCallFlake still references s
8. crash
Nix 2.3 did not have a problem with recreating EvalState.
This fixes a class of crashes and introduces ptr<T> to make the
code robust against this failure mode going forward.
Thanks regnat for the idea of a ref<T> without overhead!
Closes#4895Closes#4893Closes#5127Closes#5113
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).
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.
This allows using an arbitrary "provides" attribute from the specified
flake. For example:
nix build --flake nixpkgs packages.hello
(Maybe provides.packages should be used for consistency...)
We want to encourage a brave new world of hermetic evaluation for
source-level reproducibility, so flakes should not poke around in the
filesystem outside of their explicit dependencies.
Note that the default installation source remains impure in that it
can refer to mutable flakes, so "nix build nixpkgs.hello" still works
(and fetches the latest nixpkgs, unless it has been pinned by the
user).
A problem with pure evaluation is that builtins.currentSystem is
unavailable. For the moment, I've hard-coded "x86_64-linux" in the
nixpkgs flake. Eventually, "system" should be a flake function
argument.
If the Env denotes a 'with', then values[0] may be an Expr* cast to a
Value*. For code that generically traverses Values/Envs, it's useful
to know this.
In this mode, the following restrictions apply:
* The builtins currentTime, currentSystem and storePath throw an
error.
* $NIX_PATH and -I are ignored.
* fetchGit and fetchMercurial require a revision hash.
* fetchurl and fetchTarball require a sha256 attribute.
* No file system access is allowed outside of the paths returned by
fetch{Git,Mercurial,url,Tarball}. Thus 'nix build -f ./foo.nix' is
not allowed.
Thus, the evaluation result is completely reproducible from the
command line arguments. E.g.
nix build --pure-eval '(
let
nix = fetchGit { url = https://github.com/NixOS/nixpkgs.git; rev = "9c927de4b179a6dd210dd88d34bda8af4b575680"; };
nixpkgs = fetchGit { url = https://github.com/NixOS/nixpkgs.git; ref = "release-17.09"; rev = "66b4de79e3841530e6d9c6baf98702aa1f7124e4"; };
in (import (nix + "/release.nix") { inherit nix nixpkgs; }).build.x86_64-linux
)'
The goal is to enable completely reproducible and traceable
evaluation. For example, a NixOS configuration could be fully
described by a single Git commit hash. 'nixos-rebuild' would do
something like
nix build --pure-eval '(
(import (fetchGit { url = file:///my-nixos-config; rev = "..."; })).system
')
where the Git repository /my-nixos-config would use further fetchGit
calls or Git externals to fetch Nixpkgs and whatever other
dependencies it has. Either way, the commit hash would uniquely
identify the NixOS configuration and allow it to reproduced.
Functions like copyClosure() had 3 bool arguments, which creates a
severe risk of mixing up arguments.
Also, implement copyClosure() using copyPaths().
Previously, all derivation attributes had to be coerced into strings
so that they could be passed via the environment. This is lossy
(e.g. lists get flattened, necessitating configureFlags
vs. configureFlagsArray, of which the latter cannot be specified as an
attribute), doesn't support attribute sets at all, and has size
limitations (necessitating hacks like passAsFile).
This patch adds a new mode for passing attributes to builders, namely
encoded as a JSON file ".attrs.json" in the current directory of the
builder. This mode is activated via the special attribute
__structuredAttrs = true;
(The idea is that one day we can set this in stdenv.mkDerivation.)
For example,
stdenv.mkDerivation {
__structuredAttrs = true;
name = "foo";
buildInputs = [ pkgs.hello pkgs.cowsay ];
doCheck = true;
hardening.format = false;
}
results in a ".attrs.json" file containing (sans the indentation):
{
"buildInputs": [],
"builder": "/nix/store/ygl61ycpr2vjqrx775l1r2mw1g2rb754-bash-4.3-p48/bin/bash",
"configureFlags": [
"--with-foo",
"--with-bar=1 2"
],
"doCheck": true,
"hardening": {
"format": false
},
"name": "foo",
"nativeBuildInputs": [
"/nix/store/10h6li26i7g6z3mdpvra09yyf10mmzdr-hello-2.10",
"/nix/store/4jnvjin0r6wp6cv1hdm5jbkx3vinlcvk-cowsay-3.03"
],
"propagatedBuildInputs": [],
"propagatedNativeBuildInputs": [],
"stdenv": "/nix/store/f3hw3p8armnzy6xhd4h8s7anfjrs15n2-stdenv",
"system": "x86_64-linux"
}
"passAsFile" is ignored in this mode because it's not needed - large
strings are included directly in the JSON representation.
It is up to the builder to do something with the JSON
representation. For example, in bash-based builders, lists/attrsets of
string values could be mapped to bash (associative) arrays.
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).
That is, unless --file is specified, the Nix search path is
synthesized into an attribute set. Thus you can say
$ nix build nixpkgs.hello
assuming $NIX_PATH contains an entry of the form "nixpkgs=...". This
is more verbose than
$ nix build hello
but is less ambiguous.
Thus, -I / $NIX_PATH entries are now downloaded only when they are
needed for evaluation. An error to download an entry is a non-fatal
warning (just like non-existant paths).
This does change the semantics of builtins.nixPath, which now returns
the original, rather than resulting path. E.g., before we had
[ { path = "/nix/store/hgm3yxf1lrrwa3z14zpqaj5p9vs0qklk-nixexprs.tar.xz"; prefix = "nixpkgs"; } ... ]
but now
[ { path = "https://nixos.org/channels/nixos-16.03/nixexprs.tar.xz"; prefix = "nixpkgs"; } ... ]
Fixes#792.
Also, move a few free-standing functions into StoreAPI and Derivation.
Also, introduce a non-nullable smart pointer, ref<T>, which is just a
wrapper around std::shared_ptr ensuring that the pointer is never
null. (For reference-counted values, this is better than passing a
"T&", because the latter doesn't maintain the refcount. Usually, the
caller will have a shared_ptr keeping the value alive, but that's not
always the case, e.g., when passing a reference to a std::thread via
std::bind.)
For example, "${{ foo = "bar"; __toString = x: x.foo; }}" evaluates
to "bar".
With this, we can delay calling functions like mkDerivation,
buildPythonPackage, etc. until we actually need a derivation, enabling
overrides and other modifications to happen by simple attribute set
update.
This modification moves Attr and Bindings structures into their own header
file which is dedicated to the attribute set representation. The goal of to
isolate pieces of code which are related to the attribute set
representation. Thus future modifications of the attribute set
representation will only have to modify these files, and not every other
file across the evaluator.
If ‘--option restrict-eval true’ is given, the evaluator will throw an
exception if an attempt is made to access any file outside of the Nix
search path. This is primarily intended for Hydra, where we don't want
people doing ‘builtins.readFile ~/.ssh/id_dsa’ or stuff like that.
With this, attribute sets with a `__functor` attribute can be applied
just like normal functions. This can be used to attach arbitrary
metadata to a function without callers needing to treat it specially.
Nix search path lookups like <nixpkgs> are now desugared to ‘findFile
nixPath <nixpkgs>’, where ‘findFile’ is a new primop. Thus you can
override the search path simply by saying
let
nixPath = [ { prefix = "nixpkgs"; path = "/my-nixpkgs"; } ];
in ... <nixpkgs> ...
In conjunction with ‘scopedImport’ (commit
c273c15cb1), the Nix search path can be
propagated across imports, e.g.
let
overrides = {
nixPath = [ ... ] ++ builtins.nixPath;
import = fn: scopedImport overrides fn;
scopedImport = attrs: fn: scopedImport (overrides // attrs) fn;
builtins = builtins // overrides;
};
in scopedImport overrides ./nixos
‘scopedImport’ works like ‘import’, except that it takes a set of
attributes to be added to the lexical scope of the expression,
essentially extending or overriding the builtin variables. For
instance, the expression
scopedImport { x = 1; } ./foo.nix
where foo.nix contains ‘x’, will evaluate to 1.
This has a few applications:
* It allows getting rid of function argument specifications in package
expressions. For instance, a package expression like:
{ stdenv, fetchurl, libfoo }:
stdenv.mkDerivation { ... buildInputs = [ libfoo ]; }
can now we written as just
stdenv.mkDerivation { ... buildInputs = [ libfoo ]; }
and imported in all-packages.nix as:
bar = scopedImport pkgs ./bar.nix;
So whereas we once had dependencies listed in three places
(buildInputs, the function, and the call site), they now only need
to appear in one place.
* It allows overriding builtin functions. For instance, to trace all
calls to ‘map’:
let
overrides = {
map = f: xs: builtins.trace "map called!" (map f xs);
# Ensure that our override gets propagated by calls to
# import/scopedImport.
import = fn: scopedImport overrides fn;
scopedImport = attrs: fn: scopedImport (overrides // attrs) fn;
# Also update ‘builtins’.
builtins = builtins // overrides;
};
in scopedImport overrides ./bla.nix
* Similarly, it allows extending the set of builtin functions. For
instance, during Nixpkgs/NixOS evaluation, the Nixpkgs library
functions could be added to the default scope.
There is a downside: calls to scopedImport are not memoized, unlike
import. So importing a file multiple times leads to multiple parsings
/ evaluations. It would be possible to construct the AST only once,
but that would require careful handling of variables/environments.
Maximilian Bosch
Robert Hensing
