Our bootstrap-files unpacker has always relied on a lot of unstated
assumptions, one of them being that every library has a DT_NEEDED
for librt.so, so patchelf'ing something into the RUNPATH into
librt.so means that it will be searched for every library load in
all of the bootstrap-files.
Unfortunately that assumption is not true for libgcc.
This causes problems, because patchelf links against libgcc (and
against libstdc++, which links against libgcc). So we can't use
patchelf on libgcc, because it needs libgcc, so patchelf doesn't
work until libgcc is patchelfed.
The robust solution here is to use static linking for the copy of
patchelf that is shipped with the bootstrap-files. We don't have to
go all the way to a statically linked libc; just -static-libgcc and
-static-libstdc++ are enough to break the circular dependency.
Right now our bootstrapFiles-selecting algorithm uses the
`loongson2f.nix` bootstrapFiles (which were not built by Hydra).
These bootstrapFiles don't work anymore. They were added in 2010 by
40405d03ac.
This commit causes mipsel-linux native builds to use the Hydra-built
bootstrap files from this PR instead:
https://github.com/NixOS/nixpkgs/pull/183487
The NIX_LIB64|32_IN_SELF_RPATH environment variables control whether
to add lib64 and lib32 to rpaths. However, they're set depending
on the build paltform, not the target platform and thus their values
are incorrect for for cross-builds.
On the other hand, setting them according to the build platform introduce
pointless differences in build outputs; see #221350 for details.
This change fixes the issues by boldly removes the NIX_LIB*_IN_SELF_RPATH
facility altogether, in the hope that it is no longer necessary. They
were introduced in 2009, long before nixpkgs had good support for
cross-builds.
Fixes#221350
See https://github.com/NixOS/nixpkgs/pull/222792#pullrequestreview-1356114111
You can't just `lib.filter _ lib.systems.all` -- that throws away
important information, leading to nixpkgs disagreeing with itself
like this:
```
$ NIXPKGS_ALLOW_BROKEN=1 nix-instantiate . -A pkgsStatic.systemd
error: Package ‘systemd-252.5’ in ... is only supported on ... x86_64-linux but not on requested x86_64-linux, refusing to evaluate.
```
After:
```
$ NIXPKGS_ALLOW_BROKEN=1 nix-instantiate . -A pkgsStatic.systemd
error: Package ‘systemd-252.5’ in ... is not available on the requested hostPlatform:
hostPlatform.config = "x86_64-unknown-linux-musl"
package.meta.platforms = [
"aarch64-linux"
"armv5tel-linux"
"armv6l-linux"
"armv7a-linux"
"armv7l-linux"
"i686-linux"
"m68k-linux"
"microblaze-linux"
"microblazeel-linux"
"mipsel-linux"
"mips64el-linux"
"powerpc64-linux"
"powerpc64le-linux"
"riscv32-linux"
"riscv64-linux"
"s390-linux"
"s390x-linux"
"x86_64-linux"
]
package.meta.badPlatforms = [
{
isStatic = true;
parsed = { };
}
]
, refusing to evaluate.
```
The primary motivating example is openssl:
Before the change full package build took 1m54s minutes.
After the change full package build takes 59s.
About a 2x speedup.
The difference is visible because openssl builds hundreds of manpages
spawning a perl process per manual in `install` phase. Such a workload
is very easy to parallelize.
Another example would be `autotools`+`libtool` based build system where
install step requires relinking. The more binaries there are to relink
the more gain it will be to do it in parallel.
The change enables parallel installs by default only for buiilds that
already have parallel builds enabled. There is a high chance those build
systems already handle parallelism well but some packages will fail.
Consistently propagated the enableParallelBuilding to:
- cmake (enabled by default, similar to builds)
- ninja (set parallelism explicitly, don't rely on default)
- bmake (enable when requested)
- scons (enable when requested)
- meson (set parallelism explicitly, don't rely on default)
- waf (set parallelism explicitly, don't rely on default)
- qmake-4/5/6 (enable by default, similar to builds)
- xorg (always enable, similar to builds)
Hydra job building them: https://hydra.nixos.org/build/208909151
The bootstrap files can be reproduced on the commit 21ec906463, e.g. by:
cat $(nix-build pkgs/top-level/release.nix -QA stdenvBootstrapTools.aarch64-linux.dist)/nix-support/hydra-build-products
file tarball /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/bootstrap-tools.tar.xz
file busybox /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/busybox
and the hashes as well:
nix hash file /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/bootstrap-tools.tar.xz
sha256-aJvtsWeuQHbb14BGZ2EiOKzjQn46h3x3duuPEawG0eE=
nix hash path /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/busybox
sha256-0MuIeQlBUaeisqoFSu8y+8oB6K4ZG5Lhq8RcS9JqkFQ=
You can check this on any machine, as the builds are on cache.nixos.org
but also you can reproduce the hashes when rebuilt on aarch64-linux HW.
This was disabled here: b86e62d30d (diff-282a02cc3871874f16401347d8fadc90d59d7ab11f6a99eaa5173c3867e1a160)
h/t to @teh: b86e62d30d (commitcomment-77916294)
for pointing out that the failure that @matthewbauer was
seeing was caused by the `separate-debug-info.sh` `build-id` length
requirement that #146275 will relax
`lld` has had `--build-id` support dating back to LLVM4: https://reviews.llvm.org/D18091
This predates every `llvmPackages_` version currently in nixpkgs (and
certainly every version actually still used in `useLLVM` stdenvs) so
with the previous commit (asking `ld` for sufficiently long SHA1 hashes)
I think we can safely enable `separateDebugInfo` when using LLVM
bintools.
with structuredAttrs lists will be bash arrays which cannot be exported
which will be a issue with some patches and some wrappers like cc-wrapper
this makes it clearer that NIX_CFLAGS_COMPILE must be a string as lists
in env cause a eval failure
PR #208478 added a lot of documentation about which packages were
rebuilt in each stage of the stdenv bootstrap. However nothing
checks that these comments agree with reality; they can bitrot over
time. This PR rewrites those comments as assertions, so they cannot
bitrot.
This conversion did expose some ambiguity in our scheme for naming
the stages. Suppose that `pkgs.stdenv.name=="stdenv-stage4", then
which of these is "the stage4 coreutils"?
```
pkgs.coreutils
pkgs.stdenv.__bootPackages.coreutils
```
The choice is arbitrary, and both choices have confusing corner
cases. We should revisit this at some point.
Hydra job building them: https://hydra.nixos.org/build/208909151
The bootstrap files can be reproduced on the parent commit, e.g. by:
cat $(nix-build pkgs/top-level/release.nix -QA stdenvBootstrapTools.aarch64-linux.dist)/nix-support/hydra-build-products
file tarball /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/bootstrap-tools.tar.xz
file busybox /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/busybox
and the hashes as well:
nix hash file /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/bootstrap-tools.tar.xz
sha256-aJvtsWeuQHbb14BGZ2EiOKzjQn46h3x3duuPEawG0eE=
nix hash path /nix/store/kdpbw0plmjqlafjnpbz31ja51m4bd2dk-stdenv-bootstrap-tools/on-server/busybox
sha256-0MuIeQlBUaeisqoFSu8y+8oB6K4ZG5Lhq8RcS9JqkFQ=
You can check this on any machine, as the builds are on cache.nixos.org
but also you can reproduce the hashes when rebuilt on aarch64-linux HW.
See docs.
Follow-up work:
- Existing packages should be converted
- `defaultPkgConfigPackages` should assert on `meta.pkgConfigModules`
and let `tests.pkg-config` alone test the build results.
CC @sternenseemann
Co-authored-by: Robert Hensing <roberth@users.noreply.github.com>
platform.uname.processor seems to be what we want in many more cases
than what we were using before — it does the right thing for aarch64,
x86_64, riscv32, riscv64, mips, mips64, powerpc, and powerpc64 (the
latter three of which were broken before).
This fixes cross-compilation of systemd for PowerPC/POWER platforms.
Derivations listed as disallowedReferences or disallowedRequisites,
currently end up as build-time dependencies.
This is problematic since the disallowed derivations will be built by nix as
build-time dependencies, while those derivations might take a very long time
to build, or might not even build successfully on the platform used.
However, in order to scan for disallowed references in the final output,
knowing the out path is sufficient, and the out path can be calculated from
the derivation without needing to build it, saving time and resources.
While the problem is less severe for allowedReferences and allowedRequisites,
since we want the derivation to be built eventually, we would still like to
get the error early and without having to wait while nix builds a derivation
that might not be used (e.g. if we listed the wrong one).
A few potentially disruptive changes:
- binutils does not embed ${binutils-unwrapped}/lib as a default library
search path anymore. This will cause link failures for -lbfd -lopcodes
users that did not declare their dependency on those libraries. They
will need to add `libbfd` and `libopcodes` attributes to build inputs.
- `libbfd` and `libopcodes` attributes now just reference
`binutils-unwrapped.{dev,lib}` pair of attributes without patching
`binutils` build system.
We don't patch build system anymore and use multiple outputs out of
existing `binutils` build. That makes the result more maintainable: no
need to handle ever growing list of dependencied of `libbfd`. This time
new addition was `libsframe`.
To accomodate `out` / `lib` output split I had to remove `lib` -> `bin`
backreference by removing legacy lookup path for plugins.
I also did not enable `zstd` just yet as `nixpkgs` version of `zstd`
package pulls in `cmake` into bootstrap sequence.
Changes: https://lists.gnu.org/archive/html/info-gnu/2023-01/msg00003.html
FreeBSD doesn't use LLVM's cxxabi implementation, for backwards
compatibility reasons. Software expects the libcxxrt API when
building on FreeBSD. This fixes the build of
pkgsCross.x86_64-freebsd.boost.
Some other packages, for example ruby gems via buildRubyGem, use a
variable called "type" internally, which is overwritten here and
causes failures like:
failure: $gempkg path unspecified
Fix for changes in 11c3127e38.
this is intentional to support both structuredAttrs and non
In pkgs/stdenv/generic/setup.sh line 614:
for pkg in ${depsBuildBuild[@]} ${depsBuildBuildPropagated[@]}; do
^------------------^ SC2068 (error): Double quote array expansions to avoid re-splitting elements.
In pkgs/stdenv/generic/setup.sh line 521:
local varRef="$varVar[$((targetOffset - hostOffset))]"
^-- SC1087 (error): Use braces when expanding arrays, e.g. ${array[idx]} (or ${var}[.. to quiet).
exit -1 == exit 255 but we don't have a reason to use 255
In pkgs/stdenv/generic/setup.sh line 518:
(( hostOffset <= targetOffset )) || exit -1
^-- SC2242 (error): Can only exit with status 0-255. Other data should be wri
tten to stdout/stderr.
`builtins.baseNameOf` retains any string context, causing the test
derivation to incorrectly depend on `pkgs.glibc`. All we really want is
to know what the dynamicLinker is called, but we don't need it to be
present in store.
Thanks to Adam Joseph for spotting this.
we use [*] to support structuredAttrs and non
In pkgs/stdenv/generic/setup.sh line 1542:
for curPhase in ${phases[*]}; do
^----------^ SC2048 (warning): Use "${array[@]}" (with quotes) to prevent whitespace problems.
In pkgs/stdenv/generic/setup.sh line 101:
source "$hookName"
^---------^ SC1090 (warning): ShellCheck can't follow non-constant source. Use a directive to specify location.
In pkgs/stdenv/generic/setup.sh line 166:
mkdir -p "$out/nix-support"
^--^ SC2154 (warning): out is referenced but not assigned.
In pkgs/stdenv/generic/setup.sh line 407:
PATH=
^--^ SC2123 (warning): PATH is the shell search path. Use another name.
In pkgs/stdenv/generic/setup.sh line 452:
declare -a pkgBuildAccumVars=(pkgsBuildBuild pkgsBuildHost pkgsBuildTarget)
^---------------^ SC2034 (warning): pkgBuildAccumVars appears unused. Verify use (or export if used e
xternally).
because pkgBuildAccumVars is used
In pkgs/stdenv/generic/setup.sh line 235:
nameref="$* ${nameref-}"
^-----^ SC2178 (warning): Variable was used as an array but is now assigned a string.
because we theres a useArray conditional
In pkgs/stdenv/generic/setup.sh line 36:
: ${outputs:=out}
^-------------^ SC2223 (info): This default assignment may cause DoS due to globbing. Quote it.
After https://github.com/NixOS/nixpkgs/pull/209054 we started moving
libstdc++.so out of default glibc's paths. This exposed bootstrap
tools build failure as:
$ nix build --no-link -f ./pkgs/stdenv/linux/make-bootstrap-tools.nix
...
>
.../bin/bar: error while loading shared libraries: libstdc++.so.6:
cannot open shared object file: No such file or directory
Note that bootstrap itself did not break. The change only expands
handcrafted `-rpath` entries.
The stdenv bootstrap creates several wrappers around binutils, and
gives them the exact same drvName as the binutils package itself.
These wrappers cost almost nothing to create (they are just file
copies and patchelf runs, not builds), so we should distinguish them
from expensive binutils builds with a unique pname. This commit
does that.
`hasUnsupportedPlatform` was not updated with #37395, so it does not
understand attrsets in `meta.[bad]platforms`. In particular,
attrsets in `meta.badPlatforms` will "fail open" and be ignored.
Let's use `lib.meta.availableOn` instead of duplicating its logic.
Thanks to @alyssais for [noticing][1].
[1][https://github.com/NixOS/nixpkgs/pull/194148#discussion_r990817610]
Co-authored-by: sternenseemann <sternenseemann@systemli.org>
Right now we build gettext several times during the bootstrap.
Gettext's build process is "embarrassingly serial", so avoiding
rebuilding it speeds things up considerably.
otherwise the build just fails with 'make: *** No rule to make target 'install'. Stop.'
and update buildPhase message
i don't know if the 'makefile may have been created in buildPhase' is
true but i guess it might be possible
This change allows building new gcc during bootstrap without fear of
pulling in outdated libstdc++.so after g++ switched from bootstrapTools
to freshly built g++.
Noticed when tried to add early bootstrap stage to rebuild `gcc` before
`glibc` is fully untangled from `bootstrapTools` as a failure to built
`binutils`:
ld: dwp.o: in function `__gnu_cxx::new_allocator<gold::Dwp_output_file::Contribution>::allocate(unsigned long, void const*)':
/nix/store/...-gcc-11.3.0/include/c++/11.3.0/ext/new_allocator.h:116: undefined reference to `std::__throw_bad_array_new_length()'
The change survives existing bootstrap and unblockes early `gcc` bootstrap.
Just a few comments added:
- added a few one-liners to explore which tools are rebuilt at each
stdenv iteration during bootstrap
- explicitly listed available toolchains and their sources for on each
bootstrap step: glibc, binutils, gcc, coreutils.
- added mention of static libraries linked into gcc
Co-authored-by: Adam Joseph <54836058+amjoseph-nixpkgs@users.noreply.github.com>
The documentation for `meta.sourceProvenance` in
`doc/stdenv/meta.chapter.md` says: "the `meta.sourceProvenance`
attribute should be a list containing one or more value..."
Let's update check-meta.nix to require that `meta.sourceProvenance` is
a list, as the documentation says, rather than a single element.
Adding two extra keystrokes `[` and `]` when filling out this field is
an insignificant burden for package authors, and being able to assume
that the `meta.sourceProvenance` field is always a list greatly
simplifies any code that acts on the value of this field.
Since `meta.sourceProvenance` was just merged a few hours ago now is
the easiest time to fix this: nobody is using the feature yet.
Without the change we don't propagate `enableParallelBuilding = true`
and leave most builds sequential.
Noticed on `mythtv` package which did not specify parallelism and
`config.enableParallelBuildingByDefault = true` had no effect.
Derivations not using `__structuredAttrs` should not attempt to set
environment variables from `env`.
Derivations using `__structuredAttrs` should fail if `env` is not
exportable.
Co-authored-by: Robin Gloster <mail@glob.in>
stdenv: print message if structuredAttrs is enabled
stdenv: add _append
reduces the chance of a user doing it wrong
fix nix develop issue
output hooks don't work yet in nix develop though
making $outputs be the same on non-structuredAttrs and structuredAttrs
is too much trouble.
lets instead make a function that gets the output names
reading environment file '/nix/store/2x7m69a2sm2kh0r6v0q5s9z1dh41m4xf-xz-5.2.5-env-bin'
nix: src/nix/develop.cc:299: std::string Common::makeRcScript(nix::ref<nix::Store>, const BuildEnvironment&, const Path&): Assertion `outputs != buildEnvironment.vars.end()' failed.
use a function to get all output names instead of using $outputs
copy env functionality from https://github.com/NixOS/nixpkgs/pull/76732/commits
In specific cases, combining the `checkMeta` and `checkMetaRecursively`
config options would result in `error: infinite recursion encountered`
fixes#193296
This is needed in order to mark a certain derivation containing a Nix
expression tarball to Hydra so that it is recognised as a channel.
When I first got an evaluation error due to using this meta attribute, I
was under the impression that nobody outside of Vuizvui[1] is using this
feature and that we don't have any occurrence of isHydraChannel in
Nixpkgs.
However, when working around[2] the issue I assumed that it's not
something that should be included in Nixpkgs because we're not using it
there.
It turned out that my assumption was wrong and we *do* use the attribute
in Nixpkgs, namely via releaseTools.channel, which is similar to what
we're doing in Vuizvui.
Since we already include a bunch of undocumented attributes in
metaTypes, it only makes sense to add isHydraChannel as well since it's
actually documented in the Hydra documentation[3].
[1]: https://github.com/openlab-aux/vuizvui
[2]: https://github.com/openlab-aux/vuizvui/commit/e0685e81b3fdc43a272f0
[3]: 53335323ae/doc/manual/src/jobs.md (meta-fields)
Signed-off-by: aszlig <aszlig@nix.build>
Passing `-l$NIX_BUILD_CORES` improperly limits the overall system load.
For a build machine which is configured to run `$B` builds where each
build gets `total cores / B` cores (`$C`), passing `-l $C` to make will
improperly limit the load to `$C` instead of `$B * $C`.
This effect becomes quite pronounced on machines with 80 cores, with
40 simultaneous builds and a cores limit of 2. On a machine with this
configuration, Nix will run 40 builds and make will limit the overall
system load to approximately 2. A build machine with this many cores
can happily run with a load approaching 80.
A non-solution is to oversubscribe the machine, by picking a larger
`$C`. However, there is no way to divide the number of cores in a way
which fairly subdivides the available cores when `$B` is greater than
1.
There has been exploration of passing a jobserver in to the sandbox,
or sharing a jobserver between all the builds. This is one option, but
relatively complicated and only supports make. Lots of other software
uses its own implementation of `-j` and doesn't support either `-l` or
the Make jobserver.
For the case of an interactive user machine, the user should limit
overall system load using `$B`, `$C`, and optionally systemd's
cpu/network/io limiting features.
Making this change should significantly improve the utilization of our
build farm, and improve the throughput of Hydra.
