We shouldn’t force the user to have a C compiler in scope, just
because the derivation is forced to build locally. That can’t be
counted as “lightweight” anymore.
Co-Authored-By: Silvan Mosberger<contact@infinisil.com>
A definition I’ve been copy-pasting everywhere so far, so it’s finally
time to add it to nixpkgs.
I’m using a remote builder for my regular nix builds, so trivial
`runCommand`s which first try a substitution and then copy the inputs
to the builder to run for 0.2s are quite noticable.
If we just always build these, we gain some build time, so let’s make
it easy to switch from remote to local.
This has the same motivation as fetchFromGitHub/fetchFromGitLab --
it's cheaper to download a tarball of a single revision than it is to
download a whole history.
I could have gone with domain/group/repo, like fetchFromGitLab, but it
would have made implementation more difficult, and this syntax means
it's a drop-in replacement for fetchgit, so I decided it wasn't worth
it.
tensorflow assumes $USER to be set to something, otherwise it complains
like this:
```
FATAL: $USER is not set, and unable to look up name of current user: (error: 0): Success
Traceback (most recent call last):
File "./configure.py", line 1602, in <module>
main()
File "./configure.py", line 1399, in main
_TF_MAX_BAZEL_VERSION)
File "./configure.py", line 478, in check_bazel_version
['bazel', '--batch', '--bazelrc=/dev/null', 'version'])
File "./configure.py", line 156, in run_shell
output = subprocess.check_output(cmd)
File "/nix/store/drr8qcgiccfc5by09r5zc30flgwh1mbx-python3-3.7.5/lib/python3.7/subprocess.py", line 411, in check_output
**kwargs).stdout
File "/nix/store/drr8qcgiccfc5by09r5zc30flgwh1mbx-python3-3.7.5/lib/python3.7/subprocess.py", line 512, in run
output=stdout, stderr=stderr)
```
Spotted while changing the hash of its fixed-output derivation on
purpose.
We could also set this in the tensorflow-specific part, but very likely,
other programs will fail as well.
This cuts down the dependency tree on some rust builds where a crate not
just exposes a binary but also a library. `$out/lib` contained a bunch
of extra support files that among other information carry linker flags
(including the full path to link-time dependencies). Worst case this led
to some binary outputs depending on the full build closure of rust
crates.
Moving all the `$out/lib` files to `$lib/lib` solves this nicely.
`lib` might be a bit weird here as they are most of the time just rlib
files (rust libraries). Those are essential only required during
compilation but they can also be shared objects (like with traditional
C-style packages). Which is why I went with `lib` for the new output.
One of the caveats we are running into here is that we do not (always)
know ahead of time of a crate produces just a library or just a binary.
Cargo allows for some ambiguity regarding whether or not a crate
provides one, two, … binaries and libraries as it's outputs. Ideally we
would be able to rely on the `crateType` entirely but so far that isn't
the case. More work on that area might show how difficult that actually
is.
If an empty string is passed to `-idirafter`, it breaks gcc. This commit makes
the stdenv less fragile by expanding out the shell glob and ensuring no empty
arguments get passed.
Before, we'd always use `cc = null`, and check for that. The problem is
this breaks for cross compilation to platforms that don't support a C
compiler.
It's a very subtle issue. One might think there is no problem because we
have `stdenvNoCC`, and presumably one would only build derivations that
use that. The problem is that one still wants to use tools at build-time
that are themselves built with a C compiler, and those are gotten via
"splicing". The runtime version of those deps will explode, but the
build time / `buildPackages` versions of those deps will be fine, and
splicing attempts to work this by using `builtins.tryEval` to filter out
any broken "higher priority" packages (runtime is the default and
highest priority) so that both `foo` and `foo.nativeDrv` works.
However, `tryEval` only catches certain evaluation failures (e.g.
exceptions), and not arbitrary failures (such as `cc.attr` when `cc` is
null). This means `tryEval` fails to let us use our build time deps, and
everything comes apart.
The right solution is, as usually, to get rid of splicing. Or, baring
that, to make it so `foo` never works and one has to explicitly do
`foo.*`. But that is a much larger change, and certaily one unsuitable
to be backported to stable.
Given that, we instead make an exception-throwing `cc` attribute, and
create a `hasCC` attribute for those derivations which wish to
condtionally use a C compiler: instead of doing `stdenv.cc or null ==
null` or something similar, one does `stdenv.hasCC`. This allows quering
without "tripping" the exception, while also allowing `tryEval` to work.
No platform without a C compiler is yet wired up by default. That will
be done in a following commit.
Go beyond the obvious setup hooks now, with a bit of sed, with a skipped case:
- cc-wrapper's `dontlink`, because it already is handled.
Also, in nix files escaping was manually added.
EMP
Using wrapProgram adds a call to `bash` around every call
of `execline`, which clearly misses the basic idea behind
`execline` in the first place …
This reverts commit b64d25c447.
One issue with cargoSha256 is that it's hard to detect when it needs to
be updated or not. It's possible to upgrade a package and forget to
update cargoSha256 and run with old versions of the program or
libraries.
This commit introduces `verifyCargoDeps` which, when enabled, will check
that the Cargo.lock is not out of date in the cargoDeps by comparing it
with the package source.
Quoting from the splitString docstring:
NOTE: this function is not performant and should never be used.
This replaces trivial uses of splitString for splitting version
strings with the (potentially builtin) splitVersion.
There is a bug in this feature: It allows extra arguments to leak in
from the environment. For example:
$ export extraFlagsArray=date
$ man ls
Note that you get the man page for date rather than for ls. This happens
because 'man' happens to use a wrapper (to add groff to its PATH).
An attempt to fix this was made in 5ae18574fc in PR #19328 for
issue #2537, but 1. That change didn't actually fix the problem because
it addressed makeWrapper's environment during the build process, not the
constructed wrapper script's environment after installation, and 2. That
change was apparently accidentally lost when merged with 7ff6eec5fd.
Rather than trying to fix the bug again, we remove the extraFlagsArray
feature, since it has never been used in the public repo in the ten
years it has been available.
wrapAclocal continues to use its own, separate flavor of extraFlagsArray
in a more limited context. The analogous bug there was fixed in
4d7d10da6b in 2011.
This removes the unnecessary compiler build dependency. We also set
preferLocalBuild = true;
allowSubstitutes = false;
to not farm out the build on a remote builder or bother with trying to
find a binary substitution.
The architecture of an image should default to the architecture for
which that image is being composed or pulled. buildPackages.go.GOARCH is
an easy way to compute that architecture with the correct terminology.
These should not cause issues in practice but it is good idea to handle them.
* prefix and targetOffset are mandatory, as they are always set by the generic builder.
* wrapPrefixVariables and dontWrapGApps are now defaulting to empty value, as they are not mandatory.
