Tests from the bazelTestTargets argument will be run before the build. The new bazelTestFlags argument can be used to pass additional flags to this phase.
This adds an option to skip adding --copt and --linkopt to Bazel
flags. In some cases, Bazel doesn’t like these flags, like when a
custom toolchain is being used (as opposed to the builtin one. The
compiler can still get the dependencies since it is invoked through
gcc wrapper and picks up the NIX_CFLAGS_COMPILE / NIX_LDFLAGS.
"bazel fetch" will, by default, fetch everything that _might_ be used,
including things that will later be discarded due to the way the build
is configured.
Concretely, this means that for some builds of Java packages, this will
avoid failures where the builder tries to retrieve the JDK from /usr/share/java
(or equivalent).
This also means that for most packages we can fetch _fewer_ dependencies,
since the standard tree pruning for artifacts to fetch will take effect.
fetchConfigured is disabled by default since it changes the fetch hashes
of tensorflow/tensorflow2 (since it ends up fetching less).
flat hashes can be substituted through hashed-mirrors, while recursive
hashes can’t. This is especially important for Bazel since the bazel
fetch dependencies can come from multiple different methods (git,
http, ftp, etc.). To do this, we create tar archives from the
output/external directory, which is then extracted to build. All of
the Bazel hashes are all updated.
All bazel fixed output derivations should be specific to the bazel
version that was used to generate them. There is not guarantee that the
build will still succeed or reproduces (without the cached fixed output)
if the fetch phase wasn't rerun with a different bazel version.
In the past bazel had been bumped but not all those packages that have
fixed outputs from bazel builds. This lead to compiling and somewhat
working TF versions that couldn't be reproduced without the cached fixed
outputs.
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.
They always can be regenerated during the actual build, and they are sometimes
random, e.g in Tensorflow;
platforms -> NIX_BUILD_TOP/tmp/install/35282f5123611afa742331368e9ae529/_embedded_binaries/platforms
Turns out markers are non-deterministic after all and even our patching still
doesn't solve this problem completely. For example (tensorflow deps, this is a
complete diff so actual dependencies don't differ):
30509c30509
< bc527ff00916b15caee38793bca8f294c748df4a256de55c5199281be0489e73 result/@bazel_skylib.marker
---
> 4e0303e815c78df1e43d4b88dfe65e73046e0c6157fb10aa9a4e8b910113cd9c result/@bazel_skylib.marker
31045c31045
< fa13d04b2316214c3b4008b52546c2d5b633e006f6f019d597bb3f9745bacf7b result/@bazel_toolchains.marker
---
> b36174bf5535e5157801b6de30c35ee03a03fe57766306393c3d65dd65cbebf4 result/@bazel_toolchains.marker
31144c31144
< b0ce4a3ac29ac22528336dd3a54b5b7af9ecc43bef2a2630713c1981a5cbbb51 result/@build_bazel_rules_swift.marker
---
> 7492528068ec4f8e7ace2ecf8f933ec4e1b2235bd7426ce6f70177919f1cd05e result/@build_bazel_rules_swift.marker
36245c36245
< be2993536a8233d63251b664caf35b1e7cd57d194ab2a39a293876c232d6bbd0 result/@io_bazel_rules_closure.marker
---
> b6655cc3f2c78525e5a724d8a4e93b1e7f09f1e09fc817d231109e7f39103e88 result/@io_bazel_rules_closure.marker
36329c36329
< 087bc674c9509dfe157400d111db4a13eeb45fc76aeccd490cee9aad6771ecad result/@io_bazel_rules_docker.marker
---
> f920ec07315ec71e800b05cd22b2a341c0a80807c6e335ee81739b13c532b422 result/@io_bazel_rules_docker.marker
79544d79543
< 85893a05a817036c61f6cd9f8247757baa1654f473c494ce4fc5253c2bbd2790 result/@platforms.marker
And here's an example of differences:
$ cat result-a/@bazel_skylib.marker
7dc7472d37424ba5ec6a5532765bc911
$MANAGED
cat result-b/@bazel_skylib.marker
a8f3f577798201157128e8e9934c4705
$MANAGED
Instead of trying to patch these markers further we now completely clear them.
Nix hacks for ignoring markers are restored and expanded so that we don't even
attempt to parse the marker.
* buildBazelPackage: autodetect nix toolchain instead of Xcode on Darwin
* do not export the variables outside of Darwin
* remove unecessary parens
* move comment within the darwin check
Timestamp verification skip is no longer needed (not sure why). Generally we
better off always using the environment hack for all packages because that
ensures all NIX_* flags are correctly applied.
One possible improvement in future is to filter only NIX_* variables to
passthru in Bazel.
Previously the installPhase of the fixed ouput derivation would fail for
a package that has no markers, since `sed` would complain about having
no input files. If we use `find` instead of bash globs, that problem
goes away.
The previous behaviour would work fine as long as `symlink` is a link to
a file. If is a link to a directory though, the new `ln` wouldn't
overwrite it but would create a new link *in that directory* (with the
name of the link source).
Instead, we can precompute the target location, then first remove the
symlink and write the new one in its place.
Bazel computes the default value of output_user_root before parsing the
flag[0]. The computation of the default value involves getting the $USER
from the environment. I don't have that variable when building with
sandbox enabled.
[0]: 9323c57607/src/main/cpp/startup_options.cc (L123-L124)
A separate function for building Bazel-bazed packages. Internally it splits the
build into two phases, fetching and building.
Users are expected to provide `fetchArgs.sha256` -- checksum of fetched
dependencies. Local dependencies should be removed in `fetchArgs.preInstall`.
Overall `fetchArgs` and `buildArgs` can be used to add specific steps to fetch
and build.
