Baseline: 7845acae

Cherry picks:

   + d9e2f918:
     Avoid overcounting in ActionLookupValueProgressReceiver.
   + da23370d:
     Fix crash for top level aspects on targets with non-idempotent
     rule transitions.
   + 4714cd70:
     Reuse the built set of tools in `SpawnAction.Builder`.
   + 0a1d5053:
     Correctly handle symlinks when prefetching.

Initial release.

In https://github.com/bazelbuild/bazel/commit/5c4cf47a131c84506aad9ce0e014c6643c31a4ac, we replaced `path = path.getRelative(path.readSymbolicLink())` with `path = path.resolveSymbolicLink`. While the former one is wrong for relative symlinks, it actually works for absolute symlinks. The latter one, however, doesn't work if the target doesn't exist on local file system -- and this is the case for most of builds using our internal version of BwoB.

This CL fixes that issue by correctly handle both absolute and relative symlinks. Additionally, it correctly handles the case where a symlink points to a tree output.

Moved symlink integration tests to the base test class.

PiperOrigin-RevId: 549921885
Change-Id: If4009525499f4d0625e7725faf4bb0675984c299

Fixes a memory regression introduced by https://github.com/bazelbuild/bazel/commit/db347988dfd1b4b8cd5685de63d937f2ffa76b25.

PiperOrigin-RevId: 548689761
Change-Id: Ie0881f0c6c6ef3c351b7888057735c58cd2b6ddb

PiperOrigin-RevId: 547876237
Change-Id: I236e26e8403cc9d02c86695284abe4103297a876

Skips counting ConfiguredTargets or Actions for nodes that perform delegation.

Shifts proxy ConfiguredTarget node filtering from AnalysisTestCase to
BuildViewForTesting so the consistency check in
BuildViewForTesting.getSkyframeEvaluatedActionLookupKeyCountForTesting
against SkyFrameBuildView.getEvaluatedCounts passes. The check here is more
involved because it needs to handle nodes that are not done.

Not-done nodes occur when a subsequent incremental evaluation dirties, but
doesn't request a node.

PiperOrigin-RevId: 547192963
Change-Id: Ica01f20440ab0ba258e73c3f9cec5c7f92c99fa9

This is "Label" in both ConfiguredTargetFactory and DependencyResolver.

PiperOrigin-RevId: 546372326
Change-Id: I74a7e774f3e9b07b45e20a1f5cc8920d80ada3bc

Delegating configured target nodes do not create configured targets directly
and make these tests more brittle than necessary.

PiperOrigin-RevId: 546371644
Change-Id: I898d69fcbabec21b13553c13cbc49e523ecf5504

PiperOrigin-RevId: 546348221
Change-Id: I4237cd6439ab227ee74bd8b7e3aa29f4abc5ec7f

In support of https://github.com/bazelbuild/bazel/pull/18854.

PiperOrigin-RevId: 546346128
Change-Id: Ie3f6976b74cba351573a7f35b0d09391da10fb55

Adds support for supplying a MemProf profile, which can be supplied to LLVM to
automatically hint allocations based on profiled memory behavior. The patch
adds Blaze flag --memprof_profile=<label>, along with a memprof_profile rule
that allows specifying a profile either by absolute path or label.

The profile can either be an LLVM memprof profile with a .profdata extension,
or a zipfile containing one. The profile is symlinked and optionally unzipped,
and set in the memprof_profile_path build variable, which can be used to pass
the profile to the compiler via a feature.

PiperOrigin-RevId: 546311130
Change-Id: I896898bf3bddcf42214cde3f3f27d2ed18de6b26

PiperOrigin-RevId: 546280755
Change-Id: Ie2c79d34e1f94a003c4caf6d32f79be9a54101cc

Introduce a `LazyImmutableStarlarkList`, used for lazily flattening depsets when they are requested from `ctx.files`.

An abstract `ImmutableStarlarkList` class is extracted to share among the three immutable implementations.

`ctx.files.x` returns a list of transitive files provided by prerequisites in attribute `x`. The list is eagerly constructed when creating the `ctx` object prior to invoking the rule or aspect implementation function. This can be slow when the `FileProvider` (`DefaultInfo.files`) contains many transitive members.

Lazily flattening is especially effective in the deeply nested case because users are likely to opt for the depset-based API of `ctx.attr.x.files` anyway, leaving the flattened lists unused.

Additionally, this saves a small bit of memory by not allocating `NestedSet` memos.

It remains up for debate whether `ctx.files.x` should be deprecated in favor of accessing depsets through `ctx.attr.x.files`, but this change has pretty significant performance benefit with no need for migrations, so I think it's worth making.

PiperOrigin-RevId: 546277754
Change-Id: Ie58277f198665c53b0dc63f7830989402e899905

It has been enabled for quite a while and there is no reason to disable it.

PiperOrigin-RevId: 546235453
Change-Id: I897056960da937b7f43e1a3b6265fcd537076182

As an aside, the java toolchain label has been freely visible in Starlark some time now, so this method has little to no utility. Will clean up usages and delete in the future.

PiperOrigin-RevId: 546228890
Change-Id: I659ecfa6244c53e9dffe035f155f64893f9e763a

`grpc` 1.48.1.bcr.1 no longer depends on a yanked version of `rules_go`, which means that the temporary direct dependency on `rules_go` introduced by 6c393ec5 is no longer needed.

Closes #18857.

PiperOrigin-RevId: 546228207
Change-Id: Ie7045feb6d155045f127bf7014e4512072f0cae2

Revert "Implement `java_common.to_java_binary_info` in Starlark and delete corresponding native code"

PiperOrigin-RevId: 546222038
Change-Id: I50032e2f509a6c630c9a29d33834e3bb137b4e29

When `--experimental_cancel_concurrent_tests` is set, Bazel will cancel concurrently running tests on the first successful run by interrupting threads these tests running on.

Normally, if the test runner hasn't started running the test, or is waiting for the test to finish, the interruption can cancel that blocking operation and clear the interrupted status of the thread.

However, current code doesn't handle the case when the tests to be cancelled are already finished which leads to two problems:
  1. The interrupted status is propagated to skyframe causing the entire invocation to be cancelled.
  2. The assumption that "no test log is generated if the test was cancelled" is no longer true.

This CL fixes that by:
  1. Inside `AttemptGroup#cancelOthers`, we only cancel others if this group hasn't been cancelled to prevent the winning thread got cancelled by other threads that are already finished when receiveing the interruption.
  2. Check thread's interrupted status after the test attempt is finished.
  3. Remove the assertion that "no test log is generated if the test was cancelled".

PiperOrigin-RevId: 546214028
Change-Id: I3fc8d4e446ccabdb4c7b1794a3aeb8052912a044

We should check for a non-`None` `java_.info.compilation_info` since the field is always present.

PiperOrigin-RevId: 546203630
Change-Id: I6a60e3860e10933b10a4ad22be3f0c89605d4050

Fixes: https://github.com/bazelbuild/bazel/issues/18641

PiperOrigin-RevId: 546092727
Change-Id: I66365813a40390ec61a1a0565b06655b3ca50fcd

PiperOrigin-RevId: 546085523
Change-Id: If287e6e143f8858d185f83a1630275520db65e44

Also avoids creating a potentially large list of artifacts by computing the set of directories in the same stream that filters for shared libraries[^1]

PiperOrigin-RevId: 546081142
Change-Id: I613df2d0e5b47deae12e0f771b6b96e8d950d539

Teach DexMapper to not separate synthetic classes from their context classes when sharding dexes, like DexFileSplitter in https://github.com/bazelbuild/bazel/commit/9092303c358b87c1e027abafe135c44a27531b36. Fixes #16368

RELNOTES: None
PiperOrigin-RevId: 546041575
Change-Id: I12fa8cff15b13534ca9a5682b7e9b43e6c860ef8

Unwrapping all StatusRuntimeExceptions in in ReferenceCountedChannel when caused by IOException will discard critical tracing and retriability. The Retrier evaluations may not see an SRE in the causal chain, and presume it is invariably an unretriable exception. In general, IOExceptions as SRE wrappers are unsuitable containers and are routinely misued either for identification (grpc aware status), or capture (handleInitError).

Partially addresses #18764 (retries will occur with SSL handshake timeout, but the actual connection will not be retried)

Closes #18836.

PiperOrigin-RevId: 546037698
Change-Id: I7f6efcb857c557aa97ad3df085fc032c8538eb9a

Closes #18843.

Change-Id: Ifa07522df85cbbb39193e60e87de525efc51f492
PiperOrigin-RevId: 546031541

Just a refactoring change that is a no-op. Split off from the change that implements `merge` in Starlark for easier review.

PiperOrigin-RevId: 546020116
Change-Id: I3dad53ceac43a7d241e2a772cfcf97d59cf73a08

This is rather unfortunate but avoiding a `load` cycle has become impossible. Implementing `java_common.merge()` in Starlark will need access to the `_new_javainfo` raw constructor which can only be returned by the `provider()` call for `JavaInfo`. `JavaPluginInfo` needs _some_ function for merging `JavaInfo`s which will require said raw constructor. So if `JavaInfo` depends on `JavaPluginInfo`, we get the `JavaPluginInfo`->`JavaInfo`->`JavaPluginInfo` cycle.

If you tilt your head to the left and squint in just the right way, you can see this change captures the class relationship in Java-land where `JavaInfoApi extends JavaPluginInfoApi`.

PiperOrigin-RevId: 545993901
Change-Id: I1ad9c0f94564acb991205757afdd39c9e617f4dd

Since we can't differentiate between the absence of `JavaCompilationArgsProvider` and empty `{runtime,compilation}_classpath` depsets, we should add an empty compilation info instance to avoid attempts to dereference a `None`.

PiperOrigin-RevId: 545993802
Change-Id: I236bdcad3eb63936e428fca1c0c5ac8b2d667afc

This was a missing behavior in Skymeld, which caused some integration tests to fail when we tried turning on Skymeld by default for Bazel. Example failure:  https://storage.googleapis.com/bazel-untrusted-buildkite-artifacts/01890719-efb9-428c-80bb-5449e92383e8/src/test/shell/bazel/bazel_proto_library_test/test_attempts/attempt_1.log

This PR fixes this bug by adding this missing behavior.

Verified that the test passed afterwards.

Closes #18850.

PiperOrigin-RevId: 545973135
Change-Id: Ifa7f7e1d270c187ad40eb8b92807b37b7de3e2a3

PiperOrigin-RevId: 545971650
Change-Id: I7ca3d0c06f4c3c0586afb62ab59523bf1ecd365c

PiperOrigin-RevId: 545947586
Change-Id: I88cb1abe4cddd4a69421fef2364ad53486068593

PiperOrigin-RevId: 545935218
Change-Id: Ia35b0b4f9a7c4e156f7676f5795e1e77cb0184c6

PiperOrigin-RevId: 545928570
Change-Id: I1f61416b5356d7e931f14765e717e8be84db23ef

PiperOrigin-RevId: 545927412
Change-Id: I34e8531b8e396ccdfe0eecbee22cb68f76f969fc

PiperOrigin-RevId: 545919876
Change-Id: Ibba18b156beef00c9fff4cfd022f4b8d4c47fd65

PiperOrigin-RevId: 545906582
Change-Id: I8f46f1eda4b2b1f6c84c071286f2f5e2bad3ffff

PiperOrigin-RevId: 545879151
Change-Id: I796da306871a592e654ba046f0de0815ac4421c3

PiperOrigin-RevId: 545845712
Change-Id: I4c5f6bfb0dab201dce82cc795a8c2f146af19b39

PiperOrigin-RevId: 545818305
Change-Id: Icbebf74b73121c99cc63c7dff1d1594b03285801

*** Reason for rollback ***

This is a roll forward CL which addressed the issue reported in b/289354550#comment9. I added another `LabelInterner#enabled()` method to check whether `globalPool` was set or not. If not, then `SkyframeExecutor` should not remove any labels from weak interner when PAKCAGE node is done.

A unit test is also added to verify the condition when `LabelInterner#enable()` returns false.

*** Original change description ***

Automated rollback of commit 0bda661e.

*** Reason for rollback ***

b/289354550

*** Original change description ***

Clean up Label Interner flag and relevant unused code

PiperOrigin-RevId: 545793223
Change-Id: I4254c6ed910e31c2702a26ddec9c334470f7e1bf

An earlier change[^1] already dropped the deprecated `output_jar` parameter so we make the `output_source_jar` parameter mandatory.

PiperOrigin-RevId: 545785281
Change-Id: I96f928940b9fa3b6752f0300a3768f0e15e587b2