I tried adding a test based on --modify_execution_info, but this doesn't quite work: The flag sets the value to an empty string (https://cs.opensource.google/bazel/bazel/+/master:src/main/java/com/google/devtools/build/lib/analysis/config/ExecutionInfoModifier.java;l=113), but a "1" is expected (https://cs.opensource.google/bazel/bazel/+/master:src/main/java/com/google/devtools/build/lib/actions/Spawns.java;l=109).

@wilwell Would you consider it acceptable to loosen the latter check?

@wilwell Do you have an opinion regarding #21370 (comment)?

@fmeum , could you explain again #21370 (comment) please?

If I understood correctly you have an issue with flag --modify_execution_info because it only adds the flag key and not flag value (set it to "" by default).

And what is you suggestion? You want to change the behaviour of supportsMultiplexSandboxing?

@wilwell Yes, that would be one way of achieving my goal: I would like to see some way to enable sandbox multiplexing for Java actions that 1) doesn't require building a custom Bazel binary and 2) is considered experimental enough that you are willing to accept it before committing to fully support the feature. An alternative could be to expose this on java_toolchain as in Lars' CL, but without enabling it for the default toolchains. If you see a different way that you like better, I'm of course open to that as well.

@fmeum could you specify a little bit more details. What do you pass here? Is it modify_execution_info=Javac=supports-multiplex-sandboxing?

In general I think, that instead of fixing the value supports-multiplex-sandboxing flag we need to fix the behaviour of --modify_execution_info flag to be able to pass the key-value pair there.

cc @meisterT

Ahh, sorry I understood your alternative. Yes, let's try add toolchain flag javac_supports_worker_multiplex_sandboxing as Lars did in his CL.

My thoughts about alternatives

1. Change supports-multiplex-sandboxing check to compare with "". It makes check on worker flags inconsistent and could confuse engineers in future.
2. Change the behaviour of modify_execution_info flag to accept key-value inputs. Its more reliable solution, but it need the large change in modify_execution_info flag and checks that it will not ruin other builds, which is more complicated.
3. Create javac_supports_worker_multiplex_sandboxing toolchain flag and set it false by default. With this solution engineers who want to interact multiplex Java workers could do it by one flag set. It doesn't require large change in code. The only weakness which I see is inefficient use of remote cache, because the toolchain will be different, if I remember correctly.

@fmeum Am I right, that if you add javac_supports_worker_multiplex_sandboxing flag with default false value one could override this value for their build in their bazelrc file? What is this change?

@wilwell Done, please review again. I also added a test.

@wilwell Could you add the awaiting-PR-merge label in addition to your approval?

@bazel-io fork 7.1.0

@sgowroji Is this already in the process of being imported?

Is this already in the process of being imported?

There are some internal references to the modified code that require small updates, but it's making progress.

The changes in this PR have been included in Bazel 7.1.0 RC2. Please test out the release candidate and report any issues as soon as possible.
If you're using Bazelisk, you can point to the latest RC by setting USE_BAZEL_VERSION=7.1.0rc2. Thanks!