Use root-relative path to derive the solib name

[fmeum]

short_path is not the proper analogue of Bazel's Artifact#getRootRelativePath, root_relative_path is.

Work towards bazelbuild/bazel#28135

[wade-arista]

FWIW I confirmed the fix with my min repro from bazelbuild/bazel#28513: https://github.com/wade-arista/bazel-demo/actions/runs/21924653638/job/63313917476#step:5:31

[iancha1992]

@trybka @pzembrod @c-mita @hvadehra Could you please take a look at this one? Our 9.0.1 release depends on this. Thank you!

[trybka]

Given this surprising behavior, would it be possible to get a test for this?

[fmeum]

I'm not entirely sure what that would look like, to be honest. We could add an integration test against Bazel to verify that the two paths agree, but that would either require the integration test setup to be in rules_cc or we would have to add a test in Bazel after this PR has been merged and Azek has updated to it.

I'm open to other ways of testing this of course.

[iancha1992]

@trybka @fmeum @hvadehra @c-mita @pzembrod What's the ETA on this? We're trying to see whether we should include the fix for bazelbuild/bazel#28513 in 9.0.1 or include it in the future 9.0.2 or even 9.1.0 if it takes too long.

cc: @Wyverald @meteorcloudy

[trybka]

I'm not entirely sure what that would look like, to be honest. We could add an integration test against Bazel to verify that the two paths agree, but that would either require the integration test setup to be in rules_cc or we would have to add a test in Bazel after this PR has been merged and Azek has updated to it.

I'm open to other ways of testing this of course.

Sorry, what is Azek in this context? I'm OK with adding a test to bazel after the fact if you're willing.

[trybka]

Approving. @fmeum (or someone) to follow-up with a test in Bazel(?) or possibly rules_cc after some of the integration test infra is moved over.