Traverse CrateGroupInfo when gathering CcInfo linking_context entries for use_cc_common_link

[ObsidianMinor]

This fixes rust_library_group and rust_prost_library for experimental_use_cc_common_link

Closes #3786

[UebelAndre]

Thanks! Is there a unit test you could add somewhere in //test/unit or something? Would love to have some protection against regressions.

[ObsidianMinor]

Yeah I was wondering where I should put that, since there's only one test I think actually exercising this experimental flag and the configuration looks mildly complex. But I'll see if I can add something to it.

[ObsidianMinor]

Turns out with further testing there's more places that don't account for crate groups. So this patch was incomplete and would fix direct dependencies, but not indirect dependencies, which happens during establish_cc_info. I've updated it to factor out the crate group flattening into a separate function plus the added tests.

[xiaopeng-tranxmart]

We also encountered the same issue, and this PR works well for us.
Looking forward to it being merged. Thanks for the pr & review work! @UebelAndre

[UebelAndre]

In general I think the change looks good, I just want to always make sure there's some regression testing. For the new tests, can you show what the error looks like without the fix applied? I otherwise defer to @krasimirgg to make sure this solves for their needs as well 😄

Thanks again!

[krasimirgg]

Looks very reasonable! +1 for adding some test coverage, as these types of things tend to regress due to "spooky action at a distance" sometimes.

To share some context: internally, we don't use rust_library_group at all today, so we don't have any needs around that. We haven't looked at its compatibility with cc_common.link at detail. At a high-level, internally we focus on a minimal set of rules and make sure they work well together and are well integrated with a variety of additional internal infrastructure. Onboarding something like rust_library_group would have high costs for us for not so much benefit comparatively.

[ObsidianMinor]

Is there somewhere else I should be adding coverage that isn't covered by what I've added in this PR?

[UebelAndre]

Is there somewhere else I should be adding coverage that isn't covered by what I've added in this PR?

No, I think you found it but if this feature is ever going to move out of experimental we'll probably need to add unit tests for it at //test/unit/cc_common_link or something.

I just want to know what the failure signature looks like without the fix and then see this fix resolve that for now.