Add stamping support to go_binary

[ian-h-chamberlain]

What type of PR is this?

Feature

What does this PR do? Why is it needed?

Adds a stamp attribute to go_binary which can be used to override --[no]stamp behavior for a specific binary target.

Semantics match those of cc_binary, and are limited to just go_binary for now.

Which issues(s) does this PR fix?

Closes #2154

Other notes for review

Open questions / tasks:

• I am not 100% sure if there's additional documentation that needs to be updated, if someone can point me in the right direction I'm happy to add that.
• Should go_test have this attribute as well? I didn't think it made sense for libraries have it, since conflicts between dependencies would need to be resolved, but I could probably add it for _test without needing to address those.

[fmeum]

The ternary value for stamp is generally pretty confusing, even if consistent with cc_binary. Could you explain what you would like to use this new attribute for? I wonder whether we really need to support all three values.

[ian-h-chamberlain]

Sure, the use case we have in my organization is for iterating while manually testing one of our apps. We basically have two main use cases:

• Building for production via CI (we use --stamp in the build)
• Developer build for local testing (requires remembering --stamp every time to get a stamped build)

We don't want to stamp everything (non-Go, unit tests, etc.) all of the time, to maximize cache hits, so we don't have --stamp in our .bazelrc. However, we would ideally like to use stamping for developer builds to embed e.g. git data, build timestamp just for the final go_binary.

As it stands without this change (or equivalent), users have to remember to pass --stamp every time they build this particular target to get the embedded metadata, but with this change we could set stamp = 1 and forget it. For our purposes we would also be fine with stamp = True to force-stamp a particular target (where the default behavior for False would be to just follow --[no]stamp).

I mainly kept the ternary in case there would ever be users that want to forcefully disable stamping for certain targets, but I'm not sure if is particularly common. A quick search on GitHub shows some usage like this, but it seems fairly limited (mostly forks of llvm-bazel, I guess): https://github.com/search?q=%2F%5Cbstamp+%3D+0%2F+path%3A%2FBUILD%28%5C.bazel%29%3F%24%2F&type=code