Can you get the apparent repo name from a Label?

[JohnnyMorganz]

We are currently migrating our usage of rules_jvm_external from WORKSPACE to Bzlmod for the maven.install

We have some code that checks Label.repo_name to do some matching. In the WORKSPACE setup, Label("@maven//:com_google_guava_guava").repo_name would just return maven. However, in the new set up, it returns rules_jvm_external~~maven~maven (which matches the canonical repo name).

We are relying on the apparent repo name to do this matching. tl;dr is that we have some custom Starlark logic that allows users to pass in a "target pattern"-like syntax like @maven//... or @maven//:com_google_guava_guava, and we match labels against these patterns, similar to this code here in rules_haskell: https://github.com/tweag/rules_haskell/blob/f28afbac8b522a2ae8747367a3f8705aad982b59/haskell/private/path_utils.bzl#L632, but we first check if the parsed patterns repo name matches the label repo name. This doesn't work anymore after switching to Bzlmod, as our parsed pattern has a repo name of just maven.

Can we somehow get the apparent repo name back from a Label struct?

[fmeum]

You can't get the apparent name back for a given label, but you can transform your patterns to Label instead and then compare canonical names.

You can use the native.package_relative_label function in a macro (or rule initializer, if you assume recent versions of Bazel) to convert any label string to a Label. You can then apply the pattern matching in the macro, or alternatively stringify the Labels to canonical form via str and then pass them to a string_list attribute on your rule. The rule can then use Label to recover the parsed label.

Since @maven//... isn't valid label syntax, you could internally convert it into a representation such as @maven//:__subpackages__ to safely roundtrip through Label.

Happy to provide more details if needed.

[JohnnyMorganz]

Thank you for your swift response! That's unfortunate, but appreciate the workarounds mentioned. We shall try those out.

Do you reckon this would be something that in theory Bazel could expose on Label, or the apparent repo name is something that we shouldn't really be relying upon? In any case, maybe we can hope this would all get solved by #7763 someday

[fmeum]

We could add helper functions for this to bazel-lib.

The reason for not having an API to get the apparent name is that there is no such thing as "the" apparent name: Depending on the repo in which a label appears, it may use a different apparent name to identify the same repo.

The workaround really is the proper way to do it.