I think Gazelle has enough users that Hyrum's law would be applied here. I wonder how many users have git-ignored files that end up in the targets that depend on a flow like:

go generate -> some git-ignored files -> gazelle -> bazel build

I believe it would be wise to make it opt-in rather than opt-out.

Facing a similar issue with git submodules. Would be happy to opt in.

As a data point, I tried the linked PR on our repo and it makes our generation time go from 2.5s to 4+s. It may make sense to advise people to put their generated junk in .bazelignore (or ideally don't generate things into the source tree, though I understand half-migrated repos are "fun")

I think that would be a reason to make this opt-in, maybe with documentation pointing out that .bazelignore is better?

FWIW the .bazelignore is only faster since 5cd3717 that stopped treating entries as globs.

aspect-build/aspect-cli@a70f957#diff-b61fcd9f5596f8f5a65d234f9606c2b41a8b7c53ab2f7701e8a0baa7a6e6f218 includes a patch to gazelle that adds an extra ignore predicate check within walk.Walk. The following is the relevant walk code.

https://github.com/bazelbuild/bazel-gazelle/blob/master/walk/walk.go#L164-L184

I suppose gazelle's walk function could support user-configured hooks by, for example, reading all entries from config.Ext that implement some interface. The config loading code (getWalkConfig) could inspect c.Exts:

https://github.com/bazelbuild/bazel-gazelle/blob/master/walk/config.go#L40-L58