what

Refactor the way retries work in UpdateStatus in gitlab.

why

I noticed this while working on #5269; we can't prove that by the time we get to "return err" we'll actually have a non-nil error to wrap. In practice this won't happen, if you carefully follow the paths (and note that StatusConflict will cause err to be non-nil), but it's hard to see. Additionally there were comments like "if we get here, then..." and "return the error, which might be nil", which to me is a code smell that the logic is hard to follow.

Instead I refactored following the example of the retry library we were using https://pkg.go.dev/github.com/jpillora/backoff

Another oddity that took me a bit to understand was that, whether there was a conflict or not, we retry if there's an error and we haven't hit our max amount. This means that the conflict really just amounts to a change in logging, which the new code reflects.

tests

The unit tests have good coverage of this area of the code.

references

• Partially addresses: Remove use of deprecated package pkg/errors #5269