x/build/cmd/relnote: look for CLs that had revert CLs sent (but abandoned)

[dmitshur]

If a change has a revert CL sent, and later the revert CL gets abandoned, it's likely that the original change is significant enough to be considered for inclusion in release notes. We can detect this fairly easily with maintner's corpus.

For an example where this would've helped, see #29384 (comment). A revert CL was sent, but then abandoned. It could've helped detect that we should include the original CL in the release notes; see #37614.

/cc @cagedmantis @toothrot @ianlancetaylor @mvdan

[bcmills]

Maybe? But sometimes the revert is due to a failure in a newly-added test, and is abandoned because the test fix is very simple and applied quickly.

A corresponding issue without the Testing label would probably be a higher-confidence signal, but if we have an open issue about an API change anyway, that should be labeled Documentation and release-blocker if the change is not reverted.

[dmitshur]

Good points @bcmills. We should be mindful of false-positives this might create.

I was imagining if this is implemented, relnote would emit a new section that humans can review and make decisions on what to keep.

[dmitshur]

I don't recall seeing many other instances where this would've helped, and I suspect the noise to signal ratio would be unfavorable, so this is unlikely to be worth doing. Closing for now.