[Mlir-commits] [mlir] [MLIR][Arith] Add ExpandOps to convertArithToLLVM (PR #117305)

[Andrzej Warzyński]

banach-space wrote:

>  I could not find guidelines on how to revert a commit without going through the MR review process.

There's a few relevant parts to consider here. First, the revert policy:
*  https://llvm.org/docs/DeveloperPolicy.html#patch-reversion-policy

And indeed, it does not discuss "how". However, note:
> Reverts should be reasonably timely. A change submitted two hours ago can be reverted without prior discussion.

And then, under [obtaining-commit-access](https://llvm.org/docs/DeveloperPolicy.html#obtaining-commit-access)
> You are allowed to commit patches without approval which you think are obvious. This is clearly a subjective decision — we simply expect you to use good judgement.

>From my experience, "reverts" are where you tend to 100% rely on your judgement - it's the easiest and quickest way to fix buildbot failures. In more complex situation (e.g. no buildbot breakage upstream, but some issues downstream), we tend use PRs. But even then, if a repro has been shared, folks tend to revert. This way patch authors are _not under pressure_ to fix immediately (sometimes that's just physically not possible).

> Thanks again all for your patience.

No worries Hugo, this is standard practice ;-)

https://github.com/llvm/llvm-project/pull/117305