[libcxx-commits] [libcxx] Revert "[libc++] Avoid constructing additional objects when using map::at" (PR #160738)

[Nikolas Klauser via libcxx-commits]

philnik777 wrote:

> > > I'm not an expert in RDF.
> > > https://llvm.org/docs/DeveloperPolicy.html#patch-reversion-policy clearly indicates: "If you break a buildbot in a way which can’t be quickly fixed, please revert."
> > 
> > 
> > Yes. However, the document is just as clear that there has to be a way forward for the original patch, which is unfortunately overlooked too often: "It is not considered reasonable to revert without at least the promise to provide a means for the patch author to debug the root issue.". I don't see that here so far given that I don't even have a contact to ask for further information so far (I assume, given your statement above: "I'm not an expert in RDF").
> 
> The quote you mentioned is about situations when the revert reason is hard to reproduce. In this particular case there are clear steps to debug. Repro steps can be taken from any of the broken buildbots.
> 
> And there is a clear way forward for the original patch: fix problematic components (either by working with code owners or yourself)

I don't think that's in the spirit of the request. I'm not working on the backend. This is basically as-if I were to ask you to fix something in GCC. I also said above that I'd be happy if I know the code owner, which I still don't know. Would it have been so hard to provide that name? Again, I'm not working on the backend, so it's not at all clear to me which part the broken code is from. I've checked both RDF and CodeGen, but I couldn't figure out who the code owner is. I'd even been happy with "X should know who to talk to" or "I'll talk to Y, he probably knows what to do", but even that was apparently too much to ask for.

> and re-apply the patch. It can be done AFTER the test infrastructure is restored.

Yes, I was absolutely fine with that _if_ I know who to talk to.

> > "It is not considered reasonable to revert without at least the promise to provide a means for the patch author to debug the root issue.".
> 
> Buildbots have public configs. The sanitizer bots IIRC provide you a link to follow to reproduce their build. That's enough for you to reasonably debug it. [sanitizer-x86_64-linux-fast](https://lab.llvm.org/buildbot/#/builders/169) doesn't seem like an unreasonable target to expect contributors to be able to debug.

Again, I don't think this is reasonable. I'm basically working on a different code base and know almost nothing about the backend. I don't even build that part of the monorepo.

All I'm asking here is for some communication what to actually do. Why is that so hard?

I don't expect other people to fix things in libc++ in the rare case someone breaks something either. It's my problem as a maintainer of that part of the code base to fix things, not for them to figure out who to even talk to.


https://github.com/llvm/llvm-project/pull/160738