using negative regular expressions with {run-}clang-tidy

[firewave]

#82416 (comment) mentioned that filter options with regular expression support can use negative expressions as exclude patterns.

I thought one of the limitations of llvm::RegEx is that it doesn't support negative expressions so I wonder how that should work.

I want to use this with -source-filter for run-clang-tidy as well as bugprone-unused-return-value.CheckFunctions.

CC @carlosgalvezp @HerrCai0907

[firewave]

Use cases:

1. With run-clang-tidy I want to scan a compilation database from a CMake project but exclude all moc_*.cpp (generated by moc from Qt) files.
2. With bugprone-unused-return-value.CheckFunctions I want to check all functions (via ::.*) but need to exclude functions like std::map::insert() and std::map::emplace() where the return code is rarely checked.

[firewave]

@EugeneZelenko This is actually a question. The feature is supposedly already there.

[EugeneZelenko]

@firewave: Do you have triage role? If not, please request it, since you file issues often, it'll be definitely helpful to have possibility to adjust labels.

[firewave]

@firewave: Do you have triage role? If not, please request it, since you file issues often, it'll be definitely helpful to have possibility to adjust labels.

No, I don't. How do I request that? The contribution guides do not seem to mention that.

"often" seems a bit like a stretch though. 😀 I am happy to help out though.

[EugeneZelenko]

@firewave: I asked for commit access (roles were split only recently) as described on https://llvm.org/docs/DeveloperPolicy.html#obtaining-commit-access. My guess that procedure for triage access is same.

[carlosgalvezp]

Hi!

Sorry for the delay in answering. I have looked a bit deeper into this. It seems llvm::Regex implements POSIX Extended Regular Expression (ERE), which does not support negative lookahead patterns (which would be required for this use case). Thus my original comment was wrong, it's not possible to get this behavior out of the box. Apologies for introducing confusion!

Thus we probably need to make do with two filters, one for filtering in, and one for filtering out.

[EugeneZelenko]

@tstellar: I think will be good idea to extend https://llvm.org/docs/DeveloperPolicy.html or https://llvm.org/docs/BugLifeCycle.html with details about getting triage access.

[tstellar]

@EugeneZelenko We don't currently have a process to apply for triage access. Users should follow the process for requesting commit access.

[nicovank]

I want to use this with -source-filter for run-clang-tidy.

This will work for run-clang-tidy.py, because the source-filter there uses Python regular expressions, it's not passed to the underlying Clang-Tidy.

Just a silly example to demonstrate: all files in a clang-query directory not in tool subdirectory. clang-tools-extra/clang-query/tool/ClangQuery.cpp is not matched anymore.

python3~/llvm-project/clang-tools-extra/clang-tidy/tool/run-clang-tidy.py \
    -source-filter='.*/clang-query/(?!tool).*'

[seanm]

I think I'm looking for one of the same things as @firewave...

With clang-tidy's bugprone-unused-return-value you can opt-in to it warning about failing to use the return value of some listed functions, or you can specify the regex .* to be warned about all ignored return values.

But what you seemingly can't do is be warned about all except for a few listed exceptions. I think the only thing stopping that is lack of support for 'negative lookahead patterns'.

Relatedly, with the clang compiler, you can also only opt-in, decorating your code with the warn-unused-result attribute. There is no warning flag to warn on all ignored return values (confirmed with -Weverything). That's a pity IMHO, but seems deliberate.

Personally, I'd like warnings for almost all ignored return values. But it seem unachievable...

(Just last week, we found 2 different bugs both due to ignored return values, hence me trying to get warnings for this.)