enable clang-tidy by opt-in

[metaflow]

at the moment clang-tidy is completly disabled as I have got multiple reports that they are not usable anymore.

[ChristianKuehnel]

I would be great to have such a feature for /clang-tools-extra/clangd. We recently cleaned up a couple of findings in our subfolder and we would like to prevent backsliding.

[joker-eph]

Please bring back clang-tidy, that was really valuable to have on review when it was enabled!

I just landed > 27 commits to fix various clang-tidy issues in mlir/.

[ChristianKuehnel]

@metaflow I would be interesting in working on this as we would like to use it for clangd.

I guess this would involve:

• Investigate how much of clang-tidy-diff.py and run-clang-tidy.py we could reuse and replace our own logic.
• Unify all the clang-tidy related config options into a single yaml file to make configuration easier.
• Introduce a new config option to suppress findings using a mix of include and exclude rules using path prefixes (as today). I'm thinking of something like this to only enable for MLIR:

  
  # The first matching rule is applied in the order of their definition. 

excludes from the current clang-tidy.ignore

• exclude:
  • mlir/examples/standalone
  • mlir/include/mlir-c/Bindings
  • mlir/include/mlir/Bindings
  • mlir/lib/Bindings
  • mlir/test
  • mlir/tools/mlir-cuda-runner
  • mlir/tools/mlir-rocm-runner

    turn on for the rest of mlir

• include:
  • /mlir

    turn off for all other paths

• exclude:
  • /

• As this would be a significant refactoring, migrate the new/changes scripting to the monorepo.
  • This would be a step towards #186 .
  • However then the pre-merge scripting would be distributed across 2 repos (until we migrate the rest some day).
  • We would need to figure out where to put these files in the monorepo.
• integration of this script with the pipeline:
  • Input: git hash to analyze (or last git commit if no hash was set)
  • Output: filtered clang-tidy findings as Python object or JSON
  • The caller of this script would then need to take the findings and upload them to Phabricator
  • The new script would be run on Linux after the build, so that all necessary header were generated already.

[ChristianKuehnel]

clang-tidy-diff.py already does most of the work around filtering findings related to a patch. I added JSON output in my patch so we don't have to parse the text output in our scripting.