[flang-commits] [flang] [flang][OpenMP][NFC] Hoist variant match-info construction into Semantics (PR #204387)

[via flang-commits]

SunilKuravinakop wrote:

> > * Have you tried using LLVM idioms/utilities in your code? I am not sure if this is standard practice in flang.
> > * Have you been able to run LLVM test suite runs.
> > * Why have you not added test cases for all the scenarios of declare variant that are being supported? If you have individual test cases then adopt them into llvm-lit.
> 
> Thanks for the review. This PR is a pure NFC refactor, it hoists the existing variant match-info construction out of Lower/OpenMP into a shared Semantics/openmp-utils helper. The purpose is to share code between existing metadirective and future declare variant work. There's no new functionality or behavior change:
> 
> LLVM idioms/utilities: Yes. The moved code keeps using the standard utilities — llvm::SmallVector, the llvm::omp::* trait/context types (VariantMatchInfo, TraitSet, etc.), std::variant visitation, and CHECK for invariants. If you spot a specific line where a different helper reads better, happy to adjust.
> 
> Testing: I run ninja check-flang and all of flang tests pass. GitHub CI runs the flang test suite on every commit pushed here, so behavior changes would surface automatically.
> 
> New test cases: Since this is NFC with no new behavior, no new tests are added. The declare variant functionality itself, with its dedicated tests, will come in the separate PRs.
> 
> Happy to expand on any of these if something specific looks off.



https://github.com/llvm/llvm-project/pull/204387