[clang] [analyzer] Fix FN for UCRT fopen ignoring the lack of `restrict` qualifier (PR #236747)
Arseniy Zaostrovnykh via cfe-commits
cfe-commits at lists.llvm.org
Thu Oct 29 06:38:27 PDT 2026
necto wrote:
I recommend reviewing by commit:
[The CI run](https://github.com/llvm/llvm-project/actions/runs/40871020864/job/123437863839) on test-only commit proves that the limitation was there before the PR. The next commit fixes it as demonstrated by the change in the tests.
I went for a minimal-scope change disabling the `restrict`-sensitivity only in MSVC context. However, I am not sure honoring `restrict` is worth it:
1. There might be other libc implementation that do not apply this qualfier and they will suffer the same FNs.
2. Keeping track of `restricts` introduces quite a bunch of complexity (which I increased by adding an exception for UCRT).
3. It is hard to imagine the case where it would prevent a FP: function name and all arguments match, except for the `restrict` keyword (Note that according to C11 standard the `restrict` qualifier does not change the type compatibility).
I would be happy to simplify the code here and always ignore `restrict` for stdlib function summaries.
https://github.com/llvm/llvm-project/pull/236747
More information about the cfe-commits
mailing list