clang should diagnose `memcpy()` via `-Wunsafe_buffer_usage`

[pkasting]

When -Wunsafe-buffer-usage is on, memcpy() should trigger a warning (unless locally pragmad away), since it effectively does pointer math under the hood (takes a ptr and size).

Similar comments would apply to e.g. memset(), memmove(), etc.

[llvmbot]

@llvm/issue-subscribers-libc

Author: None (pkasting)

When `-Wunsafe-buffer-usage` is on, `memcpy()` should trigger a warning (unless locally `pragma`d away), since it effectively does pointer math under the hood (takes a ptr and size).

Similar comments would apply to e.g. memset(), memmove(), etc.

[nickdesaulniers]

Sorry, trying to parse the description (thanks for filing a bug to track this either way). Are you suggesting we enable -Wunsafe-buffer-usage when building llvm-libc? (then proceed to fix any new instances of the warning that crop up)? Is this a new-ish compiler diagnostic? Does it exist in both clang and gcc?

[pkasting]

No; I'm suggesting that authors who call memcpy() in their own code should receive warnings if they compile said code with -Wunsafe-buffer-usage. I don't have any recommendation re: how libc itself should be compiled.

[nickdesaulniers]

To answer my own questions:

Is this a new-ish compiler diagnostic?

clang-16+

Does it exist in both clang and gcc?

Does not exist in gcc, yet.

I'm suggesting that authors who call memcpy() in their own code should receive warnings if they compile said code with -Wunsafe-buffer-usage. I don't have any recommendation re: how libc itself should be compiled.

Ah I see! This issue was accidentally re-tagged as libc; this is a feature request against Clang! Restoring the initial tag which was correct. Sorry for my confusion.

[nickdesaulniers]

cc @ziqingluo-90

[ziqingluo-90]

It should already warn about memcpy and other unsafe libc functions with -Wunsafe-buffer-usage. Please let me know if this is not what you see. @pkasting

E.g., https://godbolt.org/z/hsenbs3n1