Windows MSVC/GNU function pointer check issues

[dybucc]

Windows targets under MSVC and GNU currently skip function pointer checks in ctest.

The issue is two-fold. Linking in MSVC works just fine. It's the function pointers that end up resolving unequal. Linking in GNU fails. It reports undefined references to certain library symbols.

More information may be added to this issue later on.

[mansiverma897993]

@dybucc I have raise PR #5229 plz look at once and lemme know if any changes required!1

[tgross35]

Context for this is starting at #5059 (comment)

[mansiverma897993]

Hi @tgross35,

Previously, all function pointer checks on Windows were skipped (cfg.skip_fn_ptrcheck(|_| true)) in #5059 with the comment // FIXME(windows): All functions point to the wrong addresses?.

The root cause was that when taking function addresses for DLL-imported functions on Windows, both the Rust and C compilers generate distinct local Import Address Table (IAT) jump stubs (ff 25 / jmp instructions). Because of these separate stubs, the function pointer values resolved to different local addresses, causing the comparisons to fail.

To resolve this:

1. IAT Jump Stub Resolution: Added a resolve helper in check_same_fn_ptr (inside ctest/templates/test.rs). On x86/x86_64 Windows, it checks the instruction bytes at the pointer address for the 0xff 0x25 opcode, reads the displacement/target address, and dereferences it to compare the underlying DLL export address directly. This allows the pointers to be compared correctly.
2. Re-enabled Checks: Removed the catch-all skip in libc-test/build.rs so that we verify function pointer signatures on Windows too. We only skip ctime and difftime now, as they are inline functions in the MSVC CRT headers and resolve to local wrapper functions rather than DLL exports.
3. Winsock Library Linkage: Linked -lws2_32 for Windows GNU targets in ctest/src/runner.rs to fix undefined reference link errors for socket functions.
4. MSVC Warning C4197: Disabled MSVC warning C4197 ("top-level volatile in cast is ignored") via /wd4197 in ctest/src/runner.rs as it was causing build failures under MSVC's /WX (warnings as errors).

[tgross35]

Why are the IAT stubs different, do they also differ across invocations?

Also, it feels like you may have passed along AI output. Please handwrite all communication, we are humans and would like to talk to other humans rather than bots.