Headers: prefer corecrt_malloc.h to malloc.h

[compnerd]

This allows us to target a lower level library (corecrt) rather than the higher level library (ucrt) which can be helpful in mordularising the SDK.

[llvmbot]

@llvm/pr-subscribers-platform-windows
@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-x86

Author: Saleem Abdulrasool (compnerd)

Changes

This allows us to target a lower level library (corecrt) rather than the higher level library (ucrt) which can be helpful in mordularising the SDK.

---

Full diff: https://github.com/llvm/llvm-project/pull/131668.diff

1 Files Affected:

• (modified) clang/lib/Headers/mm_malloc.h (+1-1)

diff --git a/clang/lib/Headers/mm_malloc.h b/clang/lib/Headers/mm_malloc.h
index d32fe59416277..4052995fff5cd 100644
--- a/clang/lib/Headers/mm_malloc.h
+++ b/clang/lib/Headers/mm_malloc.h
@@ -13,7 +13,7 @@
 #include <stdlib.h>
 
 #ifdef _WIN32
-#include <malloc.h>
+#include <corecrt_malloc.h>
 #else
 #ifndef __cplusplus
 extern int posix_memalign(void **__memptr, size_t __alignment, size_t __size);

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff 2e6402ca2c6c33ccf41d74383a8e3afb82489410 8415be6e59aa2f8a23a3e484b1c8053f4d0c7206 --extensions h -- clang/lib/Headers/mm_malloc.h

View the diff from clang-format here.

diff --git a/clang/lib/Headers/mm_malloc.h b/clang/lib/Headers/mm_malloc.h
index addb598f62..93c98bae27 100644
--- a/clang/lib/Headers/mm_malloc.h
+++ b/clang/lib/Headers/mm_malloc.h
@@ -13,11 +13,11 @@
 #include <stdlib.h>
 
 #ifdef _WIN32
-# if defined(__MINGW32__)
-#   include <malloc.h>
-# else
-#   include <corecrt_malloc.h>
-# endif
+#if defined(__MINGW32__)
+#include <malloc.h>
+#else
+#include <corecrt_malloc.h>
+#endif
 #else
 #ifndef __cplusplus
 extern int posix_memalign(void **__memptr, size_t __alignment, size_t __size);

[mstorsjo]

Looks ok to me.

I don't know about whether this really works as expected with corecrt_malloc.h or not, but I would expect you to have tested it.

We could look into adding corecrt_malloc.h in mingw-w64 as well, and then we could simplify this after a couple of years maybe.

[compnerd]

Looks ok to me.

I don't know about whether this really works as expected with corecrt_malloc.h or not, but I would expect you to have tested it.

We could look into adding corecrt_malloc.h in mingw-w64 as well, and then we could simplify this after a couple of years maybe.

Right. I'm not 100% certain if it fully solves the issue that I am encountering yet, however, this seems like a good thing to do nonetheless.

The header inclusion is present to avoid forward declaring _aligned_malloc and _aligned_free. Both of these are declared in the header thusly:

_ACRTIMP
void __cdecl _aligned_free(
    _Pre_maybenull_ _Post_invalid_ void* _Block
    );

_Check_return_ _Ret_maybenull_ _Post_writable_byte_size_(_Size)
_ACRTIMP _CRTALLOCATOR _CRTRESTRICT
void* __cdecl _aligned_malloc(
    _In_ _CRT_GUARDOVERFLOW size_t _Size,
    _In_                    size_t _Alignment
    );

It is easy for us to miss some of the annotations as corecrt evolves which is why the header inclusion is preferable to retain if we can manage.

[AaronBallman]

The CI failure on Windows looks potentially related. Also, can you validate that this change actually solves the issue for you regarding modules?

[compnerd]

I don't think that this fully fixes the issue as it doesn't completely break the cycle.