Le 30/10/2026 à 13:56, Jerry DeLisle a écrit :
See attached patch. This is 1 of 2 to fix some loose ends in this effort.
Fairly self explanatory.
Regression tested on x86_64
OK for mainline?
This is mostly good; two cosmetic points below.
Regards,
Jerry
diff --git a/libgfortran/caf/shmem.c b/libgfortran/caf/shmem.c
index 964c7449e02..88b800e5247 100644
--- a/libgfortran/caf/shmem.c
+++ b/libgfortran/caf/shmem.c
@@ -266,7 +274,8 @@ _gfortran_caf_register (size_t size, caf_register_t type,
caf_token_t *token,
if (type != CAF_REGTYPE_COARRAY_ALLOC_ALLOCATE_ONLY)
*token = malloc (sizeof (struct caf_shmem_token));
- size = alignto (size, sizeof (ptrdiff_t));
+ size = size > SIZE_MAX - sizeof (ptrdiff_t)
+ ? SIZE_MAX : alignto (size, sizeof (ptrdiff_t));
Can you put a comment somewhere telling that saturating to SIZE_MAX will
end up being caught by the allocation code, resulting in a failed
allocation? You mention it in the commit message, but it remains
somehow implicit in the code itself.
switch (type)
{
case CAF_REGTYPE_LOCK_STATIC:
@@ -309,7 +319,8 @@ _gfortran_caf_register (size_t size, caf_register_t type,
caf_token_t *token,
allocator_lock (&local->ai.alloc);
mem = alloc_get_memory_by_id_created (
- &local->ai, size * local->total_num_images, next_memid, &created);
+ &local->ai, mul_size_sat (size, local->total_num_images), next_memid,
+ &created);
Please use a variable for the multiplication result, here...
if (created)
memset (mem, 0, size * local->total_num_images);
... and here, so that it remains apparent that we are clearing the same
amount as allocated.
allocator_unlock (&local->ai.alloc);
OK with these changes.
