https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88372
Martin Sebor <msebor at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2018-12-05
CC| |msebor at gcc dot gnu.org
Component|c |tree-optimization
Ever confirmed|0 |1
--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
This attribute is already accepted on the function pointer so it should work
just like alloc_align does. The reason why __builtin_object_size doesn't
report the size is because it only considers attributes on the called functions
in function calls and doesn't try to look for attributes on the types of
function pointers. So I'd consider this limitation a bug rather than
enhancement request. With that, accepting [[gnu::alloc_size(N)]] should make
sense as well.
With the very lightly tested patch below __builtin_object_size reports the same
size in both functions in the test case.
Index: gcc/tree-object-size.c
===================================================================
--- gcc/tree-object-size.c (revision 266799)
+++ gcc/tree-object-size.c (working copy)
@@ -414,8 +414,18 @@ alloc_object_size (const gcall *call, int object_s
gcc_assert (is_gimple_call (call));
+ /* Lopok for the called function. */
callee = gimple_call_fndecl (call);
if (!callee)
+ {
+ /* If there is no function, look at the type of the called
+ expression in case it's been declared attribute alloc_size. */
+ callee = gimple_call_fn (call);
+ if (callee && TREE_CODE (TREE_TYPE (callee)) == POINTER_TYPE)
+ callee = TREE_TYPE (callee);
+ }
+
+ if (!callee)
return unknown[object_size_type];
alloc_size = lookup_attribute ("alloc_size",
@@ -429,7 +439,8 @@ alloc_object_size (const gcall *call, int object_s
arg2 = TREE_INT_CST_LOW (TREE_VALUE (TREE_CHAIN (p)))-1;
}
- if (DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL)
+ if (DECL_P (callee)
+ && DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL)
switch (DECL_FUNCTION_CODE (callee))
{
case BUILT_IN_CALLOC:
