https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117289
Joseph S. Myers <jsm28 at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |david.faust at oracle dot com,
| |ibhagat at gcc dot gnu.org,
| |jason at gcc dot gnu.org
--- Comment #1 from Joseph S. Myers <jsm28 at gcc dot gnu.org> ---
I'm not sure how exactly CTF is meant to handle deduplication where there are
multiple equivalent versions of the same type (POINTER_TYPE, FUNCTION_TYPE
etc.) that are supposed to end up the same in output. Here, we have two
versions of the function pointer type void (*) (struct callback_head *head):
one for the member of the structure, and one in the subsequently defined
typedef.
DWARF debug info seems to get things right even with -std=gnu23: the debug info
for the typedef points to the same type as used for the structure member.
However, CTF outputs the pointer and function types twice, so resulting in the
failure of this test with -std=gnu23.
An actual difference between the two function type definitions is that the one
for the member has TYPE_STRUCTURAL_EQUALITY_P set, whereas that for the typedef
has TYPE_CANONICAL set. It appears, judging by the comment on
c_update_type_canonical, that it may be deliberate for a type created while the
structure was incomplete to retain TYPE_STRUCTURAL_EQUALITY_P. I don't know if
this difference is the actual cause of the problem, or if something else is
meant to ensure that types with no semantic difference as far as CTF is
concerned don't end up duplicated in the CTF output, even if there are multiple
type nodes at the tree level (as noted above, it seems that DWARF ends up
getting the output right, without duplication).
