On 05/25/15 21:18, Jason Merrill wrote:
On 05/25/2015 04:55 PM, Nathan Sidwell wrote:
else if (TYPE_CANONICAL (to_type) != to_type)
TYPE_CANONICAL (t)
= build_reference_type_for_mode (TYPE_CANONICAL (to_type),
mode, false);
But we're passing 'false' in as 'can_alias_all', rather than pass the
value passed into us.
Yes, I actually just changed that a month ago because we were hitting this same
ICE from a different direction (bug 50800). Since TYPE_CANONICAL (to_type)
doesn't have the may_alias attribute, the canonical pointer shouldn't have
TRCAA.
Ok, so IIUC a canonical pointer to a may_alias type should have TRCAA but a
canonical can_alias_all pointer to a non-may_alias type should not have TRCAA?
(i.e. one where CAN_ALIAS_ALL was passed true). Or are you saying that no
canonical pointers should have TRCAA?
Hmm, are you seeing a case where TYPE_CANONICAL (to_type) has the may_alias
attribute?
Yes. This occurs when the newly created TRCAA pointer is to a self-canonical
type. The
else if (TYPE_CANONICAL (to_type) != to_type)
is false, so the newly created pointer is self-canonical too (and has TRCAA).
If the canonical type should not have TRCAA we need to change the if condition
to:
else if (TYPE_CANONICAL (to_type) != to_type || could_alias_all)
where COULD_ALIAS_ALL is the incoming CAN_ALIAS_ALL value. Does that make
sense?
Making that change does stop the ICE I was seeing, but I've not done a full
test yet.
nathan
