--- Additional Comments From reichelt at gcc dot gnu dot org 2004-10-23 01:21
---
Btw, the first testcase causes mainline to segfault.
I just opened PR 18121 to track this problem.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18110
--- Additional Comments From reichelt at gcc dot gnu dot org 2004-10-22 15:47
---
Just two more remarks:
* The relevant part of the standard is 14.8.2 [temp.deduct],
especially 14.8.2.1 [temp.deduct.call].
* The reason why 3.3.5 doesn't error out might be that it confuses
friend P
--- Additional Comments From carlo at gcc dot gnu dot org 2004-10-22 15:39 ---
Ok, thanks Volker. Our last posted crossed, sorry.
I'll close the PR because I suppose you are right.
I want my library to accept (a + b) * c however,
so I guess I'd have to add new operator* prototypes :(.
--- Additional Comments From carlo at gcc dot gnu dot org 2004-10-22 15:32 ---
Created an attachment (id=7400)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7400&action=view)
Maximal stripped down test case.
This snippet still shows the same difference.
If the 'friend' is removed f
--- Additional Comments From reichelt at gcc dot gnu dot org 2004-10-22 15:28
---
Here's something shorter:
struct B {};
template struct P;
template P operator* (const P&, const P&);
template Boperator+ (const P&, co
--- Additional Comments From carlo at gcc dot gnu dot org 2004-10-22 15:01 ---
Created an attachment (id=7399)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7399&action=view)
Stripped down test case.
I managed to strip it down to a more reasonable size.
Still not sure what is going
--- Additional Comments From carlo at gcc dot gnu dot org 2004-10-22 13:56 ---
Created an attachment (id=7398)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7398&action=view)
bzip2-ed preprocessed source code without line numbers (s/^# .*//).
--
http://gcc.gnu.org/bugzilla/show
