http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55910
Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |ebotcazou at gcc dot
| |gnu.org
--- Comment #6 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2013-01-09
09:55:35 UTC ---
> IMHO, I've localized the issue clearly enough. The code that generates
> instances for methods may sometimes consider some of those instances as having
> no parameters in method instance itself and as having some nonzero size of
> parameters (namely 0x38) in caller that re-adjusts stack after method instance
> has cleared the parameters.
> If those numbers would be taken from single variable that kind of bug would be
> impossible. So please examine the code and see where do you get total
> parameter
> size for 'ret' instruction from and where do you get total parameter size for
> stack re-adjustment from. And then please try to realize why those locations
> could possibly have different numbers in them.
When -O2 is enabled, there are dozens of different passes massaging the initial
code and trying to optimize it. It is essentially impossible to predict how
they interact with each other given the complexity explosion, that's why being
able to replay the compilation is required.
The rules are clear: if you want the GCC team to look at your problem, you must
provide a reproducer. Otherwise this PR will be closed as WONTFIX.
