On Wed, Feb 17, 2010 at 7:46 PM, Mark Mitchell <m...@codesourcery.com> wrote: > Steven Bosscher wrote: > >> Mark just made an ICE in the compiler with non-default options a P1 >> bug for GCC 4.5 (xf. >> http://gcc.gnu.org/ml/gcc-bugs/2010-02/msg01695.html). >> >> Can someone please explain why this kind of bug should be of >> release-blocking priority? > > As I wrote in the PR, I want to understand what kind of "broken" applies > to this pass. To be clear, I have no idea whether the pass is perfect, > totally wrong, or just a bit buggy. I'm not casting any aspersions > whatsoever. I'm responding to your comment in the PR that the patch is > broken. > > Shipping a compiler with an option that we know is just a piece of junk > is a bad idea. It's one to ship an experimental option, or a > "technology preview"; it's another to ship something that's no good to > anybody. As a responsible software distributor, we should disable such > things in our releases. GCC *developers* can always hack the source if > they want to play with the feature, but GCC *users* shouldn't look at > the manual, turn on some option that we know doesn't work, and then have > the compiler blow up. > > I consider it P1 to understand what the situation is here. That doesn't > mean we should fix the ICE. The right outcome might be to disable the > pass, or to do nothing at all.
I agree. Note that all regressions from 4.4 that are visible with release checking and valid input should be considered P1 first - we can decide to downgrade them later anyway, but it's good to get those top on the radar. Richard. > -- > Mark Mitchell > CodeSourcery > m...@codesourcery.com > (650) 331-3385 x713 >
