On Sun, Mar 23, 2008 at 08:11:31PM +0000, Matthias Klose wrote: > > > This will last up to the lenny release, and the toolchain is to be > > > freezed next week. So I don't ... > > ... think you'll have to support that patch actively for too long. > > wrong, once released with the patch you'll get bug reports for the > compiler with the patch applied. For every report you need to recheck > with a compiler without this patch. If you do want to do that until > lenny+1 is released, please go ahead. So that everyone can follow at home, the discussed patch comes from the GCC people themselves[1], and for having read the thread, it appears that nobody was against the patch, that introduce an -mcld option so that people can force gcc to emit a CLD before functions that use string operations (stos and friends). All that is done in the current patch in Debian is to not let it be a gcc CLI switch and to be active by default. If you look at the patch, it merely reinstate the prior GCC behaviour which has a probability 0.0001 to generate surprising code as it's what it did prior to that. Of course, without this patch, GCC has a 0.9999 probability to generate code that would break if string ops are used in an interrupt handler, but if I read that comment right, since we know where those bugs come from, we don't care about them ? Is that it ? I would have argued that prodding GCC people to include the patch would have been time better spent than arguing about its inclusion in Debian. Debian is probably not the sole distribution that will need a behavior overlap on this one. And if all you need is a release team member to maintain the patch, then okay, I will do that, I can deal with x86(_64) asm. But unlike others that doesn't seems to care a lot about which compiler is default in lenny, I would be sad to not see gcc-4.3 our default compiler. It comes with a lot of enhanced warnings and error checking, that I've seen in action, and hence know to be worthwhile (the array bound checking e.g. is really nice). And the g++ strictness wrt headers is important to have, else code developed on Debian would need porting to work on other distributions. I believe Debian to be an excellent development platform, and we should not lag behind wrt stable development tools. [1] http://gcc.gnu.org/ml/gcc-patches/2008-03/msg00417.html -- ·O· Pierre Habouzit ··O madcoder@debian.org OOO http://www.madism.org
Attachment:
pgpNysS75jmu0.pgp
Description: PGP signature