> On Sun, Feb 21, 2010 at 08:42, Pavel Heimlich > <tropikhajma at gmail.com> wrote: > > Hi, > > the recent crashes in kdeinit4 were resolved (send > flowers to ade who found out the cause first). > > > > They were caused by the use of > '-xalias_level=compatible' flag in qt, which we did > not use globally before. As a quick counter-measure > I've removed the flag from our includes. > > > > After reading about it in the Studio User's Guide I > think it's quite a dangerous flag to use on a code > that is not under our control and I am not sure if > the performance benefits would outweigh increased > risk of weird crashes. > > Do you understand what that flag does ?
No, I haven't read the Studio code. Since the description at http://docs.sun.com/app/docs/doc/820-7599/bkayq?a=view concerns only SPARC, I can only hope it's doing the same on x86. On the other hand, the code is buggy. Every code is buggy. The Qt code alone has 7 millons lines of code, which a conservative estimate translates to around 7 000 bugs. There is no way people here would find and make patches for all of them. If a change in the compiler flags hides some of them without a significant performance drop, why not use it? The questions IMO are: How frequent are bugs affected by this flag? How big will the performance drop be? How big of a burden would patching and maintaining the bugs be if we try to fix them? How likely is upstream fixing such bugs? (BTW have you gotten any response from the Qt developers regarding the patches you sent them a year ago?) > > A bunch of dramatic statements with no basis in fact, > which do not > explain why pointer aliasing was causing the software > to crash, and > justify the removal of the flag with "ohhh, it's > really really > dangerous!". > > The fact that removing the flag has stopped the > crashes does not fix > the underlying problem. It only hides the most > visible symptom of the > problem. true . But if it hides it for long enough, it's the same as fixing it. > > Also, anyone can claim "ohhhh, it's really really > dangerous" about any > other compiler flag. > > --Stefan > > -- > Stefan Teleman > KDE e.V. > stefan.teleman at gmail.com -- This message posted from opensolaris.org
