Le 06/06/2012 22:29, David C. Rankin a �crit :
> On 06/06/2012 12:42 PM, Francois Andriot wrote:
>> Hello, luckily I found the fedora 17 beta iso to download (it's hard to find now
>> !).
>> I installed it inside a virtual machine.
>> I installed TDE 3.5.13 (compiled on final F17): no bug.
>>
>> Unlike I said previously (my mistake !), there were lots of updates to GCC
>> between beta release and final release.
>> So I focused on compiling with different GCC versions.
>>
>> Test 1:
>> I rebuilt kdelibs inside the virtual machine, installed the resulting package:
>> the bug is back !
>>
>> Test 2:
>> I've upgraded GCC inside the virtual machine, with the gcc package from F17 final.
>> I've upgraded nothing else.
>> I rebuilt kdelibs inside the virtual machine, installed the resulting package:
>> the bug has gone !
>>
>> Conclusion: it is really a GCC problem.
>>
>> You can find the source package for GCC 4.7.0-5 (used in Fedora 17 final) in
>> case you want to play with it:
>> ftp://distrib-coffee.ipsl.jussieu.fr/pub/linux/fedora/linux/releases/17/Everything/source/SRPMS/g/gcc-4.7.0-5.fc17.src.rpm
>>
> Thank you Francios!
>
> That is both encouraging and discouraging at the same time. Arch uses the
> current (latest) gcc and I'm still dead. I'm glad it is working with the Fedora
> final release. I wonder if Fedora final includes any patches to unicode or the
> basic font libraries. I don't know what is causing the race condition, but it
> seems like whatever tries to compute the line width to determine if it
> needs_wrap, something is blowing up that computation. I'll pick around when I
> get a chance, but I'm not really sure what I'm looking for.
>
> I would like to see what the status of some of the other distros are with this
> issue and gcc 4.7. Right now we have 3 data points - Arch, Fedora and Slack.
>
> Every distro showed the bug. The only distro that has successfully built
> without it is FC 17 after the move from beta->final. What also causes me pause
> is that you report FC 17 successfully built on gcc 4.7.0-5. However, Arch is
> already (and has been for ~month) using gcc 4.7.0-6. So if it was a gcc update
> that fixed it, I would expect to see the fix with gcc 4.7.0-6 (unless there is a
> gcc regression in 4.7.0-5 -> 4.7.0-6)
>
> Dunno, still crashing on Arch. I'll see if we get an other updates. As of
> today, there were none.
>
Hello, the "release" number (-5, -6 etc ...) is distro-dependant, so it
does not mean that Arch "-6" is newer than Fedora "-5" package. It does
only mean that Fedora has released the package 5 times, and Arch 6
times. You cannot compare Arch release number with Fedora release number.
The Fedora GCC final package (4.7.0-5) is built from a GCC SVN snapshot
dated 20120507, SVN revision 187244.
Then Fedora adds it's own patches. I've attached these patches in this mail.
I still do not know exactly where the providential fix is (inside the
GCC upstream, or in a Fedora patch ?)
Francois
Attachments:
- gcc47-java-nomulti.patch (text/x-patch, 1.6kb)
- gcc47-ppl-0.10.patch (text/x-patch, 5.0kb)
- fastjar-0.97-filename0.patch (text/x-patch, 0.5kb)
- fastjar-0.97-segfault.patch (text/x-patch, 1.0kb)
- gcc47-libgomp-omp_h-multilib.patch (text/x-patch, 0.5kb)
- gcc47-no-add-needed.patch (text/x-patch, 2.1kb)
- gcc47-libtool-no-rpath.patch (text/x-patch, 1.2kb)
- gcc47-hack.patch (text/x-patch, 4.5kb)
- fastjar-CVE-2010-0831.patch (text/x-patch, 3.0kb)
- gcc47-c++-builtin-redecl.patch (text/x-patch, 4.5kb)
- gcc47-cloog-dl.patch (text/x-patch, 25.7kb)
- gcc47-pr33763.patch (text/x-patch, 3.3kb)
- fastjar-0.97-len1.patch (text/x-patch, 0.5kb)
- gcc47-sparc-config-detection.patch (text/x-patch, 1.9kb)
- gcc47-libstdc++-docs.patch (text/x-patch, 1.3kb)
- gcc47-rh330771.patch (text/x-patch, 1.1kb)
- gcc47-i386-libgomp.patch (text/x-patch, 0.8kb)
- gcc47-pr38757.patch (text/x-patch, 3.7kb)
- gcc47-libitm-fno-exceptions.patch (text/x-patch, 1.0kb)
- fastjar-man.patch (text/x-patch, 1.1kb)
- gcc47-ppc32-retaddr.patch (text/x-patch, 1.9kb)