6 Jun
2012
6 Jun
'12
8:57 p.m.
Le 06/06/2012 22:29, David C. Rankin a écrit :
On 06/06/2012 12:42 PM, Francois Andriot wrote:
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
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.