View previous topic :: View next topic |
Author |
Message |
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Tue May 02, 2006 3:32 pm Post subject: Updating to gcc-4.1.0 |
|
|
UPDATE-2006-05-08 :Make sure that USE="-objc-gc", then openoffice will also build.
Following the guide from http://gentoo-wiki.com/HOWTO_gcc4, there are some difficulties for anyone having also updated to portage-2.1.x. I made the error of updating to portage-2.1_pre7-r5. This version had serious flaws with checksum mismatches that were fixed by updating to pre9-r5.
When emerging system, there was initially a problem with groff, solved by su to root and
Code: | #CXXFLAGS="-ffriend-injection" emerge groff
and then exit
#emerge --resume |
For some reason the problem did not recur during 'emerge -e world'
For anyone using utf-8, slang will give a sandbox violation, but there is a simple patch at https://bugs.gentoo.org/show_bug.cgi?id=114730
In addition, there is a problem for all KDE users. 'emerge -e world' lists qt then a number of kde apps, but not kdelibs. You will most likely need to emerge kdelibs manually once qt has been done. I recommend using a different user for that, as you can then 'emerge --resume' and continue.
There is a problem with libggi. I haven't found a fix for it yet. As only gnuplot and mplayer depend on it, I'm not overly worried.
Last edited by davidgurvich on Tue May 09, 2006 4:49 pm; edited 4 times in total |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Wed May 03, 2006 12:35 am Post subject: |
|
|
swig does not compile, not sure of the reason yet. |
|
Back to top |
|
|
fb l33t
Joined: 08 Dec 2003 Posts: 636 Location: New Zealand
|
Posted: Wed May 03, 2006 12:46 am Post subject: |
|
|
I updated to gcc4.1.0 a while ago now.
I remember having to remove a file for slang as indicated in the sandbox
violation bug. Not that this bug is not gcc4.1.0 specific, it hits everyone
who has already a version of slang installed.
Swig is on my list off packages that failed (I have a sheet of paper with
everything that broke during my emerge -eD world), bumping it to ~ppc solved
the problem. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Wed May 03, 2006 1:27 am Post subject: |
|
|
Thanks. swig-1.3.25 compiled with no problems.
I haven't included all the java packages that fail due to having ibm-jdk-bin-1.5.0 as system JVM
I would suggest either setting USE="-java" or installing ibm-jdk-bin-1.4.2 and using that as the system JVM while doing this update.
gdome2-0.8.1-r1 has the same sort of sandbox violation as slang. gdome2-0.7.4 compiled with no problems.
cdrdao has failed. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Wed May 03, 2006 12:42 pm Post subject: |
|
|
Octave has failed because of a bug in octave or gcc-4.1.0, I'm not sure which is at fault. The octave ebuild is looking for g77, but gcc-4.1.0 has gfortran instead. I will try and add a symlink from gfortran to g77 in the gcc-4.1.0 bin directory and see what happens.
UPDATE: The compile has gotten through the problem with not having g77, but fails for a different reason now.
CMatrix.cc:2262: error: conversion from âComplexâ to non-scalar type âComplexMatrixâ requested
Will need to look at octaves documentation or source code for this I suppose.
Looks like octave needs to go ~ppc also. Compiling octave-2.1.72, this version is patched for gcc-4.1 |
|
Back to top |
|
|
fb l33t
Joined: 08 Dec 2003 Posts: 636 Location: New Zealand
|
Posted: Wed May 03, 2006 7:54 pm Post subject: |
|
|
davidgurvich wrote: | Octave has failed because of a bug in octave or gcc-4.1.0, I'm not sure which is at fault. The octave ebuild is looking for g77, but gcc-4.1.0 has gfortran instead. I will try and add a symlink from gfortran to g77 in the gcc-4.1.0 bin directory and see what happens.
UPDATE: The compile has gotten through the problem with not having g77, but fails for a different reason now.
CMatrix.cc:2262: error: conversion from âComplexâ to non-scalar type âComplexMatrixâ requested
Will need to look at octaves documentation or source code for this I suppose.
Looks like octave needs to go ~ppc also. Compiling octave-2.1.72, this version is patched for gcc-4.1 |
Yes it does. There is a bug somewhere about it. Support for gfortran is in octave-2.1.72.
If you are using scientific software, netcdf, blas-lapack, lapack-config and scilab also need to be marked ~ppc
from memory. And there is quite a few other. The only program with which I still struggle is maxima, the lisp
compiler breaks during the compilation of maxima - I finally got it to compile in ~ppc with some new use flags
but I am not sure it is stable - needs more testing. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Wed May 03, 2006 11:56 pm Post subject: |
|
|
I am also having trouble with ghc, but am not at my machine just now, so can't tell exactly what went wrong. |
|
Back to top |
|
|
fb l33t
Joined: 08 Dec 2003 Posts: 636 Location: New Zealand
|
Posted: Thu May 04, 2006 1:25 am Post subject: |
|
|
I am unfamilliar with ghc I am afraid. I searched for bugs without success, however the changelog
revealed this: Code: | *ghc-6.4.1-r3 (26 Mar 2006)
26 Mar 2006; Duncan Coutts <dcoutts@gentoo.org> +ghc-6.4.1-r3.ebuild:
Add support for ia64. Change the way we do the CFLAGS and GHC_CFLAGS again.
Make limited use of users CFLAGS. Support gcc-4.1. Fix parallel make.
|
This version is still masked, so you will need a ~ppc keyword in package.keywords and
an entry in package.unmask. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Thu May 04, 2006 2:50 am Post subject: |
|
|
Yes it looks like many of these packages need ~ppc. JoseJX suggested that mixing ppc and ~ppc would not work well. I suppose I'm just a control freak in this respect and like to know exactly what is going on. I've just taken a look at the size of my package.keywords file and it's starting to become quite large. |
|
Back to top |
|
|
JoseJX Retired Dev
Joined: 28 Apr 2002 Posts: 2774
|
Posted: Thu May 04, 2006 3:23 am Post subject: |
|
|
If you're playing with gcc-4.1, unfortunately, ~ppc is really the way to go. Most of the patches for gcc-4.1 aren't yet included in the stable tree.
We're looking at unmasking gcc-4.1 shortly, which will put it in the ~ppc arch tree. Once that's done, we'll focus on getting the stable tree ready for gcc-4.1. _________________ Gentoo PPC FAQ: http://www.gentoo.org/doc/en/gentoo-ppc-faq.xml |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Thu May 04, 2006 4:17 am Post subject: |
|
|
Interestingly, enough lapack-atlas requires using an emerge option I have never seen before.
Code: | interactive=1 emerge lapack-atlas |
There follow a series of questions , but be careful to choose to continue despite lapack-atlas not being able to find gcc-3.3 or later on your system. |
|
Back to top |
|
|
fb l33t
Joined: 08 Dec 2003 Posts: 636 Location: New Zealand
|
Posted: Thu May 04, 2006 9:03 am Post subject: |
|
|
davidgurvich wrote: | Interestingly, enough lapack-atlas requires using an emerge option I have never seen before.
Code: | interactive=1 emerge lapack-atlas |
There follow a series of questions , but be careful to choose to continue despite lapack-atlas not being able to find gcc-3.3 or later on your system. |
Interesting option. A few ebuild have actually stopped on me asking question, usually
on errors. I have specifically been facing interactive questions when lisp compilers or packages
build with lisp break (clisp in ppc break with gcc4.1.0 and stop to ask questions, ~ppc version
build fine. Maxima in ppc _AND_ ~ppc breaks, with a stack error if I remember correctly, and ask
question if build with gcl or clisp, ~ppc version can be build with a new lisp compiler which seems
to be fine, but it needs more testing as I had strange behavior.)
I don't remember doing anything special for lapack-atlas. I got a warning but it went fine, it was
lapack-atlas 3.7.11 which goes with blas-atlas 3.7.11 (I had trouble with 3.6.0-r1 even before the
shift to gcc4.1.0 to be honnest).
As for mixing ppc and ~ppc, you should see my package.keywords... Between kde3.5 and the
requirements of gcc4.1.0 plus a few packages for which I like to be on the edge (and the one I
test) it starts to be a bit long. I possibly need to clean it a little bit.
I got bitten once or twice on a pure ~ppc system so I have stayed mainly on ppc with a few exceptions.
Only recently I have decided to be a bit wild and break my computer with kde3.5 and then gcc4.1.0,
and now glibc2.4-r2 (compiled and installed fine this afternoon). Apparently I won't have to go ~ppc to
have Xorg 7.0, unless it doesn't cooperate with gcc4.1.0......... |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Thu May 04, 2006 12:13 pm Post subject: |
|
|
I am also using kde-3.5.2, largely due to problems with earlier versions. Oddly enough 3.5.2 had random crashes in various programs that disappeared as soon as I added "debug" to their USE flags.
I had the same experience earlier when I used ~ppc in my make.conf, so my reasons are actually the same.
I actually put a typo in my package.keywords where modular xorg started being built and decided that I liked the idea of incremental updates to xorg.
Once that worked, I noticed JoseJX made a comment about how most distros were currently using gcc4 as the base. That convinced me that I would need to change to gcc4 soon anyway, as most maintainers will most likely be using it.
Incidentally, kopete-3.5.2 does not buiild, nor does octave-forge. Neither is a show stopper. ghc-6.4.1-r3 has specific patches for gcc-4.1 and built without problem.
kopete-3.5.2 is having a problem with kdelibs libkhtml, for some reason can't use the version compiled with gcc-4.1. konqueror and other kde apps don't have any problems with it though. So far here are the things I've tried
fix_libtool_files.sh
revdep-rebuild
There in fact seems to be nothing wrong with kdelibs, except that kopete won't build against it. |
|
Back to top |
|
|
corsair Retired Dev
Joined: 06 Oct 2003 Posts: 451 Location: Aachen, Germany
|
Posted: Thu May 04, 2006 2:58 pm Post subject: |
|
|
davidgurvich, could you post the exact error, please? I am running ppc64 and kopete compiles just fine using gcc-4.1.0. (libkhtml is compiled using gcc-4.1.0, too). |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Thu May 04, 2006 3:50 pm Post subject: |
|
|
/bin/sh ../../libtool --silent --mode=link --tag=CXX powerpc-unknown-linux-gnu-g++ -Wno-long-long -Wundef -ansi -D_XOPEN_SOURCE=500 -D_BSD_SOURCE -Wcast-align -Wconversion -Wchar-subscripts -Wall -W -Wpointer-arith -g3 -fno-inline -O2 -mtune=G4 -maltivec -mabi=altivec -fno-strict-aliasing -pipe -Wformat-security -Wmissing-format-attribute -Wno-non-virtual-dtor -fno-exceptions -fno-check-new -fno-common -DQT_CLEAN_NAMESPACE -DQT_NO_ASCII_CAST -DQT_NO_STL -DQT_NO_COMPAT -DQT_NO_TRANSLATION -o kopete -no-undefined -L/usr/kde/3.5/lib -L/usr/qt/3/lib -L/usr/X11R6/lib -R /usr/kde/3.5/lib -R /usr/kde/3.5/lib -R /usr/qt/3/lib -R /usr/X11R6/lib -lktexteditor main.o kopeteapplication.o kopeteiface.o systemtray.o kopeteballoon.o kopetewindow.o kopeteaccountstatusbaricon.o kimifaceimpl.o kopeteeditglobalidentitywidget.o kopeteiface_skel.o kimiface_skel.o addcontactwizard/libkopeteaddcontactwizard.la addaccountwizard/libkopeteaddaccountwizard.la contactlist/libkopetecontactlist.la config/plugins/libkopetepluginconfig.la ../libkopete/libkopete.la ../libkopete/ui/libkopeteui.la
/usr/kde/3.5/lib/libkhtml.so: undefined reference to `__cxa_get_exception_ptr@CXXABI_1.3.1'
collect2: ld returned 1 exit status |
|
Back to top |
|
|
JoseJX Retired Dev
Joined: 28 Apr 2002 Posts: 2774
|
Posted: Thu May 04, 2006 4:01 pm Post subject: |
|
|
Do you have an older version installed on your machine? There's a KDE bug that screws up the link order so the older versions get used or something like that. The bugs aren't very clear and it keeps getting duped by the KDE team to a semi-unrelated bug. https://bugs.gentoo.org/show_bug.cgi?id=131523
Basically, try removing the old version first and see if that fixes it. _________________ Gentoo PPC FAQ: http://www.gentoo.org/doc/en/gentoo-ppc-faq.xml |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Thu May 04, 2006 4:22 pm Post subject: |
|
|
I do not have any older version, but will quickpkg kopete, unmerge, and emerge. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Thu May 04, 2006 4:59 pm Post subject: |
|
|
It looks like a similar problem to some that occured when gcc-3.4.5 was installed. I've never uninstalled gcc-3.4.4.
So that I actually have 3 versions of gcc installed. Forgot to account for that before.
Instead of Code: | fix_libtool_files.sh 3.4.5 |
I tried Code: | fix_libtool_files.sh 3.4.4 |
I don't know if that will fix the problem, but a number of kde libraries did show up in the list.
UPDATE: That seems to have fixed that problem. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Fri May 05, 2006 7:43 pm Post subject: |
|
|
So, it looks like most packages either compile with no problems or a version exists that can be.
An interesting situation has arisen. I decided to look into the gcj-overlay. Part of doing so requires that gcc not have gcj compiled. During recompilation of gcc-4.1.0 by gcc-4.1.0 the ebuild fails with the fortran module. |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Sat May 06, 2006 3:31 pm Post subject: |
|
|
Could not get gcc-4.1.0 to build with gcc-4.1.0 with or without gcj,fortran, etc.
Decided that there must be a problem in the toolchain. Perhaps some part was still using gcc-3.4.5 instead of gcc-4.1.0 even after
Code: | #emerge binutils glibc gcc
#emerge -e system
#emerge -e world |
- Check gcc-config for all previous compiler versions, confirming that the current one is gcc-4.1.0
- Use fix_libtool_files.sh [3.4.4] and [3.4.5] on my system.
- Uninstalled gcc-3.4.4-r1 keeping only 3.4.5 and 4.1.0
- Take all the packages that would be emerged with system prior to and including gcc-4.1.0 and remerged them.
This seems to have worked. Will try some packages later that require fortran.
Perhaps the migration instructions should be modified slightly.
Code: | #emerge binutils glibc gcc
#emerge -e system
#fix_libtool_files.sh previous_compiler_version
#emerge -e system
#emerge -e world |
|
|
Back to top |
|
|
fb l33t
Joined: 08 Dec 2003 Posts: 636 Location: New Zealand
|
Posted: Sun May 07, 2006 11:33 pm Post subject: |
|
|
davidgurvich wrote: | Could not get gcc-4.1.0 to build with gcc-4.1.0 with or without gcj,fortran, etc.
Decided that there must be a problem in the toolchain. Perhaps some part was still using gcc-3.4.5 instead of gcc-4.1.0 even after
Code: | #emerge binutils glibc gcc
#emerge -e system
#emerge -e world |
- Check gcc-config for all previous compiler versions, confirming that the current one is gcc-4.1.0
- Use fix_libtool_files.sh [3.4.4] and [3.4.5] on my system.
- Uninstalled gcc-3.4.4-r1 keeping only 3.4.5 and 4.1.0
- Take all the packages that would be emerged with system prior to and including gcc-4.1.0 and remerged them.
This seems to have worked. Will try some packages later that require fortran.
Perhaps the migration instructions should be modified slightly.
Code: | #emerge binutils glibc gcc
#emerge -e system
#fix_libtool_files.sh previous_compiler_version
#emerge -e system
#emerge -e world |
|
Somehow I didn't suffer from that problem at all. I am finding that the upgrade to glibc 2.4 is may be
more a problem than gcc 4.1.0. Most things I have tried work OK after a reboot. I still have to recheck
my printing system as before the reboot when I tried printing I got a warning about some lib it had trouble
finding, but I discovered that it was printing nonetheless. Before the reboot kmail wouldn't fetch my mail,
it would report that my pop3 server was unknown.
And it doesn't play well for compiling some package (xorg-x11-6.8.2-r7): Code: | g++ -c -I../nurbtess -I../../libutil -I../../include -I../../../../include -I../../../../include/GL -I../../../.. -I../../../../exports/include -Dlinux -D__powerpc__ -D_POSIX_C_SOURCE=199309L -D_POSIX_SOURCE -D_XOPEN_SOURCE -D_BSD_SOURCE -D_SVID_SOURCE -D_GNU_SOURCE -D_EXTENSIONS_ -DLIBRARYBUILD -DNDEBUG -fno-strict-aliasing arcsorter.cc -o unshared/arcsorter.o
/usr/include/bits/stdio-ldbl.h:53: error: type of 'dprintf' is unknown
/usr/include/bits/stdio-ldbl.h:53: error: invalid type in declaration before ';' token
/usr/include/bits/stdio-ldbl.h:53: error: 'int dprintf' redeclared as different kind of symbol
/usr/include/stdio.h:394: error: previous declaration of 'int dprintf(int, const char*, ...)'
make[5]: *** [arcsorter.o] Error 1
make[5]: Leaving directory `/var/tmp/portage/xorg-x11-6.8.2-r7/work/xc/lib/GLU/libnurbs/internals'
make[4]: *** [libnurbs/internals] Error 2
make[4]: Leaving directory `/var/tmp/portage/xorg-x11-6.8.2-r7/work/xc/lib/GLU'
make[3]: *** [all] Error 2
make[3]: Leaving directory `/var/tmp/portage/xorg-x11-6.8.2-r7/work/xc/lib'
make[2]: *** [all] Error 2
make[2]: Leaving directory `/var/tmp/portage/xorg-x11-6.8.2-r7/work/xc'
make[1]: *** [World] Error 2
make[1]: Leaving directory `/var/tmp/portage/xorg-x11-6.8.2-r7/work/xc'
make: *** [World] Error 2
|
|
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Sun May 07, 2006 11:59 pm Post subject: |
|
|
My biggest issues right now are these
- emerging gcc-4.1.0 with gcc-4.1.0
- emerging openoffice2
Both are having problems. I have rebuilt the toolchain and managed to get gcc-4.1.0 built once, but not since, failing every other time. revdep-rebuild keeps wanting to build gcc-3.4.5-r1 and gcc-4.1.0 despite both having already been built, and no other packages in that mix.
Openoffice.org-2.0.1-r1 compiles with no problems, but then fails during installation. Some problem with pyuno.
I will try gcc with other USE flags and see what happens. My goal is to be able to emerge gcc twice in succession. |
|
Back to top |
|
|
fb l33t
Joined: 08 Dec 2003 Posts: 636 Location: New Zealand
|
Posted: Mon May 08, 2006 12:34 am Post subject: |
|
|
davidgurvich wrote: | My biggest issues right now are these
- emerging gcc-4.1.0 with gcc-4.1.0
- emerging openoffice2
Both are having problems. I have rebuilt the toolchain and managed to get gcc-4.1.0 built once, but not since, failing every other time. revdep-rebuild keeps wanting to build gcc-3.4.5-r1 and gcc-4.1.0 despite both having already been built, and no other packages in that mix.
Openoffice.org-2.0.1-r1 compiles with no problems, but then fails during installation. Some problem with pyuno.
I will try gcc with other USE flags and see what happens. My goal is to be able to emerge gcc twice in succession. | The openoffice problem I also had with gcc3.4 last time I tried to rebuild it. I am
trying to have 2.0.2 building (started this moment - and stopped while I was writing this even earlier
than before). About gcc what does "equery d gcc" report? There may be some ebuild still requiring a
specific version of gcc.
Since my openoffice ebuild failed I will try rebuilding gcc to see what happens.
And can you post the error you get as well? |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
Posted: Mon May 08, 2006 4:29 am Post subject: |
|
|
Interesting situation. After researching the options decided to compile gcc-4.1.0 with USE="-fortran" and try and install the g95 fortran compiler instead.
g95 failed to find ansidecl.h from binutils.
Not sure what that means.
Last edited by davidgurvich on Mon May 08, 2006 4:36 am; edited 1 time in total |
|
Back to top |
|
|
davidgurvich Veteran
Joined: 23 Apr 2004 Posts: 1063
|
|
Back to top |
|
|
|