GCC 4.3.0 testing!

[unK]

I suppose gcc-4.3 for x86 and x86_64 has its own problematic packages (eg. openssl)

Btw openssl compiled fine with gcc-4.3.0-alpha20070928, so now whole system can be compiled with gcc-4.3 (at least on x86)

Also, cabextract and exiv2 both hang on 'checking for working mktime... ' during configure.

Any bright ideas?

Did you wait a moment?
It just stops for one minute max and then goes again.

[eremini]

Path to make tar compile

Code: Select all

--- lib/argp.h.orig     2007-09-25 11:04:18.000000000 +0200
+++ lib/argp.h  2007-09-25 11:06:24.000000000 +0200
@@ -580,7 +580,11 @@
 # endif
 
 # ifndef ARGP_EI
-#  define ARGP_EI extern __inline__
+#  if defined __GNUC_STDC_INLINE__
+#    define ARGP_EI extern __inline__ __attribute__((__gnu_inline__))
+#  else
+#    define ARGP_EI extern __inline__
+#  endif
 # endif
 
 ARGP_EI void
--- lib/argp-fmtstream.h.orig   2007-09-25 11:06:49.000000000 +0200
+++ lib/argp-fmtstream.h        2007-09-25 11:07:35.000000000 +0200
@@ -198,8 +198,12 @@
 #endif
 
 #ifndef ARGP_FS_EI
+#ifdef __GNUC_STDC_INLINE__
+#define ARGP_FS_EI extern inline __attribute__((__gnu_inline__))
+#else
 #define ARGP_FS_EI extern inline
 #endif
+#endif
 
 ARGP_FS_EI size_t
 __argp_fmtstream_write (argp_fmtstream_t __fs,

[eremini]

Yep, openssl fails here as well

[laughinggnome]

Did you wait a moment?
It just stops for one minute max and then goes again.

After 9 3/4 hours it's still just sitting there!

Most of the the other problems are, I think, due to x86_64 and missing headers.
It's fun though, this messing up a perfectly good and working system with something new and exciting!

[b33fc0d3]

are we allowed to post these patches on bugs.gentoo.org?

Only file bugs for Gentoo specific bugs, like glibc-2.6.1 contains a gentoo specific patch for gcc-4.3. Send generic patches upstream.

[b33fc0d3]

Did you wait a moment?
It just stops for one minute max and then goes again.

After 9 3/4 hours it's still just sitting there!

Most of the the other problems are, I think, due to x86_64 and missing headers.
It's fun though, this messing up a perfectly good and working system with something new and exciting!

Try the GCC "-fwrapv" compiler flag for packages that do this at configure. It's a temporary hack and very ugly.

[b33fc0d3]

about:buildconfig

Build platform
target
x86_64-pc-linux-gnu

Build tools
Compiler Version Compiler flags
x86_64-pc-linux-gnu-gcc gcc version 4.3.0-alpha20070921 (experimental) (GCC) -Wall -W -Wno-unused -Wpointer-arith -Wcast-align -Wno-long-long -march=athlon64 -fPIC -Wno-return-type -w -pthread -pipe
x86_64-pc-linux-gnu-g++ gcc version 4.3.0-alpha20070921 (experimental) (GCC) -fno-rtti -fno-handle-exceptions -Wall -Wconversion -Wpointer-arith -Wcast-align -Woverloaded-virtual -Wsynth -Wno-ctor-dtor-privacy -Wno-non-virtual-dtor -Wno-long-long -march=athlon64 -fPIC -Wno-return-type -w -fpermissive -fshort-wchar -pthread -pipe -DARON_WAS_HERE -fpermissive -I/usr/X11R6/include

Configure arguments
--enable-application=browser '--enable-optimize=-O3 -ftree-vectorizer-verbose=3' --disable-installer --disable-pedantic --enable-crypto --with-system-jpeg --with-system-png --with-system-zlib --disable-updater --enable-default-toolkit=gtk2 --enable-pango --enable-svg --enable-svg-renderer=cairo --enable-system-cairo --disable-strip --disable-strip-libs --enable-single-profile --disable-profilesharing --disable-profilelocking --disable-ipv6 --disable-xinerama --disable-xprint --disable-freetype2 --disable-debug --disable-tests --enable-reorder --enable-application=browser --enable-image-encoder=all --enable-canvas --with-system-nspr --with-system-nss --enable-official-branding --enable-oji --enable-mathml --disable-jsd --disable-xpctools --with-default-mozilla-five-home=/usr/lib64/mozilla-firefox --enable-extensions=default,typeaheadfind,-venkman --prefix=/usr --host=x86_64-pc-linux-gnu --mandir=/usr/share/man --infodir=/usr/share/info --datadir=/usr/share --sysconfdir=/etc --localstatedir=/var/lib --libdir=/usr/lib64 --build=x86_64-pc-linux-gnu

[unK]

Ok, I fixed all packages which failed for me and I have full gcc-4.3.0 system, except kernel (asm error) and minefield :)

All patches are available here. I also added eremini's patch for tar and cpio, I think it is nice to have them all in one place, but if you mind, I'll remove it.

ps. I am not sure about safety of patch for jpeg and id3lib.

cheers.

[eremini]

unK: of course thats fine

Confirmed OpenSSL compiles with the latest gcc snapshot

[eremini]

Yep, mm-sources don't compile

CC arch/i386/lib/usercopy.o
arch/i386/lib/usercopy.c: In function ‘__copy_to_user_ll’:
arch/i386/lib/usercopy.c:777: error: ‘asm’ operand has impossible constraints
make[1]: *** [arch/i386/lib/usercopy.o] Error 1

[enderandrew]

Yep, mm-sources don't compile

CC arch/i386/lib/usercopy.o
arch/i386/lib/usercopy.c: In function ‘__copy_to_user_ll’:
arch/i386/lib/usercopy.c:777: error: ‘asm’ operand has impossible constraints
make[1]: *** [arch/i386/lib/usercopy.o] Error 1

Send an email to Andrew Morton and/or the LKML. I'm sure someone will post a quick fix.

[unK]

I'm quite bored so I started playing around with cflags and I just noticed that using gcc-4.3 and flag -ftree-loop-linear mesa compiles. But with xorg-server it still fails.

Btw I changes my sane cflags to these: "-Os -march=native -pipe -fomit-frame-pointer -freorder-blocks -fno-ident -freorder-blocks-and-partition -fgcse-sm -fgcse-las -fgcse-after-reload -fweb". I'm going to recompile world with them and gcc-4.3 and see what happens

[nbvcxz]

Btw I changes my sane cflags to these: "-Os -march=native -pipe -fomit-frame-pointer -freorder-blocks -fno-ident -freorder-blocks-and-partition -fgcse-sm -fgcse-las -fgcse-after-reload -fweb". I'm going to recompile world with them and gcc-4.3 and see what happens

Should work ok - see my table to avoid possible conflicts, but first try them all - gcc-4.3.0 is still being improved and some data in my table can be not up to date.

[eremini]

ghostscript-gpl

i686-pc-linux-gnu-gcc -DHAVE_MKSTEMP -DHAVE_HYPOT -DHAVE_FILE64 -O2 -fPIC -Wall -Wstrict-prototypes -Wmissing-declarations -Wmissing-prototypes -Wwrite-strings -Wno-strict-aliasing -fno-builtin -fno-common -DHAVE_STDINT_H -DGX_COLOR_INDEX_TYPE="unsigned long long" -march=native -O2 -pipe -fomit-frame-pointer -ftree-vectorize -ftracer -ftree-loop-linear -I./obj/../soobj -I./src -o ./obj/../soobj/gdevtfnx.o -c ./src/gdevtfnx.c
./src/gdevtfnx.c: In function ‘tiff12_print_page’:
./src/gdevtfnx.c:98: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
make[1]: *** [obj/../soobj/gdevtfnx.o] Error 1
make[1]: Leaving directory `/var/tmp/portage/app-text/ghostscript-gpl-8.57/work/ghostscript-8.57'
make: *** [so] Error 2

Edit:
Never mind it fails on gcc 4.2 as well, its the cflags (-march=native -O2 -pipe -fomit-frame-pointer -ftree-vectorize -ftracer -ftree-loop-linear).

[nbvcxz]

@ eremini - try with no -ftree-* flags

[unK]

using -ftree* on x86 can only cause trouble imo (especially ftree-loop-linear).

Edit: Ok, -Os is not really good, xorg-server and compiz-fusion compiled with it have horrible performance, now I'll try -O3 (and -O2 for qt&kde).
Edit2: LoL, It seems messing with cflags using experimental gcc doesn't give good results at all, so I set my defaults again.

[eremini]

patch to make in portage antlr compile

Code: Select all

--- lib/cpp/antlr/CharScanner.hpp.orig	2007-10-06 12:10:48.165620503 +0100
+++ lib/cpp/antlr/CharScanner.hpp	2007-10-06 12:11:51.487277141 +0100
@@ -55,6 +55,7 @@
 	}
 }
 #else
+#include <string.h>
 #ifdef NO_STRCASECMP
 ANTLR_C_USING(stricmp)
 #else

Haven't submitted this upstream because I don't know if the latest upstream version has the same bug

[eremini]

libglade also fails for me with gcc 4.3. Anyone can confirm?

[unK]

libglade also fails for me with gcc 4.3. Anyone can confirm?

yeah. use b33fc0d3's patch from here to fix it.

[b33fc0d3]

*** REMOVED ***

[unK]

Openssl compiled with gcc-4.3 is broken (apache compiled with ssl won't start and subversion won't checkout from https://*).

[Dottout]

hi, thank you for your patches, Ive almost finished to emerge world with gcc-4.3.0-alpha20071005 (CFLAGS="-O2 -march=core2 -pipe -fomit-frame-pointer")
well, does anyone have a fix for amsn? it dies at login with 'illegal instruction'
I also confirm again svn issue, checkout from https doesnt work

[tomt21]

Sorry if this is a stupid newbie question but how do you install gcc 4.3? i got 4.2 to install by unmasking it but there doesnt seem to be an ebuild for 4.3.

I don't want to use it as my main compiler, i just want to use it to build python 2.5 and compile some .so files because i'm having problems getting openmp to work using ctypes in python with 4.2 (it complains about dlopen not being able to open libgomp)

[cruzki]

you need to use layman and fetch the overlay toolchain.

#emerge layman
#layman -f
#layman -a toolchain

Then unmask (or keyword, I don't remember) gcc-4.3 and install

Good lock

[tomt21]

Thank you! I looked for an overlay but I didn't know gcc was in the toolchain one (I realise how stupid that sounds.. in my defense I was looking for an overlay called gcc!)