| View previous topic :: View next topic |
| Author |
Message |
cwr
Veteran
Joined: 17 Dec 2005
Posts: 1484
|
 Posted: Fri Apr 06, 2012 8:52 am Post subject: GCC 4.6.2 build flags |
 |
|
GCC 4.6.2 is still masked for ~x86, but I need that specific version to build a (working)
cross-compiler. I can unmask it, but the build then fails with:
| Code: |
Comparing stages 2 and 3
warning: gcc/cc1plus-checksum.o differs
warning: gcc/cc1-checksum.o differs
Bootstrap comparison failure!
gcc/varasm.o differs
make[2]: *** [compare] Error 1
|
Does anyone know of any specific build flags that I could add to make.conf to enable
GCC 4.6.2 to compile? (I'm building it with a current 32-bit x86 system and GCC 4.5.3).
Thanks - Will |
|
| Back to top |
|
 |
mv
Advocate
Joined: 20 Apr 2005
Posts: 3226
|
| Posted: Sat Apr 07, 2012 6:44 am Post subject: |
|
|
| gcc-4.6.* needs a rather recent glibc to compile properly. I forgot which version, but it may very well be 2.14 or even 2.14.1 |
|
| Back to top |
|
|
Veldrin
Veteran
Joined: 27 Jul 2004
Posts: 1937
Location: Zurich, Switzerland
|
| Posted: Sat Apr 07, 2012 9:00 am Post subject: |
|
|
I was able to build gcc-4.6.2 with glibc-2.13-r4 and binutils-2.21.1r1 back in November.
If that still fails, try upgrading glibc or even binutils to the latest unstable.
V.
_________________
read the portage output!
If my answer is too short, just ask for an explanation. |
|
| Back to top |
|
|
cwr
Veteran
Joined: 17 Dec 2005
Posts: 1484
|
| Posted: Sat Apr 07, 2012 9:15 am Post subject: |
|
|
Yes, that's what I'm using, but after three attempts with different USE flags I gave up and
downloaded the package from gnu.org. So far, that's compiling ok ...
Thanks for the confirmation of the version numbers; it means something is off in my system.
Since its a new and clean Stage 3 install I'm not sure there's much I can do about that.
Will
l |
|
| Back to top |
|
|
mv
Advocate
Joined: 20 Apr 2005
Posts: 3226
|
| Posted: Sat Apr 07, 2012 8:19 pm Post subject: |
|
|
I can confirm now some problems: On one of my 3 systems (-march=pentium3) the gcc does not compile properly (differing stages as in your case), even if originally built with gcc-4.6.2.
It built on this system previously. On a very similar system (-march=native with an athlon) there is no such problems.
It might be related with graphite which is the only difference in the CFLAGS (besides -march) on these systems.
Since gcc-4.7 will come out soon, I make no further tests now. |
|
| Back to top |
|
|
cwr
Veteran
Joined: 17 Dec 2005
Posts: 1484
|
| Posted: Sun Apr 08, 2012 12:11 pm Post subject: |
|
|
| mv wrote: | I can confirm now some problems: On one of my 3 systems (-march=pentium3) the gcc does not compile properly (differing stages as in your case), even if originally built with gcc-4.6.2.
It built on this system previously. On a very similar system (-march=native with an athlon) there is no such problems.
It might be related with graphite which is the only difference in the CFLAGS (besides -march) on these systems.
Since gcc-4.7 will come out soon, I make no further tests now. |
Yup, that's the one. I've tried pentium3 and native, and neither build (but with different problems). OTOH both
4.6.2 and 4.6.3 build correctly from the original gnu.org packages; the trouble is I'm not sure how to package
the results. I'm not sure where Gentoo gets its gcc patches, but it looks as if one of them may be the problem.
I'll see if the graphite flag makes any difference; it's a nuisance, since I only want 4.6 as the basis for a
cross-compiler - I don't want to build any Gentoo software with it.
Will |
|
| Back to top |
|
|
mv
Advocate
Joined: 20 Apr 2005
Posts: 3226
|
| Posted: Sun Apr 08, 2012 1:37 pm Post subject: |
|
|
I have now tried the new gcc-4.7 from the hardened-dev overlay, and it built on all of my systems without any problems.
There is also gcc-4.6.3 in that overlay (but I have not tried). So it is probably not worth looking for the bug in gcc-4.6.2 |
|
| Back to top |
|
|
cwr
Veteran
Joined: 17 Dec 2005
Posts: 1484
|
| Posted: Sun Apr 08, 2012 9:35 pm Post subject: |
|
|
| mv wrote: | I have now tried the new gcc-4.7 from the hardened-dev overlay, and it built on all of my systems without any problems.
There is also gcc-4.6.3 in that overlay (but I have not tried). So it is probably not worth looking for the bug in gcc-4.6.2 |
Well, the hardened 4.6.3 doesn't build for me - once again it fails with varasm.o differing between stages 2 and 3.
Since I've compiled the original package several times now it must be a Gentoo patch that's breaking things,
but I've no idea which.
Will |
|
| Back to top |
|
|
Hu
Watchman
Joined: 06 Mar 2007
Posts: 7679
|
| Posted: Sun Apr 08, 2012 10:01 pm Post subject: |
|
|
| cwr: could you post your emerge --info so that others can more readily try to replicate the specific failure? |
|
| Back to top |
|
|
cwr
Veteran
Joined: 17 Dec 2005
Posts: 1484
|
| Posted: Mon Apr 09, 2012 9:59 am Post subject: |
|
|
| Hu wrote: | | cwr: could you post your emerge --info so that others can more readily try to replicate the specific failure? |
I've filed a bug under GCC 4.6.2; I had another shot at rebuilding that, since the ebuild is (masked) in the standard tree.
The results are pretty consistent - it always fails with a mismatched stage2-gcc/varasm.o and stage3-gcc/varasm.o
Will |
|
| Back to top |
|
|
mv
Advocate
Joined: 20 Apr 2005
Posts: 3226
|
| Posted: Thu Apr 12, 2012 7:04 pm Post subject: |
|
|
| I can confirm now that gcc-4.6.3 from the hardened-development overlay also fails with -march=pentium3. It works on the same system with -march=native (which is on this system an intel core2 in 32bit mode). In both cases, graphite flags were active, so graphite seems to be unrelated. So it seems that only -march=pentium3 is broken. |
|
| Back to top |
|
|
cwr
Veteran
Joined: 17 Dec 2005
Posts: 1484
|
| Posted: Sun Apr 15, 2012 12:36 am Post subject: |
|
|
| mv wrote: | | I can confirm now that gcc-4.6.3 from the hardened-development overlay also fails with -march=pentium3. It works on the same system with -march=native (which is on this system an intel core2 in 32bit mode). In both cases, graphite flags were active, so graphite seems to be unrelated. So it seems that only -march=pentium3 is broken. |
Unfortunately I tried -march=native with 4.6.2, and it also failed - as I recall a slightly different faillure to -march=pentium3, but
it still failed to build. I've managed to find a binary package of version 4.5.1 of the gcc-avr cross-compiler I wanted,
(the 4.5.3 cross-compiler builds but produces buggy code), so I'm in no rush for 4.6.2. Since the package taken directly
from gnu.org, with no patches, builds correctly the problem is somewhere in the gentoo patches, but I've no idea where.
Will |
|
| Back to top |
|
|
alexbuell
Guru
Joined: 18 Jul 2002
Posts: 433
Location: "Hemp"shire, UK
|
| Posted: Wed Aug 01, 2012 10:26 am Post subject: |
|
|
| cwr wrote: | | mv wrote: | | I can confirm now that gcc-4.6.3 from the hardened-development overlay also fails with -march=pentium3. It works on the same system with -march=native (which is on this system an intel core2 in 32bit mode). In both cases, graphite flags were active, so graphite seems to be unrelated. So it seems that only -march=pentium3 is broken. |
Unfortunately I tried -march=native with 4.6.2, and it also failed - as I recall a slightly different faillure to -march=pentium3, but
it still failed to build. I've managed to find a binary package of version 4.5.1 of the gcc-avr cross-compiler I wanted,
(the 4.5.3 cross-compiler builds but produces buggy code), so I'm in no rush for 4.6.2. Since the package taken directly
from gnu.org, with no patches, builds correctly the problem is somewhere in the gentoo patches, but I've no idea where.
Will |
Folks, temporarily change your CFLAGS to use i686 and 4.6.3 should build correctly. Change CFLAGS back once it's built. Worked for me.
_________________
Cheers,
Alex.
Linux - the best text adventure game ever. |
|
| Back to top |
|
|
|
Portage & Programming
