GCC 5 is coming with a massive rebuild

chaonaut · 2015-03-05 10:28:08

glibc, file, binutils, gmp, mpfr, libmpc, gcc-multilib-git (rebuild against recompiled libraries), bzip2, pkg-config — OK.
this time i stuck building ncurses. the build fails here:

gcc -DHAVE_CONFIG_H -I../ncurses -I../../ncurses-5.9/ncurses  -D_GNU_SOURCE -DNDEBUG
 -I. -I../include -I../../ncurses-5.9/ncurses/../include -mtune=native -O2 -fstack-protector-strong
 -pipe -fomit-frame-pointer --param=ssp-buffer-size=4 -fdiagnostics-color=always 
 --param max-inline-insns-single=1200 -fPIC -c ../../ncurses-5.9/ncurses/base/lib_getch.c
 -o ../obj_s/lib_getch.o
_14547.c:1364:22: error: expected ‘)’ before ‘int’

Octoploid · 2015-03-11 13:18:37

chaonaut wrote:

glibc, file, binutils, gmp, mpfr, libmpc, gcc-multilib-git (rebuild against recompiled libraries), bzip2, pkg-config — OK.
this time i stuck building ncurses. the build fails here:

gcc -DHAVE_CONFIG_H -I../ncurses -I../../ncurses-5.9/ncurses  -D_GNU_SOURCE -DNDEBUG
 -I. -I../include -I../../ncurses-5.9/ncurses/../include -mtune=native -O2 -fstack-protector-strong
 -pipe -fomit-frame-pointer --param=ssp-buffer-size=4 -fdiagnostics-color=always 
 --param max-inline-insns-single=1200 -fPIC -c ../../ncurses-5.9/ncurses/base/lib_getch.c
 -o ../obj_s/lib_getch.o
_14547.c:1364:22: error: expected ‘)’ before ‘int’

See: http://gcc.gnu.org/gcc-5/porting_to.html (Preprocessor issues)

The fix is to add -P to the preprocessor invocation.

(See also https://lists.fedoraproject.org/piperma … 7549.html)

Allan · 2015-03-11 14:34:28

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61832

Octoploid · 2015-03-11 15:09:09

And BTW something like "internal compiler error: in address_matters_p, at symtab.c:1908" isn't a crash,
the compiler just hit an assertion.
A crash looks like "internal compiler error: Segmentation fault"

chaonaut · 2015-03-11 15:50:03

thanks for pointing there, i’ll continue the experiment soon.

hoschi · 2015-04-07 09:22:37

Thanks for the script Allan
Nothing to worry about as user. I "survived" this in the past - as user - on Gentoo two times (GCC 3.2 and GCC 3.4), were you have to recompile basically everything and it worked out well. On Archlinux the complete burden is on the shoulder of the maintainers and developers of Archlinux* and GCC, with a few execptions which boil down to maybe installed packages from AUR (vim-youcompleteme-git, ...).

* Thanks for the upcoming work

headkase · 2015-04-22 16:56:48

Well, GCC 5.1 is officially released: Announcement and Changes.

5.1 is the production release, 5.0 is the development release. So, now with 5.1 out things are well underway.

smirky · 2015-04-22 16:57:28

You mean 6.0 is the dev release.

headkase · 2015-04-22 16:59:55

smirky wrote:

You mean 6.0 is the dev release.

No, see: Here.

5.0.0 (experimental) during active development of GCC 5 (stage 1 to 3)
5.0.1 (prerelease) during the stabilization period of GCC 5
6.0.0 (experimental) during active development of GCC 6 (stage 1 to 3)
5.1.0 for the first release from the GCC 5 branch
5.1.1 during development on the branch post the 5.1.0 release
5.2.0 for the second release from the GCC 5 branch
5.2.1 during development on the branch post the 5.2.0 release
6.0.1 (prerelease) during the stabilization period of GCC 6

5.1.0 is the first production release of the 5 series.

jasonwryan · 2015-04-23 23:11:32

In [testing] now...

kokoko3k · 2015-04-24 07:51:20

Ok, let's see if i understood properly:

* New libstdc++ now introduces a new ABI, but the old ABI is not dropped -> dual ABI.
* Binaries compiled with default gcc 5.1 options will use the new ABI.
* but gcc in [testing] is compiled with --with-default-libstdcxx-abi=c++98
...that means that new compiled code will still use the old ABI.

At this stage, all the old code will still be compatible right?

If/When --with-default-libstdcxx-abi will be removed from the PKGBUILD, what will happen?
I mean, since libstdc++ has this dual ABI concept, if there is a binary compiled with gcc 4.9 that uses libstdc++, will still use the old ABI by default?

Thanks, and sorry for the noob question.

Last edited by kokoko3k (2015-04-24 07:52:07)

AnAkkk · 2015-04-24 08:37:22

kokoko3k wrote:

Ok, let's see if i understood properly:
* New libstdc++ now introduces a new ABI, but the old ABI is not dropped -> dual ABI.
* Binaries compiled with default gcc 5.1 options will use the new ABI.
* but gcc in [testing] is compiled with --with-default-libstdcxx-abi=c++98
...that means that new compiled code will still use the old ABI.
At this stage, all the old code will still be compatible right?
If/When --with-default-libstdcxx-abi will be removed from the PKGBUILD, what will happen?
I mean, since libstdc++ has this dual ABI concept, if there is a binary compiled with gcc 4.9 that uses libstdc++, will still use the old ABI by default?
Thanks, and sorry for the noob question.

Apparently everything will use the old ABI unless you force it to use the new one when compiling. No massive rebuild apparently.

smirky · 2015-04-24 08:46:55

There will be a massive rebuild. It's just a matter of time. Rushing things like this has a huge chance to end with shooting yourself in the foot.

Allan · 2015-04-24 08:48:29

Rebuild will probably happen in two weeks...

kokoko3k · 2015-04-24 09:46:50

AnAkkk wrote:

kokoko3k wrote:
[..]
If/When --with-default-libstdcxx-abi will be removed from the PKGBUILD, what will happen?
I mean, since libstdc++ has this dual ABI concept, if there is a binary compiled with gcc 4.9 that uses libstdc++, will still use the old ABI by default?
Apparently everything will use the old ABI unless you force it to use the new one when compiling. No massive rebuild apparently.

...that is the case when gcc 5.x is compiled with --with-default-libstdcxx-abi=c++98
If i understood properly, new compiled code will use the NEW ABI by default if the switch --with-default-libstdcxx-abi=c++98 is absent.

I was asking about OLD compiled code and his relation with the DUAL ABI concept when no --with-default-libstdcxx-abi=c++98 switch is given.

-EDIT-
will arch drop --with-default-libstdcxx-abi=c++98 in the future or will mantain it and select the default ABI in each package?

Last edited by kokoko3k (2015-04-24 09:49:22)

Allan · 2015-04-24 11:04:56

--with-default-libstdcxx-abi=c++98 will be dropped in a couple of weeks (assuming no big issues with the compiler are found) and then we will do a rebuild.

anatolik · 2015-04-24 20:02:18

Just pushed arm-none-eabi-gcc 5.1.0 to [community-testing]. If you use it for embedded development please test this version. Feedback is welcome.

pauldcox · 2015-04-24 23:44:45

Excuse my ignorance in all this, but after the rebuild is complete, things moved out of testing, yadda yadda, will the old ABI still be available somehow for prebuilt binaries such as Steam and plex etc.?

Last edited by pauldcox (2015-04-24 23:48:34)

Allan · 2015-04-25 02:59:12

Yes - libstdc++ is one binary, two ABIs.

Jristz · 2015-04-25 22:53:08

gcc5 hit testing yay, but was compiled with the old gnu89, buh.
but now my list is bigger

aacskeys
apng-utils
apngopt
codecs
command-not-found
copy-agent
doukutsu
dropbox
gog-beneath-a-steel-sky
gog-dragonsphere
grub-customizer
htmlcxx
jdk
lgogdownloader
love07

grub-customizer, love07, htmlcxx, lgogdownloader, aacskeys, command-not-found, apng-utils, apngopt re the only builded from source, the rest is binary or closed.

I'm conserned for the future... the gnu11

chaonaut · 2015-04-30 10:07:03

jasonwryan wrote:

In [testing] now...

cool.
btw, after gcc-5.1 installation there are no updated symlinks to lto plugin in /usr/lib/bfd-plugins (but dead symlinks to 4.9.2 lto plugin). should i report a bug about that?

Allan · 2015-04-30 10:30:25

No... You should check what package owns those symlinks.

smirky · 2015-04-30 11:08:04

What is the criteria of gcc 5.1.0 getting out of [testing]? Some time without complaints, patches, 5.1.1 or probably all of them?

Allan · 2015-04-30 11:38:51

There are a couple of bug that need fixed.

chaonaut · 2015-04-30 13:06:56

Allan wrote:

No... You should check what package owns those symlinks.

seems like they're made manually.
pacman reports that they don't belong to any package.
quite strange, because they must be there to make that fancy LTO feature work.

Arch Linux

#51 2015-03-05 10:28:08

Re: GCC 5 is coming with a massive rebuild

#52 2015-03-11 13:18:37

Re: GCC 5 is coming with a massive rebuild

#53 2015-03-11 14:34:28

Re: GCC 5 is coming with a massive rebuild

#54 2015-03-11 15:09:09

Re: GCC 5 is coming with a massive rebuild

#55 2015-03-11 15:50:03

Re: GCC 5 is coming with a massive rebuild

#56 2015-04-07 09:22:37

Re: GCC 5 is coming with a massive rebuild

#57 2015-04-22 16:56:48

Re: GCC 5 is coming with a massive rebuild

#58 2015-04-22 16:57:28

Re: GCC 5 is coming with a massive rebuild

#59 2015-04-22 16:59:55

Re: GCC 5 is coming with a massive rebuild

#60 2015-04-23 23:11:32

Re: GCC 5 is coming with a massive rebuild

#61 2015-04-24 07:51:20

Re: GCC 5 is coming with a massive rebuild

#62 2015-04-24 08:37:22

Re: GCC 5 is coming with a massive rebuild

#63 2015-04-24 08:46:55

Re: GCC 5 is coming with a massive rebuild

#64 2015-04-24 08:48:29

Re: GCC 5 is coming with a massive rebuild

#65 2015-04-24 09:46:50

Re: GCC 5 is coming with a massive rebuild

#66 2015-04-24 11:04:56

Re: GCC 5 is coming with a massive rebuild

#67 2015-04-24 20:02:18

Re: GCC 5 is coming with a massive rebuild

#68 2015-04-24 23:44:45

Re: GCC 5 is coming with a massive rebuild

#69 2015-04-25 02:59:12

Re: GCC 5 is coming with a massive rebuild

#70 2015-04-25 22:53:08

Re: GCC 5 is coming with a massive rebuild

#71 2015-04-30 10:07:03

Re: GCC 5 is coming with a massive rebuild

#72 2015-04-30 10:30:25

Re: GCC 5 is coming with a massive rebuild

#73 2015-04-30 11:08:04

Re: GCC 5 is coming with a massive rebuild

#74 2015-04-30 11:38:51

Re: GCC 5 is coming with a massive rebuild

#75 2015-04-30 13:06:56

Re: GCC 5 is coming with a massive rebuild

Board footer