[Buildroot] Announcement: availability of autobuilder toolchains

[Buildroot] Announcement: availability of autobuilder toolchains

[Thomas Petazzoni]

Hello,

One common complaint about the autobuilders was that the builds were
not reproducible because they were using external toolchains that were
not available to others.

I have fixed this now for all the toolchain configuration used on the
Free Electrons build server, and I guess Peter will follow with the
toolchain configurations he is using on the gcc10 and gcc14 build
machines.

So, what I've done is:

 *) For publicly available external toolchains (Sourcery CodeBench,
    Blackfin), I've changed my configurations so that they point to the
    official toolchain instead of pointing to a local directory on the
    build machine. Originally, I was pointing to a local directory to
    save the extraction time of the toolchain, but this time is very
    small compared to the overall build process, and it allows the
    configurations to be buildable by anyone.

 *) For specific external toolchains (built with Buildroot or
    Crosstool-NG), I've uploaded their tarballs at
    http://autobuild.buildroot.org/toolchains/tarballs/, and the
    configuration directly point to them, thanks to the possibility
    added by Thomas De Schampheleire to specify a custom download URL
    for an external toolchain. I did a bit of tricks for the Buildroot
    toolchains to make sure they are relocatable and usable by
    Buildroot through tarballs (I'll post a different e-mail about
    this).

This means that from now on, *all* configurations used on the Free
Electrons build server are independent from the build machine and can
be reproduce by anyone.

The Buildroot base configuration I'm using are all available at:

   http://autobuild.buildroot.org/toolchains/configs/free-electrons/

I hope these new informations will help people reproduce the build
issues and help in fixing them.

Peter, could you:
 * Give me your base Buildroot configuration that you are using on gcc10
   and gcc14
 * Give me the external toolchain configuration you're using on gcc14
   so that I can build a relocatable Buildroot toolchain for it,
   generate a tarball, and put it in
   http://autobuild.buildroot.org/toolchains/tarballs/, and then have
   you update your base Buildroot configuration so that it downloads
   this tarball instead of using a local directory?

Best regards,

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

[Buildroot] Announcement: availability of autobuilder toolchains

[Peter Korsgaard]

>>>>> "Thomas" == Thomas Petazzoni <thomas.petazzoni@free-electrons.com> writes:

 Thomas> Hello,

 Thomas> One common complaint about the autobuilders was that the builds
 Thomas> were not reproducible because they were using external
 Thomas> toolchains that were not available to others.

 Thomas> I have fixed this now for all the toolchain configuration used
 Thomas> on the Free Electrons build server, and I guess Peter will
 Thomas> follow with the toolchain configurations he is using on the
 Thomas> gcc10 and gcc14 build machines.

Cool, thanks!

 Thomas> Peter, could you:

 Thomas>  * Give me your base Buildroot configuration that you are using
 Thomas>  on gcc10 and gcc14

The (internal) gcc10 config is present in every build result (grep -v
BR2_PACKAGE defconfig).

gcc14 toolchain config is:
BR2_powerpc=y
BR2_powerpc_603e=y
BR2_HOST_DIR="/home/peko/br/ppc603e-wl6"
BR2_JLEVEL=8
BR2_TOOLCHAIN_BUILDROOT_LARGEFILE=y
BR2_TOOLCHAIN_BUILDROOT_INET_IPV6=y
BR2_TOOLCHAIN_BUILDROOT_WCHAR=y

gcc110 toolchain config is:
BR2_x86_i686=y
BR2_HOST_DIR="/home/peko/br/i686-wchar"
BR2_JLEVEL=8
BR2_PTHREAD_DEBUG=y
BR2_TOOLCHAIN_BUILDROOT_WCHAR=y

But notice that this is a power7 machine, so an x86 toolchain won't work.

uname -a
Linux gcc1-power7.osuosl.org 3.3.1-3.fc16.ppc64 #1 SMP Wed Apr 11 09:05:31 MST 2012 ppc64 ppc64 ppc64 GNU/Linux

-- 
Bye, Peter Korsgaard

[Buildroot] Announcement: availability of autobuilder toolchains

[Thomas Petazzoni]

Le Sat, 25 Aug 2012 12:03:55 +0200,
Peter Korsgaard <jacmet@uclibc.org> a ?crit :

> The (internal) gcc10 config is present in every build result (grep -v
> BR2_PACKAGE defconfig).

Thanks, added to
http://autobuild.buildroot.org/toolchains/configs/gcc10/toolchain.config.

However note that I haven't see build results coming from gcc10 since
about two days.

> gcc14 toolchain config is:
> BR2_powerpc=y
> BR2_powerpc_603e=y
> BR2_HOST_DIR="/home/peko/br/ppc603e-wl6"
> BR2_JLEVEL=8
> BR2_TOOLCHAIN_BUILDROOT_LARGEFILE=y
> BR2_TOOLCHAIN_BUILDROOT_INET_IPV6=y
> BR2_TOOLCHAIN_BUILDROOT_WCHAR=y

Is this running on a x86/x86-64 machine? If so, then what we need to do
is generate a tarball from the toolchain you have
in /home/peko/br/ppc603e-wl6/, and then use this tarball as the
external toolchain instead of making a reference to this local
directory. This will allow everyone to reproduce the build.

Would you agree to do this?

> gcc110 toolchain config is:
> BR2_x86_i686=y
> BR2_HOST_DIR="/home/peko/br/i686-wchar"
> BR2_JLEVEL=8
> BR2_PTHREAD_DEBUG=y
> BR2_TOOLCHAIN_BUILDROOT_WCHAR=y
> 
> But notice that this is a power7 machine, so an x86 toolchain won't work.

Argh. Then what can we do to allow other people to reproduce the build?
I agree that doing a build targeting x86 on a !x86 machine is really
an interesting test, so I'd like to keep it. Or maybe we just say that
the build results from this machine cannot easily be reproduced by
others and we live with it?

Thanks!

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

[Buildroot] Announcement: availability of autobuilder toolchains

[Peter Korsgaard]

>>>>> "Thomas" == Thomas Petazzoni <thomas.petazzoni@free-electrons.com> writes:

 Thomas> Thanks, added to
 Thomas> http://autobuild.buildroot.org/toolchains/configs/gcc10/toolchain.config.

Great.

 Thomas> However note that I haven't see build results coming from gcc10 since
 Thomas> about two days.

Thanks. It was stuck in midori configuration (python waf?). Perhaps
because gcc10 uses python 2.5.


 >> gcc14 toolchain config is:
 >> BR2_powerpc=y
 >> BR2_powerpc_603e=y
 >> BR2_HOST_DIR="/home/peko/br/ppc603e-wl6"
 >> BR2_JLEVEL=8
 >> BR2_TOOLCHAIN_BUILDROOT_LARGEFILE=y
 >> BR2_TOOLCHAIN_BUILDROOT_INET_IPV6=y
 >> BR2_TOOLCHAIN_BUILDROOT_WCHAR=y

 Thomas> Is this running on a x86/x86-64 machine? If so, then what we
 Thomas> need to do is generate a tarball from the toolchain you have in
 Thomas> /home/peko/br/ppc603e-wl6/, and then use this tarball as the
 Thomas> external toolchain instead of making a reference to this local
 Thomas> directory. This will allow everyone to reproduce the build.

 Thomas> Would you agree to do this?

Yes, x86-64. I'll take a look at your instructions and set it up.


 >> gcc110 toolchain config is:
 >> BR2_x86_i686=y
 >> BR2_HOST_DIR="/home/peko/br/i686-wchar"
 >> BR2_JLEVEL=8
 >> BR2_PTHREAD_DEBUG=y
 >> BR2_TOOLCHAIN_BUILDROOT_WCHAR=y
 >> 
 >> But notice that this is a power7 machine, so an x86 toolchain won't work.

 Thomas> Argh. Then what can we do to allow other people to reproduce
 Thomas> the build?  I agree that doing a build targeting x86 on a !x86
 Thomas> machine is really an interesting test, so I'd like to keep
 Thomas> it. Or maybe we just say that the build results from this
 Thomas> machine cannot easily be reproduced by others and we live with
 Thomas> it?

I think we can just document the toolchain config for this one and not
archive the toolchain.

-- 
Bye, Peter Korsgaard