[buildd-tools-devel] Bug#609151: [Fwd: Re: Bug#609151: sbuild reported architecture is wrong when using a chroot with a personality]

[Jamie Strandboge]

On Fri, 2011-01-07 at 19:30 +0000, Roger Leigh wrote: 
> > An easy fix is to start the chroot and set the Arch before before
> > calling open_build_log(), which is what earlier versions of sbuild did.
> > Attached is a patch to do just that.
> 
> I'm assuming from the above that you have an i386 chroot on an amd64
> host system?
> 
Yes.

> The arch is initially set from the host architecture.  You can set this
> using the --arch option, which you should arguably be doing if the
> host and build arches differ.
> 
Hmm, this used to work with 0.60.0-2, but I guess it was changed due to
the issues you mention below.

> Thanks for the patch.  I've had a look at it, and I'm afraid it can't
> be applied in its current form because it breaks cleanup on failure.
> You have moved a chunk of the setup code earlier in the run() function,
> but it's still using the old label.  Also cleanup_log is now out of
> order.
> 
> Those can be fixed, but the order of the log opening is important.  If
> the build log is not open when we start the session, any errors will
> not be logged correctly.  They will go into the main build log rather
> than the package build log.  This is something we've got wrong in the
> past and which caused great annoyance to the buildd admins.  Even
> harmless informational messages get into the main log, and these are
> then spammed to the admins for each build (the main log is normally
> completely empty except on failure, and this change breaks this
> requirement).  The intention is that the main log is only filled when
> there's a major failure prior to starting or after ending a package
> build, and which the buildd admin must be made aware of.
> 
> The bug isn't in the position of open_build_log(), it's in the
> misdetection of the architecture when we open the log.
Yes, I should have probably explained that better -- the arch isn't
properly detected when the log is opened.

> Now, we have
> not currently got any way of specifying this automatically; I may add a
> suitable configuration option to schroot, I've been considering it for
> a while.  But I think that at present you should be using the --arch
> option which is also used to search for the correct chroot.
> 
> i.e. you should be using
> 
>   -d maverick --arch=i386
> 
> rather than
> 
>   -d maverick-386
>
> since the latter also rewrites the architecture in the .changes; if
> you want to specify a specific chroot, then use the --chroot option
> rather than --dist.  --dist specifies the exact distribution, while
> --chroot specifies the build chroot (which normally defaults to a
> chroot based upon the distribution and architecture set e.g.
> maverick-i386-sbuild).

This makes sense. I'll adjust our build scripts accordingly. Thanks for
your consideration and thoughtful response.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part
URL: <http://lists.alioth.debian.org/pipermail/buildd-tools-devel/attachments/20110107/464a7351/attachment-0001.pgp>