Bug#852814: Fails at 'policy-rc.d already exists'

[James Clarke]

On 27 Jan 2017, at 19:10, Guido Günther <agx at sigxcpu.org> wrote:
> 
> Hi Mattia,
> On Fri, Jan 27, 2017 at 07:27:09PM +0100, Mattia Rizzolo wrote:
>> On Fri, Jan 27, 2017 at 06:41:33PM +0100, Guido Günther wrote:
>>>> # To use a chroot with glibc 2.13 or older (Debian wheezy or older)
>>>> # with Linux 4.8 or up, you may need to add this to /etc/default/grub:
>>>> # GRUB_CMDLINE_LINUX="rootdelay=5 syscall.x32=y vsyscall=emulate net.ifnames=0 kaslr"
>>>> # (well, the “vsyscall=emulate” part, although the others don’t hurt)
>>>> 
>>>>> This was reported properly as http download error (since apt noticed
>>>>> that it's helper died) in 0.227. But 0.228.3 simply dies without any
>>>>> hint what the error could be and where it happened to the user. It would
>>>>> be great if pbuilder would at least print the command that failed.
>>>> 
>>>> Agreed. I’d lower the severity since apparently, the vsyscall=none
>>>> switch was reverted (especially as wheezy is still very much in use).
>>> 
>>> I'll let the maintainers decide if they want to treat is as a regression
>>> or enhancement / wishlist bug .
>> 
>> I think I'd rather close it, tbh.
>> The failure comes at this line
>>    ARCHITECTURE="${ARCHITECTURE:-"$($CHROOTEXEC dpkg-architecture -qDEB_BUILD_ARCH)"}"
>> catching a segfault there would just be noise in the code for a very
>> weird situation.
> 
> It would just be totally awesome if one could see that _without_ looking
> at the code (e.g. by being able to toggle set +x easily). In order to
> make $builder, git, etc invocations easier to debug in gbp we have:
> 
>    https://github.com/agx/git-buildpackage/blob/master/gbp/scripts/buildpackage.py#L540
> 
> in git-buildpackage and for the pbuilder/cowbuilder case specifically:
> 
>    https://github.com/agx/git-buildpackage/blob/master/bin/git-pbuilder#L284
> 
> so we have the exact invocation on update and
> 
>    https://github.com/agx/git-buildpackage/blob/master/bin/git-pbuilder#L336
> 
> on pdebuild. Having something similar in pbuilder would just be great so
> one can dig out why pbuilder exited with ret != 1.
>> 
>> BTW, that thing was reverted, as you can see.  I think that if you just
>> update your kernel (and reboot) it should just work.  At the very least,
>> I could use a wheezy chroot without any issue last week.
> 
> I know, I'm not referring to the initial error cause anymore. See
> above. I looked at the docs only found --loglevel in pbuilder's manpage
> which isn't accepted by cowbulder.

There's --debug for pbuilder, which cowbuilder passes on. This will do a
`set -x`. It will also keep your chroot build directory around rather than
removing it on failure.

Regards,
James