Bug#782553: sbuild: 'sbuild --source --no-arch-all' fails in a very confusing way

[Johannes Schauer]

Hi,

On Mon, 13 Apr 2015 23:06:35 -0700 Dima Kogan <d...@secretsauce.net> wrote:
> The --[no-]arch-all and --source options all touch the dpkg-buildpackage -b
> and -B options, and it is possible for the user to mistakenly request an
> incompatible combination that results in very confusing failures. In my case
> I had $sbuild_source=1 in the config file, and I was running 'sbuild
> --no-arch-all'. This was ignoring the --no-arch-all partially:
> 'dpkg-buildpackage -B' was being invoked, in direct contradiction with the
> --no-arch-all request. However later on, sbuild would say
> 
>     E: Package builds hello-all_2.10_all.deb when binary-indep target is not 
> called.  This is a bug in the packaging.
> 
> and then it tries to do stuff with this package (that wasn't supposed to
> be built), and says stuff like
> 
>    dpkg-deb: error: failed to read archive 
> `/«CHROOT»/«BUILDDIR»/hello-all_2.10_all.deb': No such file or 
> directory
> 
>    dpkg-deb: error: failed to read archive 
> `/«CHROOT»/«BUILDDIR»/hello-all_2.10_all.deb': No such file or 
> directory
> 
>    mv: cannot stat '/«CHROOT»/«BUILDDIR»/hello-all_2.10_all.deb': 
> No such file or directory
>    E: Could not move hello-all_2.10_all.deb to .
>    du: cannot access '/tmp/hello-all_2.10_all.deb': No such file or directory
> 
> and exits with 'Status: attempted'.

I am not able to reproduce this behaviour. I have $sbuild_source=1; in my
~/.sbuildrc but neither do I see any of the above error messages nor do I get
'Status: attempted'. Can you confirm that this issue was fixed with the recent
sbuild release?

> This was really confusing to the user (me) because on the commandline I was
> explicitly requesting --no-arch-all. The --source was being given in the
> config file, and even then, it's not immediately obvious it's related. Can I
> request we detect such cases and barf early with a clear error message?
> 
> It's also possible to issue more clearly illegal requests, such as
> 'sbuild --arch-all --no-arch-all'. This succeeds doing something, but one of
> the requests was ignored. Things like this should be illegal too.

If you can confirm that the first issues is no more, then instead of closing
this bug, I'd like to make it into one to track improvements to the
documentation and behaviour of argument parsing. Does that sound good to you?

Thanks!

cheers, josch

signature.asc
Description: signature