Invalid/ignored FE builds 2006-09-16

[Michael Schwendt]

On Sat, 16 Sep 2006 15:55:50 +0300, Ville Skyttä wrote:

> Packages apparently rebuilt without any NEVR changes and thus *not
> published* even if the build report says otherwise:
> 
> FC-5:
>   hamzy AT us.ibm.com
>     sblim-cmpi-base 
> 
> devel:
>   hamzy AT us.ibm.com
>     sblim-cmpi-base    
>     sblim-cmpi-devel
>     sblim-testsuite
>   paul AT all-the-johnsons.co.uk
>     fuse-emulator-utils
>     genchemlab
>     gonvert 
>     lib765
>     libdsk
>     libopts
> 
> More info:
> http://www.redhat.com/archives/fedora-extras-list/2006-September/msg00293.html
> 

For some minutes I thought we could fix this easily at the level of the
push-script by excluding such invalid builds. But I didn't think about the
way we push individual build-jobs from the need-sign queue -- it is not an
atomic operation. It could happen that disk space or file permissions
break in the middle of a push, and then the next run would need to
continue there (which the script does fine, but it cannot guess too much
based on files which exist already in the repository, hence the warnings
we see).

So, in the meantime it's best if the packagers do the thing.