----- "Will Woods" <wwoods(a)redhat.com> wrote:
On Tue, 2010-08-10 at 15:32 +0200, Kamil Páral wrote:
> The hooks can now use their process_testdata method to check the
> correctness of provided options and arguments and abort the program
if
> something is wrong. It also allows hooks to have some mandatory
options
> defined.
This patch looks fine, but one question:
> -def process_testdata(opts, args, arch, **extra):
> +def process_testdata(parser, opts, args, **extra):
Where'd 'arch' go? Was it just not needed?
That was part of my control.autoqa patch:
http://git.fedorahosted.org/git/?p=autoqa.git;a=commitdiff;h=55ddd92dfffe...
If you see the current workflow, hook.process_testdata() is now called
just once for the whole job, not several times for every arch. This is
needed, because hook.process_testdata() must be executed before
control.autoqa is evaluated, and control.autoqa needs also to be run
just once for the whole job, not once for every arch (otherwise
control.autoqa wouldn't be able to modify the archlist).
The result is the hooks are now not able to modify test input values
according to a specific architecture. They have to pass some general
value or the autoqa harness must be run several times for different
architectures if needed (in the aforementioned patch I have modified
watch-composes.py in exactly that way - it runs autoqa several times,
because different archs need different arguments).