Integration per-component tests in Fedora

Nick Coghlan ncoghlan at redhat.com
Tue Oct 28 07:04:52 UTC 2014


On 10/22/2014 12:06 AM, Matthew Miller wrote:
> On Tue, Oct 21, 2014 at 01:53:43PM +0200, Honza Horak wrote:
>>> There was previously some discussion on this (I think on -devel? I'll
>>> look for the thread)
>> Ah, thanks, I probably found it:
>> https://lists.fedoraproject.org/pipermail/devel/2014-January/193498.html
> 
> Yes, there it was. Thanks. :)
> 
> 
>> So, the only disadvantage I found was that keeping tests further
>> from components' sources might make users not keeping tests in sync
>> with packages.
> 
> Adam's point in that thread was (and I think I'm representing it
> fairly) that spec files are already horrendous enough, and that we
> don't want to add "learn how to package RPMs" to the bar to cross in
> order to contribute tests.
> 
> (And I guess along the same lines, if you're not a maintainer of the
> package, adding tests causes you to add extra packaging work for the
> people who are the maintainers.)
> 
> Once we have disposable/relatively-safe test sandboxes, I was thinking
> that possibly anyone in some low-threshold test-writer group could have
> commit access to all packages, and all tests would run, but each test's
> impact would be decided by a higher-threshold group. For example, if I
> create a test for httpd, it would first send results just to me and to
> some big bucket of all results. Then, it could be increased to send
> failures to the maintainers and/or to bodhi and/or to the devel or
> testing lists, and with varying levels of impact in bodhi (block, -1,
> 0, or whatever.)

+1 for being able to initially develop new tests in a low impact
advisory only mode - it provides a chance to check for intermittent
problems in the test itself before making it passing a barrier to
publishing an update.

Regards,
Nick.

-- 
Nick Coghlan
Red Hat Hosted & Shared Services
Software Engineering & Development, Brisbane

HSS Provisioning Architect


More information about the env-and-stacks mailing list