[waffle] Checking In

[Chad Versace]

On Wed 22 Oct 2014, Emil Velikov wrote:
> On 11/09/14 15:43, Chad Versace wrote:
> > On 09/07/2014 05:37 PM, Jordan Justen wrote:

> > Having a year between 1.3 and 1.4, I'm reluctant to introduce delays. 
> > To avoid
> > delays like that in the future, I'm thinking that waffle should adopt a
> > short,
> > time-based release cycle. Maybe every 4 to 6 weeks. With a cycle that
> > short...
> > 
> >     1. There would be no pressure to delays a release for a feature.
> > Because
> >        a new release is right around the corner.
> >     2. If there's nothing interesting to release for a given cycle, then we
> >        could just skip doing that release.
> >     3. The release process would, by necessity of frequency, become mostly
> >        automated. That automation would allow anyone (not just me) to do
> >        a release.
> > 
> > I haven't *decided* on any of the above. Just thinking that it may be a
> > good
> > idea. What do you think?
> > 

> All of those sounds ok with me, yet I hope that the level of automation
> is better than the one we have in mesa.

I began writing a python script to fully automate that process. The
script builds waffle, runs its testsuite, makes a git tag, creates
a tarball and signs it, uploads everything, and then writes out
a template announcement email.

I should dig that script up, clean it up, and commit to the repo.

> > So... let's try not to delay 2.0. And let's make releases more frequent.
> > Then
> > Emil's work on issue #9 will land quickly enough.
> >
> 
> Afaict one of your concerns may be bumping the library major version due
> to the waffle_get_proc_address API change.
> Admittedly I have not checked if piglit and the wfl utils will work if
> we leave the interface as is and I'm planning to do so tomorrow. If
> possible I'll avoid the API change, otherwise I might consider adding
> another entry point in order to preserve backwards compatibility - ie.
> waffle_get_proc_address2().
> 
> Either one of those will get is in waffle-1.5 land which does not sound
> as scary :P

I would like to avoid breaking ABI compatibility unless the alternatives
are really really bad. Because breaking ABI compatibility, being
really^3 bad, is even worse than really^2 bad. I discussed introducing
waffle_get_proc_address2() with Jordan in person, and (if I recall
correctly), he was ok with the idea.