On Fri, 17 Feb 2012 00:45:00 +0100, Marek Olšák <mar...@gmail.com> wrote: > I think the problem is not in the process itself, but in the fact that > this project lacks manpower (at least outside of the Intel camp) and I > am afraid it will not get better without another 5-10 people working > full-time on the project. Independent contributors like myself will > never have enough time to implement tests for every possible corner > case which may occur.
I don't think I want the project to demand a complete set of tests (nor did Ian's post ask for that), but I think some sort of assessment along with the first pile of code of "here's a summary of what's in the spec that I know I haven't finished for testing and implementation" is entirely reasonable and would have would have been a big deal to have. I came into EXT_texture_integer thinking that things were supposed to be about half done in core, since it said ~50%. I think today it might be close to 50%, and piglit testing is maybe at 25%. (I only think that because of our internal testsuite, where we're at about 50%) On the other hand, after my experience with the "50% done" GL3 features, I started poking at MSAA (supposedly already done in core) assuming that nobody had done anything and that everything was broken. So far, the record is that >50% of tests I write show things that are broken, even though this is something that was supposed to have already been done in core. Similarly, when I hit a bug in ARB_copy_buffer, I decided to go write tests for the spec instead of just patching up the driver: it took two hours, and at the end I posted a list of things that I knew were weaknesses in our testing that the next person picking up ARB_copy_buffer might want to think about. That's the model I'd like to see, except that it should come with the first implementation.
pgpNPZysi0Cxg.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
