Phoronix: Mesa's "Stupid Development Model" Discussed
Red Hat's David Airlie has started a new mailing list discussion that's surrounding the "stupid development model" of Mesa. Their accepted policy of developing in stable branches and then pulling the code into the master code-base periodically (rather than just working directly on master) is causing many frustrations for Dave in being able to back-port fixes to existing stable branches of Mesa...
I assume these problems are not adressed with libv's new model?
For the problem statement that Dave (curiously i am sure he will maintain that this sudden change of ideas was in no way related to unified driver stacks and the dri work) came up with in the start of the second thread, separate dri/gallium drivers and unified driver stacks would indeed be the best way forward.
But of course, Dave and some others are still unwilling to acknowledge that. I wonder how long it will take them to properly start thinking in this direction.
In any case, it looks as if there was a lot of noise, that the wrong tree is being barked up, and that in the end nothing major will be changed and that at best, only very minimal improvements will come from this discussion.
Mesa development should just reorganize and follow the Linux development model. It's tried and tested - it just works.
I still think Dave's unfortunately true oxymoron is hilarious, development in stable.
I think that sums up the whole problem.
The linux kernel model simply does not apply for graphics hw related stuff. It's great for (relatively) simple devices, but the complexity of graphics hardware, and the fact that all parts are spread between kernel, libdrm, xserver, mesa, and external utilities, mean that there will be no half-decent development model until they split parts out of their current mother-projects.
The fact that it isn't split out is the root cause of this and many more issues we are seeing with graphics drivers.
I can't seem to follow either individual. But I know the standard linux development model doesn't go back and correct mistakes.
Let's say the scheduler worked great in 2.6.21 but 2.6.22 was terrible. A few bug fixes come out for 2.6.22 as 22.214.171.124, ..., 126.96.36.199. The scheduler is still bad. But the reports are ignored as being people using old hardware. 2.6.23 is released and they realize there is problem with the scheduler. It get's fixed as 188.8.131.52.
The problem comes in that 184.108.40.206 should also get a fixed scheduler but it doesn't. Everyone is recomended that they should move to 220.127.116.11.
This is stupid development.
Are the network adapters ever ported back to older releases? Only RedHat does this. I'm not sure what they do for graphics drivers. All the I915 changes seen in the last 10 months probably aren't in the collection.