Mesa Developers Discuss Dropping Older Drivers, Removing Untested Code

[phoronix]

Phoronix: Mesa Developers Discuss Dropping Older Drivers, Removing Untested Code

Last week Mesa developers had their once-in-a-while discussion about cleaning up the code-base and potentially dropping the older drivers...

http://www.phoronix.com/scan.php?pag...nup-Discussion

[dungeon]

Maybe AMD should make 17 cores / 34 threads, just to match 17 year and then every year after year adding 10 cores more

Who knows maybe we will have 256 cores normal in 10 years, of course that would be minimum for proper VR anyway

[Azrael5]

it would be desirable in terms of efficiency

[microcode]

I think this could be cool, maybe they could achieve this by having an LTS branch for Mesa with support for hardware which is more or less stable now. As long as they backport fixes I'm happy.

[theriddick]

splitting the older stuff off could benefit it actually, lesser build times and a more narrower driver set could allow for some focused fixes vs having it all mashed into a super sized build like they have now. Doesn't matter if no one at this time wants to work on it, just split it off to legacy branch, and in time interest will appear no doubt.

[tildearrow]

Typo:

This original plan would also deperecate the old DRI driver loader.

[Adarion]

IF the drivers were working and stable some legacy branch would be okay. As long as security fixes or nasty bugs are still fixed and the drivers still compile. But one cannot exchange HW on embedded systems that are still good. Even some older systems still do live but don't have e.g. PCIe for a modern chip (that might also be way beyond the needs of that very system).
Also not-"first"-world countries still have a lot of aging hardware still in use and it would be beneficial if they could reasonably use it (instead of falling back to cruelties like VESA drivers).

[Girolamo_Cavazzoni]

I would prefer if they kept i915 because my netbook with an Intel Atom N450 processor and GMA3150 GPU depends on it. It may be very slow but it still works flawlessly.

[DanL]

Maybe AMD should make 17 cores / 34 threads, just to match 17 year and then every year after year adding 10 cores more
Who knows maybe we will have 256 cores normal in 10 years, of course that would be minimum for proper VR anyway

WTF are you talking about and what does this have to do with deprecating old mesa drivers?

[Veerappan]

The idea seems to make sense. Most of the classic drivers that are still in mesa support DRI2-only, don't use gallium, and receive very little testing/support. Most of the changes to these drivers are to prevent them from breaking the build as API changes, but don't really add new features.

Moving them off to a legacy repository/branch makes sense. It would isolate these drivers from accidental breakage in the master branch. Mesa already has effectively done something like this in the past with the old DRI1 drivers that were dropped back in the Mesa 7.x days. Mesa has kept ABI compatibility in the DRI loader for those drivers, but they haven't gotten new features as time progressed.

I'm not saying that it'll happen, or that it will be perfect, but there is precedent, and those old DRI1 cards do still work using the old driver binaries loaded through a newer Mesa. It's not that these cards won't still work, they'll just be serviced by a driver that changes much less, and is more insulated from upstream changes that are more likely to break them than fix them.