"Ask ATI" dev thread

[Vighy]

My question:
Could there maybe be two branches for the fglrx driver?
One driver, where _only_ fixes go in (and it is branched off the other one in bigger intervals), so something like a stable branch.
8.40.4 was _a bit_ like that.
And one branch where features go in (AIGLX), but which might have more bugs and for which not every release might work for everyone.

Mainly what I'm asking here is:
Can we get fixes for already released driver versions?
The current policy to not touch drivers, that have been released is really not the best.

So, lets consider, that 8.40.4 maybe was a version, that worked for most people, so there could be a verion 8.40.4-rev1, which brings kernel 2.6.23 (Yes, that is a fix, not a feature!), or if it had this resolution thing (that 7.12 currently has), then there could be a -rev2 for that.

do you mean something like the kernel way?

could be nice, but with reduced man power it's the end of the development.

[Berniyh]

do you mean something like the kernel way?

could be nice, but with reduced man power it's the end of the development.

Yes, and No.
The kernel people stabilize every release, that's not what I asked for.

Every once in a while (lets say every about every 6 monts) there is a driver version, that seems to work for most people quite good, but as always there are some issues.

Take 7.11 for example, that seemed to work for most people, but it has bugs, that simply prevents it from going stable on some distros, like the soname bug. I guess that one is not a really big deal, 7.12 fixed it. Now for most people 7.12 was a disappointment, because if the resolution bug and various other issues. If now ATI would backport the soname fix (maybe a few others), that are known to not open one of the new bugs, that 7.12 has, there would be a more stable 7.11, maybe one, that most distros could use as a stable driver.
Now as I said, not every driver version has to get fixes, obviously 8.42.3 and 8.41.7 were not good enough to deserve a living beyond there release.

Would it really be that hard, to maintain (in addition to the normal release) lets say two branches, that are known to be worth maintaining?
I don't think, that someone would really care if they get dropped after a year (I mean, who cares about 8.36 or 8.30 these days?).

[Vighy]

Yes, and No.
The kernel people stabilize every release, that's not what I asked for.

Every once in a while (lets say every about every 6 monts) there is a driver version, that seems to work for most people quite good, but as always there are some issues.

Take 7.11 for example, that seemed to work for most people, but it has bugs, that simply prevents it from going stable on some distros, like the soname bug. I guess that one is not a really big deal, 7.12 fixed it. Now for most people 7.12 was a disappointment, because if the resolution bug and various other issues. If now ATI would backport the soname fix (maybe a few others), that are known to not open one of the new bugs, that 7.12 has, there would be a more stable 7.11, maybe one, that most distros could use as a stable driver.
Now as I said, not every driver version has to get fixes, obviously 8.42.3 and 8.41.7 were not good enough to deserve a living beyond there release.

Would it really be that hard, to maintain (in addition to the normal release) lets say two branches, that are known to be worth maintaining?
I don't think, that someone would really care if they get dropped after a year (I mean, who cares about 8.36 or 8.30 these days?).

and what about 8.28? :-P the last one for r200 chipsets :-)

but I think they don't have enough manpower to do what you suggest...

but let's see what they say :-)

[givemesugarr]

Yes, and No.
The kernel people stabilize every release, that's not what I asked for.

Every once in a while (lets say every about every 6 monts) there is a driver version, that seems to work for most people quite good, but as always there are some issues.

Take 7.11 for example, that seemed to work for most people, but it has bugs, that simply prevents it from going stable on some distros, like the soname bug. I guess that one is not a really big deal, 7.12 fixed it. Now for most people 7.12 was a disappointment, because if the resolution bug and various other issues. If now ATI would backport the soname fix (maybe a few others), that are known to not open one of the new bugs, that 7.12 has, there would be a more stable 7.11, maybe one, that most distros could use as a stable driver.
Now as I said, not every driver version has to get fixes, obviously 8.42.3 and 8.41.7 were not good enough to deserve a living beyond there release.

Would it really be that hard, to maintain (in addition to the normal release) lets say two branches, that are known to be worth maintaining?
I don't think, that someone would really care if they get dropped after a year (I mean, who cares about 8.36 or 8.30 these days?).

to me the idea is not good. your idea would split the development more, than actually focusing it. having different driver versions to maintain is stupid. if i were to choose how to set the development schedule, i'd say:
1. ok for monthly releases of patches for open bugs and compatibility with new software versions.
2. immediate retirement of patches that cause regressions in the driver
3. 2 major releases twice a year with new features.
so the schedule would be:
on june the driver would be released with schedule for the 15th of the month. the second major release with new features would came on 23rd of december as a christmass present.
the other months of the year the driver would remain as is and will only get monthly patches if they would correct bugs. otherwise there wouldn't be any patch released. this new type of development that amd has adopted is somehow stupid and useless. the differences between 7.10 and 7.11 have been only in the known bugs actually.

[Berniyh]

to me the idea is not good. your idea would split the development more, than actually focusing it. having different driver versions to maintain is stupid. if i were to choose how to set the development schedule, i'd say:
1. ok for monthly releases of patches for open bugs and compatibility with new software versions.
2. immediate retirement of patches that cause regressions in the driver
3. 2 major releases twice a year with new features.
so the schedule would be:
on june the driver would be released with schedule for the 15th of the month. the second major release with new features would came on 23rd of december as a christmass present.
the other months of the year the driver would remain as is and will only get monthly patches if they would correct bugs. otherwise there wouldn't be any patch released. this new type of development that amd has adopted is somehow stupid and useless. the differences between 7.10 and 7.11 have been only in the known bugs actually.

What you are suggesting is exactly the same, you just name it differently.

[bridgman]

The difficulty with that kind of release model is new ASIC support. If we are going to support new ASICs in a timely manner we can't go six months between major releases... and it's not really practical to push support for anything but the most trivial ASIC spin into a side branch while maintaining development in a mainline.

[Kano]

Wow, that took really long get the problem... Maybe use your biggest competitor as example how it could be done.

[bridgman]

Not sure I understand. Neither of our big competitors release Linux drivers on a six month cycle (unless I'm thinking of the wrong competitors ). Both average a couple of months between releases and both branch off a mainline like we do.

[Kano]

Well usually you don't get drivers every X month but as soon as there are problems with the current one or when new hardware is out. For 8800 GT it took a bit long (compared to former new cards) to get updated drivers but you could at least adopt that there will be _public_ beta drivers and ones marked as stable. Next will there will be new ATI cards, I hope you don't forget the ASIC for that in the new driver. Also you should think about crossfire and multi gpu support - NV has SLI and mulit gpu running on Linux since long time.

[bridgman]

Ahh, OK. We're already doing a bit of that -- you probably noticed the release numbering changed a while back -- and we can cover that more in the Q&A.