Oh, now I’m finally getting a reaction :)
I would have preferred the professional way and getting response to my emails, but if I have to make a lot noise in the public eye, so shall be it :p
I know, and you’re lucky that this still works as you expect it. (until module loader will be rewritten …)
As already mentioned long ago, that is the actual problem.
This whole idea is very fragile and causing a lot of extra work to prevent this evaporating in free fall.
A slight change in some struct or function prototype makes the whole thing explode. And we do have those changes across ABI major versions.
Do you really expect us to carry tons of technical debt for all eternity just because you don’t want to compile your driver properly ?
Come one, it’s so simple: just compile the driver against the several ABI versions separately. It’s just two .so’s that have to be shipped in several per-ABI builds.
For all practical purpose, you only need 23.x, 24.x and upcoming 25.x. Three different CI jobs really can’t be such a big deal.
And once you’re already there, you could also set some more jobs that are building actual distro packages instead of this weird “installer”.
This is just a few days of work. If I had access to the source code, I already would have done this long ago.
You’d do everybody a huge favor. 90% of the hassles people have with it on daily basis would be just gone.
Do you need an DevOps coaching ? Feel free to call me ;-)
15 years ago. And the DevPrivateKeyRec changed back then, so you must recompile anyways - otherwise you’ll end up with out-of-bounds writes.
Crazy.
Those kind of changes happen over and over.
Are you really reviewing all commits for such issues and take specific actions in your driver ? Seriously ?
Oh, and didn’t you want to clean up this old stuff anyways ?
And why didn’t you just copy over the obsolete mi-overlay code that I had to reintroduce just for you in the latest driver release ?
Do you really expect us unpaid volunteers to do your work ?
Do you really think that is fair ?
NVidia made 44 BILLIONS (!) net profit this year. Don’t tell me there’s no room for hiring some experts for few month to clean up all that mess once and for all.
I really hope so !
But still this causes trouble for us, upstream: we just have now idea what you really need.
If you just would look through the MRs that are already laying around for long time.
I’ve spent so much time for making those changes like unexporting symbols or dropping dead code in small and
easily digestable commits, to make it easy for you, so you can just look the subject list. It still didn’t help :( :( :(
I’m on the brink of killing all support for NVidia and announcing people shouldn’t stop buying Nvidia,
because I’m really sick of making workarounds for your mess - for free.
I’m just sick of having a mega corporation with 44 Billions profit, that’s probably spending more for toilet paper
than for driver development, shifting work on unpaid volunteers.
Frankly, I cannot understand these crazy corporate politics at all. NVidia is the most hated corporation (even more than Microsoft)
in FOSS community - because their drivers are so bad, always have been. Torvalds once told it loud and clearly.
Keeping source code secret is bad and expensive enough. But then there at least should be some professional cooperation
with the community, so we can make the products actually work, reliably, for professional/industrial environments.
What is the most powerful GPU practically worth, when one just cannot use it reliably ?
Quite my only reason for caring about Nvidia at all is the multi-port cards, because my clients having huge
screen walls. Maybe I just should tell them to use a bunch of cheap chinese cards and revive Xinerama.
That’s still more than 10 years old.
I happen to be one of the folks who’re called for maintaining really old legacy stuff (sometimes even reviving or fixing
ancient PL/M code), but I haven’t seen such ancient Xserver, running on non-ancient Nvidia hardware in the field.
Don’t get me wrong, I really honor anybody maintaining old tech - there should be more of this.
BUT: those installations are frozen, they don’t get upgrades ever, just - if necessary - minor bug fixes.
They obvious solution is branches.
Which version exactly ?
(Do you meanwhile have nightly build against master and incubate branches ?)
And then how do you even get those old symbols that aren’t even existing anymore ?
You must have your own prototypes somewhere, that’s how C works.
I’m now making a last attempt to get into active cooperation. Please respond to my mails, and let’s get the stuff fixed together.
Next release is scheduled for summer solstice. I really wish to have the nvidia driver bugs fixed here, but I can’t do that
without your cooperation. If I don’t get it, there can’t be any Nvidia support anymore.
Let’s cooperate instead of working against each other.
