Mesa, llvm & kernel after pontostroy

[dungeon]

At the end of the week, ~agd5f/linux/?h=drm-next-4.12-wip kernel is stable. Developers do test their code against that kernel.

Stable kernels are released on average 5 times per year by Linus Torvalds and these are on kernel.org, GKH marks one longterm kernel per year. llvm stable happens 2 times per year, mesa stable happens 4 times per year, debian stable and ubuntu lts happen once in 2 years, etc...

And what you claim there every day on every commit as stable is random AMD's driver development branches, so if that is stable your ass is stable too

[PuckPoltergeist]

At the end of the week, ~agd5f/linux/?h=drm-next-4.12-wip kernel is stable. Developers do test their code against that kernel.

I'm always laughing when noobs are explaining to developers, how the developers work. But I'm afraid that other less experienced users believe this shit and harm themselves.

[dungeon]

See bugzillas.

When i look at bugzillas i only see how you are trolling there too

[PuckPoltergeist]

How do i debug it? The pc crashes after some minutes after gaming, sometime after an hour, most of the times at around 20 minutes

I don't use OpenSuSE, so my help is limited. If it's really a hard crash without any reaction afterwards, it's hard to debug. And this really sounds like a kernel bug. If you still have network access, you should get some info what's going on. Maybe SuSE is writting crashdumps that could help debugging this. You should open a bugreport, so the maintainers are aware of the problem and can work on it.

https://en.opensuse.org/openSUSE:Submitting_bug_reports

Describe the bug as good as possible. What are you doing when the bug happens, what exactly happens (e.g. system completely frozen or is network access possible?), logs, if you have some. The maintainers/developer will ask you for more information, if they need.

[dungeon]

Use the latest code. Old software is buggy.

No way, people should stay with fglrx or switch to nvidia... whatever else, just to not listen you

People should do entirely the opposite of what you recommending anywhere

Fuck off with your's amdgpu drivers, old or new buggy or not

[bridgman]

For the amdgpu driver, "stable" kernels and mesa do have bugs that are fixed in the ~agd5f/linux/?h=drm-next-4.12-wip kernel and mesa dev git. See bugzillas.

You may be using a different meaning of the word "stable" from most people. Yes the development branches will contain the most bug fixes, so IFF your system's stability is affected by one of those fixed bugs then you may get best results from the newest code.

There are periods of time (typically after the introduction of a big slug of new functionality) where latest code is the best (and we will recommend that) but in between those periods the benefit from most recent bug fixes is less and the risk of temporary breakage becomes something worth considering.

The development branches tend to be fairly solid because of good per-commit developer testing and code review practices, but there *are* going to be regressions in the development trees which would probably be found and fixed in a released kernel. So sometimes it does make sense to pick up latest WIP code from a development tree (when you have an urgent need for specific fixes) but until we have more automated testing in place on the development trees you probably should not think of them as "stable" in the generally accepted sense (free from, or at least very low risk of regression breakage).

No way, people should stay with fglrx or switch to nvidia... whatever else, just to not listen you

Hey, let's not go that far please

What debianxfce is saying will sometimes be the right guidance, depending on timing & individual circumstances, I just don't want people thinking that those branches have the kind of testing or limits on commits that are normally associated with the term "stable branch".

[PuckPoltergeist]

What debianxfce is saying will sometimes be the right guidance, depending on timing & individual circumstances, I just don't want people thinking that those branches have the kind of testing or limits on commits that are normally associated with the term "stable branch".

He's advertising work in progress code for stable systems. And this is the same kind of wrong like advertising debian testing for stable systems. Code that is only in the wip-branch is missing in drm-next for a good reason. Somebody, who is using this, really should know what he is doing. It was more than once that wip-code even didn't boot for me.

Moreover, debianxfce is advertising to use the branches from Alex. That's really dangerous, cause the have the latest code for AMD graphix but are missing every update for the other subsystems. This way you will suffer from serious bugs for several weeks. Everybody who wants to use drm-next or wip code should do this on top of linus/master or one of the stable branches. But this way you have to know what you're doing again. You have to deal with merge errors, compiler errors or maybe subtle runtime problems.

So using this code is a really bad advise, as long as you don't have someone who knows what he is doing and maintains this.

[dungeon]

Hey, let's not go that far please

What debianxfce is saying will sometimes be the right guidance...

No, troll is never right . He speak of not someting which sometimes could be right guidance.. quite the opposite - he presenting that as universal and always right guidance

[bridgman]

Agree that presenting development branches as the universal solution is never right; but what I do agree with is that *sometimes* even we will recommend using development branches in specific situations so it's also not correct to say that people should *never* use the development branches. Default should be released kernels, but there will be times when picking up something faster moving will be "right" for specific users at specific times.

I am also disagreeing with the "it's wrong just because debianxfce said it" line of thinking, although I realize I am on thin ice there.

[dungeon]

Agree that presenting development branches as the universal solution is never right; but what I do agree with is that *sometimes* even we will recommend using development branches in specific situations so it's also not correct to say that people should *never* use the development branches. Default should be released kernels, but there will be times when picking up something faster moving will be "right" for specific users at specific times.

I am also disagreeing with the "it's wrong just because debianxfce said it" line of thinking, although I realize I am on thin ice there.

That is large difference, good and bad advices. Example of this behavior i might say like this... you don't recommend euthanasia every day as universal solution to fix everybody's problem even that provenly will really fix any problems that people have. Altough you might also agree that also might be sometimes do good at very specific cases, but that is still considered as really bad general advice...

If he stop spreading that as good general advice i would be fine with that... but i never spot in his comments some other side of it, these "sometimes" does not exists in his vocabular when he didn't recommend it That is total wrong, so troll is always wrong

Fuck off debianxfce