@pixx @AnachronistJohn @kabel42 @rl_dane

Admittedly most crashes are from come from running out of RAM/modern computers sucking at handling Swap latency. But even when programs properly crash without running out of RAM, even if the system doesn't _technically_ go out with it (which it often still does), there's still rarely any chance of recovering the system without (if you're lucky) a reboot or (more likely) a hard reset. Even xkill doesn't help all that much a lot of the time

@OpenComputeDesign @pixx @kabel42 @rl_dane You might have hardware problems, then.

I’m compiling perl on a system with 24 megs of memory, so the system is basically entirely in swap. If that can run like that for a week or two and be fully fine afterwards, then the VM system is doing what it should.

I can’t speak for Linux - it’s becoming the Windows of the open source world - but I also thrash the heck out of memory and swap on modern high memory systems, too, without issues.

@AnachronistJohn @pixx @kabel42 @rl_dane

I used to be able to live out of swap both on Linux and the BSDs. But these days, neither Linux nor the BSDs like touching swap _at all_. Linux is still much worse about it. But on every computer I have, touching swap is like running through a minefield blindfolded.

It's way too widespread of a problem to be a hardware issue

@OpenComputeDesign @pixx @kabel42 @rl_dane Let’s reproduce it so it can be reported.

I have an amd64 system here running NetBSD. I can force the memory down from 32 gigs to whatever I want with a kernel config change.

Can you come up with a recipe for software to install and run, and perhaps sites to visit and do things, that you’re pretty sure will result in a non-responsive system?

@AnachronistJohn @pixx @kabel42 @rl_dane

Yeah, if I load up firefox, log into all my chats and emails, and play a couple youtube videos, that's easily enough to use up all my RAM, dip into Swap space, and cause the system to start freezing and hitching, and eventually become completely unresponsive.

But erm, I'm guessing you meant that _other people_ could use to reproduce my issue. So uhm, let me find some sites that don't require other people to log into all my stuff...

@OpenComputeDesign @AnachronistJohn @kabel42 @rl_dane

...how much ram did you say you have? That's kinda ridiculous unless you have a _lot_ more chats than I think you do O_o

@pixx @AnachronistJohn @kabel42 @rl_dane

4GB RAM, 4GB Swap space. I have mastodon, matrix, gmail/gchat, protonmail (actually, usually close protonmail so I can have more youtube), and youtube is pretty much all I ever have open on this computer. I only ever have a web browser open, no other programs except terminals, and I reboot twice a day

@OpenComputeDesign @AnachronistJohn @kabel42 @rl_dane

...I know firefox sucks but 4G for that is a bit insane imo

@pixx @AnachronistJohn @kabel42 @rl_dane

Well, it runs fine on first boot. But after a while (A few hours, less if I'm doing research and opening extra tabs), all the webapps have leaked enough memory for it to really slow down.

Just as an experiment, I've opened some extra tabs to accelerate the usage (I'm on a fairly fresh reboot) and just 600mb of Swap used is enough for the system to lag really hard when switching windows

@OpenComputeDesign @AnachronistJohn @kabel42 @rl_dane

huh

Well I just ran out of RAM on Linux twice, _without swap_, and system recovered fine. Was responsive enough for me to go kill the system update (which was using all RAM to compile, uh, vscodium, i think. used that one fucking time lol.)

This tab in firefox got killed, as did steamwebhelper and several other FF tabs, but the system is just, fine

@pixx @AnachronistJohn @kabel42 @rl_dane

weird :o

@OpenComputeDesign @AnachronistJohn @kabel42 @rl_dane

I wonder if swap makes it _worse_ lol

@OpenComputeDesign @AnachronistJohn @kabel42 @rl_dane

TBH I want to see a memory usage breakdown of your system when this happens, I think that's what you'd really need to know what's going on :/

Firefox shouldn't generally be leaking memory. I've left 100 tabs open for weeks and memory usage never just randomly goes up

@pixx @OpenComputeDesign @AnachronistJohn @kabel42

Firefox is good about suspending inactive tabs to save RAM.

@rl_dane @OpenComputeDesign @AnachronistJohn @kabel42

agreed, other than the part where you started by saying "firefox is good" /snark

@pixx @OpenComputeDesign @AnachronistJohn @kabel42

I think that Firefox is objectively, ethically the least bad among any modern web browsers that can load a page like youtube or amazon.

Horrible bar to pass under, of course, but it is what it is.

@rl_dane @OpenComputeDesign @AnachronistJohn @kabel42

sure, it's still shit software thuogh

I'm not even talking ethics, just as a pure matter of code

it's bad code

@pixx @rl_dane @AnachronistJohn @kabel42

Bad ethics, bad code, still better in both respects to Chrome imo :P

Still don't like it, though

@OpenComputeDesign @pixx @AnachronistJohn @kabel42

I'm not so sure that Firefox is better code than Chromium. But definitely more ethical. Or at least, up until a couple years ago.

@rl_dane @pixx @AnachronistJohn @kabel42

I dunno, chromium based browsers have always been buggy as _fuck_ in my experience. Yes, even worse than firefox.

@kabel42 @rl_dane @pixx @AnachronistJohn

Too be fair, wayland is also absolute _crap_ :P

@kabel42 @rl_dane @pixx @AnachronistJohn

Hey, as we've all been told, on a preemptive multitasking system, it is _impossible_ for one crashed app to effect the rest of the system. So you must just be imagining things, anyway :P

@OpenComputeDesign @kabel42 @rl_dane @pixx That’s a bit disingenuous.

Obviously, on a Sinclair QL or an Amiga, a rogue program can take down the whole system.

On Windows, with tons of design issues and decades of bad decision history, a rogue program can take down the whole system.

With Linux and the BSDs, this shouldn’t happen, but it’s possible, and most often it happens when trusting stuff like video hardware to do its thing where the rest of the OS has not as much control over things as it does with the rest of the computer.

With Linux and the BSDs, if you can reliably crash the whole computer using a userland program, that’s a big bug and should be reported.

On the other hand, sometimes it feels good to vent, and if that’s the purpose of what you’re saying, that’s fine, but understand that your generalizations aren’t correct.

@kabel42 @AnachronistJohn @pixx @rl_dane

Ok too be fair drivers are literally the worst thing ever and it would genuinely be better for everyone if we just standardized hardware so drivers could be abolished completely

@kabel42 @AnachronistJohn @pixx @rl_dane

What possible downsides could there be to no drivers?

@kabel42
Theoretically you wouldn't even need that but really, part of the problem with standardization is that if the standard way is bad, you'll just have people ignore it

Opengl is a standard and we still need drivers.

The only way around that is what rpi and, apparently, nvidia?? are soing, where the hardware has a coprocessor in it that implements the driver and then the api on the cpu is just a proxy

Which means you still have a driver but now you're dependent on the vendor for it and it's opaque and hidden

This is not better

@OpenComputeDesign @AnachronistJohn @rl_dane

@pixx @kabel42 @OpenComputeDesign @AnachronistJohn

Disagree. If there's a single hardware standard, and the vendor's implementation is crap, then the blame is on the hardware/firmware, rather than the OS/drivers.

@rl_dane

Yeah but it doesn't *matter* who the blame is on

When the standard is implemented as a driver in software it can be _fixed_

When it's a driver in firmware then bugs are permanent the instant the vendor stops caring

A valve dev has been improving Amdgpu support for older cards recently

If the driver was in firmware then that wpuld be completely impossible

@kabel42 @OpenComputeDesign @AnachronistJohn

@pixx @rl_dane @kabel42 @AnachronistJohn

Counter point:
If there were no drivers, and everything was standardized, then there would be no moving targets, and application optimization could be perfected in perpetuity without the threat of obsoletion that currently plagues the computing world

@OpenComputeDesign

No, because now you've just shifted bugs into hardware where they're harder to fix

Drivers aren't buggy because drivers are a bad idea, it's because it's a hard problem. Implement an algorithm entirely in hardware and it's impossible to fix if the silicon is wrong.

The only possible way to have bugs be fixable is to have software controlling the hardware, so that you can route around bad or incorrect hardware. You can't just say 'well stop fucking up the hardware then,' that isn't how anything works

@rl_dane @kabel42 @AnachronistJohn

@pixx @rl_dane @kabel42 @AnachronistJohn

Well, given that drivers keep bricking hardware, I don't really trust drivers to fix anything.

I think the fundemential problem is, humans are unfixably incompetent

@OpenComputeDesign
I've never seen drivers brick hardware lol

I've had many gpu crashes over the years. The long term trend for both intel and amd gpus - the two most complicated drivers i interact with - is higher performance _and_ fewer bugs (crashes, glitches, misrendering) over time, and it's pretty obvious.

For a normal computing experience, mesa and the linux drivers are, uh, pretty good actually.

A lot of the code is a stinking pile. Not all, but a lot.

But to sit here and act like it doesn't work and isn't getting better seems patently absurd to me

@rl_dane @kabel42 @AnachronistJohn

@pixx @rl_dane @kabel42 @AnachronistJohn

Well, the main way drivers make hardware not work is by flat out not existing, and being virtually impossible to make exist. But there have been genuine instances of drivers not only screwing up Flash ROM and and excessively wearing out batteries and stuff like that, but even going so far as to melt hardware. (printers, GPUs, and depending on how you define a driver, CPUs)

@OpenComputeDesign

Sure but "doesn't work on my os" is not the same as bricked and that's a really ridiculous equivalence

Rpis gpu does the coprocessor proxying driverless thing; it's just as useless to me as tje amd ones, on 9.

Not having the ability to talk to hardware is not the same as the hardware being trash

@rl_dane @kabel42 @AnachronistJohn