The only arguments I've ever heard in favor of wasm/webgpu were that using native graphics/GUI toolkit APIs are a pain. That's definitely true, because I've written stuff with gtk and it sucks, but that doesn't mean we should just shovel an entire tech stack into the browser.
Just because we can, doesn't mean we should. I'm tired of these BigCos shitting everything up.
Why does this spill over? Unlike CPU which is multiplexed by the kernel's scheduler (so infinite loops can't lock out other programs), is the GPU not multiplexed in the same fashion?
Often there's shared resources that are statically allocated to shaders (register space, local memory etc.) that means you often can't "just" add a new task if those shared resources are already in use. But not using those resources to their full would cause performance issues.
And the internal state of a GPU is often very large, much larger than a CPU, so suspending the current tasks, saving out their state and replace it with a "higher priotity" one can be very expensive - so often an afterthought of support at best.
Locked up my entire M1 Macbook Pro, held power button and I was back into chrome in <20s but I did kinda go "why did I just do that?"
Of course, plenty of other uses. Disable your adblocker or we crash your computer. Watch the whole ad or we crash your computer. Click the follow button or we crash your computer.
Maybe I'm crazy, but "crash your computer" as a building block seems powerful enough to be a security issue. Is denial of service not a security thing anymore?
I encountered the same type of death freeze when trying (and failing) to run models in browser tabs, but didn't spend much time trying to understand how severe it is.
Hope they don't disable WebGPU...
while (true) console.log('this will freeze/crash dev tools')
It is pretty egregious though, I hope they fix this. I expect there'll be a Radar tracking this now that it's made it to the HN front page.
It did do it to all tabs though.
"It's just" dismissals are annoying when they're blatantly wrong. An infinite-loop counter in Orion.app shouldn't cause my entire machine to freeze, down to being unable to force quit.
I think near any anti-fingerprinting efforts though presume some floor level of system security and stability. If some particular hardware exposure feature lets attackers run arbitrary low level timing and hardware testing code or crash the system or break the sandbox the game is likely over for most people.
An extra bit of entropy isn't meaningless sure, but at some point there should be some weighing of absolute attack surface against it right? Some features just seem inherently anti-privacy/anti-security and one might just have to try to deal with that via other approaches.