@Boris_yo returns, i dont really do service or warrenty most the time. You know i mod, change , alter, and research before hand. Warrenty is such a pain, unless they are going to service in-house, and that is usually prohibitivly expencive in some way.
wtf. I just put the CD (supposedly) containing the manual for my fridge into my computer and it's the most fucked up thing ever. There's like Apache with a MySQL database and PHP on it and it wants to run some website locally and then opens a browser to http://localhost
@nhinkle 1500 MB just for Firefox rendering that SVG? whaaa....
well, the fact that it isn't outright crashing is definitely different behaviour than Ruda's seeing on Linux, but 1500 MB is way too much for just that
@allquixotic whoa.. get this. So the memory usage calmed down a while after I switched to a different window (Firefox still open in the background)... when I reopen it, it shoots back up and starts oscillating again.
it's still got to be hardware specific, because I'm running a very old Nvidia driver from ~2010 (I don't have control over it) on Win7 with the latest Firefox and my memory is absolutely stable, and FF is eating a "normal" amount of memory at all times, with fairly low CPU usage
@nhinkle OK, that definitely tells me that this is driver related; Firefox must be pushing "frames" of that SVG (each frame is the same, but hey, it could be recalculating it each time) to the backend renderer (2D canvas or whatever), and each frame requires it to draw all of those damn spheres
but why is this working 100% fine on my FF 24 and/or 25.0.1 on Win7?
but from what little I know of SVG rendering, the browser is technically rendering the SVG as a sequence of "frames"
@nhinkle thanks, that helps... but we've already detected at least very significant anomalies across two major, distinct platforms: Windows 8.1 and GNU/Linux (Fedora 19)
At least in Vista and later. Each process gets its own canvas to draw on essentially, and the whole system is graphics-accelerated and managed by the DWM.
@terdon you're a smart guy; having read the huge backlog on this, any insight into what's going on? we have similar-ish symptoms on two platforms now, but I'm a weirdo because I can't reproduce the problem at all o_O
we've narrowed it down to one specific SVG rendered in Firefox, causing an undue spike in FF memory usage, although the extent of that spike is much worse on GNU/Linux
@nhinkle Just pointing out that in my system (Debian) the OOM killer does not step in cause I have enough memory to deal with what FF is requesting. @ThatBrazilianGuy's system probably doesn't which is why the entire system freezes and not just the browser.
@Boris_yo Come to think of it, the car stereo did the same thing. There was a blue led (if i remember right) on one side of the display, and the usual diffusion layers to spread the light out. A 3mm led had failed because it was blue (and blues fail in time sooner) , and it was overdriven , and it got hot in that player. Pulled it apart and put a slow chaging RGB chip led in, because they waste power with the curcuit. it has worked ever since, but noxiously changes colors :-)
Android is probably different enough from Windows/Linux in its rendering API to not hit that case
Nvidia and AMD use very similar driver cores for their Windows and Linux proprietary graphics drivers, so it might be specific to one (or both) of those; the Android driver stack is completely different and shares no common heritage with either
i still need glxinfo from @terdon and @ThatBrazilianGuy
nope, still working fine here even with FF set to disable hw accel
my system just can't reproduce the problem, period.
@terdon your about:support copied as text from FF would be useful, because we already have that from a few people, so we can start to correlate the results
it might be something specific to the cairo renderer because i can't reproduce it with direct2d... i wonder which one @nhinkle is using
@Braiam brought up a great point that depending on the version of our graphics drivers and our platform, Firefox is going to detect and switch on different rendering backends for the canvas and the content
that's very, very likely to influence the reproduceability of the bug
we are like detectives -- Mozilla is going to be so impressed when we file our bug report
@terdon it'll be extremely helpful if you can narrow it down or even just simplify the image by throwing out stuff within the svg xml that isn't contributing to the problem