01:11imirkin: kisak: while i'm not aware of anything in particular that would cause steam to not work with the nv30 driver (which drives nv3x and nv4x hardware), i can also say that the driver is crap wrt conformance, and not-great wrt stability.
01:12imirkin: there are lots of innocuous ways to get off the "things work" path
01:12imirkin: that people just didn't do back when the hardware was cool, but newer software will happily do that stuff
01:14imirkin: nv50+ should be much stronger in terms of stability and conformance
01:14imirkin: although hardly a beacon of perfection either
02:04kisak: imirkin: yeah, without a box with SSE3 and an AGP port, I can't tinker with that. might just be something back in 18.0
02:04imirkin: kisak: conceivable. but the driver gets VERY few updates.
02:04imirkin: tbh i can't remember the last time i did anything other than maintenance with it
02:05imirkin: (but i bet "git log" remembers...)
02:05imirkin: 17.1 -- fix for a crash if the very first thing you do is a clear.
02:07imirkin: my last batch of work on it landed in 13.0...
02:08imirkin: i do have a nv34 plugged in, but unfortunately that'll only get you GL 1.5 coz of the stupid NPOT textures
02:08imirkin: kisak: you can also get pcie nv4x boards for about $10 on ebay, depending on your caring level
02:10imirkin: e.g. https://www.ebay.com/itm/NVIDIA-Quadro-FX-3500-Video-Graphics-PCIe-Card-256MB-GDDR3-Tested-Working/202545391138
02:12kisak: I don't have any disposible income to throw towards old hardware, whatever I have laying around is all I can use
02:17imirkin: intel has a hilariously slow SIGILL-wrapping piece of software that allowed me to play with swr even though my cpu doesn't have AVX. not sure if it'll do SSE3 though.
02:17kisak: fwiw, the minidump is a crash completely in nouveau, but it's a pain to get http://ddebs.ubuntu.com/pool/main/m/mesa/libgl1-mesa-dri-dbgsym_18.0.5-0ubuntu0~18.04.1_i386.ddeb (or the 64 bit variant) unpacked to then convert it to what minidump_stackwalk will consume
02:17imirkin: oh, i have no doubt it's some sort of shortcoming in that driver
02:18kisak: the crashing thread only goes back 4 calls and it's all nouveau_dri
02:18imirkin: kisak: do you know if steam will load with a GL 1.5 driver that supports GLSL but not NPOT textures?
02:18imirkin: (and how about ES 2.0?)
02:19imirkin: kisak: alternatively if you can get the individual to make an apitrace, that could tell us what's going on
02:19kisak: hypothetically it was supposed to work on 1.5 hardware, but I've never seen pre GL 2.0 hardware actually work
02:20imirkin: let's try it...
02:20imirkin: oh wait... hrm. that won't work so easily. no prime.
02:20imirkin:will ponder this
02:22imirkin: alright... i've got xeyes
02:25imirkin: ok. that thing has a LOT of problems.
02:25imirkin: but i also got a crash
02:25imirkin: but even the password prompt... the font is missing -- it's just a white box
02:26imirkin: kisak: how do i run steam inside gdb?
02:26imirkin: it's a wrapper script, so it's non-trivial i guess
02:27imirkin: aha, probably set STEAM_DEBUGGER...
02:29imirkin: that didn't work
02:29imirkin: STEAM_DEBUGGER=gdb steam didn't dump me into gdb when it crashed
02:30imirkin: aha. DEBUGGER=gdb
02:31imirkin: ha! it's the issue i thought i fixed back in 17.1!
02:32imirkin: apparently with insufficient fervor...
02:35imirkin: right. ok. so that's mostly the NPOT thing...
02:36imirkin: (which wouldn't be an issue on nv4x)
02:40imirkin: [ok, no, not exactly the same issue as i fixed in 17.1...]
02:41imirkin: the fact that steam is a 32-bit app is ... annoying
02:43kisak: at least you get a consolation prize that you're not going crazy
03:09imirkin:is slowly building a 32-bit mesa
03:13imirkin: kisak: how do i force steam to load my libGL? doesn't seem to accept LD_LIBRARY_PATH...
03:17imirkin: got it with LD_PRELOAD. this is _not_ designed to be easy...
03:18imirkin: right. no bufctx bound. i know this issue... i even had a hackfix for it
03:20imirkin: this was the bug where i tried to figure it out -- https://bugs.freedesktop.org/show_bug.cgi?id=102349
03:20imirkin: however i was only going on thought rather than being able to poke around
03:21imirkin: kisak: --^
03:27kisak: I'd stick symlinks in ~/.local/share/Steam/ubuntu12_32/steam-runtime/pinned_libs_* for any libraries you want prioritized
03:28kisak: if something goes terribly wrong, you can nuke the pinned_libs_* folders and steam will regenerate them on next application start
03:44imirkin: w00t, fixed it
03:45imirkin: remove a line of code, solve all your problems
03:45imirkin: kisak: thanks for bringing the issue to my attention
04:04imirkin: kisak: looks like the issue is triggered by steam creating and destroying contexts left and right
04:04imirkin: without actually using them for any drawing
04:05imirkin: i have a local fix that i want to validate a bit before pushing to the world
04:07HdkR: imirkin: So you're saying I should be super rude and create/destroy contexts on every texture upload for fun :P
04:07imirkin: well, it happens to trigger a bug in nv30
04:08imirkin: basically we have 2 parallel things tracking the same thing
04:08imirkin: one gets updated, the other doesn't
04:08imirkin: hilarity ensues
04:08imirkin: at least it totally explains the mystery of how this impossible scenario could occur
04:09imirkin: now ... what can i run on this stupid nv34 that would actually work on there
04:09airlied: quake? :-P
04:09imirkin: that's a little more invovled than glxgears
04:09airlied: openarena might I suppose
04:09imirkin: hopefully they have a GL 1.5 renderer
04:10imirkin: some day i'll pipe enough bits through to get ES 1.9999 on there
04:10imirkin: (how much are separate rgb and alpha blend factors worth?)
04:22imirkin: wow, xonotic is a lot of fail
04:22imirkin: that's surprising
04:23imirkin: ffs. util_copy_rect is failing...
04:24imirkin: don't they know that this isn't the thing i'm debugging...
04:29imirkin: and of course nv30 doesn't support multi-layer transfers
04:37imirkin: and it's broken at every level... sigh
04:41imirkin: at this point it's easier to fix util_copy =/
04:52imirkin: and it looks even more broken with NV30_SWTNL. sigh...
15:23imirkin: in case anyone's curious, i did get xonotic to work ok on the nv34. now to figure out what bits of that are different from nv4x.
15:30imirkin: the final issue is in the layout of s3tc data, which appears to want reduced strides for deeper levels, while the current logic enforces a uniform stride for all levels. but perhaps that's just nv30 vs nv40
16:49kisak: imirkin: heh, nouveau being reliable or not is kinda irrelevant when the competing vendor has completely dropped support.
16:50kisak: That fact will get progressively more interesting when Tesla and Fermi get the same scenario
16:52kisak: thanks for looking into that