As of 5919, and after lots of mucking about in my driver settings, etc., the main errors are:
On my rig (7800GT):
1. CEGs don't draw, unless an object is being built in the gameworld. Not *all* CEGs, mind you- that's the weird part. Some particle systems draw- the vast majority won't. What's weird is that a Unit, when exploding, will show all of the particles, then they get zapped exactly when the unit dies. Like it's the only thing that can see them, and when they're removed from the gameworld, poof, there they go. However, when anything's being built, they operate normally. I almost think that Units that are being built are the only thing being assigned AirLos...
2. After /give all, I see corruption of the UI- specifically button textures getting overlaid with TextureAtlas.
3. TextureLODBias needs to be -1 (one mip lower than standard) for textures to look entirely correct. At 0, it's clearly a step down from where it should be. -1 is correct. This should be an easy fix.
4. Compared to jK's compile of 5850+patch, performance is terrible- merely 125FPS, with no Units on the screen, 3D trees and Water 0. With 5850+patch, it was almost 270.
That's about it, on the 7800 right now, and I have no idea whether the fact that the situation has improved was the result of patches over the last week, changes to my driver settings, or tweaks to the Registry values affecting Spring. Too many variables, too many things have been changed.
What I do know is that, without altering anything, I can run 5840+patch, and Spring runs almost spooky-fast.
Faster with UsePBO = 1, CompressTextures = 0 than with UsePBO = 0, which is interesting. Revert to 5919, massive problems ensue. It ain't my hardware or drivers
What it appears is that something between 5850 and 5879 caused bitrot to jK's patch, or something went awry when Tobi applied it, or a compiler error is causing this. I can't rule any of these things out. I did a comparison between Tobi's version and jK's, and there aren't a lot of differences. jK mentioned that he thought 5872 was to blame, but I don't see why he thinks that, given that it changes very little, and doesn't touch the rendering code. I've tested all of the Buildbot builds between 5875 and 5879, and 5879 is where things went off the rails- I'm 100% certain.
Radeon 9600:
1. Spring is completely unplayable. Any combination of UsePBO / CompressTextures results in crashes, or at best crippled performance.
2. Turning the UI off, with F5, results in a working game, or what looks like one, shortly before it crashes.
3. I haven't tested 5850+patch on the Radeon, so I have no idea whether that's better or not.
Intel chipset (on a laptop, about 2 years old- I don't know what chipset it is, tbh, but it's all right for low-demand gaming):
Pretty much the same as the Radeon 9600, only it crashed faster and ran slower.
That's pretty much it for now. Just about everybody has seen the UI corruption bugs, or the map-drawing bug (which, for whatever strange reasons, I haven't seen in recent revisions). GMN saw the massive-slowdown bug. I think I'm the only one who's seen actual crashes, but maybe I missed that in people's notes.
<pant, pant> Ok, that's just about everything. There are a lot of user reports now, so just about everything is now corroborated, etc., now we "just" need to find the bugs and squish them
[EDIT]
Zerver, the best thing I can tell you is to always use the same map, the same test conditions, etc., and to not move the camera for the first test, then move around a bit, etc. I always use Tropical, I see the scene and FPS before I hit Enter (i.e., no game UI, no Units on screen) then I hit Enter and compare the two, etc. Unless you use a map that's completely flat and voidwater, you can never really take the tesselation of the map mesh out of consideration. Normally, for me, this is just a minor cost overall though.
Which commit btw?
