This might be related to some other problems I've been having for a while with tyrlight. Here's just sunlight, no dirt:

http://lunaran.com/pics/fitz0063.jpg

Opposite corner from my previous screenshot. Note that those blocks are not floating.

I would really like to say when I started seeing this kind of thing, because I know it happened after I tried to get the latest compilers six or eight months ago, but I've tried to reproduce the collection I had before that and I still get those gaps and bleeds, so I can't actually prove it's a change in some compiler that's done it.

eric, what are you BSPing with? Do you think it matters? Utter conjecture here, but could some subtle difference in point quantization or double rounding in the BSP phase cause the lightmap sample origins to fall on the wrong side of a face or become coincident with an edge in a microscopically unhealthy way? 

p.s. the poorly shadowed corner in 0061 is the -x,-y corner, and in 0063 it's the +x,+y corner. 

I think I understand the dirtmapping glitches on my box room, at least. The problem seems to be the raytracing code in tyrutils is unreliable when sample points lie exactly on a plane separating the void from the map interior. (but unreliable in a consistent way, so in some directions it works) Unfortunately, you get these quite a lot of these void-straddling points with the way the sample points are positioned:
The sample points for the west wall of a box room, for example, are distributed in a grid that's pushed 1 unit into the middle of the room, so most of the points are free-floating in the map interior as desired. The problem is, the columns of points on the north and south edges lie exactly in the plane of the north and south walls, and same goes for the row of points along the floor and ceiling.

The one thing I tried as a fix, which is admittedly a coarse-grained fix, was replacing the trace code with LordHavoc's from his hmap2 compiler, which seemed to magically fix the dirtmapping glitches. It seems it can consistently deal with sample points straddling the void and treat them as within the map, which would be good.

Could you send me the map for 0063.jpg? (or just that section of it?) I'd like to see if I can fix that bleeding is the same way as the dirtmapping glitches.

For bsp I'm using snapshot I posted in #113, osx version. I don't think that's a factor, though I can double check on windows. 

Looks like that glitched shadow in fitz0063.jpg was introduced in TyrUtils 0.7, in commit cc36d8e. I reverted part of that commit and it seems to have cleared up the holes in sunlight shadows as well as fixed up the AO.

win32 osx source

The dirtmapping looks way better in this snapshot than the previous ones! It looks reasonable now without -extra4. 

https://www.dropbox.com/s/5hmx6nortmhkksb/fixed.jpg?dl=0 

God, OUTSTANDING. 

I KNEW IT

i just didn't test all the way back to 0.7 because I had no idea the version of tyrlite I was using was that bloody old. :( 

Also this is what the last snapshot looked like:
https://www.dropbox.com/s/9wy9aky8u9j2myr/tyr-trace.jpg?dl=0

It was totally broken! The rays to test occlusion were not impacting the world properly and the measurement of the distance to impact was also messed up. Would love to see what your screenshots look like recompiled with this fixed build, WarrenM :D 

Only because you're my hero ...

Unlit/Dirt Only/Full Lighting
https://dl.dropboxusercontent.com/u/161473/Misc/Dirty_Shot1.jpg

I turned down the dirty setting to 1.5 since the new code makes it a lot dirtier!

And a random shot:

https://dl.dropboxusercontent.com/u/161473/Misc/Dirty_Shot2.jpg 

 

Hmm, the floor under the chain in the middle shot looks a bit odd. https://dl.dropboxusercontent.com/u/161473/Misc/Dirty_Shot1.jpg

Could just be the occlusion test rays are hitting the chain, combined with high dirt settings, in which case it's expected I guess. If you think it might be a bug, I'd be happy to have a look at the .map for that chunk. 

Sure ...

https://dl.dropboxusercontent.com/u/161473/Misc/MapJam4.map

And no, this won't be finished for the jam. :) 

Seems to just be shadows from the chain! 

 

Hmmm wonder what my map would look like with this AO pass now... 

that is some sexy looking dirty brushwork! 

ooo 

Is very nice. The lighting and shadows are very believable. Maybe we need an "ikebony" type texture set. 

Leveled lighting : https://dl.dropboxusercontent.com/u/161473/Misc/FixedLevels.jpg

Really nice ... !! 

that's awesome! Would love to see a whole Quake level that looked like that 

I have to think that perhaps the ikwhite jam maps would have looked about 10x better with these tools *HINT HINT* 

 

eh, the geometry sticking out over the shadowed areas looks all wrong or are those bars actually floating and not connected to walls and floor? 

How did you do that? What compilers did you use? 

Photoshop + Levels adjustment layer I believe.