They still aren't perfect, but now they are at least not quite so obviously just translucent polygons over the level. A mixture between partially modulating the background colours and adding the fog colour. Notably white fog blocks look like they're brightening what's behind them.
Additive was also setting noalphatest before, can probably decide that depending on what it needs anyway. I don't think it's currently used anyway.
Sorts all translucent sprites and MD2s so they're drawn after all the opaque ones. Fixes most of the observable issues between translucent MD2s and opaque sprites/MD2s.
Only use glCopyTexImage2D when first creating the screen texture, use glCopyTexSubImage2D anytime after that as it does not define a new texture each time.
Flushing of the screen textures has been implemented for when the screen size changes (so that the screen textures don't stay at a wrong size) and the game is closed, I believe they would leave a memory leak before.
The Far clipping plane did not need to be nearly as high as it was, the new value is 32768, which I suspect is about how far software can render before it completely falls apart.
It is desirable to increase the near clipping plane to between 6-10, but it can introduce more issues with close geometry not being drawn when the player or camera is scaled or viewheight is set to MIN in first person view. It would also stop sprites from being drawn ever so slightly too early, but this isn't too much of an issue and isn't too noticeable with those values. Might look into scaling near clipping plane in accordance to camera scale in the future.
The reason for wanting to increase the near clipping plane is because the small value can cause very noticeable Z-fighting where there shouldn't be on older GPU's, usually Intel ones, that don't support 24-bits for the depth buffer.
Also removed the + 1 from newtime, since there was never really any need for it. It just offset the interpolation so it went like (1 -> 2] instead of [1 -> 2), so you never saw the base appearance for each frame except at the end of any frames interpolating to it
Changed DrawMD2Ex's duration/tics type to INT32 so -1 comparisons work, probably want to change the signs elsewhere too but this is fine for now
Draw Textures and Flats that have holes in them like a solid polygon so they use the depth buffer and don't need to be sorted
Disable all linear filtering on textures and flats that have holes in them, the linear filtering introduces translucency into the textures where the edges are. Leaving them with either a black border, or causing pixels behind the slightly translucent areas to not be drawn. Doesn't apply to sprites and the HUD as they are always already sorted properly.
Make the Alpha Testing more strict on non-translucent blend modes. This makes it so any transparency below 0.5 is discarded instead. Would make anything that is blended and has holes in it look slightly better, only the HUD and MD2s where the texture has holes are effected currently.
Set TF_TRANSPARENT on flat texture flags when there are holes in the texture.
Minor fix to make sure MD2s always set the right blend mode
OpenGL slope FOF lighting fix
This fixes some issues with sloped FOFs that affect lighting in OpenGL (as in, those that cast a shadow or have a colormap). Particularly, they can do strange things to any wall textures adjacent to them, as we've noticed ourselves in levels for 2.2. =P
See merge request !194
Polyobject seg render fix
This fixes both Software and OpenGL renderers so that polyobject segs aren't drawn if the game is drawing the actual subsectors they're from (outside the main level, where the polyobject walls were pre-spawn). They should only appear as part of the polyobject itself in-level.
This means a few glitches with polyobjects are probably fixed: for instance in Software mode, polyobject walls sometimes appear through level boundaries (and make everything above/below vanish, turning into HOM or skybox), if the BSP rendering code happens to find one of the subsectors said segs came from outside the level. I don't think anything similar happens in OpenGL, though I'm sure some unwanted typecasting is happening as a result of attempting to draw the segs. (And it fixes a crash in 2.2 anyway.)
See merge request !195
Polyobject segs should ONLY be drawn if the polyobject itself is in the polylist of a subsector being rendered. That way you won't sometimes see polyobject walls through level boundaries, if you happen to be close enough to their pre-spawn locations outside the level (or in them, if you decided to go on a noclip journey).
Some texture-related fixes
Bugs fixed in this branch:
* upper/lower/middle textures with non-existent texture ids being capable of crashing the game. For instance, RVZ1 has colormap codes on non-colormap linedefs, which causes them to wind up with invalid texture ids because of how the game tries to interpret lower/upper textures with "#" followed by characters on normal linedefs. Fortunately these "textures" are normally not visible anyway (since they're all in control sectors) unless they are swapped with in-level textures by some crazy Lua script of some sort...
* animated single-patch textures with holes displaying garbage on first viewing (see this thread: https://mb.srb2.org/showthread.php?t=42195)
* the heights of the lighting (shadows or colormapping) from water/translucent/shadowcasting/etc FOFs become messed up when displayed on repeated midtextures.
See merge request !144
I added similar checks for the other num* but it seems some MD2s break the other limits without knowing anyway ...so I've commented these checks out for now, unless we have further discussion regarding them later on
xorshift* PRNG
This needs testing to ensure I didn't mess anything up switching function names around.
Our PRNG sucks. This is probably obvious. I wish I had known better at the time I implemented it, but oh well.
The replacement is an xorshift* PRNG variant with period 2^32 - 1 (meaning that the PRNG state will loop after four billion calls ... that's not likely to happen), versus the old PRNG's period of about 2^22 (?). The output is also much more random and less predictable; the old PRNG would fall into a predictable loop of output after about 4000 numbers were generated, which isn't much.
The PRNG here also outputs numbers as fixed point from [0,1) (that's 0 to FRACUNIT-1, in other words) instead of single bytes at a time. This makes it much easier to calculate things for, say, P_RandomRange and P_RandomKey. A new macro, P_RandomChance(p), is now in use that returns true _p_ percent of the time, where _p_ is a fixed_t probability from 0 (0%) to FRACUNIT (100%).
This doesn't affect netgames at all; the code for seed saving and restoring is identical (aside from a check to prevent seed being set to 0, which breaks xorshift PRNGs). Demos break, but A: _duh_ and B: they're already broken by all the changes to physics to accommodate slopes.
P_Random is deprecated in Lua, as the function was renamed to P_RandomByte. Aside from that, nothing special.
See merge request !64
P_RandomChance is now a macro for something that should happen a
certain percentage of time.
P_SignedRandom was moved to a macro. Nobody cared.
# Conflicts:
# src/p_inter.c
the screen texture does not need an alpha channel.
the fact that it had one made OGL copy the topmost pixel of the screen texture's alpha channel.
which, naturally results in the screen becoming partially transparent and letting you see the working texture in the background.
