Qrack Coronas Fixed How?

[Baker]

R00k, if you read this ... how did you end up fixing Qrack's coronas?

Last year when my laptop died in September, I lost all my old Qrack source codes so I don't quite have the ability to backtrace through the versions.

If I recall, the historical issue with Qrack coronas is that they specifically would appear through world entities (doors, the fake start map floor) so I am guessing they used vis.

Without knowing, I'm going to guess you used the traceline that the QMB particle stuff uses that also collides against ... ugh ... I can't remember what it also collides against ... but it was an improved traceline of some sort. QMB particles have to collide with everything and use an enhanced traceline that isn't quite the same as the standard traceline, but it has been 9 months since I've thought about what makes it different (because it was crashing my engine at one point ... well not directly, but it wasn't fond of being connected to a server without the map ).

/A somewhat crappy quality post, but then again I'm trying to aggressively attack my "to-do" list

[Spike]

you can also trace against either cl_visentities (I think it was) or cl_entities+cl_staticentities.
remember that sv_recursivehullcheck will only check ONE bsp object.

the alternative is to use GL_ARB_occlusion_query (the idealistic method, though somewhat awkward), or just directly sampling the depth buffer (the cheap hack that will stall the gpu, but is meant to work on pre-geforce2 cards and requires some matrix know how).

[mh]

Occlusion query is the way I'd do it, but it should also be possible to hack something up using polygon offset (may need careful tweaking though).

[leileilol]

readpixels is slow but accurate, but don't have too many of those. They can be a big performance hit on modern hardware.

q3 and TF2 have readpixels flares. TF2 sometimes had updates which used ludicrous amounts of them leading to slowdown, making players blame something else for it (like model LODs or particles in the soldier 2011 update, when it's really just the flares coming off the trails of the cow mangler projectiles)

[mh]

A single readpixels of even a 1x1 region will cause a pipeline stall. You could probably do it asynchronously to a PBO then grab the result in the following frame (on the principle that things don't change too much between any two consecutive frames) but that ends up being not very different to what you need to do with occlusion queries, so you may as well just use occlusion queries.

Hmmm - a 3D texture may be good for this too.

[r00k]

Code: Select all

	GL_Bind(corona_textures[f->coronatexnum]);

	//glDisable(GL_DEPTH_TEST); //R00k: removed this so coronas are not visible through doors/plats
	glDepthMask (GL_FALSE);	// don't bother writing Z
	glBlendFunc(GL_SRC_ALPHA, GL_ONE); 
	glEnable (GL_BLEND);
	glTexEnvf (GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);

[revelator]

Hehe elegant it also reared its ugly head on some other code i was working on and i was like wtf this works nicely though.
Funny thing related to depth checking if you try realm notice the water at r_wateralpha 1 if theres a lightsource (torch etc) on the other side the light from it is still visible (kinda a poor mans wall hack hehe) you cant see much else though.

[Baker]

Code: Select all

	GL_Bind(corona_textures[f->coronatexnum]);

	//glDisable(GL_DEPTH_TEST); //R00k: removed this so coronas are not visible through doors/plats
	glDepthMask (GL_FALSE);	// don't bother writing Z
	glBlendFunc(GL_SRC_ALPHA, GL_ONE); 
	glEnable (GL_BLEND);
	glTexEnvf (GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);

That was anti-climatic. And anti-climatic is good

Since Qrack had the problem for maybe 3 years, I thought somehow the issue was "hard to fix".