cloudwalk [Mon, 8 Jun 2020 14:46:12 +0000 (14:46 +0000)]
Reimpl. the cvar alias system. Reimpl. the cvar hash table as a distinct type
It is now type-safe, cleaner, and doesn't require new static cvar_t's
for each alias. The hash table being implemented as a distinct type
allows multiple hashes to point to the same cvar without having to
create a new cvar_t for each alias.
cloudwalk [Sat, 6 Jun 2020 18:11:31 +0000 (18:11 +0000)]
Initial implementation of cvar aliases, with slowmo as a starting example
This allows cvars to be renamed, and keep the old cvar by pointing it to
the new one. Pretty much every cvar finding function has been adapted to
return a the aliased cvar or the alias itself depending on context.
For example, if a value is needed, it'll dereference the alias. If the
cvar is needed to be listed somewhere or to check if the cvar merely
exists, it will not. Other than that, it's pretty straightforward.
cloudwalk [Sat, 6 Jun 2020 01:35:55 +0000 (01:35 +0000)]
Disable occlusion query on Intel. Can be force-enabled. Hopefully temporary
The previous occlusion query crash is still occurring on Intel
for reasons totally elusive to me. I'm tempted to think it's a driver
bug, but honestly I'm not sure.
cloudwalk [Fri, 5 Jun 2020 14:55:28 +0000 (14:55 +0000)]
Draw coronas with the same cscale whether gl_flashblend or not. Fixes brightness
This should fix the ridiculously high brightness of coronas in pretty
much every game, when gl_flashblend is 1. This brings it closer to what
coronas looked like in 2014.
cloudwalk [Thu, 4 Jun 2020 12:11:20 +0000 (12:11 +0000)]
Fix the +map crash once and for all. Defer stuffcmds until it's safe to run.
Check if the literal quake.rc file exists rather than using stuffcmds as
an indicator of its existence, and use a new qboolean host_init to defer
stuffcmds.
cloudwalk [Wed, 3 Jun 2020 15:28:17 +0000 (15:28 +0000)]
Implement cvar callback functionality
With these, cvars can run arbitrary functions to bounds-check themselves or
perform some other task such as cleanup. Use the function Cvar_RegisterCallback.
havoc [Wed, 3 Jun 2020 02:26:30 +0000 (02:26 +0000)]
Fix gamedir command by removing the forced video subsystem shutdown, it was trying to load the loading screen without a running video subsystem and erroring out on the texturearray having mempool == NULL.
havoc [Mon, 1 Jun 2020 03:02:26 +0000 (03:02 +0000)]
Fix a bug with r_batch_debugdynamicvertexpath in R_DrawTextureSurfaceList_Sky where it was trying to use the surface->num_firstvertex rather than the batch info.
havoc [Sun, 31 May 2020 05:41:15 +0000 (05:41 +0000)]
Fix multiple bugs with r_viewscale in r_shadow_deferred and r_bloom - they were relying on vid.width/vid.height rather than r_fb.screentexturewidth/r_fb.screentextureheight.
Thanks to Cloudwalk for debugging this a fair bit already.
cloudwalk [Thu, 28 May 2020 13:53:30 +0000 (13:53 +0000)]
Implement experimental hook-like system, with working example.
This system allows you to call functions that may or may not exist at runtime
and compile-time. You create hook_t pointers and register them during
initialization, and point them to a function. If they're not registered,
no problem. This could potentially allow different subsystems to function
independently of each other, such as the client and server.
It's using a union to allow functions to have any return type they want.
However, for obvious reasons, functions cannot have any parameters they want
and must accept a pointer to the union type where the actual args are stored.
This is experimental and there may be bugs, but the current working example
should be stable.
cloudwalk [Wed, 27 May 2020 01:07:28 +0000 (01:07 +0000)]
Fix major regression preventing connect to any server using curl for custom CSQC
The command interpreters must be executed in sequence with clientfromserver
being the first, otherwise curl won't get a chance to run before the client
continues connecting. A few more commands were also missing from the
interpreter and were added.
havoc [Tue, 26 May 2020 08:34:12 +0000 (08:34 +0000)]
Added lightgrid texture based lighting in q3bsp maps by uploading it as a 3D texture, to make more accurate and continuous model lighting throughout a scene. Can be disabled via the mod_q3bsp_lightgrid_texture cvar.
The lightgrid texture based lighting can also be enabled on world surfaces if one desires (mod_q3bsp_lightgrid_world_surfaces cvar) which gives levels a dream-like quality that is globally consistent, can also be enabled for bsp models that are not the world (mod_q3bsp_lightgrid_bsp_surfaces cvar) which may be more useful for ammo pickup models that are bsp or similar.
This frobs the shader glsl crc so expect to update any modified combined*.glsl files as per usual with updates to the embedded shader code.
havoc [Mon, 25 May 2020 22:46:47 +0000 (22:46 +0000)]
With MATERIALFLAG_ALPHAGEN_VERTEX we should basically always use MATERIALFLAG_ALPHA | MATERIALFLAG_BLENDED | MATERIALFLAG_NOSHADOW - this should fix UI rendering issues with alpha being ignored on DrawQ_Pic et al.
Minor shuffling of flag order in lightning beam texture loading to be consistent with the other uses - no behavior changes.
cloudwalk [Mon, 25 May 2020 13:51:59 +0000 (13:51 +0000)]
Fix most Q3 & QL glow maps
From bones_was_here: "Very few glow maps work in unmodified Q3 maps as they do not use the _glow darkplaces convention, instead Q3 shaders specify the file name to use which is not currently supported.
Fortunately there is a naming convention started in id software's pak0 and continued in QL and most 3rd party maps. It uses these forms: foo_texture.blend.ext and bar_texture_blend.ext
It's not possible to fix 100% of glow maps without parsing Q3 shaders and using the specified file name, but this is a big improvement.
This also fixes a few glow maps in nexcompat, which are originally Q3 textures that were not renamed to _glow."
cloudwalk [Mon, 25 May 2020 12:51:08 +0000 (12:51 +0000)]
Fix visual regression where parts of the Xonotic menu lacked transparency.
Set ent->render.alpha to its old value of 0.999999f. This doesn't seem to break
anything else, or anything that was recently fixed. There's probably a better
fix, but this is simple and seems to work for now.
havoc [Mon, 25 May 2020 08:13:16 +0000 (08:13 +0000)]
Forcibly disable the alphatocoverage feature because we don't currently use MSAA antialiasing and it only works for that, resulting in broken rendering.
havoc [Mon, 25 May 2020 08:11:22 +0000 (08:11 +0000)]
Allow r_viewscale to be set to values higher than 1 for super-sample antialiasing, also integrate vid_samples into this calculation to restore old behavior somewhat (super-sampling is more expensive than multisampling, so this isn't a great solution yet, but it brings back the quality). Still need to implement a better SSAA filter for when the antialiasing is in effect.
havoc [Mon, 25 May 2020 08:10:04 +0000 (08:10 +0000)]
Add csqc_polygons_defaultmaterial_nocullface cvar which defaults to 1 in nexuiz and xonotic derived games, this restores the compatible behavior with older versions of the engine. It defaults to 0 in other games (and would be recommended to set to 0 in those games when they make a new release) as it seems like that would be the better default for efficiency (and matches FTEQW engine).
Fixed a bug in CL_MeshEntities_FinalizeRenderEntity where it skipped finalize when the entity has no surfaces, but since it is still included in the render entities list it was rendering garbage when unused.
Use the "$whiteimage" texture for CSQC/UI polygons (VM_CL_R_PolygonBegin) that specify texture "", to be compatible with older versions.
Clarify the name of a parameter on Mod_Mesh_GetTexture prototype to match the implementation as it was misleading otherwise.
havoc [Mon, 25 May 2020 04:18:02 +0000 (04:18 +0000)]
Reworked VM_CL_PolygonBegin/Vertex/End functions to do the commit to the mesh in PolygonEnd, this allows it to examine if alpha is used on vertices and request different variants of a material according to such discoveries.
Refactored CL_MeshEntities_AddToScene and CL_MeshEntities_Reset to be the more distinct stages of CL_MeshEntities_Scene_Clear (which is called in VM_CL_R_ClearScene), CL_MeshEntities_Scene_AddRenderEntity (which is called during VM_CL_R_AddEntities), and CL_MeshEntities_Scene_FinalizeRenderEntity (which is called during VM_CL_R_RenderScene) - this fixes issues with CSQC predraw functions adding polygons which were not subject to Finalize and thus not rendered.
Fixed Mod_LoadTextureFromQ3Shader handling of texture name "" to properly set basematerialflags, this fixes invisible polygons in some games such as Xonotic.
Refactored CL_Mesh_CSQC and CL_Mesh_Debug into CL_Mesh_Scene - not really a need for more than one mesh to overlay on the world.
cloudwalk [Sun, 24 May 2020 14:18:15 +0000 (14:18 +0000)]
Fix FS_Seek for compressed file from PK3
From Slava: "Currently FS_Seek works incorrectly with compressed files from PK3 but this issue isn't noticed because in most cases engine is using FS_LoadFile which reads file sequentially without seeking."
havoc [Sun, 24 May 2020 07:06:25 +0000 (07:06 +0000)]
Rework r_shadow_shadowmode to be more maintainable and understandable - it now caches the values of r_shadow_shadowmapping and r_shadow_deferred, and is now updated when those change, rather than just at startup, and changing those now also causes a recompile of world lights (which always had different bounds when shadowing was on/off, toggling shadows had weird behavior as a result, so this is a bugfix).
This fixes a massive fps drain with settings that disabled shadows entirely by disabling shadowmapping (where r_shadow_shadowmapping 0 and r_shadow_deferred 0 together caused it to reallocate the shadowmap texture every frame and do other bad stuff)
cloudwalk [Sat, 23 May 2020 22:22:30 +0000 (22:22 +0000)]
Fix Windows-specific use-after-free causing crash after disconnecting
This explicitly deregisters CSQC commands every time CSQC shuts down. This patch
also avoids Z_Malloc'ing the cmd names for each CSQC command and avoids forcing
a Cmd_AddCommand if the command exists.
CSQC was force-adding a command even if it already existed, and when it did,
it would pass a pointer from a mempool that would later get freed when
CSQC shuts down, and Windows' strcasecmp doesn't like dangling pointers.
The Z_Malloc wasn't much better because it could have caused a memory leak.
So the best solution was to only pass the pointer but make sure the commands
are freed when CSQC shuts down.
projects / xonotic / darkplaces.git / log