simplify the face structure; now, "fragments" just mean the output windings (if you opt out of tjunction fixing, then there will only be 1 fragment with the same values as original_vertices)
move MakeTangentAndBitangentUnnormalized to qvec.hh since we will use it later
-tjunc allows for more fine-grained control now (-notjunc still exists)
-maxedges now throws if you specify a bad value (1 and 2)
remove the checks to OmitFaces - we can tell if a face was omitted because it will have an empty winding
remove static variables from tjunc
move light-specific "load textures from BSP" routine to light
fix a couple bugs with external wad textures (there should always be at least a 40-byte miptex in there)
light can now load external textures
move -paths to common settings
fix bug with missing texture not filling miptex name
- for Quake-likes, WADs will be added as file system archives directly when the first brush is encountered since the wad key is ready to go by that point.
- after loading the .map and before constructing the BSP, what I call secondary textures (animation frames for Quake-likes) are added to the list, and the BSP's miptex data is filled.
- BSP miptex data is now opaque to the BSP process; the BSP simply copies the raw miptex information from the input. This simplifies the BSP structures and reading/writing routines.
- the img subsystem supports reading from raw miptexes now, although in practice this only happens for Quake-likes atm
- instead of a "wal cache", maps now have a "meta cache" which stores the metadata information for all games instead of just Quake II
- texture meta can now be ref'd out rather than needing to be copied every time (mainly only a gain for the std::string)
- wad.cc/wad.hh is gone; wads are now pushed into the filesystem archive
- fixed broken wad_archive
- fs subsystem iterated on a little bit; supports marking archives as external (only used for wads currently; see `LoadTextureData`)
- load_tga has better error handling
- add in foot-gun protection to the BSP readers; now they assert on failure to read instead of silently propagating that error throughout the rest of the process
- bsputil uses img stuff now, so it should handle palettes better
remove byteswap as it wasn't used any more
re-introduce imemstream and omemstream to make it harder to accidentally use the wrong operators
add dynamic padding stream manipulator
use memstreams for BSPX
add new parallel logging stuff in common/parallel which wraps parallel_for(_each) and provides automatic percentage indicator
fix issue with light not pulling in settings
move logging stuff into its own namespace and simplify
* cmd
* cmake change for tbb DLLs
add in test for double-hyphens
* cmd
* game version
* settings!
* bsp texture usage
decompile fix for q1bsp
* finish up the options for qbsp
* light finished?
* finish up the few missing bits
* move scalar/int32 up
* Run clang-format pass
* natvis for our mathlib
use slightly faster storage for winding
fix light tests
* allow dictionary to be a base type
* re-run style
* Remove cfg_static
Fix options summary
* natvis for lockable_value<T>
* last-ish pass?
* format
* forgot last few lockables
imglib now lives in common
fs 'load' split into two functions, where and load. the former allows you to query for and return archive & path for a given file, whereas the latter handles that for you.
use log verbose in light rather than own verbose flag
- there's still a couple failure cases - specifically base1 in the outdoor area has a side that wasn't able to match to a face
- the clip brush at the start of base1 gets cut into a bunch of pieces because of the whole "check for texinfo, compare and split into multiple pieces" thing, but I'm pretty sure it's just one single brush - we need to ignore the splitting thing on clip brushes since they don't care about texture
- triggers don't pull in a temp texture
- origin brushes?
Adjusted makefile to work with both clang and msvc on win32
Removed "WIN32", use the standard _WIN32 instead
Fixed bad VectorCopy's
Use qvec4b instead of special color_rgba type
qvec now follows built-in type procedures for handling operators; this means that (qvec3i + qvec3f) will now properly produce a float, as one would expect, instead of adding the float to the integers without casting to the common type.
Add a json.hh file which includes common JSON serialization types
surfflags_t::extended uses booleans instead of a bitflag for simplicity
JSON serialization for surfflags
qmat follows other type conventions (T, N, M) and is annotated differently so it's not as confusing to follow
Removed qmat::operator[] - it's not clear from calling it that it's column-major (as evidenced by the comments needing to explicitly make clear at call sites that it's column, not row), so I've collapsed them into two separate functions that explicitly fetch either a column or a row. This also ties them nicely into qvec<N>.
Moved vector-related operations from mathlib to qvec, moved a few into qv::
Use clamp/max/min directly instead of #define'ing them
Move a few single-use functions directly above their usages
simplified constructor for qvec(T...) that uses fold expressions
simplified rgba_miptex_t
fixed a lot of broken winding_base_t behavior
merged all winding types into winding_base_t
removed now-empty files
- bsp store vectors now
- vis stored in a way that is compatible with both Q2 and Q1
- binary reading of stuff, so swapping is done at load time now
- vis cleanup, although there's still a lot of raw long modification that should get simplified later
- main mbsp types are declared first in the files, and types that were single-member are now not structs
- conversion is handled by the sub-types themselves. Currently no diagnostic for *what* members overflow, but, overflow is handled via thrown exceptions on all overflowable members
- aabb type is used where appropriate now, simplifying a few areas
Cleanup/rework of logging system. I'm not too happy about it still, but, it's at least a bit easier to follow now.
Started cleanup of file-related stuff.
In preparation for converting all of the vec_t[3] to qvec's, the Vector routines are now templated and work with any type that provides indexing operators.
Polylib.cc is gone, since this all needs to be in header now
qvec can now implicitly convert from T[N]
constexpr where appropriate
static inline -> inline
side_t is now an enum, moved into mathlib
stdbool not required
_vsnprintf and _snprintf are standard in MSVC since 2015, so I just use them directly now. They'll be replaced by std::fmt later anyways.
use <thread> for threading, instead of pthreads/windows.h thread directly
typedef -> using
include guards -> pragma once
typedef struct x { } y; -> struct y { }
some fixes to extremely old code that shouldn't be required any more
* Add QBSP - the BSP2-esque variant to Q2BSP - to bspinfo, and all of its accompanying structures.
* pass around ident, since Q2 needs it - admittedly it's a bit ugly, but it works for now.
conversion for QBSP
* Fix light
* _qbsp_ -> _qbism_
* Introduced bspversion_t, a struct that holds pertinent information about different BSP versions and also acts as a tagged pointer type for direct comparisons. This makes a lot of code paths simpler. I'm not entirely set on the wordings or usages yet, and maybe we can stuff boolean flags inside of them for different behaviors (for instance Q2, QBism and HL would have the "colored lightmap" boolean set to true, which replaces the check-for-all-three in the lightmapper)
Swapped arguments to ConvertBSPFormat to have the conversion target last instead of first
Finished rename of qbsp -> qbism
Tested:
- bspinfo on various BSPs I had laying around (Q1, Q2, Qbism)
- bsputil converting between Q2 and Qbism, and that they loaded in engine/roundtripped properly
Not tested:
- vis/rad on anything major (I still can't run rad due to embree being weird)
- bsputil conversion of Q1-esque maps
Jonathan
Eric Wasylishen
Ozkan Sezer