The choice of the various epsilon values is critical to having geometrically
complex maps compile correctly.
The one which seems most important (so far) is the ANGLE_EPSILON. This one
needs to be quite tight to avoid "bad things" happening when you create small
angles between adjacent surfaces.
Another epsilon inter-relationship I discovered, is that it seems
CONTINUOUS_EPSILON needs to be larger than EQUAL_EPSILON. It seems to play
havoc with the t-junction code otherwise.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Offsets pointing to clipnodes were originally used as signed short (16-bit)
numbers, so that negative numbers could represent the various content types of
leaf nodes. However, only a small number of these types are required, so we
now make use of the rest of the clipnode range.
Signed-off-by: Tyrann <tyrann@disenchant.net>
When writing out portals for a node, we can skip over portals that separate
volumes with the same contents. The existing logic I find very hard to follow,
so I've separated that out in an attempt to make it more understandable.
While we're at it, equivalent logic is required in NumberLeafs, so re-use.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Remove the big union in mapentity_t, collecting together the information for
each lump in a struct. Still not 100% sure where I'm going with this, but I
think this will help me with removing some of the reliance on globals further
along (which will be important once it comes to multi-threading).
Signed-off-by: Tyrann <tyrann@disenchant.net>
Ugh, looks like this extraneous line snuck in with commit 6bb3ad32
While I'm here, I see no reason to zero pWorldEnt->cTexdata just above
either. It should already be zero'd memory from the malloc of map.rgEntities.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Fix the logic that decided which edges were to be added to smooth the clipping
hulls. The original logic is completely backwards and throws away all the hull
edge clipping planes you want, while potentially keeping the worst corner
cases.
This patch corrects that logic - for that reason, regressions are expected as
we will now be adding more planes to the maps and clipping more things at
wierd angles. Compile times will probably increase and clipping errors will be
more common, at least until some more tuning of the various epsilons is done.
While we're hacking the function, hoist some calculations from the inner loops
up to the outermost loop. Not that this function is that performance critical.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Small cleanup to TestAddPlane - re-use the PlaneEqual and PlaneInvEqual
functions and replace the "counts" array with simple booleans to record which
side of the plane has points.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Use a hash to speed up FindPlane for maps that have huge numbers of
planes. Not many maps have enough planes that this is a win, but those maps
are the ones where speed matters most. :)
Signed-off-by: Tyrann <tyrann@disenchant.net>
Instead of just assuming that 9999 or similar is "big enough", use the
implementation defined type limits.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Make ChooseMidPlaneFromList a bit more picky about the non-axial planes it
selects. This can matter a lot for maps which have a lot of non-axial-aligned
brushwork (e.g. terrain, etc.) The number of clipnodes can skyrocket if planes
are not chosen with at least a small amount of care.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Factor out the common split plane metric code from the SelectPlane*
functions. While we're at it, make it also work for non-axial planes using the
DivideBounds function.
Signed-off-by: Tyrann <tyrann@disenchant.net>
To make testing easier, log the output of the job to mapname.log, rather than
qbsp.log. Probably more desirable to make the logfile selectable in the long
term, but this will help for now. Note that we now don't open the log file
until after parsing the command line options, so the Message function needs to
check the file handle before writing to it.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Add the "-wadpath" command-line argument to allow the user to specify a
directory where wad files can be found. If no wadpath is given, the path
defaults to the directory where the map file is located. Absolute path names
in the worldspawn "wad" key should also work.
Signed-off-by: Tyrann <tyrann@disenchant.net>
If we run out of memory, then too bad. Qbsp never _really_ takes so long that
you'd really care too much about starting again. Simplifies portability.
Signed-off-by: Tyrann <tyrann@disenchant.net>
The options are currently parsed as a big text array, so that the parsing
function can be shared with the code to load options from a file. Anyway,
dynamically allocate the array to copy in the command line arguments rather
than restricting it to 512 bytes.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Fix numerous places where epsilons have been hard-coded, substituting one of
the defines from qbsp.h.
A couple of epsilons have actually been changed in this patch; the one used to
detect degenerate QuArK texture coordinates has been loosened (so the test is
actually a little more strict). The other is the test for flipped portals. If
my understanding is correct, then result would actually be < 0 if the portal
winding normal flipped, so this change _should_ be harmless.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Add the ability to extract texture information from a BSP file and write it
out as a WAD file. Not very robust in the face of errors or corrupt BSP files,
but it does the job for now.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Save some memory in CSGFaces by not splitting faces outside the brush being
processed. Normally these faces will be split by the planes of the brush, even
if the brush doesn't intersect with it at all. This also saves some time in
MergeFaces, since we don't need to re-assemble these faces again.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Each of ClipWinding, DivideWinding and SplitFace need to calculate on which
side of the split plane each winding point lies. Split this operation out into
a shared function.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Rather than using it's own type of winding, make face_t use the existing
winding_t struct to define it's edges. I tried dynamically allocating the
winding as needed, but it was a net performance loss (though much better for
memory usage). For now we'll just embed a winding_t struct inside face_t
instead.
Signed-off-by: Tyrann <tyrann@disenchant.net>
The number reported for peak memory usage by winding_t structs was not
accurate, due to just using the sizeof(winding_t) for accounting. Track the
_bytes_ used by each type in a separate array. This is pretty much redundant
for everything except windings, but it's the simplest way to implement it for
now.
Also tidied up the verbose memory output a bit - the total column is pretty
useless, so just don't print it. Re-align the other columns and special case
the "Total" line, as only the "Peak Bytes" value is really interesting.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Make all WADList functions take a pointer to the first wad_t, rather than
embedding it inside the struct wadlist. wadlist_t is no longer needed.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Begin breaking down the wadlist_t structure by having the Init function return
the number of wads loaded to the caller, rather than relying on just setting
it internally.
Signed-off-by: Tyrann <tyrann@disenchant.net>
If the map has no wad key, or none of the wad files found in the key can be
loaded, then try loading mapname.wad as a default.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Add back the warning message about invalid wads. Don't bother to differentiate
between filesystem errors and bad magic, etc. The file is already open, so if
we can't read it's either too small or there might be some hardware problem
(not worth dealing with separately).
Signed-off-by: Tyrann <tyrann@disenchant.net>
Move the logic from WADList_LoadLumpInfo into WADList_Init. Now the init
function will open all the wad files, testing for validity. Note that I still
need to tidy up the error handling and warnings a little bit to handle the
difference between filesystem/read errors and invalid/corrupt wad files.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Renaming a few variables to keep things nice and clear:
- w -> list ; the wadlist_t
- list -> wadstring ; the string from the map's "wad" key
- wad -> w ; local pointer to a single wad
Signed-off-by: Tyrann <tyrann@disenchant.net>
Rename a couple more structure members to make their purposes clearer;
- wad_t::Wad -> wad_t::file - it's a file handle
- wadlist_t::wadlist -> wadlist_t::wads - yes, we already know it's a list
Also, get rid of fileT from WADList_LoadLumpInfo.
Signed-off-by: Tyrann <tyrann@disenchant.net>
The wad_t struct is actually a list of wads, and the wadlist_t struct
represents the info for a single wad. This is kind of backwards, so reverse
the names. A few function renames went along with this to make things look
saner.
Signed-off-by: Tyrann <tyrann@disenchant.net>
wad_t::name isn't required outside of some temporary usage in WAD_InitWadList
so doesn't really belong in the struct.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Memory stats are reported inaccurately because the memory allocated to
windings is not accounted correctly (when freed, in particular). Use the usual
technique of recording the allocated size in the same block of memory, just
before the portion returned to the caller.
Signed-off-by: Tyrann <tyrann@disenchant.net>
Fix a bug introduced in my class File -> stdio conversion. Don't use the len
variable to save the result of fread, since it's still needed as part of the
loop condition!
Signed-off-by: Tyrann <tyrann@disenchant.net>
Tyrann