in_dga -> 0 (until X is fixed)
gl_multitexture -> 1 (why was this 0? not enough support back then?)
cl_usleep -> 1 (seems to be ok)
host_mem_size -> 40 (even 32 isn't enough these days)
rate -> 10000 (we're not in the modem era any more)
cl_mem_size -> 32 (16 is not enough, 32 sounds better than 24)
Move the texture coordinates in 1/4 of a pixel. To avoid unnecessary
calculations, pre-caclulate the character cell texture coordinates and
blast them into the the texture coordinate array.
r_skyname now acts as the default sky to use when no sky name is specified
by other means ("none" is still no sky). 'loadsky foo' will load the
"foo*" sky textures, 'loadsky none' gives the default sky, and 'loadsky
""' causes uses r_skyname.
MOD_TraceLine now behaves the same as id's SV_RecursiveHullCheck (from
WinQuake). This means that even if the trace would escape from solid space
into non-solid space, the trace is treated as allsolid if it crosses from
one solid space to another before hitting the empty space.
trace-id.c is used only for establishing the behaviour of id's code.
This reverts commit 10232acdfe.
The problem was really in the trace code, but it got fixed by "accident"
when I had similar problems in hipnotic a couple of years later. Now to
figure out just what the trace could really should be doing.
I've been experiencing funny trace issues while playing Soul of Evil. While
I think the boxclip dregs shouldn't be causing problems, it's probably best
to clean them out properly before trying to fix anything.
When R_AddEfrags is used (as is the case in nq), this function is
redundant. Brush models in qw are currently broken (invisible), but that's
just a matter of getting qw to use R_AddEfrags instead of R_NewEntity.
This removal should speed up the software renderers a little bit.
o Check the return value of XF86DGADirectVideo.
o Use input_grabbed instead of in_grab for checking whether to enable dga
mouse and other grabbed actions.
This is an extension "wrapper" (no such C function). This allows ruamoko
programs to read property lists without worrying about the memory required
to store the string for the property list being read.
Much more testing is required (oh, but for qc-valgrind), but there is now
a ~945kB block of free data in the menu progs heap :).
Also, correct the printed size of the memory block to not include the
block header size.
It's probably nowhere near right, but probably ok for now (I need to study
the GNUStep code). I'm unhappy with the menu code hook, but it will have
to do for now.
The plist code was written long before I thought of resource handles, and
then it was forgotten. This is much nicer and safer than storing C
pointers in progs memory space (*shudder*).
Here's a patch to NET_SendToAll() which was always
broken: it never skipped non-connected clients.
Depending on the compiler, it would wait the whole
5 seconds of its blocktime before it gave up.
While there, changed its blocktime argument to
double (the comparison is against a double.)
