Moved from reference counting over to a mark and sweep method. Objects
are now required to implement an allRefs method which returns a pointer
to a list of all other referenced objects. How the list is maintained
is up to each object. The retain/release mechanism still works; objects
with any references will not be garbage collected, even if the collector
thinks they should be.
an incremental, reference counting garbage collector for QFobjects. A
different kind of collector will eventually be needed to run once in a
while to sweep up cyclic structures.
GIB will be reworked to use it, and hopefully Rua will also so that the
two languages can share objects, events, etc.
Warning: This uses quite a few hacks and tortured macros, it might cause
breakage.
all (or most, there are limits still) of the maps on a server with many
maps.
move the optional progs funcs into sv_funcs_t and add UserInfoChanged,
ChatMessage and LocalinfoChanged callback support.
clean up PF_setinfo (and SV_SetInfo_f and SV_Localinfo_f) using shared code
where possible und to use the UserInfoChanged and LocalinfoChanged
callbacks.
add chat message callback to SV_Say. if it returns zero, normal chat
handling is done, otherwise it's assumed to have been handled by the progs.
provide a hook for unkown user commands. non-zero return means it's been
handled.
cache the value.
don't call Qopen directly in the engine. instead call QFS_Open (generic) or
QFS_WOpen (write only, zip flag).
rework QFS_NextFilename to use a dstring (avoiding a potential buffer
overflow), support 10000 files and work from the top-level fs_userpath
directory. adjust QFS_WriteFile and QFS_WriteBuffers etc to suit.
make sw32 screenshots actually get written.
hopefully everything gets written to the right places :)
of leaf names in a stem variable via %var, the builtin functions 'count'
and 'contains', and an initial interface between qw-server and GIB to allow
querying of clients and their info strings. Also cleaned up the chat event
interface a bit. Renamed a few builtins.
All the details of game directories will be fully configurable (search
paths and gamecode file name currently supported). Downloading of skins to
qw/skins only currently broken (will download to $gamedir/skins for now).
below, or relative above (uses .. to ascend the filesystem). Changed
file functions in GIB to use this. GIB can now be initialized in a
non-sandboxed mode, which at the moment means that GIB scripts run with
carne can access the entire filesystem.
time to what the server thinks is correct when too much inaccuracy
builds up. This nearly eliminates all speed cheating, both normal and
reverse, except for a brief burst before the protection kicks in. Might
need cleaning up/tweaking.
the SDL target files for Despair since he's working on the currently). Also
make progs segfault handling much more robust (and general: sigill etc too:)
to prevent conflicts with math expressions. All clients and servers are set
to create buffers with the normal parser. However, scripts ending in .gib
are now executed in a GIB buffer, and the export command can be used to
make specific GIB functions available to normal command buffers (i.e. the
console).
physics and progs, that's what sv.time is for. Things seem to work nicely,
including map changing, and this /should/ make long uptime servers work so
long as the map gets changes occasionally.
realtime back caused the problems with clients not reconnect on map change
and after my preliminary mucking around with time, I'm convinced there's a
better way.
directly calling sys_doubletime () - managed to remove six calls. Proboably I
should look through the code some more and find more ways to remove more calls
... The server works fine with them on my system as is, but I've gotten the
impression from people in the know that spamming the clock this often is bound
to cause non x86 arches to have problems.
Tim McGrath (Misty)
Also of note, I found a line I missed for logging from the last checkin. Doh.
Did not test heartbeats, although the code is so dead simple it ought to
function.
I did not test logging, although the code is rather simple so unless I missed
something, it ought to work this time. ;)
Tim McGrath (Misty)
not. Your choice.
if sv_frametime is less than or equal to zero, progs does not enjoy life.
Don't make progs suicide, join the > 0 club today!
Tim McGrath (Misty)
done every maxtic instead of every *mintic* which is what it was supposed
to be doing. Ooops.
Also minor whitespace cleaning in sv_init.c
Tim McGrath (Misty)
and also prevents old_time in sv_main.c from getting screwed up in the head
and pausing the physics indefinitely (check and see if there is a faster way
to do it :)
What does this let us do? LEAVE THE SERVER RUNNING! Imprecision due to the
server being on for long periods of time should now no longer be a problem,
so long as you have a map rotation going at least once a day. :)
I plan on committing updated versions of my glspeed cfgs next, and then
looking at timeleft - just to make sure when sys_dead_sleep is 1 it can't
overflow accidentally.
Tim McGrath (Misty)
add dstring_replace. this replaces a string of lenth rlen at position
pos with data of lenth len, growing, shrinking and shuffling data as
appropriate. At this rate, the dstring `class' will get buffer gap
editing capabilities :)
cmd.c:
Cmd_TokenizeString builds cmd_active_buffer->line again.
Cmd_Process bails out instantly if cmd_active_buffer is a legacy buffer
and uses dstring_replace to modify the parameters in
cmd_active_buffer->line. This last change results in drastic
simplification (and accuracy) of the commandline reconstruction code,
both in Cmd_TokenizeString and Cmd_Process.
now recycled, not freed. Fixed some bugs in exp.c. Ready to add embeded
functions (read: function calls with return values) and for loops.
Probably some other misc. fixes, I tend to go on debugging streaks.
Changed Cmd_TokenizeString to accept a flag that controls the application
of filters (tags, variables, escape characters) to the tokens and modified
a few places in the source that called it. Added a secondary command
buffer that is parsed without filters for legacy command support.
Currently, it is only used for commands stuffed into the console from the
server. It is hacky, and I hope to eventually generalize the console
interface to support any number of buffers and audit the rest of the code
to recognize it. For now, the legacy buffer at least keeps escape
character parsing from destroying info strings.
code cleanups and general performance work to be developed in relative peace.
While cleaning up the networking code /is/ important, fixing QF's perfomance
issues is of much higher priority.
