- In G_AddBot, try to allocate clientNum before doing anything else.
- In G_AddBot, don't set SVF_BOT and inuse. It's done in ClientConnect, plus inuse causes ClientDisconnect to be run for no reason.
- In G_AddBot, only set skill in bot useinfo once.
- Avoid using cl->ps.clientNum to check if cl is a bot.
Fix passing args[11] to args[15] from vm to engine on ppc64 and sparc64. Some of the args are used by game bot prediction syscalls. May have been causing bugs. Note: This was fixed for x86_64 in r2163.
really fix the confusion with game entity and refentity numbers
for any natural number M, the following is logical as a whole:
- the array size for refentities is M;
- the refentity number limit is M-1, ie., each refentity number is in [0..M-1];
- the special number for the world is M.
before r1429, the code was roughly the following:
// constants related to the game, should not be used by the renderer
// renderer stuff
refEntity_t refEntities[MAX_ENTITIES];
int numRefEntities = 0;
void addRefEntity(refEntity_t re)
{
if (numRefEntities >= ENTITYNUM_WORLD)
return; // full
refEntities[numRefEntities++] = re;
}
void render(int num)
{
if (num == ENTITYNUM_WORLD)
renderWorld();
else
renderRefEntity(refEntities[num]);
}
so before r1429,
- the array size for refentities was 1023;
- the refentity number limit was 1021, ie., each refentity number was in [0..1021]; and
- the special number for the world entity was 1022.
this was a small waste of memory, as the last array element wasn't used.
r1429 changed
if (numRefEntities >= ENTITYNUM_WORLD)
to
if (numRefEntities >= MAX_ENTITIES).
this creates the following configuration:
- the array size for refentities is 1023;
- the refentity number limit is 1022, ie., each refentity number is in [0..1022]; and
- the special number for the world entity is 1022.
r1429 just makes things worse: it allows 1 more refentity to be added, but that entity doesn't get drawn anyway, as its number will be equal to the special number for the world. this is a small waste of not only memory, but also processing time.
perhaps in XreaL,
ENTITYNUM_WORLD is a game entity constant, and has nothing to do with refentities. a new REFENTITYNUM_WORLD constant should be added to denote the special number for the world, and that constant should be used in the renderer code in place of ENTITYNUM_WORLD. so define such a constant, and let it be equal to MAX_ENTITIES, which is 1023.
Bots set their team later, setting it here causes some bots to change team and die later. Can cause extra skulls to be spawn at beginning of harvester (bug #5740).
- A stupid bug where bots re-trigger jumppads if they fell onto it.
- A small "memset" bug concerning player animations.
- Reward sounds were never cleared and thus they are played on a map restart.
- Safer and more secure handling of disconnected clients and clients with
malformed or illegal info strings.
- first_gauntlet_hit.wav was not played (ops/ps) bug
- capturelimit not hit (from OAX)
the SOCKET type is unsigned on Windows, and should be casted to an int before comparing with the highestfd variable (note: ``int highestfd = -1;'')
From /dev/humancontroller.
################################################################################
This Patch fixes:
################################################################################
- The "fraglimit warning" was not played at all, if on the blue team.
- The "where" console command was broken.
- Obelisk explosion wasn't drawn if no Rocketlauncher was loaded.
- Impact marks sometimes didn't draw at all.
- IMPORTANT BUGFIX: No killing for cheaters with Lightning gun and Gauntlet.
- If two doors are close to each other a spectator couldn't fly through them.
- More robust, efficient and logical respawning routine.
NOTE: The game.qvm will get notable smaller and will use LESS MEMORY!
- Drowning sounds are fixed. Now they are played as intended. (as the id
comment
in the source code shows).
- Some AI bugs (OVERFLOW!) in the bot movement code.
- Several "Team Arena" Overload and Harvester bugs.
- Stops bots from attacking a team mate (player) who only changed teams.
- Some voice chats and CTF commands fixed.
- "Team_ReturnFlag" was called twice, which did wired things sometimes.
NOTE: (G_RunItem checks CONTENTS_NODROP already!)
- A bugfix for Gauntlet animation.
- Incorrect CTF scoring.
- A bunch of corrected comments and print lines ("\n").
- Some regularity of expression and some small trivial bugs.
################################################################################
Details:
################################################################################
********************************************************************************
BUG: in gamemode GT_TEAM the fraglimit warning will not be played if joining
the
blue team!
--------------------------------------------------------------------------------
Solution: In "CG_CheckLocalSounds": if cgs.scores2 > highScore, highScore
should
be cgs.scores2.
********************************************************************************
BUG: the "where" console command doesn't work as expected (it's always 0 0 0)
but not in id Quake 3 Arena. It seems that now Ioquake3 is affected!
--------------------------------------------------------------------------------
Solution: In Function "Cmd_Where_f" ent->s.origin should be
ent->r.currentOrigin.
********************************************************************************
BUG: in gamemode GT_OBELISK obelisk explosion won't be drawn if there is no
Rocketlauncher loaded. (The "maps without Rocketlauncher" bug)
--------------------------------------------------------------------------------
Solution: in "cg_main.c": cgs.media.rocketExplosionShader should be registered
if gamemode is GT_OBELISK.
********************************************************************************
BUG: Impact marks sometimes doesn't draw at all. Not easy to reproduce if you
don't play (io)Quake3 every day and know the places where it happens! ;)
But anyway...
Test: start q3dm12 go to "Long Jump Canyon" (where the small platform
teleporter for the BFG is) place yourself at the point where the railgun
spawns, look in the direction where the red suspended armor is. Now shoot
at the sloped wall on the out/leftside of the door you see. (the sloped
wall should be nearly in the center of your screen now). If you choose the
correct brush face and shoot up and down at this brush face, the impact
marks sometimes aren't visible.
There are hundreds of custom maps where this can happen!
--------------------------------------------------------------------------------
Solution: I replaced the function "SnapVectorTowards" with the one from
"Wolfenstein - Enemy Territory (GPL Source Code)"
********************************************************************************
BUG: Normally "NOCLIP" cheaters are logically not allowed to fire a gun.
Unfortunatly the Gauntlet (and Lightning gun) was forgotten and not
restricted to that. All weapons except those two were handled correct.
--------------------------------------------------------------------------------
Solution: Make Gauntlet and Lightning gun not firing for someone who cheats
with "NOCLIP" (like all other weapons).
********************************************************************************
NOTE: A few bugfixes are not mine and are reported here:
http://www.quake3world.com/forum/viewtopic.php?f=16&t=9179.
Thanks to Quake3world, for all those years and the good guys there!
********************************************************************************
BUG: During making a mod I found a very strange bug, which mainly occurs if
someone tries to implement a lot of singleplayer monsters which should
walk
slowly (like the "Crash" bot). So if someone wants to make slow down bots
or monsters when they are walking towards a goal and alter the function
"BotMoveInGoalArea" then the bots/monsters do stupid things. Otherwise and
this is the default (also buggy) behavior they start running although they
shouldn't (as seen with the "Crash" bot and will not be fixed here).
--------------------------------------------------------------------------------
Solution: Fix overflow in bot user command. BUGFIX from "Hunt" mod by J.
Hoffman.
********************************************************************************
BUG: in function "BotMoveToGoal" the special elevator case doesn't make sense.
--------------------------------------------------------------------------------
Solution: in "be_ai_move.c": ((result->flags & MOVERESULT_ONTOPOF_FUNCBOB) ||
(result->flags & MOVERESULT_ONTOPOF_FUNCBOB))
should be ((result->flags & MOVERESULT_ONTOPOF_ELEVATOR) ||
(result->flags & MOVERESULT_ONTOPOF_FUNCBOB)).
********************************************************************************
BUG: in function "BotWantsToRetreat" and "BotWantsToChase" this is wrong:
"(bs->enemy != redobelisk.entitynum || bs->enemy !=
blueobelisk.entitynum)"
--------------------------------------------------------------------------------
Solution: "... redobelisk.entitynum) && (bs->enemy != blueobelisk.." is
correct.
********************************************************************************
BUG: in gamemode GT_OBELISK there are too many node switches for bots
(test: mpq3tourney6 with many bots). If that happens, game becomes
unplayable. I don't know if this is the best solution but here it is:
--------------------------------------------------------------------------------
Solution: In function "AINode_Battle_Fight" right after:
if (!BotEntityVisible(bs->entitynum, bs->eye, bs->viewangles, 360, bs->enemy))
{
I added this:
#ifdef MISSIONPACK
if (bs->enemy == redobelisk.entitynum || bs->enemy ==
blueobelisk.entitynum)
{
AIEnter_Battle_Chase(bs, "battle fight: obelisk out of sight");
return qfalse;
}
#endif
********************************************************************************
BUG: in gamemode >= GT_TEAM, after team change, bots will (sometimes) not stop
shooting at you, although you are on their team now. It seems that the
configstrings are f***** up or not reliable in this case!
--------------------------------------------------------------------------------
Solution: In function "BotTeam" and "BotSameTeam" get the real team values.
********************************************************************************
BUG: Some of the bots voice commands are wrong. They are commanded to attack
the
enemy base but they say "Okay, I will defend!"
--------------------------------------------------------------------------------
Solution: Corrected some voice commands in "BotCTFOrders_FlagNotAtBase" and
"Bot1FCTFOrders_EnemyDroppedFlag"
********************************************************************************
BUG: Spectators couldn't fly through doors if they are very close to each
other.
You can test it with some regular id maps (q3dm14, q3dm12) but there are
also many custom maps where this can happen! This is annoying because in
the worst case you can't move at all and are caught inside a door.
--------------------------------------------------------------------------------
Solution: There is a solution in a mod called "Hunt" by J. Hoffman.
Bugfix is included in this patch!
********************************************************************************
BUG: During making a mod I found it very hard to implement some of my ideas
(something like "Limbo" or "Meeting") because of the way the player spawn
effect, intermission and spawning on victory pads is handled. I reworked
it
a bit and simplified it so that the effect is handled when a client
respawns
(as the name says) and not when a client begins. I think this will help
more
mod makers everytime they want to make changes to spawning of players,
bots
on victory pads or monsters... and want to avoid spectators with
Machineguns
which can kill and score... :()
NOTE: I also renamed the poorly named function "respawn"
to "ClientRespawn". If someone searches the code base for "respawn"
it was really hard to find the correct place for what was
meant. "respawn" is used so often, that you really get headache ...
now with "ClientRespawn" it's easier!
IMPORTANT: The whole respawning, moving to intermission point and
everything related to that is now done in a more reliable way
without changing the default behavior. (How critical the whole
spwaning mess was did you see by yourself (ioquake3 rev. 2076).
With this patch it's safer.
Trust me, I spent hours of fixing silly problems...
--------------------------------------------------------------------------------
Solution: Simplified "ClientBegin" and moved the teleport event
to "ClientSpawn".
********************************************************************************
BUG: If a player is dying or hurted under water the hurt/dying sounds AND the
drowning sounds are played together. This is silly. Moreover it's no good
idea to let the server play client sounds! There was a solution in a mod
called "Q3A++" by Dan 'Neurobasher' Gomes which fixes the problem.
--------------------------------------------------------------------------------
Solution: Created a "CG_WaterLevel" function to play the appropriate sounds.
********************************************************************************
################################################################################
When /connect to the same server is issued while already connected, an initial call to CL_Disconnect will remove all pak file references
and reset the pak order.
Reordering only occurs through FS_Restart, which in turn is called when checksum feed changes. Because we reconnect to the same server,
checksum feed never changes and pak file order is not restored to server order again. With certain pak file constellations between client/server,
this may result in an inability to load files from paks which are not correctly detected as referenced paks.
- Get rid of ugly cvars sv_heartbeat and cl_gamename and replace with single com_gamename
- Remove sv_flatline. Flatlines are ignored by dpmaster and are considered to be insecure because flatlines can be udp-spoofed.
- Fix Ryan's FIXME and have voip packet buffer on the server dynamically allocated via Z_Malloc and store pointers in a circular buffer
- Improve voip target parsing on top of Ben Millwood's patch
- Add new "spatial" target where speaker is spatialized in 3d space and can be heard by all clients in hearing range (s_alMaxDistance)
(#4467)
- Decrease voip sound lengths from 240ms to 80ms per voip packet to mitigate udp packet loss and decrease latency
- Protocol version incremented to 71
- Make server send packet fragments and queued packets when server is idle
- Voip protocol detection is tied to com_protocol making past-end-of-message reading unncessary
- Use Hunk_AllocateTempMemory() for buffering VOIP packets and fix buffering scheme that ryan hates so much
- Disable packet scrambling for new protocol as it is useless now
- Get rid of the old packet scrambling functions predating latest point release
- Use Hunk_AllocateTempMemory() for netchan packet queue to fix memory leak when client gets disconnected with packets in the queue
- Use Hunk_AllocateTempMemory() for download blocks to fix memory leak when client gets disconnected with download blocks in the queue
- Fix SV_RateMsec to account for udp/udp6 packet lengths
- Fix demo selection in team arena menu on case sensitive file systems
- Some changes in the way how vanilla q3 demo file lists are compiled in the menu
Show other players' railgun color and their firetime state.
Show snapshot client's color on world item models of railgun.
Fix the impact mark using color2 (spiral) rather than color1 (beam).
Credits go to Ensiform and Harekiet for the refire portion.
- Bug 5057 - Two weapon number range checks allows invalid number
- Bug 5058 - Railgun explosion cannot be colorized
- Bug 5059 - Client's railgun does not glow in third person and mirror
- Bug 5060 - Warning in RAVENMD4 R_GetAnimTag cross compiling with amd64-mingw32
- Fix problem where users could change values of CVAR_INIT variables after the call to Cvar_Get() via Com_StartupVariable()
- Move sound shutdown after client shutdown so VMs don't hold invalid sound handles in memory
Check if uivm is loaded before using it
Patch adds a check for uivm before using it, fixes running dedicated server
using client binary.
Missing uivm check pointed out by Ensiform.
- Bug 5049 - TA UI forces overstrike mode on when focusing an edit box, by Ensiform
- Bug 5052 - Missing podium timer initialization, by Eugene C.
* differing screen resolutions and network settings are now honoured when changing fs_game
* Fix hunk memory leak on game_restart
* Move cls.state and cls.servername to clc so connection state is fully preserved over game_restart
* Revert back to previous fs_game after disconnecting from a server that triggered a game_restart
* Fix error dialog popping up after every game_restart if an error happened previously (reported by Ensiform)
- Fixed that not all commands added by CL_Init() would be removed by CL_Shutdown()
This patch moves cg_particles.c out of cg_marks.c to the actual file that was
already there but not used. By default WOLF_PARTICLES definition is commented
out if someone wants to use the old cg_particles.c changes. cg_particles.c has
been added to the makefile and msvc projects.
Clear game logfile handle after closing file.
Game does not clear logFile handle when closing file. Might be an issue if
G_LogPrintf is called after this.
Patch from tremulous http://svn.icculus.org/tremulous?view=rev&revision=2041
- Add opStack protection
- Fix dataMask check for OP_BLOCK_COPY
- Add instruction number check for conditional jumps
- Make errors in VM_PrepareInterpreter nonfatal
- Make VM loading more robust, change loading order: when vm_* == 0 first try loading DLL, then QVM in *each* search directory/path
- Fix FS_FileForHandle that would return a FILE pointer to invalid file handle 0
- Clean up ftol()/snapvector() mess
- Make use of SSE instructions for ftol()/snapvector() if available
- move ftol/snapvector pure assembler to inline assembler, this will add x86_64 and improve support for different calling conventions
- Set FPU control word at program startup to get consistent behaviour on all platforms
- Bug 5029 - IQM skin support for upper case surface names
- Bug 5030 - IQM version 2 support
- Bug 5031 - Animated IQMs cause segfault
- Bug 5032 - IQM does not setup tag axis/origin correctly
- Bug 5033 - IQM bounds swap and segfault issues
- Add a third pass to VM compilation for computation of jump address offsets, removing the need to dereference into vm->instructionPointers[] for every call/jump
- Make use of 0F opcode prefix relative jumps
- Call Com_Memcpy for OP_BLOCK_COPY
- Fix several missing range checks for vm->dataBase memory access
- Add all GPR to clobber list as GCC function callAsmCall seems to change register r11 without saving it to stack first, which may result in SIGSEGV in VM_CallCompiled()
- Fix iss8, iss16 and iss32 which would not correctly detect 32-bit signed integer range
- Add support for "addb", "subb" etc. assembler mnemonic
- Change opStack protection for x86_64 VM: No overflow checks necessary anymore as offset register is 1 byte long only
- Changed model_t::md4 to model_t::modelData
- Fix R_ModelBounds for MD4, MDR and IQM models (#4966)
- Support Model format fallback similar to image formats in tr_image.c, patch by Zack Middleton (#4967)
- Retains full compatibility to the old but unsecure protocol between clients and servers
- Harden the connection process against DoS attacks, possibly connected to UDP spoofing
- Add capability to load demos with com_protocol suffix, partially applied patches from Simon McVittie
- Fix demo loading if protocol number has more digits than 2
- Minor refactoring, replace all occurances of suffix "dm_" with global macro DEMOEXT
