The fallback logic is: r1q2 style URLs -> q2pro style URLs -> UDP. In
case that an r1q2 or q2pro server is missing files, we may fall through
to UDP and download files over abysmal slow UDP that are available over
HTTP. To work around this problem we reset the precacher iterations back
to 0 and start over with r1q2 style URLs. This works as long as we're
able to download all files. But if one or more files are unavailable
over all three download options we'll run into an endless loop. Mitigate
this by starting over only once. This comes with the drawback that on
servers that miss more than one file downloads might done over UDP were
HTTP is available. In that case the server should be fixed.
This closes#429.
This prevenits the worlds from advancing during client connect. The
player won't get attacked by monsters or hurt by the environment. Note
that in baseq2 there're still 4 world frames processed by the game and
100 world frames if the player enters a level that he or she already
visited. Both aren't a big problem, 4 world frames are hardly enough
for monsters starting to attack and in most levels the starting area
can't be reached by monsters and is free from environmental effects.
Pause mode is only entered for local servers and only in single player
mode. This should prevent problems with coop and deathmatch games.
The behaviour can be controlled by `cl_loadpaused`:
* `0`: Do not enter pause mode, Vanilla Quake I behaviour.
* `1`: Enter pause mode at load and leave it at first regular server
frame.
* `2`: Enter pause mode at load, never leave it. The player must leave
it by hand.
This was requested in issue #417.
cvar operations are special commands that allow the programmatic
manipulation of cvar values. 'reset' resets a given cvar to it's
default value, e.g. `reset r_mode' would reset `r_mode` to `4`.
'resetall' resets all known cvar with the exception of `game`.
The code is based upon q2pro.
This is part of issue #414.
44472722e added some sanity checks to the AI code. The checks in
ai_run() are likely wrong because the enemy entity might be already
NULL if we arrive their. By aborting early the code is unable to
determine a new enemy or return the monster to idle state, so the
monster will wait forever for an enemy that'll never come.
This happens only in monster vs. monster fights. Never in monster vs.
player, that game ends if the player dies.
In theory this change should be harmless, because if the enemy entity is
gone it won't generate sound targets now be visible. If the game crashes
by self->enemy being NULL we've got a problem elsewere.
This was reported by @BjossiAlfreds in #483. He also suggested the fix.
When searching for the player FinTarget() always goes after sound
targets and aborts as soon as it finds one. So if the player is
constantly generating sounds - for example firing the machine gun -
there's a high chance that monsters will only hear but never see
him. Work around this by adding a small timeout to player noises, make
sure that at least 3 frames passed since the last noise. This gives
monsters 2 frames to see the player.
This bug was present in the original code, this is a small gameplay
change.
The problem was analysed by @BjossiAlfreds in #436. He also suggested
the fix.
M_MoveFrame() calls first the AI functions that decide if a monster
should attack or not. After that the monsters think function is called
which walks through berserk_frames_stand[]. Even if the AI function found
an enemy and decided to attack, the monster is still standing for this
frame and berserker_fidget() is called. It may override the earlier
earlier decision, aborting the attack. Even worse this may let the
berserker stuck, because AI_STAND_GROUND may be cleared which prevents
further attacks.
This bug was present in the original code, so this is small gameplay
change. It's likely also present in both addons.
Reported and analyzed by @BjossiAlfreds in issue #433. He also suggested
the fix.
Like most other cvars 'sensitivity' allowes for float values. But until
now mouse events were handled as integers which led to some confusing
problems. This was especially noticeable at values lower than 1, small
mouse movements were cut to 0 and discarded. Since the clients movement
code is written in floats and we're already using floats for joystick
movement switch the mouse event handling over to them, too.
This should have any impact on configs were 'sensitivity' is ste to
integral values. If it was set to decimal values the behaviour is now
correct.
This fixes#419.
Check if we're in the requested mode after the fullscreen window was
created. If not: Try to switch again in the requested mode by calling
SDL_SetWindowDisplayMode(). If that's successfull set the new window
size with SDL_SetWindowSize(). That shouldn't be necessary, at least to
SDLs crappy doku, but without the subsequent SDL_GetWindowDisplayMode()
call fails with 'Invalid Window'. Use that call to check if we're now
in the requested mode. If yes, process. If not abort and trigger the
fallback magic. It'll set `r_mode 4` and `vid_fullscreen 0`.
Caveat: In the worst case this will switch the display mode 3 times.
To create the window, to work around the bug and to set a refresh rate.
No problem for flat panels, but my unforgotten Trinitron CRT would
have cried in pain.
* Normaly SDL chooses a sane refresh rate for fullscreen windows. Users
may want to override that, so provide a new cvar `vid_rate`. If it's
set to a value greater than 0, we're trying to get a mode close to the
requested resolution and refresh rate and switch to that.
* A bug in SDL may leave us in the wrong mode, detect that condition and
abort. See https://bugzilla.libsdl.org/show_bug.cgi?id=4700 for details.
This is part of issue #302.
Print a list of all available modes as soon as SDLs video backend
initializes and the real display mode after the window was created
or altered.
This hopefully helps debbuging problem with display mode selection, see
issue #302 for an example.
-Add back use of last_position_x and last_position_y
-last_position_x and last_position_y will be set to undefined when the window is shutdown IF the current display used is not the desired display
-last_display will be set to desired display at window shutdown if not the same
-vid_displayindex clamped using ClampDisplayIndexCvar() at startup and window shutdown
-We only need to init the display indices once in GLimp_Init
-We only need to clear the display indices once in GLimp_Shutdown
-Remove extra 'displayindex' variable
-SDL_GetNumVideoDisplays() will always remain the same after the call to SDL_Init(SDL_INIT_VIDEO), so it makes sense to set in GLimp_Init where we do this.
Otherwise the entities origin might be in the surface, which causes it
to be rendered in full black. This fix is a port from KMQuake2, reported
by @m-x-d. Closes#407.
There's an "enable alt joy keys" command now. If a key is bound to that
command, all joystick buttons (incl. hat and triggers) are turned from
K_JOYx into K_JOYx_ALT, which allows two keybindings on the same key,
one with the altselector pressed and one without.
If there's no keybinding for K_JOYx_ALT, it will use the binding for
just K_JOYx as a fallback (if it exists).
This is especially handy to create direct bindings for all the weapons
on the (limited amount of) Joystick buttons.
Seems like AMDs Windows driver doesn't like it when we call
glBufferData() *a lot* (other drivers, incl. Intels, don't seem to
care as much).
Even on an i7-4771 with a Radeon RX 580 I couldn't get stable 60fps
on Windows without this workaround (the open source Linux driver is ok).
This workaround can be enabled/disabled with the gl3_usebigvbo cvar;
by default it's -1 which means "enable if AMD driver is detected".
Enabling it when using a nvidia GPU with their proprietary drivers
reduces the performance to 1/3 of the fps we get without it, so it
indeed needs to be conditional...
use GL3_BufferAndDraw3D() instead of glBufferData() and glDrawArrays()
in each place it's needed.
This by itself doesn't make anything faster, but it will make trying out
different ways to upload data easier.
The developers tested their maps without the fix and decided that it
looked good. Add a new cvar gl_fixsurfsky defaulting to 0 that enables
the fix if someone really want it.
The software renderer already did this, but not the GL renderers. Maybe
the logic was lost somewhere on the long way... Without this change a
fullbright lightmap is generated for SURF_SKY surfaces and without the
SURF_DRAWSKY flags the surfaces aren't skipped in RecursiveLightPoint()
and GL3_LM_CreateSurfaceLightmap(). This isn't a problem under real
skyboxes, but in cases were SURF_SKY is abused fpr interior lightning.
rmine2.bsp in rogue is a good place to see the problem
Reported by @m-x-d, fixes#393.
At least with MinGW on Windows vsnprintf() treats buffer < size as an
error, returning -1 instead of the number of characters that would have
been printed without size restrictions. Therefor msgLen may be wrong,
leading to all kind of funny mistakes further down below... Buffer
overflow included. Work around this by handling the msgLen < 0 case and
adding an explicit terminating \0.
This is another case of "I wonder why nobody has never noticed this",
the GL1 renderers extension string triggered the buffer overflow each
time the game started.
I guess it makes sense to apply gamma to the color, we do the same
for the standard round particles.
Also, this way the fragment shader for square particles references the
uniCommon UBO (gamma is part of it) - apparently the Intel HD4000
(from Ivy Bridge) GPU driver for Windows has a bug that uniform blocks
that exist in the shader source but aren't actually used can't be found
(with glGetUniformBlockIndex(prog, name)), which we treat as an error
in gl3_shaders.c initShader3D().
fixes#391
If the vsync is enabled missuse it to slow the client down, e.g.
calculate the target framerate, add an security margin of 20% and
let the vsync handle the rest. This hopefully solves some problems
with frametime spikes. This is an idea by @DanielGibson.
If the vsync is disabled use a simple 1s / fps calculation.
GetSystemTimeAsFileTime() is okay as long as the game runs fullscreen.
For some reasons it's resolution degraded to ~16ms as soon as the game
runs widowed... Better use GetPerformanceCounter(), its more reliable
and the recommended API for timecounters.
Apparently the lightsource for exploding rockets/grenades is very close
to the surface, so the dot-Product between surface-normal and the
vector between the light and the pixel returns 0, basically disabling
the dynamic light for that surface.
As a workaround, move the lightposition (only for that dot product)
a bit above the surface, 32*surfaceNormal looks good.
fixes#386
The real needed size can't be derived from the .bsp file size, because
* many generated structs contain pointers
* there's lots of data generated per face..
* _especially_ for warped faces that are subdivided
introduced FS_GetFilenameForHandle(fileHandle_t) for this
this helps if a map has been started with "wrong" case, which doesn't
immediately fail if it has been loaded from a pack, but will result
in invalid savegame names that (with case-sensitive FSs) will fail to
load (when going back to a formerly played level)
The r1q2 code prioritized pak files over all other files, e.g. as soon
as a pak file was requested no more file were added to the download
queue until it finished downloading. That way one could be sure that
assets included in the pak file weren't downloaded in parallel as single
files.
This is a better, bugfixed and more robust implementation of the same
logic. With this back in place we can switch back to parallel downloads
which gives a nice speedup on Windows. Maybe, just maybe some day
Microsoft will fix Windows crappy I/O...
Working with getter and setters was a good idea as long as we had one or
two quirks. Now we're at three with maybe more to come so it's easier to
use a struct to communicate quirks between the precacher and the HTTP
download code.
The stores it's text in the key_lines array which is NUM_KEY_LINES *
MAXCMDLINE chars long. The code never checked for overflows, it just
assumed that a line will never be longer then 256 chars * 8 = 2048
pixel. With modern displays we can have higher vertical resolutions,
so the array will overflow sooner or later.
Fix it by clamping the maximum line width to MAXCMDLINE - 2 chars (1
for the prompt and 1 for the terminating \0). While at it increase
MAXCMDLINE to 1024 chars * 8 = 8192 pixel, which is more then 8k
resolution and should be enough for the years to come.
This is belived tot fix at least a part of issue #368.
While loading a savegame the global edict arrays is free()ed and newly
malloc()ed to reset all entity states. When the game puts the first
client into the server it sends it's entity state to us, so as long as
there's only one client everything's okay. But when there're more
clients the entity states if all clients >1 are dangeling. Hack around
that by reconnecting the clients >1 entity states "manually".
I'm not 100% sure if this is okay for q2pro, but at least in my simple
tests r1q2, q2pro and now yq2 generate the same URL. Nevertheless it's
somewhat inconssistent to search generic files at /moddir/... and the
filelist at /moddir.filelist
This closes issue #370.
Until now the softrenderer calculated the fov relative to a hard coded
aspect of 4/3. That's wrong, because we're supporting arbitrary aspects
and we aren't calculating a fov but just a scaling factor to the global
fov which takes the aspect into the account.
Fix this by not taking any aspect calculations into account. BUT: While
this renders the gun with a correct perspective it's positioned much
nearer to the camera / player then in the GL renderers. The GL renderers
work around that problem by enforcing a minimal Z distance of 4 units,
which can't do because we're just calculating a scaling factor...
This fixed bug #357 - the problem was that
client->resp.coop_respawn.weapon and .lastweapon (pointers to gitem)
were not properly initialized when loading a savegame.
Now those fields are saved (=> we had to bump the savegame version)
and for old savegames client->resp.coop_rewspawn is initialized
from client->pers, as a hack for backwards-compatibility.
Until now the UDP download code prohibited downloading of maps from all
pak files. That was some kind of copy protection, without the limitation
demo users could download assets from the full version. Don't apply that
protection for all paks, but only for numbered .pak files.
This could be enhanced by limiting the protection to pak0 to pak2 for
baseq2 and pak0 for both xatrix and rogue.
The r1q2 URL generator was, like everything in this game, buggy. It took
cl.gamedir into account when generating the URLs, but overlooked that it
is only set when 'gamedir != BASEDIR'. So baseq2 assets ended up with
/maps/foo.bsp and mod assets with /mod/maps/foo.bsp. q2pro fixed that
to always include the gamedir...
Work around this by refactoring the HTTP -> UDP fallback logic to be
more generic: Count the number of iterations and depending on the
iteration set the gamedir to be used by the URL generator or force
UDP downloads.
Cleaning the download queue as soon as a file finished downloading leads
in combination with only one parallel download and multiple precacher
runs to an ugly problem: The generic filelist is requested several time
which can lead to cycles. Hack around this by rembering if we already
requested it and reset set reminder as soon as the precacher finished.
Yes, I'm feeling dirty.
If the server sends us an URL with trailing slash we're generating URIs
like http://example.com//maps/foo.bsp. While double // are perfectly
valid they might me rejected by some servers. So let's play save.
While at it replace the crappy Z_TagMalloc() with the standard malloc().
Z_TagMalloc() ist just a wrapper arround malloc(), there's no functional
change.
This looks easy, but is rather hacky... Downloading is implemented
through the precacher. The server sends an asset list, while loading
the map another one is generated. CL_RequestNextDownload() goes
through this list, in the order models / maps -> sounds -> images,
calls CL_CheckOrDownloadFile() for each file. CL_CheckOrDownloadFile()
checks if the file is already there, return true if it is and false
if not. If the return code is false CL_RequestNextDownload() itself
returns, it's called again by CL_ParseDownload() as soon as the just
queued file finished downloading. This way all missing files are
downloaded one after the other, when CL_RequestNextDownload() finally
reaches it's end (all files are there) it send 'begin' to the server,
thus putting the client into the game.
HTTP downloads are parallel, so CL_RequestNextDownload() cannot track
which files are there and which are missing. The work around for that is
to queue the file but have CL_CheckOrDownloadFile() return true. So
CL_RequestNextDownload() thinks the file is already there, continues
with the next one, until all missing files are queued. After that it
polls CL_PendingHTTPDownloads() and sends the 'begin' as soon as all
HTTP downloads are finished.
If a HTTP download fails we cannot just queue it as UDP download,
because the precacher things that the file is already there. And we
can't tell the precacher that it's not because the precacher tracks
files only by the number of downloaded files per asste type and not
their name. Just decreasing the number of downloaded files isn't
possible since the precacher may have progressed to the next asset
type.
So: On the HTTP side it's tracked if there was an error or not. After
CL_RequestNextDownload() has queued all files and waited for all HTTP
downloads to finish it checks the HTTP error status. If there was an
error the precacher state is reset and CL_RequestNextDownload() recurses
into itself to take another run. All files that couldn't be downloaded
are queued again, this time as UDP downloads.
My solution to this problem is somewhat hacky. A newly added function
OGG_SaveState() can be called to save the state and OGG_RecoverState()
at a later time to restore it. There's only one state and it works
only iff OGG is state playing.
This closes#347.
If we're passing file handles to libcurl to write the data into, the
game may crash under Windows due to incompatible C runtimes between cURL
and quake2. This is even mentioned in the official cURL doku.
Since the moment I took a very first look at the download code I wasn't
a friend of parallel downloads. There're several reasons for that:
- Parallel downloading needs some ugly hacks. For example downloading a
pak file has a high chance to make asset downloads running in parallel
unnecessary.
- Parallel downloads are hard to debug.
- There's just no need for them. I've tested several connection, 1
GBit/s LAN, 50 MBit/s DSL, 6 MBit/s DSL, and there wasn't a
significant difference between 1, 4 or even 16 parallel downloads.
I'm leaving the parallel download code in place. I someone really wants
parallel downloads he can bump the MAX_HTTP_HANDLES define.
The signal handler was always fishy since it modified the global process
state but worked on all common platforms. libcurl turns the process into
a multithreaded environment, thus breaking that fragile construction.
After the signal handler was called the global state is inconsistent and
there's a high chance of things going wrong. For example at the net curl
download or when setjmp() is called in Qcommon_Frame().
Until now download queues entries were created for each file, but only
removed in the unlikely event of changed download server. This leaked
about 4200 byte per file. Fix this by:
- Create a new function CL_RemoveFromQueue() that removes a queue entry
and call it everytime when a download either finished, failed or
aborted.
- Retire the 'download done' state, just the queue entries associated
with a download handle to NULL to communicate that the handle is
unused.
- Cleanup the full queue at shutdown.
Loading libcurl at runtime instead of linking it at compile time makes
things a lot easier and more reliable on Windows. On other platform
libcurl can be installed as optional dependency instead as an hard one.
We're printing only the two relevant informations: A download was queued
and a download finished or failed. That's enough to see what's going on
and not too noisy.
...and fix the bugs, that were worked around with that crap, instead.
This removes some corner cases like cancelation of all HTTP downloads
and fallback to UDP if too many 404 errors were generated. If this is
still a problem in reality - for example HTTP servers blocking the
client after too many 404 or even crashing HTTP server - fix the server
and don't force the clients to work around that.
We aren't in 1997 anymore, todays broadband connections are fast enough
to handle multiple large files. This may download some assets twice if
the server provides both a pak file with all assets and the assets as
plain files.
There's no need to parse the HTTP header on our side of things, CURL can
do that. And the old buffer code was overcomplicated, simplify it. While
at it switch to normal malloc().
