bugs.!!!

- a.bat, map tokyo, nosound 0, snd_restart -> crash with "cannot read file ... .wav"
  - the file is corrupt, so should simply ignore it

- q3a map "mythology": has areas with very slow lighting (shoot from hyperblaster inside a long corrydor with disabled lightgrid ...
  fps will fall to 15, 'ent light' will be 50ms)

- gl_light.cpp: when lightstyle #0 (any slow lightstyle?) changed, light from grid not changed; to fix should:
  a) clear lightgrid
  b) store into lightgrid light as style[0]==1.0, and rescale after GetCellLight()
- gl: tris2D not displayed for whole scene (currently - display info about 3D scene, when have 2D drawings)
- r_showBrush, r_surfInfo with brush display: brush can be moved (doors etc), but will be displayed in initial place
- when typing 'bind key var=value', it will be treated as 'bind key var'=value
- r_showBrush/r_surfInfo=8: often DropError() or not show brush (have error message on screen)
- "version" cvar: created in common.cpp, but updated when this file or qcommon.h changed only (date is not updated!)
- check: under Wine: errors does not catched (vc-win32: TRY_S/CATCH_S do works, but TRY/CATCH does not works) -- Wine bugs?
- map "flashfire":
  - when paused, will display network errors (newnum!=oldnum ...)
  ? may be, set numClusters=0 for CM_Xxxxx() funcs (current: set for renderer only)

- a bug with collision detection against moving/rotating brushes (move through model with collision INSIDE model)
  - appeared after sv_world.cpp::LinkEdict/UnlinkEdict/AreaEdicts uses num[Solid|Trig]Edicts; sometimes this var becomes <0 ...
  - seems, this is a bug in game source
- game, "dead body": after player respawned, body will re-appear, and its position will be lerped from {0,0,0} (other coords?); reproduce:
  1) start server, connect to it with 2nd q2 copy, look at some place, record demo
  2) on 1st (or any another, but not demo-recording) q2, die few times behind player, recording demo
  3) disconnect, view recorded demo ...
  ?) for best effect, should dead somewhere near to {0,0,0} point
  ?? -- is it game or my bug? may be, bug in original Q2; when bug will be reproduced: try to record same demo with ORIGINAL q2,
    view demo with original and with my q2

- "dirty": "newgame" from menu: "loading" will appear, but new game not started (no alias)
?! savegame shot: when loading game/show "load menu" from non-baseq2, shot will be displayed when it is not exists in current mod,
   (saved without shot) but exists in baseq2 (incorrect shot; BUT: situation, when save exists but shot -- not, may be treated as incorrect)
- explosions are invisible through non-transparent water (incorrect server-side PVS cull ?) -- should be at least dlights ?? (or not?)
? game, base1, kill all grunts -- dead bodies will fall through floor (not always)
- jumpy demo "tower" (at home, all Q2 versions -- at least from p10) -- low fps on original server computer ?

- q3 player model animation:
  - when standing on a movable entity - legs will move in entity direction:
    ? should analyze usercmd
    ? check ground entity and compare player speed with its speed (dirty, map "lumber", moving tape -- CONTENTS_CURRENT_NNN|CONTENTS_SOLID)
  - should detect teleports (respawns) and reset rotating submodels (current: will 'slowly' rotate from old direction to new one)

- ATI: BK_EndFrame(): if add GL_SetMultitexture(0) to the end of function, sky will be corrupted (NV -- OK); _should_ not be any effect
  at all

- timing:
  - after power sleep/resume FPS counter will be incorrect
  ? may be, use appCycles() in most places (at least, in WinMain()::MainLoop) instead of appMilliseconds() -- will require to implement
    this for non-Pentium (RDTSC) platforms
  ? check sync of float/int versions of appMilliseconds
  ? for testing: may set small timescale, and display graph of local time (frame frac) -- should be smooth (not ladder/sawtooth)

- bugs with deathmatch mode
  - map "dm_dukecity_beta":
    - disco: neon lamp text should not disappear (bug with "dirty" only, all OK with "baseq2" -- in func_wall+func_timer implementation)
  ? g_misc.c::SP_light() -- disabled lights with style>=32

? crash on map change/game load with Anachronox map : crash when door/elevator moving somewhere on level -- absolute pointers ?

- z-buffer bugs with glass floor on "omd_map1"

- command completion: complete cvar with quotes - should use Com_QuoteString() before inserting into edit line; bugcheck:
  1) a="""a""" b """c"""; 2) a->complete+ENTER; 3) a->complete --> will be a different line

- in windowed mode, when launching console/menu, mouse will be deactivated, but it will not be usable at all (no cursor); to use mouse,
  should deactivate window

- new pmove.cpp errors:
  - "ground3" cafe, on the street: cannot jump onto a corner column (horiz. speed set to 0 due to position snap)

! Win9x (98SE): when running + rotating view, when releasing <forward> key, but still rotating with mouse,
  player will continue to run for a few seconds (bug present in original q2 too)
  ? try to close most background apps - may be, the reason is there ...

GAME DLL:
~~~~~~~~~
- after teleporter_touch() all angles will be set to null (but pmove.delta_angles[] contains angle values), and angles will be
  corrected in PMove() call, which will appear only on the next call to ClientThink(). This call will be made in 2 situations:
  1) when server receives another usercmd after teleport (most frequent case) - called from SV_MoveClient()
  2) when teleport was occured within last usercmd, angles will be corrected in PMove() on the next server frame only
     (visual appearance: after teleport player will look at yaw=0 for 1/10 of second, then will rotate to correct view angle)

UNRECOVARABLE:
~~~~~~~~~~~~~~
- rogue demo3: ".../items/defender/tris..." - linked NULL-model entity (modelindex2 > max available model index)
  reason: old rogue used linked model ".../shell/...", but link was created using "modelindex|0x80"; now rogue is new, but this
  demo was recorded using old game


BUGGY MAPS:
~~~~~~~~~~~
- "reboot" (from KP, user-created) pos(-658,-242,-178) dir(-0.91,0.39,0) -- incorrect backface culling in gl (OK in oldgl); reason:
  gl and oldgl have different backface culling (gl -- dot(videdir,normal); oldgl -- using BSP only, w/o dot()), this
  map have incorrect normal info for this surface (if look at r_surfinfo, can see: incorrect brushes too - to simplify map navigation ?)
  To fix: can rebuild normals for draw surfaces
- "winter" (AQ2) pos(-1023,463,55) dir(0,1,0) - incorrect backface culling in gl (OK in oldgl) (same as "reboot")
- game=dirty, map "mine4" (baseq2), laser cannon -- when shoot rock, invisible entity painted (should not be entity at all, "hall of mirrors"
  effect appears); check with baseq2 (sp+deathmatch); when do it 2nd time (wall respawned in DM) - all OK
- "team_harbor" (KP, 3rd-party) - PF_SetModel() -- NULL name (game=dirty; with baseq2 - no problems)
  ? try to create empty (dummy) model and assign it from PF_SetModel(name==NULL) calls
- "flashfire" (q2): 1) have cluster info per leaf, but no visinfo (so: cluster info is useless); 2) have no lightmaps, but set lightmap
  sources (OpenGLDrv will display "disabled lm" + "SetLM(NULL) for non-vertex lightmap shader")


-----------------------------------------------------------------------------------------------------------------------

HARDWARE:
~~~~~~~~~
- ATI:
  - when uploading rectangular texture (GL_TEXTURE_RECTANGLE_NV) in compressed internal format, app will crash in ICD GL dll
  - (old drivers) when game launched fullscreen, <Alt-Enter> will switch to windowed mode with incorrect viewport (visible region);
    vid_restart will not fix situation
  - (Catalyst 4.9) GL_SGIS_multitexture: GL_MAX_TEXTURES_SGIS == 0, but extension is present
  * gamma: after app startup (screen resolution changing) drivers will set gamma==1; tech info: main window will get messages:
    WM_SYNCPAINT, WM_NCPAINT, WM_ERASEBKGND, WM_PAINT; Q3 have no this problem, UT2004 (GL) have.
    * FIXED: set TOPMOST wnd attribute with SetWindowPos(), not with SetWindowLong(); when restore from minimized state - update
      gamma in AppActivate() using "r_gamma->modified=true" i.e. in 1st rendering frame, not using UpdateGamma() call
    - this issue appears, when pressing Alt-Tab from game few times until app will be selected again (no switch to another app
      will be performed, but gamma will be damaged)
  - during video playback (different task/app, using overlay), map textures will disappear (lightmaps only); when fire from
    blaster (dlight) - some textures appear; may be reversed effect: have textures, but with dlight disappears;
    no bug, when no 2D display at all, and when not using multitexture; vid_restart does not helps; when closing video player,
    effect will stay until fire from blaster; additional bug: additive blended texture (multitexturing) may have incorrect
    color (white -> green); this effect will not be corrected even after player quit and quake exit/launch (corrected after
    sleep/resume or Windows restart)
    - gl_interface.cpp::GL_ResetState() does not helps
    - textures appears, when looking in 3rd person at some shadered models (with additive glow - "dangergirl/unpretty")
- NV
  - (NV:bitdepth change, Blade3D:vid_restart): cannot restart GL with Vid_DestroyWindow(force=false) -- require to recreate window
  - screenshot is black after changing color depth, but all OK when make one more vid_restart (TNT2, Win2K, 41.09)
  - GF FX5200 + det.56.72, compiled_vertex_array: have geometry bugs when mixing non-CVA & CVA rendering (disappear with
    gl_showtris=1); workaround: CVA is ALWAYS used (i.e. for 1-pass rendering too); ATI have no this bug
  - GF FX5200: GL_EXT_texture_compression_s3tc: font image, compressed into DXT1 (with 1-bit alpha) have visual bugs (pink color on
    letters); no bug with DXT5 (same bugs with GF2MX)
  - when app not focused (and. may be, during video playback), wglSwapBuffers() may fail (turned fatal error -> warning)
? texture rectangle (ATI; NV untested): when "*video" created as regular texture and uploaded as rect texture, after displaying
  video (exec alias "d1" -- video+demo) when level loading, displayed black rect instead of "loading" banner
? notebook keyboard: when pushing arrow keys with <Shift>, <Shift> will be unpressed while other keys is down (unable to move
  crouched)

COMPILERS:
~~~~~~~~~~
- (VisualC++ 6)
  - when compiling renderer with !STATIC_BUILD:
    - "unguard" will be linked to "NORETURN struc.appUnwindThrow()"; VC ignores NORETURN here,
      and will display warning C4715 (not all control paths return value)
    - FS_LoadFile(filename, &len) have arg "unsigned *len"; using this via structure will produce warning C4700
      (local variable used without having been initialized)
  ? "dot(const CVec3 &, const CVec3 &)" is not inlined, but "DotProduct(const float*, const float *)" inlined ...
    When mark function as "__forceinline" - it will be inlined.
    Reason: "dot()" uses overloaded inline "operator []", and compiler prevents from inlining nested inline functions.
    Compiler even ignores "#pragma inline_depth(8)" and "/Ob1" options ... Workaround: do not use "operator []" - access
    to members directly (as field)
    VisualC++7 have no such problem, but when VC6 inlined dot() everywhere, VC7 inlined most of calls, but some calls
    made as call to separate function.
  - not understand template classes as friends (VC7 - OK)
- (VisualC++ 7)
  - cannot use default arguments for functions, which are part of structure
    Workaround: should use `inline Struc_Func()' (with default args) instead of struc.Func() call
  - not inlined most TString<> calls (e.g. 'operator=') unless specify "__forceinline" (see dot() for VC6 above ...)
- (VisualC++ 8)
  - compilation: should define _CRT_SECURE_NO_WARNINGS before any includes; when defined in Core.h, have
    warnings for Win32 and OpenGLDrv files (when _CRT... not defined -- LOTS of warnings)
  - VC 8 have worse issues with not-inlined functions (see VC7 bugs)
  - AppName.exe.manifest issues
- GNU C++ 3.4: optimization: "-O2"-level: (all bugs - from FP hacks!?)
    ! all these bugs are covered with "strict-aliasing" optimization; can use "-fno-strict-aliasing" to allow FP-hack constructions.
    - incorrect code, when use uint_cast():
	float appRsqrt (float number)
	{
		float x2 = number * 0.5f;
		uint_cast(number) = 0x5F3759DF - (uint_cast(number) >> 1);  //!! this line will be ignored with this optimization!
		number = number * (1.5f - (x2 * number * number));
		return number;
	}
      (appearance): bugs with movement, light; when adapt appRsqrt(), movement will be correct, but lighting - not
    - gl_bspmodel.cpp, LoadSurfaces2()
	if (surfs->side)
	{
		s->plane.normal.Negate ();       // inlined; -> FNegate() *3 times -- FP hack
		s->plane.dist = -s->plane.dist;
		s->plane.SetType ();             // here: few "if (normal[] == 1|-1)" ...
	}
      Bug: normal.Negate() is not finished, but already analyzed in SetType(): normal==(1,0,0)=>(-1,0,0) will be analyzed as (1,0,0)
      (appearance): bad light: no dlights on walls with normal {-1,0,0}; no lighting for models from this wall's SURF_LIGHT
    * Also, check GCC manual:
       - "3.8  Options to Request or Suppress Warnings" : '-Wstrict-aliasing' to detect possible errors -- our situations was not detected!
       - "5.33 Specifying Attributes of Types" : 'may_alias' attribute
       - "6.3  Restricted Pointer Aliasing": '__restrict__' keyword