NOTE: This file has been retired, and contains changelog entries
from v3.35.92 up to the release of v3.39.20.
See haleyjd_changelog_5.txt for more recent updates.
================================================================================
Changes since Eternity Engine v3.37.00
================================================================================
--------------------------------------------------------------------------------
10/09/10
Added iwad_hacx variable and added the HACX titlescreen to startup.wad, and
made modifications to the IWAD picker to allow choosing it, to finish up
support for the HACX 1.2 release.
Updated the changelog and version info for upcoming 3.39.20 "Resheph" release.
--------------------------------------------------------------------------------
10/05/10
I added an r_voxels module and code to load .vox format voxel models.
--------------------------------------------------------------------------------
10/03/10
CSonicGo found a bug in the d_fastrefresh tweak which caused -timedemo to break
completely - it would dump you to the console, where you could enter commands
that would be acknowledged but never executed, and where you could access the
menus, but changing certain variables would cause the game to crash. Indeed,
the game was stuck in GS_STARTUP state because the first gametic could never
be run with TryRunTics returning immediately every time.
The fix is to not do the early return from TryRunTics if the timer is being
used unlatched, as in -timedemo mode.
--------------------------------------------------------------------------------
09/07/10
I made some changes to d_fastrefresh which make it even faster and allow it to
behave as proper unlatched framerate support - though note that there is still
no interpolation support and so the apparent FPS is 35 even though the drawn
framerate is 200 or higher. I can definitely tell the difference though,
especially since d_fastrefresh=false cannot attain even 35 FPS under Windows 7
due to some kind of bizarre lag in getting the refresh to the display. I
suspect this is related to Aero somehow, and seems to affect all SDL apps when
running under Vista or 7.
I also added a new d_drawfps cvar that will show the drawn FPS on screen,
rather than the count of gameloops per second, which caps at 35.
--------------------------------------------------------------------------------
08/15/10
Added a logical warning output system for EDF which, with use of the
-edf-show-warnings parameter, can print out warnings to the Win32 (or Linux)
console window. Warnings will additionally always appear in the verbose log
even when this parameter is not used. A count of EDF warnings is always shown
on the console.
printz found another DECORATE states bug; I needed to allow transition back to
the NEEDLABELORKWORSTATE state after accepting a 'bright' keyword as well, and I
somehow failed to notice this oversight while adding support for making the
transition after parsing the tics token.
It is now possible to edit the default values of sync-critical variables while
a demo is in progress, rather than being told by the menu system that you cannot
do something because a demo is playing. This required some special checks in
the console system which switch over to setting only defaults when the cvars
are in a state where their runtime value should not be modified.
The end of a demo will automatically propagate defaults of sync-critical vars
back into their runtime versions, courtesy of the BOOM code which has always
worked this way, and so this is just exploiting a facility that already exists
and was only overlooked by fraggle while implementing the console in SMMU.
--------------------------------------------------------------------------------
08/14/10
After attempting to implement modification-safe blockamp list traversal, I
realized that nullification of blockmap and sector lists in P_UnsetThingPosition
is actually unnecessary and causes many more problems than it solves. Instead
it suffices entirely to do the addition of MF_NOBLOCKMAP|MF_NOSECTOR when
removing the thing, so that it cannot be relinked into the lists. Not nullfiying
the links does at least still allow the object to be traversed through while
completing traversals during the current gametic, while the object is still
valid in memory even though it has been removed from the gamesim.
--------------------------------------------------------------------------------
08/13/10
Added some safety tweaks when demos are not being played. Sector and blockmap
links should be nullified in P_UnsetThingPosition, and the old_sectorlist in
mobj_t should be cleared when calling P_RemoveMobj. MF_NOBLOCKMAP and
MF_NOSECTOR *must* be set on objects when they are removed, or other code in the
engine could force them to be relinked into the blockmap and sector lists while
they are pending to be freed.
The sound engine is also now allowed to increment and decrement reference counts
on objects, in the unlikely event that a thing plays a sound AFTER having
S_StopSound called on it but before actually getting freed. This may come in
handy later for supporting transient sound origins.
--------------------------------------------------------------------------------
08/09/10
Another major bug was found in edf-branch code, this time by printz. I was using
the wrong ehash_t specialization for the new system of state hashing which would
cause lower-case state names to inappropriately go unrecognized. Fixed by
changing E_StateNumForName to use the case-insensitive specialization.
I also added additional use of ZONEVERBOSE in z_native.c to #if out some more
dangerous message prints which can wipe out the original cause of a memory error
when the game is trying to report it.
--------------------------------------------------------------------------------
08/08/10
I removed the SMMU limit on console argvs and tokens, including a fix to an
unchecked crash condition if a command had more than 64 tokens. The token and
argv arrays are now fully dynamic and will reallocate if their initial size
is exceeded.
--------------------------------------------------------------------------------
07/29/10
A long-time fear of mine was confirmed today when I discovered that DOOM sets
the game into the GS_LEVEL gamestate too early - at the start of G_DoLoadLevel
instead of at the end of P_SetupLevel when everything has been created. To fix
this I added a new GS_LOADING gamestate which is entered in G_DoLoadLevel, and
then I do not set GS_LEVEL until everything is actually ready to be manipulated
externally.
I have rewritten (already) the DECORATE state parser so that it is semi-tolerant
of errors, rather than bombing out for any little mistake. No states defind in
an erroneous DECORATE state block can be used, however, and the DSO will be
disposed of when any error occurs, causing a NULL return value from
E_ProcessDecorateStates.
DECORATE state processing errors will now appear in the EDF verbose log file.
--------------------------------------------------------------------------------
07/26/10
esselfortium found a major bug from edf-branch which disabled the ability to
load EDF from wads - this was caused by returning lump->index (the hash chain)
instead of i (the lump number) from E_FindLumpInclude.
--------------------------------------------------------------------------------
07/25/10
Implemented an experimental change to dynasegs that hopefully fixes the problem
of inconsistent splitting of polyobjects with vertices that are too close to
node lines found by Graf Zahl in ZDoom. I also added QStrFindFirstOfChar and
QStrFindFirstNotOfChar routines which are similar to std::string's
find_first_of in C++, and employed them immediately in the console. I rewrote
tab completion in the console to use qstrings and to eliminate some dangerous
unchecked static limits from SMMU.
--------------------------------------------------------------------------------
07/24/10
I have eliminated most of the rest of the static limits in the console via
additional application of qstrings.
--------------------------------------------------------------------------------
07/18/10
I added a QStrCDupAuto method for creating an automatic-allocation copy of a
qstring_t's contents, so that temporary, disposable copies of qstrings can be
made for one-off operations
I made E_SpiltTypeAndState in e_things.c global so that it can be used by the
state evaluator in e_args.c, and implemented a new E_ArgAsStateLabel evaluator
with support for the :: scope operator in state jump targets. This allows the
A_Jump codepointer to jump to any foreign object state.
All other state evaluators also now support DECORATE state labels, when the
jump target is not found as a global state name first (this makes DECORATE
labels a fall-back in pointers that use these routines). I also added an
args[2] parameter to A_Scratch which allows the pointer to play a sound when it
is invoked from a DECORATE state.
--------------------------------------------------------------------------------
07/16/10
I gave the qstring module another massive overhaul. I changed all size and index
variables/parameters to use size_t rather than unsigned int, changed the
QStrBuffer, QStrSize, and QStrLen functions into macros for efficiency, and
added new macros and functions QStrConstPtr, QStrClearOrCreate, QStrBufferAt,
QStrCNCopy, QStrInsert, QStrMakeQuoted, QStrNCaseCmp, QStrNCmp, QStrQCat,
QStrQCopy, and QStrToDouble.
--------------------------------------------------------------------------------
07/12/10
I have added a basic inheritance mechanism to metatype_t's so that they can call
into their parent class's methods. For example, in the new toString method for
metastrings, the metaobject base class copy method is invoked first to do a
shallow copy, and then the referenced string is sent to strdup so that the new
metastring gets an independent copy of the string.
I also added a toString method for speedsets, the only metaobject type that was
currently missing this method.
Tweaked the w_levels module so that the w_masterlevels command cannot be
exploited to load PWADs in shareware games.
--------------------------------------------------------------------------------
07/10/10
I found a problem in Z_Realloc that would have crashed the game if a memory
outage occurred and the game tried and failed to remove purgable blocks. I was
also able, with a realization that the currently reallocating block is detached
from the zone block list beforehand, to restore support for reallocation of
purgable blocks.
--------------------------------------------------------------------------------
07/08/10
I got sick of typing M_ everywhere just to do string operations, and so I
dropped this prefix from all qstring routines, leaving them with names such as
QStrCopy. I think this is a vast improvement.
While in m_qstr.c I took the opportunity to reorganize the module, giving each
function a proper categorization.
I rewrote the native heap's zone alloca system to exploit the native heap's
blockbytag list so that it no longer has to add an extra header into the
allocation or do its own linked list tracking. This allows me to add a
Z_Realloca for reallocation of automatic blocks.
--------------------------------------------------------------------------------
07/07/10
I re-removed the error message length limitation in E_EDFLogPrintf by using two
va_list objects. An initial implementation of M_QStrPrintf, a function that can
do formatted printing into a qstring, has been added. I started replacing all
instances of psnprintf/M_QStrCpy pairs with calls to this new function.
--------------------------------------------------------------------------------
07/05/10
Added support for the mythic, unimplemented vt_toggle cvar type mentioned in
comments from SMMU, and then I changed all actual boolean variables to use that
cvar type, so that they are not using int * to write into boolean locations
(even though this is generally perfectly safe under C, it's still semantically
poor).
I have created a stopgap solution for associating cvars with their corresponding
default_t in the config file, so that the console now has access to defaults.
The default value is now printed when the cvarhelp command is used. I also
rewrote cvarhelp so that it properly uses the console's hashing mechanism,
rather than doing an inefficient linear search over all console command objects.
Console.argv and Console.argc have been changed from char arrays into qstring_t
objects, eliminating all SMMU static limits on console token and command length.
Fixed undefined strcpy behavior in C_SetVariable in the process.
cvars can now be reset to their default value by using the syntax:
<variablename> *
That is, following the variable name with a star character.
--------------------------------------------------------------------------------
07/03/10
I rewrote the config file system to use a pseudo-object-oriented design for
default_t objects, default_t's are populated with methods from their
corresponding static interfaces at startup.
--------------------------------------------------------------------------------
07/02/10
I have added full support for floating point cvars to menu slider controls,
including the snazzy ability to display the current value of the slider above
the slider in a framed box while the slider is the currently selected menuitem.
--------------------------------------------------------------------------------
06/25/10
Work continues on the LevelInfo rewrite but I'm running into some problems that
may take a good while to overcome.
--------------------------------------------------------------------------------
06/24/10
edf-branch has been merged into trunk, bringing in support for dynamic states,
sprites, and DECORATE states.
Some post-merge cleanup work was required, including resolving the issue in
w_wad.c so that all wad files, even ones loaded into private directories, are
given a unique source identification number.
The A_Jump DECORATE pointer is now fully functional.
I finally got around to changing all EMACS mode select strings in the source
code so that automatic source code analyzers such as ohloh will not misidentify
the code as C++.
--------------------------------------------------------------------------------
06/21/10
Started work on rewrites that will hopefully enable LevelInfo to be cached, so
that Hexen-style MAPINFO can be supported properly within the same system.
--------------------------------------------------------------------------------
06/20/10
Added a GIF_HUDSTATBARNAME flag to GameModeInfo which controls the position of
the automap levelname widget, rather than branching on GameModeInfo->type in
HU_Init. I also added a clear method to the leveltime widget which will move
the widget up if the automap level name is so long that it would overlap with
the timer.
--------------------------------------------------------------------------------
06/19/10
I finally got around to changing the array of gamemission paths in d_main into
a field in GameModeInfo. Now when new gamemissions are added I don't have to
remember to add an item to an array in another module.
--------------------------------------------------------------------------------
06/18/10
The Master Levels directory can now be re-enumerated if the path is changed at
runtime.
--------------------------------------------------------------------------------
06/17/10
Master Levels support is now complete with the ability to set the directory to
use, and support for reading the author names and map titles out of their wad
template files when such are present.Added w_masterlevels command to bring up
the special loading menu.
--------------------------------------------------------------------------------
06/16/10
Additional work has been done on the w_levels module. It is now possible to save
and reload the game while on a level loaded out of a managed private wad
directory. Did another round of GCC warning fixing.
--------------------------------------------------------------------------------
06/15/10
Added a new w_levels module for maintaining managed private wad directories.
Made modifications to mn_files that will allow the file selector widget to be
used externally on an arbitrary directory structure, rather than being
restricted to operating on a single data structure private to that module.
Fixed a bug whereby the IWAD picker module could access freed memory while
closing startup.wad.
--------------------------------------------------------------------------------
06/14/10
I have added support for ZDoom's new uncompressed extended node format, which
has support for maps too massive for the original format. Also fixed a problem
I caused in the new texture system that temporarily broke runtime loading of
PWADs with new textures in them.
Mouse motion events are now finally ignored for real at startup, so you once
again shouldn't start out facing a random direction and looking up at the
ceiling when starting games with parameters such as -warp.
Finally, I have gotten support into p_setup.c for loading levels out of private
wad directories. This is going to come in handy for an idea I have for the
implementation of native Master Levels support.
--------------------------------------------------------------------------------
06/13/10
Simplified the screenshot IO macros in m_shots.c. Also made an experimental
change to span projection to use a Mozilla routine from SpiderMonkey for what I
hope is a portable means of converting doubles into unsigned integers. It works
on PC, but needs to be tested on Linux and MacOS.
--------------------------------------------------------------------------------
06/09/10
Static analyzer fixes continue, though they are all very minor now. I fixed the
low-precision span drawers to use the proper fixedunit (65536). This was lost
during the experimental rewrite when SoM decoupled the x and y scaling factors
in order to support non-symmetric flat sizes. The low-precision drawer for
64x64 flat drawing requires special treatment in do_draw_plane to fix this.
--------------------------------------------------------------------------------
06/08/10
I have done more work on I_Error-class functions to ensure that they never
return to the caller. Allowing this to happen in recursive error situations is
incorrect, as the calling code will almost certainly assume exiting as a post
condition of I_Error, and will therefore crash when that post condition is
violated. I_Error etc. will now unconditionally abort if a recursive error
occurs.
I have also added GNU C's __attribute__((noreturn)) to all of these functions,
as it is recognized by the Clang analyzer.
--------------------------------------------------------------------------------
06/07/10
MP2E ran the Clang static analyzer over EE again and it found a new host of
issues, mostly minor, most of which I have fixed. One major problem it uncovered
was an ability to crash in Z_Realloc if the function were invoked on a PU_CACHE
block.
To fix this, I have made reallocation of all purgable resources illegal, since
this has never been supported. The fact that reallocation can cause purgable
resources to be freed makes this inherently unsafe. It may also create issues
with the user pointers for cachable resources.
--------------------------------------------------------------------------------
06/06/10
I have continued to work on honing EE's error/shutdown behavior. Error codes
should only escalate, and not degrade, if reentrancy is detected through any of
the various shutdown routines. Also, each routine invoked from I_Quit should
recheck for the occurance of a fatal error, as some of the routines called invoke
the I_FatalError function.
I found a gigantic memory leak being caused by private wad directories. While I
have been freeing the lumpinfo array, that array is an array of pointers, and
the pointers themselves point to lumpinfo_t objects which are allocated in
separate chunks for each file or group of lumps added to the directory.
This has been repaired by adding a reallocating array to waddir_t which keeps
track of all lumpinfo_t allocations that belong to the wad's directory. A new
routine in w_wad.c can then be invoked when closing the wad file which will
delete all of the lumpinfo_t allocations properly.
--------------------------------------------------------------------------------
06/05/10
SDL_MOUSE_X1 and X2 identifiers don't seem to exist on SDL versions prior to
the latest few, so I've added a define for people who cannot for some reason
update their system headers on Linux. Compiling EE under such cirucmstances
will result in no support for the 6th or 7th mouse buttons.
I fixed a problem caused by my previous I_Error tweaks which was causing any
non-fatal error message to simply not appear. I also added an I_ExitWithMessage
function which can quit the game without causing a logical error condition.
This means that all normal shutdown will proceed in full and the game will not
wait for input at the console in Windows.
--------------------------------------------------------------------------------
06/03/10
I have at last crossed off an ancient FIXME/TODO item by eliminating the
grotesque use of K&R function declaration syntax in m_cheat.c to circumvent ANSI
C type checking, which was being done in order to call functions of multiple
prototypes through the same function pointer member of the cheat_t structure.
Instead, cheat callback functions now accept a void *, which they convert to the
proper type of data inside. This isn't particularly any more type-safe, but at
least it colors within the lines of valid language constructs.
K&R function declaration syntax isn't even supported in the C99 standard as far
as I know, so I have no idea why GCC hasn't been complaining about this for a
long time now.
--------------------------------------------------------------------------------
06/02/10
By user request I have changed the default value of process_affinity_mask to
1. This will have to remain until SDL_mixer gets its crap in order, which at
this point may be never.
I have also enabled the idclev cheat to warp you to *any* ExMy or MAPxy level
(depending on the gamemode), so long as that map actually exists and appears to
be in a valid directory format. This allows directly visiting such locales as
SoD MAP33 and Crucified Dreams MAP99 without use of the console.
--------------------------------------------------------------------------------
06/01/10
The normal "IWAD found" message will now appear when a disk file is in use, and
the canonical resource path and name from the disk file will appear.
After user complaints, I have disabled the ability for the unbind command to
automatically release bindings in the kac_menu and kac_console key action
classes when executing the unbind action with only one argument. It will now
be necessary to explicitly specify the key action class in order to unbind such
actions from the console.
This helps avoid a serious problem that could develop if the user inadvertently
releases both the "toggle console" and "toggle menu" bindings at the same time.
>_>
I have also added support, by request from Gez, for SDL's MOUSE_X1 and MOUSE_X2
mouse button identifiers, which support logical mouse buttons #6 and 7. On my
Logitech Click! mouse, the "application change" button is recognized as X1 under
the default Windows 7 USB mouse driver.
--------------------------------------------------------------------------------
05/31/10
Disk file loading and version recognition have been extended for Ultimate DOOM
disk files.
IdentifyVersion has been split into IdentifyIWAD and IdentifyDisk functions to
resimplify the logic and to allow the eternity.wad file to be consistently
loaded before the IWAD, regardless of whether it comes from a disk file or not.
--------------------------------------------------------------------------------
05/30/10
Support has been added for parsing specific "metadata" files inside disk files,
which can override defaults for LevelInfo, as well as set certain properties
such as the episode name.
--------------------------------------------------------------------------------
05/28/10
I have added support for ".disk" files, which may contain an IWAD and one or
more additional PWAD files. These files have a 72-byte-entry directory structure
at file offset 0 which appears as follows:
00000000: DWORD - number of files (big endian)
00000004: Start of directory
????????: DWORD - total size of file minus header (big endian)
Directory entry:
00000000: char[64] - file path/name, null-terminated and extended with junk
00000040: DWORD - offset of file, relative to end of header (big endian)
00000044: DWORD - length of file in bytes (big endian)
Game mode and mission recognition characteristics have been extended to the one
currently known disk file.
--------------------------------------------------------------------------------
05/27/10
Changes have been made to sidedef loading to match those made to sector loading,
allowing R_SearchWalls to also avoid using a temporary string buffer.
My previous fix to ga_victory avoidance on ExM8 with LevelInfo.killFinale set
needed an additional tweak to avoid returning early from G_DoCompleted.
Otherwise you'd never exit the level, which was quite bizarre.
--------------------------------------------------------------------------------
05/26/10
A Doomworld user found there was a problem with LevelInfo.killFinale on ExM8
levels in Doom 1. I needed to add code which overrides the automatic triggering
of gameaction "ga_victory" in this circumstance.
I have changed the reading of sectors to be done on a field-by-field basis
rather than via structure mapping, so that the flat names may be stored into
char [9] arrays rather than char [8]. This allows proper null-termination of
the strings before they are sent into R_SearchFlats, allowing the hack of
copying the argument into a temporary buffer in that function to be removed.
--------------------------------------------------------------------------------
05/23/10
SoM completed tolerance for missing patches in multipatch textures under the
new "experimental" texture system.
--------------------------------------------------------------------------------
05/22/10
I earlier noticed while doing heap debugging for the V_CopyRect problem that
there were a huge number of 32-byte allocations on the heap stemming from
e_args.c, and I then realized that all states are being given an argument list,
which is in itself a problem that still needs to be addressed.
However, on top of that, all the states are being assigned 5 empty argument
strings when their arguments default via a cmp field. This means that there are
tons of copies of the empty string "" sitting on the heap doing nothing.
I have fixed this by adding code to the e_args module which interns any empty
strings added to arglists by setting them equal in value to a private constant
static empty string. The code to free argument values checks for this before
freeing anything so that it remains safe. This saves a buttload of memory.
--------------------------------------------------------------------------------
05/21/10
I have overhauled I_Error further to add an I_FatalError function, which can be
given an error code as well as a message. This function is now called for
severe errors such as heap corruption, failures in the config file writing
process, etc. which must be immediately fatal; ie., the shutdown code in I_Quit
should not attempt to execute any further.
This restores the ability to save the configuration files when minor errors such
as W_CacheLumpNum occur.
I also audited all I_Error/I_FatalError message strings for \n's at the end, and
added them where needed.
--------------------------------------------------------------------------------
05/20/10
I repaired some more bad problems with interactions between the menu and console
systems from SMMU that were causing some menu widget types to not result in
values being saved to the configuration file if the cvars in question had
default variables.
I have moved almost all c_ variables (such as c_showprompt, currenttarget,
currentheight, c_argv, c_argc, etc) into a new global Console structure.
--------------------------------------------------------------------------------
05/18/10
SoM put in some fixes for slope precision, as well as odd-sized flats such as
some that are found in Heretic and Hexen. Those have to be treated as 64x64
even though their resources are differently sized.
--------------------------------------------------------------------------------
05/17/10
redrawborder *must* be set after the screenwipe routines have finished drawing
for the final time. I have no idea how long this has been broken; possibly
since SMMU.
--------------------------------------------------------------------------------
05/15/10
SoM fixed three more problems with the new experimental branch code:
* The BAADF00D placeholder texture is no longer displayed in the menu background
selection dialogue.
* V_DrawBackground now renders flats with the proper orientation again. They
were, for a short time, rotated 90 degrees due to change in their storage
for interchangeability support.
* r_swirl can now be toggled off at runtime again. It was previously sticking
"on" due to experimental branch changes.
I fixed -solo-net to allow cheats in the same manner as single-player games.
I also rewrote the W_ReadLumpHeader routine so that it caches the target lump.
R_Init was significantly slower under experimental branch conditions due to
constant fseeking through the IWAD to re-read the same patches multiple times.
This fix should bring it back up to pre-experimental spec.
--------------------------------------------------------------------------------
05/11/10
I fixed some problems with warnings from experimental branch found by Gez, and
also fixed a problem with R_MakeMissingTexture having an improper return
statement in it. Added support for experimental code files to the Visual C++
6.0 project as well.
--------------------------------------------------------------------------------
05/09/10
SoM fixed a bug in the experimental flat code that caused 0-length lumps such
as F1_START to be treated as 64x64 flats. The menu would crash quite nicely if
you tried to select such an item as a menu background, since my previous safety
checks were removed or accidentally circumvented in experimental branch.
--------------------------------------------------------------------------------
05/08/10
SoM has merged his experimental branch into trunk, bringing in support for
flat and texture interchangeability, tall textures (>= 256px), and other
awesome features. EE's texturing abilities are now about 99% on par with ZDoom
and PrBoom+, minus the following that may be eventually implemented:
* Support for loading TEXTUREx lumps additively. This is likely to happen very
soon.
* Support for PNG-format textures. This is probably pretty far away right now.
* Support for ZDoom's textual texture definitions. This will probably happen
fairly soon. This will include support for exotic features such as texture
translations, scaling, etc.
* Support for TX_START - TX_END. This should be doable immediately. I hope this
will be in place before the next release.
--------------------------------------------------------------------------------
05/07/10
xlut/ylut pointers are now cleared when freeing VBuffer objects, and are checked
for validity before trying to free them, to avoid potential problems with
dangling pointers and double frees.
--------------------------------------------------------------------------------
05/05/10
Repaired a problem with massive heap corruption originating from V_CopyRect due
to a bad assumption I made about the scaling arrays being consistent regardless
of x/y position (they are not).
--------------------------------------------------------------------------------
05/04/10
I finished a definitive rewrite of SoM's new V_CopyRect after I noticed some
problems in its scaling and clipping code; this should fix multiple issues.
--------------------------------------------------------------------------------
05/02/10
Cleanup work on the menu system continues. I have finally gotten around to
renaming the MN_WriteTextColoured function to MN_WriteTextColored, for
consistency with the rest of the source. My apologies to fraggle ;)
The menu contents widget has been modified to give off sounds upon its
appearance and dismissal, to give proper auditory feedback matching the rest of
the menu system.
I also finally got around to adding code to set the initial index of the
selected item in the contents widget to match the item for the currently
displayed page of the multipage menu. The pointer previously always began on the
first item.
My 4/30 change to add drop shadows to the v-box has required a modification to
V_DrawLoading so that it calls C_Drawer when the game is in startup state.
Otherwise the translucency adds up over the non-redrawing background and just
becomes solid black. This is a problem because of the fact that the main
drawing loop is not yet active while the code is still inside D_DoomInit ;)
--------------------------------------------------------------------------------
05/01/10
Did some major work in the menu system replacing the gigantic switches on menu
item types in the responder and drawer code with tables of function pointers.
After future extension, these modifications may eventually coalesce into a sort
of menuitem pseudoclass structure, bearing method function pointers similar to
what the HUD code already does for HUD widgets.
--------------------------------------------------------------------------------
04/30/10
I added a drop shadow to boxes drawn via the BOOM/SMMU V_DrawBox routine, which
really makes them pop out from the menus, and avoids visual clashes.
The left and right miniskull pointers are now synchronized so that they begin
and end their rotations (which move in the opposite direction) at the same time.
--------------------------------------------------------------------------------
04/27/10 - 4/29/10
Consistency in use of segclip vs ds has been added when determining wall
lighting. I added #if 0 around the now unused R_DrawSprite function which has
been dead since the integration of portals.
Version information is now printed to the console when starting a demo.
The p_markunknowns variable is now saved in the config file - no clue how I
missed this.
Further sound engine tweaks:
* The low-level sound engine is now limited to numChannels, so that it
responds absolutely to the user's settings, and can't fudge it. This
will also allow the variable to offer a real performance improvement
potential, as the mixing callback will no longer loop around all 32
channels even if it's only using 8 or 16 of them.
* snd_channels is now stored in system.cfg
* Hexen thunder sounds are now played from a private sound origin object with
ATTN_NONE, rather than playing on the "NULL" global sound channel. This means
that if, for example, a Heresiarch awakens, thunder will not cut off his
wake-up sound.
* Positioning of sector sound origins is now done in such a way to avoid
numeric overflow when dealing with very large sectors. Thanks to entryway.
* Avoid calling I_SoundID and potentially memset on all inactive sound
channels during every main loop iteration.
* Volume limiting is now performed in the high-level sound code to avoid
oversaturation of the sound mixer when too many sounds are playing.
--------------------------------------------------------------------------------
04/25/10
SoM has attempted to fix flat drawing on MacOS, but we remain unaware of
whether or not the fix actually works.
After getting frustrated with EE's framerate in Speed of Doom MAP28, I have
implemented the drawsegs_xrange optimization from PrBoom+, which cuts down
the number of cache misses generated during vissprite clipping dramatically
(it turns out this code is completely cache bound, and cannot be optimized
effectively in ANY way which does not directly address the motion of cache
RAM). Saves 6 FPS at a particular spot on the map, making the difference
between an almost unplayable 11 FPS and a passable 17 FPS.
--------------------------------------------------------------------------------
04/22/10
3-band EQ should not be run on silence, as it chews a ton of CPU (an amount
drastically disproportionate to running it on actual sound, which must be some
weird consequence of floating-point math on x86).
Fixed a serious bug caused by using a %g specifier to sscanf with a double
variable. Unlike printf, scanf requires explicit notice that the variable it is
writing into is a double vs a float. This is due to lack of promotion semantics
when passing by reference.
--------------------------------------------------------------------------------
04/21/10
After listening to some vanilla Doom sound output and noticing its manly bass
as compared to EE's horrific harsh rasp, I have implemented an experimental
3-band equalizer into the SDL digital sound refresh callback which has support
for bass boost and low-pass filtering.
In order to support this, I have implemented the mythical vt_float console
variable type, which was declared in SMMU but never implemented or supported in
any way. Matching support has also been added to the configuration file, though
groundwork was already laid for this during the previous rewrite to eliminate
pointer-cramming there.
The cvarhelp command now takes unlimited bounds for float and integer variables
into account properly when printing the range of values they can take. It would
previously display the values 123456789 or -123456789 literally. I also added
proper support for string-type variables.
--------------------------------------------------------------------------------
04/19/10
I have been forced to finally fix the "ninja spawn" phenomenon, caused by
integer overflow in G_CheckSpot when spawning players at deathmatch spots. The
code there treats signed mapthing_t::angle as unsigned, and as a result,
generates out-of-bounds indices into the finesine and finecosine table.
Fortunately for purposes of compatibility, this behavior in vanilla has been
reverse engineered by entryway. This is important because ninja spawns can have
an impact on gameplay during intense deathmatch rounds.
I have also repaired a problem from vanilla Doom which involves your spawn
sound getting lost. There was a bizarre hack in G_CheckSpot trying to somehow
magically defer execution of the S_StartSound call - of course this didn't
work, as C doesn't execute in such a way (I really have no idea what they were
thinking, it's mind-boggling).
Instead the sound must be played after the player has been fully spawned and
positioned, back in P_SpawnPlayer.
--------------------------------------------------------------------------------
04/16/10
Changed tmbase and tm to clip and pClip, and then I changed most of the old
clipping engine to simply use clip directly rather than the clip stack, as SoM
says the plan for portals no longer involves implementing recursive clipping in
this code path. This should improve efficiency, as the compiler can compute
constant addresses for members of the global clip struct which it cannot do
when dealing with the pClip pointer.
Sector link collection has been modified to use the clip stack, however, outside
of old demos, replacing the previous fix from PrBoom by cph which simply saved
several members of the tm* family before overwriting them, and then restored
them afterward.
Hexen mapthing z position setting has apparently been broken forever. It should
work now.
I also eliminated the redundant clip.flags field, which is now gotten instead
from clip.thing->flags, as in PrBoom.
--------------------------------------------------------------------------------
04/15/10
THE DEMO IS IN SYNC!
* Changed EE's JUSTHIT fix to match PrBoom+'s, possibly breaking support for
EE 3.35 demos.
* P_TeleportMove compatibility check changed to match PrBoom+.
* PrBoom+ PIT_CheckThing return value compatibility fix applied.
* Added an mobj_t position logging system.
* Dropoff logic, which proved to be the problem causing the Cyberdemon's
position to be incorrect, has been split up into different functions which
are called via a pointer depending on the demo version, because it's simply
too complex to combine all of this code into a single path. This alone was
what finally fixed the demo.
--------------------------------------------------------------------------------
04/13/10
I have not yet succeeded in restoring sync to Okuplok's demo. The PrBoom+ fix
to 2.02/MBF incompatibility in P_LookForPlayers must use the P_SetTarget
function or else risks corrupting the mobj_t reference counts. This is probably
a bug in PrBoom+.
I have found that a previous refactoring in P_TryMove has been responsible for
bad dropoff behavior in EE for quite some time now, so I have had to remove
the refactoring altogether - this is because I cannot understand the original
code. It is impossible to read due to impenetrable abuse of conditional
expressions with side effects, to the point where applying order of operations
rules doesn't even work unless you are a machine.
--------------------------------------------------------------------------------
04/11/10
I have set out on a maddening quest to restore/improve BOOM 2.02 demo sync to
EE, at least well enough that Okuplok's amazing demo of Speed of Doom MAP33
will remain in sync.
Aggravatingly, this demo stays in perfect sync almost all the way to the end,
when the Cyberdemon appears on the "descent to nowhere." As the Cyberdemon
dies, the player appears to become hung up on him for a moment longer than in
PrBoom+, and then appears to continue on as normal.
However, the player's position being just slightly off alters the behavior of
the monsters in subtle ways that begin to add up over the next 30 or so seconds
of gameplay, culminating in an Imp who is in the wrong spot firing a fireball
which shouldn't have been fired, which knocks the player off course, causing him
to miss the Soul Sphere in the central pillar. This ensures his untimely death
shortly thereafter in a hail of Hell Knight plasma balls.
I have started with the following fixes and tweaks:
* G_SetCompatibility has been adapted from PrBoom+ to facilitate tracking of
the versions in which various compatibility tweaks were first added to the
engine, and first became optional as opposed to being hard-coded to the
current demo_version.
* PrBoom+ fixes for incompatibilies between BOOM 2.02 and MBF in P_Move and
P_LookForPlayers.
* BOOM 2.02 friction model code restored for use only during 2.02 demo playback.
* p_map3d functions made to I_Error if ever invoked accidentally during old
demo versions.
* PrBoom+ fixes for incompatibilities between 2.02 and MBF in player bobbing,
particularly when related to friction changes and the effects of voodoo
dolls.
--------------------------------------------------------------------------------
04/10/10
Player color and name settings have finally been restored to full functionality.
Turns out this was accidentally lost by cutting out C_InitNet during the MBF
netcode restoration.
This fix is only good for single player, though, as EE still lacks any ability
to multicast settings to all nodes in order to allow variables which have
unique values on a per-player basis. This is one of the major deficiencies of
Eternity's current netcode and needs a fix badly. Players will still have
auto-generated names and default colors at the start of any netgame.
--------------------------------------------------------------------------------
04/09/10
I have added support for analogues of two more DECORATE actor flags:
SYNCHRONIZED - things with this flag are not subject to having their first
spawnstate's tics randomized in P_SpawnMobj. This is normally
done so that monsters don't all start out doing the exact
same thing, and torches flicker differently, etc.
NORANDOMIZE - As a projectile, a thing marked this way will not have its
spawnstate and first death state tics randomized. Unfortunately
this flag's name and semantics had to be negated from ZDoom, or
else major problems with EDF backward and forward compatibility
would have been created. So in EE, this flag disables the
randomization of projectiles, whereas in ZDoom, RANDOMIZE is
required to enable it.
Because the VPO warning hud widget is entirely inaccurate and has outlived any
usefulness it may have ever had, I have removed that feature in order to spare
ptoing the need to make a new graphic for it. I also made some minor fixes for
menu sliders in Heretic, and drawing of the current crosshair on the HUD options
menu.
--------------------------------------------------------------------------------
04/08/10
SoM rewrote V_CopyRect after deficiencies were found in its clipping abilities.
The header for the frags hud widget is now positioned properly regardless of
the graphic's dimensions, so that ptoing's new graphic for Heretic will appear
in the proper location.
--------------------------------------------------------------------------------
04/07/10
The ZONEVERBOSE define has been added in z_native.c in order to control output
of potentially dangerous fields from the zone memblock structure during error
conditions, because printing these fields when the heap is in a corrupt state
generally results in an immediate segmentation violation, obliterating the
ability to get useful error message output. The primary field in the memblock
struct which is a culprit in this respect is block->file when INSTRUMENTED is
defined.
--------------------------------------------------------------------------------
04/06/10
As I have serious doubts about the sustainability of using floating point code
in the gamecode as far as demo sync goes, I have eliminated use of
tantoangle_acc within the gamecode. It is now only used within the renderer;
P_TanToAngle uses only the old precalculated (and therefore reliably valued)
tantoangle table.
I eliminated most uses of "unsigned" as a type by itself without any further
qualification.
--------------------------------------------------------------------------------
04/04/10
ptoing has joined the team as a graphics guy, and is setting about improving
various things in eternity.wad for both Doom and Heretic. He has given me some
advice that has lead to some minor changes in drawing of systems like the
console.
Blzut3 has contributed CMake scripts and code modifications that enable EE to
successfully build on Mac OS X, though it is currently only functional on x86,
and furthermore, flat drawing is horribly broken due to some undefined
conversion behavior in Cardboard :/
TheGreenHerring found a WAD in which EE played the music incorrectly. This was
due to the MUS lump in the wad actually being embedded within another wad
file - a wad inside a wad. It became obvious to me that DMX must be capable of
scanning forward at least some distance when looking for the MUS\x1a header,
because vanilla Doom plays this music normally. I have made this adjustment to
EE's mmus2mid.
--------------------------------------------------------------------------------
04/01/10
Once again, there will be no April Fool's jokes in the changelog ;) I have
added support for the -solo-net parameter to match PrBoom+. This allows starting
EE from the command-line with gametype set to cooperative. While it was already
possible to set this within the game, it required use of the menus before you
could get started, making it no good for demo recording etc.
This was implemented mainly so that I could play the amazing new UAC Ultra in
cooperative mode easily.
--------------------------------------------------------------------------------
03/28/10
Doomworld Forums user Six somehow stumbled onto a brand-new (well, actually
ancient) vanilla Doom bug involving voodoo dolls which has a spectacular effect
in some source ports which I have compared to the Super Metroid Space-Time /
Murder Beam glitch.
If a voodoo doll is injured by an explosion just enough to kill the real player
but not the doll, and the doll is then pushed by the force of the explosion
into a health-imbuing item on the very next gametic, the player's weapon
changing logic will malfunction.
The weapon change starts on the tic the voodoo doll was injured on, and the
player's weapon is set to state S_NULL as usual. However it takes several
frames for the player's weapon to finish lowering. The voodoo doll picking up
health during this time causes the player's HP to be replenished. The next call
to A_Lower sees this and assumes, incorrectly, that the player's current weapon
change must have been caused by explicit player input.
However, because it was not caused by input, the player_t::pendingweapon value
is still set to wp_nochange. This is a special-purpose ordinal which is greater
in value than NUMWEAPONS. Setting player_t::readyweapon to wp_nochange causes
P_SetPSprite to access the weaponinfo array out of bounds, reading undefined
data as a weapon.
This includes using undefined values for the maxammo, current ammo, and most
importantly, the *raisestate* of the weapon.
In PrBoom+ this resulted in the player toting around an Imp with negative 6
billion maxammo. In Chocolate Doom, the player obtains a "Fist of Death," which
while it appears like the normal fist, crashes the game should the player try
to punch or change to a different weapon. In TASDoom, the player obtains an
exploding Revenant rocket. In Eternity, the game simply crashes immediately due
to states[] having been changed into an array of state_t pointers for EDF -
the read automatically goes to a bad address trying to get to the actual state.
This is easily fixed by checking explicitly against wp_nochange in A_Lower
instead of making an assumption about player input versus forced weapon change
relying on the player's health value.
--------------------------------------------------------------------------------
03/27/10
SoM fixed more problems with the clipping of screen patches (it's never-ending,
I think) which fixed patches with posts that extended off the bottom of the
screen. The FreeDoom cast call was one such wad that had problems with this -
columns of garbage would streak up and down the screen while the tiny skeleton
replacing the Cyberdemon was displayed.
--------------------------------------------------------------------------------
03/20/10
I finally fixed deep water beneath skies, which had been broken since I changed
the handling of skies to use an internal sector flag. Because deep water
rendering is not compatible with sky hacks in the renderer, it is necessary to
make sure the SIF_SKY flag is cleared from the temporary sector created by
R_FakeFlat.
SoM fixed more sky problems, including one with slopes.
--------------------------------------------------------------------------------
03/19/10
The last IWAD chosen via the IWAD picker will now be remembered in system.cfg,
as per user request by Gez and others.
--------------------------------------------------------------------------------
03/18/10
I have tweaked I_Quit so that saving of configuration files is not done when the
game exits via I_Error. This is necessary because the game may be in an unstable
state. Even if written successfully, the configuration files could be corrupted
with incorrect data. More often however, the config saving code crashes the game
by causing a recursive call to Z_Malloc.
Because Z_Malloc is not reentrant, this can cause it to enter an infinite loop
attempting to dump PU_CACHE blocks.
--------------------------------------------------------------------------------
03/12/10
SoM further fixed problems with seg scaling assumptions in Cardboard, as well as
an off-by-one error which had been causing a single pixel of HOM at the bottom
of certain skies for a very long time now.
--------------------------------------------------------------------------------
03/11/10
I have made some minor optimizations to R_AddLine by collapsing the numerous
mark* variables into a flag field. I also added code which reinforces the
implicit condition that f_window or c_window must be valid if the corresponding
mark* flags are set. This allowed elimination of multiple redundant checks on
the validity of these pointers later on.
I modified R_FakeFlat to return NULL when passed NULL, for safety purposes.
The seg_t::offset field is always being used a float within Cardboard via
casting, so I have changed it to float, and it is converted to such at map load.
I have also added tag names to 99% of structure types in EE. A shocking number
were relying only on typedef names, leaving the structures themselves anonymous.
This is bad when you need to use "struct foo_s" somewhere but end up only having
the corresponding foo_t typedef.
--------------------------------------------------------------------------------
03/09/10
Another fix for Linux 64-bit involved changing the heightmask variable in the
quad column drawer to signed, matching the other column drawers.
--------------------------------------------------------------------------------
03/08/10
GhostlyDeath helped me figure out why EE was crashing on 64-bit Linux. It was
in fact dropping out of Z_Init due to failure of the sanity check making sure
that the zone block header size was greater than or equal to the sizeof the
memblock_t structure. The error message wasn't appearing in the console because
it lacked a \n on the end, which is apparently a quirk of stderr in Linux not
being implicitly flushed when a program exits normally.
I had honestly forgotten about the zone header problem. I fixed this by doing
away with BLOCKHEADERSIZE entirely and determining the appropriate zone header
size to use at runtime by rounding up sizeof(memblock_t) to the nearest higher
multiple of 32 bytes. This formula still gives a 32 byte header on 32-bit, and
gives a 64-byte header on 64-bit, and should maintain alignment assumptions on
just about any platform.
I added packing pragmas to the post_t structure, because it is mapped directly
onto patch lumps loaded into memory. C data type agnosticism has been increased
throughout texture and map loading code, as well as by replacing most instances
of "short" with int16_t.
--------------------------------------------------------------------------------
03/07/10
Version name recognition and game mission tweaks where needed are now in place
for all of the official FreeDoom IWADs, which are recognized via the FREEDOOM
lump during IdentifyVersion.
SoM fixed a bug in Cardboard which incorrectly assumed that a seg scaled to
nothingness at one end would be similarly zero-height at the near end. This
fixed CSonicGo's school-shooting simulation wad :P
--------------------------------------------------------------------------------
03/04/10
Catoptromancy found a wad (AMADEUS.WAD) in the /idgames archive which contains
a corrupt blockmap lump in which several of the block cap words have a single
bit flipped, making them appear as FF FE in a hex editor rather than the
appropriate FF FF. This seems to fly in vanilla DOOM, since it can access just
about any memory address it wants, but in EE the program crashes trying to
access the lines array at an outrageous address.
To fix all such wads, several more of which have appeared afterward, I have
added a P_VerifyBlockMap routine which does complete verification on blockmap
data read from wads, including checking all linedef references, checking that
all blocklists terminate before the end of the lump, and that all blocklist
offsets in the header are within valid range.
If any of these checks fails outside of demo compatibility mode, we will rebuild
the blockmap using Lee's code from MBF.
--------------------------------------------------------------------------------
02/27/10
To celebrate SVN r1000 I have enhanced the graphical DOOM HUD with support for
translucency, controlled by the st_fsalpha variable, as well as long-requested
support for displaying the keys the player is carrying.
--------------------------------------------------------------------------------
02/23/10
I have de-Killoughified the branches in P_UpdateThinker that determine the
thinker class, and removed a minor FIXME on something that I consider a
non-issue in p_user.c.
--------------------------------------------------------------------------------
02/21/10
I have added preliminary support for the recording of v1.9-compatible demos, a
long-requested feature. There is still some work to be done in making sure that
EE only executes compatibility codepaths while recording, and that all optional
overflow emulation code is enabled for maximum fidelity.
--------------------------------------------------------------------------------
02/19/10
I changed E_HashInit so that functions passed to it as methods are all passed
inside a new sort of structure I call an interface which only consists of
function pointer members.
I have reorganized a lot of the code in P_SetupLevel to clean up that function
considerably, as it was way too long (several hundred lines). Several operations
that need to be performed pre-Z_FreeTags had found their way into the
G_DoLoadLevel function rather than P_SetupLevel, where they really belonged,
so I have moved all of them there and into their own function,
P_PreZoneFreeLevel. This way they will all stay together from now on.
--------------------------------------------------------------------------------
02/18/10
I have placed a more reasonable limitation on the growth of metatables so that
each individual table will not grow to more than 400 KB on 32-bit platforms.
At the number of hash chains this implies (around 96000), there would need to be
around 65000 objects in the table to cause another overload condition with the
current rehash criterion of 66% loadfactor. If EE is hashing 65000 objects per
metatable, there are going to be worse concerns for performance than the load
factors :/ I do not anticipate that even with Aeon, table loads should ever
become so vast.
--------------------------------------------------------------------------------
02/16/10
I have improved the cooldemo feature from SMMU so that it creates its own
dynamic camera spots which follow the target player from one of his own
previous viewpoints or from the viewpoint of his current attacker.
I found out that EE's demosync was trashed because I accidentally called
Table_SetTanToAngle at the wrong place in G_DoPlayDemo.
--------------------------------------------------------------------------------
02/15/10
I have rewritten W_CheckNumForName and split off a new W_CheckNumForNameNS
function, which eliminates the macroziation of W_CheckNumForName. The new
function is now called everywhere that a specific namespace needs to be searched
for a lumpname, which is how BOOM team should have done it in the first place
rather than being lazy.
An additional function, W_CheckNumForNameNSG, has been added to support
searching first inside a specific namespace, and then in ns_global if the lump
was not found in the first search. This is used by the new code I have put inside
the playdemo console command handler to correct a problem caused by the previous
change that puts command-line-added demos into their own ns_demos namespace
which caused this command to be unable to play such demos.
--------------------------------------------------------------------------------
02/09/10
E_HashRebuild has been rewritten so that it maintains the original order of
objects on each hash chain when resizing the table. This is required to make
sure that modules that expect newer defintions to precede older ones in the
hash table continue to work when they are converted over to use e_hash.
This was made possible by my realization that the m_dllist data structure
could support tail insertion simply by passing the address of the previous
object's next pointer into M_DLListInsert as the list head pointer, which makes
sense when you think about the inductive definition of a list.
I fixed the old vanilla bug where the Doomguy would evilgrin on the status bar
when picking up any item after loading a savegame. This was caused by the game
failing to wake up the status bar again after loading the saved state over the
blank-slate state created by the previous call to P_SetupLevel.
--------------------------------------------------------------------------------
02/07/10
SoM suggested an optimization to e_hash table rebuilding which involves caching
the unmodulated key value in each node. This removes the necessity of
recomputing hash keys for every object in the table when it is grown.
--------------------------------------------------------------------------------
02/07/10
I have added an e_sprite module, which is currently unused. This will become the
new home of sprite processing and maintenance once it is rewritten for DECORATE
state support.
--------------------------------------------------------------------------------
02/03/10
In order to make test-building Eternity under my XP Pro virtual machine less of
a pain , I went ahead and used some macros to re-enable building under Visual
Studio 6 without the Win2k3 PSDK installed.
I am preparing for a rewrite of sprite name handling in order to support
implicit definition of sprite names in DECORATE states. The first thing I have
done in this regard is to change coallocated sprite names to align to an 8-byte
boundary, which should make accessing them faster, particularly on 64-bit
architectures which care about this kind of thing a lot more than x86.
--------------------------------------------------------------------------------
01/28/10
I moved the test on the line->nodraw field, which is used to keep PolyObject's
parent linedef's segs from rendering, as they are replaced at runtime by
dynasegs, down into R_AddLine and out of the main loop in R_Subsector. This
kind of virtually-always-false branch should usually *not* be moved up, contrary
to your normal expectations, as it causes more overhead than making a few
unnecessary function calls.
I figured out that the rendering glitches in Sunder MAP10 were due to numeric
overflow, and used intuition to figure out that the overflow must have been
coming from the R_PointToAngle2 function. I have adapted similar changes to
those in PrBoom+ and ZDoom that allow this function to call the numerically
stable standard atan2 function when its parameters would cause an overflow.
This has necessitated the splitting off of a P_PointToAngle function that is
used by the playsim only. That function will continue to use the old imprecise
tantoangle table unconditionally during old demo playback.
--------------------------------------------------------------------------------
01/24/10
SoM noted a long time ago that Lee Killough's visplane hash function, though
touted for being well-tuned and tested empirically, did not take any bits of
the visplane's height into account, by virtue of more or less all visplanes
having integral fixed-point heights, meaning the first 16 low bits were all 0.
I got around to fixing this by shifting the height down by FRACBITS before
adding it to the hash key.
--------------------------------------------------------------------------------
01/11/10
CSonicGo found ANOTHER bug, this time the problem was that the status bar would
not be woken up after a screenwipe unless the wipe engine was set to use the
normal DOOM melt wipe. I fixed this by moving the wake up from wipe_MeltDrawer
to Wipe_Drawer.
I deleted most of the left-over dead SMMU menus which were commented out during
the previous menu system redesign.
--------------------------------------------------------------------------------
01/10/10
CSonicGo found an interesting SMMU bug by changing the value of the chasecam
variable during a screenwipe. Turns out SMMU could run the camera thinker code
outside of GS_LEVEL gamestate, which would crash when the chasecam would try
to access the displayplayer's currently invalid body.
--------------------------------------------------------------------------------
01/09/10
Rulesets have been added to GameModeInfo to determine boss special defaults.
Defaults for finale types have been merged into the pre-existing finale
rulesets, and the code to set finale type has been merged in p_info.
--------------------------------------------------------------------------------
01/07/10
Fixed an oversight in EV_ThingSpawn that would have let 256 pass as an ACS
spawn number. Until UDMF is implemented, 255 is the limit on spawn ids.
Rulesets have been added to GameModeInfo for determination of default sky
textures, to replace more branches on gamemode and mission ids.
--------------------------------------------------------------------------------
01/06/10
stretchsky now defaults to off since mlook and keylook are disabled by default.
--------------------------------------------------------------------------------
01/04/10
All xrefs to the Small scripting engine have been corralled into conditional
statements, using the symbol EE_NO_SMALL_SUPPORT. Defining this when building
will result in an EE exe with no Small scripting support. This will be required
for 64-bit builds.
Resolved a long-standing TODO by changing M_FileLength to use fseek and ftell
to determine file sizes rather than using the POSIX fstat routine, which is less
portable and is actually usually slower due to its need to populate the stat
structure fields by calling mulitple operating-system-specific functions.
As a better solution to the problem of BOOM's jumbling up of comments in the
configuration file whenever variables are added or removed from it, I have added
an m_resetcomments console command. Issuing this command will clear all comments
cached from the configuration file at startup, and this will cause the proper
comments to then be written on exit so long as config_help is enabled.
--------------------------------------------------------------------------------
01/02/10
Fraggle has tweaked our main loop so that it can be limited to running at 35 Hz,
consuming less than 100% CPU. However, this is causing bad frame skipping on my
Windows 7 box, so I have had to make it optional on the value of the
d_fastrefresh variable. The older powersaver variable has been eliminated.
--------------------------------------------------------------------------------
01/01/10
Added a new GS_STARTUP gamestate to prevent the gamestate from starting out set
to GS_LEVEL, which could cause any code checking for that state to malfunction
during startup. This was causing a crash in R_DoomTLStyle when the -devparm
command line parameter was issued.
================================================================================
Changes since Eternity Engine v3.35.92
================================================================================
--------------------------------------------------------------------------------
12/29/09
As an exercise of the new screenshot module, and because doing so requires
fewer than 100 lines of code, I added Targa as a supported screenshot format.
Targa, although relatively very old, is still in extensive use in the graphics
and game industries due to its simplicity of implementation, as well as
features it offers that make it useful for display on TV screens (obviously
not a factor to EE of course).
Hopefully PNG support is not far off :)
--------------------------------------------------------------------------------
12/28/09
I have rewritten the screenshot code into a new mini-screenshot engine that
can support multiple formats with minimum addition of code, and also makes use
of a new buffered file output module which was initially written as part of my
experimental zip file writing code that will eventually become part of our new
savegame code, which is still in the planning stages.
--------------------------------------------------------------------------------
12/27/09
EE is now internally consistent with its use of strcasecmp and strncasecmp
versus the nonstandard stricmp and strnicmp that Visual Studio expects. This was
getting totally out of hand, and our defines were bordering on circularity under
GCC.
_CRT_SECURE_NO_WARNINGS and _CRT_NONSTDC_NO_WARNINGS have been added to the
VS2008 project to prevent 394 some-odd warnings of no consequence. I actually
attempted to address the NONSTDC warnings by providing some defines, but
Visual Studio predictably threw a fit since its headers contain defines for
those function names already which cannot be sorted out in preference for our
own without a major bunch of work that is completely not worth it.
Thanks, Microsoft, for supporting portable development so well....
psvnsprintf has been hacked up once again to avoid undefined behavior with
respect to the passing down of va_list objects to multiple functions, to which
GCC 64-bit strongly objects due to some standards-acceptable but highly
questionable implementation decisions on that platform.
Protip: To avoid breaking 90% of C programs, do not make varargs functions use
the AMD64 fastcall convention....
A similarly undefined use of va_list was also eliminated from the E_EDFLoggedErr
routine, although doing this imposed a static limit on the error message length
that I would have rather avoided. C'est la vie.
Linebreak styles have been made consistent in multiple files, as it seems
some of our more Linux-minded developers in the past have used tools that are
not DOS linebreak-aware, and added Linux linebreaks into their new lines
without first converting the existing ones. I have no opinion as to whether
the majority of the code should use Linux or DOS linebreaks, so long as the
individual files remain self-consistent. To me, that is an absolute must.
--------------------------------------------------------------------------------
12/26/09
A Visual Studio 2008 project has now been added for Eternity, currently
supporting 32-bit builds only. I added a tools directory to the SVN project,
and wrote up a qbin2c tool that creates C header files from binary files, as
part of an attempt to get EE to have an icon on its titlebar. However, SDL's
support for this seems to be unacceptably out of date and possibly even glitchy,
so for now EE's icon will only display on the taskbar when built with the VS
2008 project. The 6.0 project doesn't currently provide an icon, but I hope to
be able to rectify that as well soon.
As per a recommendation on MSDN, I have changed the use of GetWindowLong and
SetWindowLong used in the code to support disabling of Alt+Space to use
GetWindowLongPtr and SetWindowLongPtr instead, as these routines are 64-bit
compatible. Sadly, this has caused problems for Visual Studio 6.0, because it
is missing these functions from its headers, even though MSDN states that these
functions have been supported since Windows 95.
This has introduced the need for Eternity to build with the Windows Server 2003
Platform SDK installed and active when using Visual Studio 6.0, something that
I had previously hoped to avoid. This is especially unfortunate because the
PSDK will no longer install properly due to a bug in either the ridiculous
ActiveX installer it uses, or in Internet Explorer 8.0. This requires the
enduser to do a run-around the installer by clicking on the proper .msi files
that are "hidden" in the setup directory.
Doing this precludes any future use of the PSDK update feature, according to
Microsoft, so don't do it if you are using a box on which you intend to do
active future development with newer SDKs. This would probably be best reserved
for that XP box you have over in the corner for a back-up system, or maybe a
virtual machine specifically set up for the purpose of supporting VS6.0-based
development.
--------------------------------------------------------------------------------
12/22/09
Implemented a MetaSetInt routine to match MetaSetDouble, as it might be useful
in the implementation of features such as droptypes and bloodtypes.
--------------------------------------------------------------------------------
11/24/09
I made some important changes to sprite handling so that NUMSPRITES is no longer
used as a value to mean "no sprite" or "no next sprite," but instead the value
-1 is used for the same purpose. This is required to move sprites toward a
fully dynamic model so that they can be implicitly defined in DECORATE states.
This will eventually almost entirely obsolete the EDF spritenames array,
though it will probably have to be retained for definition of the original
DOOM sprites in order to guarantee retention of DeHackEd compatibility.
--------------------------------------------------------------------------------
11/22/09
SoM and I have cooperated to add sprite scaling.
--------------------------------------------------------------------------------
11/21/09
Got around to another long-standing TODO, which was the addition of an r_tlstyle
variable which controls the default presentation of thingtypes that were
modified by BOOM to be translucent, a change some people object to, and others
feel was not radical enough.
Under the current system, which is based on a list of EDF thingtype mnemonics
and tries to avoid stepping on the toes of any existing mods, you can set
r_tlstyle to one of three values, which are explained here:
* None - this value makes all BOOM translucent thingtypes opaque again. This
is vanilla DOOM's presentation mode.
* Boom - pretty obvious I think. The things use BOOM-style translucency.
* New - The thingtypes are selectively set to opaque, translucent, or additive
blending depending on what I myself, with advice taken from others, believe
looks the best. This is Eternity's presentation mode, and is very similar to
the behavior of ZDoom in fact.
--------------------------------------------------------------------------------
11/18/09
Applied a long needed compatibility fix: DeHackEd's maxhealth property is only
supposed to affect the behavior of health potions, not *ALL* health items as
it did in BOOM. This broke HacX, amongst other mods.
z_native.c has been extensively cleaned up via the use of some intelligent new
macros near the head of the file that get rid of what was previously #ifdef
hell (this can still be seen in z_zone.c if you are curious). The code is now
infinitely more readable.
--------------------------------------------------------------------------------
11/13/09
EE now expects to build and link against SDL 1.2.14 and SDL_mixer 1.2.10, which
may prove to be the final 1.2 versions of SDL if what is written on the website
is to be believed.
After doing this, I found the need for a new option to disable Alt+Space in
Windows, since SDL makes no effort to intercept this event. It causes the
window system menu to open, making the game pause and any active keys get
stuck in the down state. This is controlled by a new system.cfg option.
--------------------------------------------------------------------------------
11/12/09
I wiped out the last remaining uses of gitextmetric_t objects, which stored
metrics for game fonts and mapped their access through the GameModeInfo
structure. This was previously replaced by the addition of EDF dynamic fonts,
which store the exact same information, but within the font objects instead of
a separate global location.
After adding the ability to have post-W_InitMultipleFiles adjustments made to
GameModeInfo based on the missioninfo object in effect, I added support for the
playing of DEMO4 in TNT and Plutonia game modes, contingent upon the presence
of an actual DEMO4 lump.
TNT and Plutonia were based on Ultimate DOOM instead of DOOM II as one would
ordinarily suspect, and because of this they used the Ultimate DOOM demo loop
which showed demos for the four episodes. However id evidently forgot about
this and only provided 3 demos in each IWAD. This means that watching the
demo loops of those games in the vanilla engines would result in a
W_CacheLumpNum error upon reaching the fourth demo.
Eternity would not error out, because Eternity used the DOOM II demo sequence
for these modes, and additionally because Eternity ignores non-existent demo
lumps (a message simply appears in the console, and the demo sequence, if it
is running, is auto-advanced to the next state). However, Eternity would also
not display the fourth demo even if it existed, which was technically incorrect
behavior.
A DEMO4 lump is provided in the Plutonia 2 mod, which is what prompted this
adjustment.
A new "open" method has been added to menu objects which allows individual
menus to have customized opening behaviors. This was needed to restore support
for the defaultskill variable from BOOM, which, although it still existed in
the configuration file, was "lost" in SMMU and has been ignored ever since.
In an effort to fix bizarre palette issues that are plaguing DirectX modes on
startup, I have added an I_SetPaletteDirect routine which ducks under our
relatively new palette caching scheme to set the palette to the video
surface immediately. This is needed at startup, as we need a palette on the
surface before any drawing is done, otherwise SDL decides to do an improper
remapping operation once the palette IS set, resulting in a REALLY
attractive screen full of what can only properly be described as technicolor
puke.
Some additional modifications were made that seem to improve the behavior of
DirectX modes under older SDL versions. Mainly, blits should be done after any
pending palette sets and not before, and second, the screen should always be
flipped even if the palette is set, as some types of surfaces violently object
to the missing update otherwise.
--------------------------------------------------------------------------------
11/11/09
I eliminated the irritating need for an "index" parameter to the full table
iteration method (MetaTableIterator) of the metatable and its corresponding
underlying e_hash method, E_HashTableIterator. The index of the current chain
being searched is now cached inside the ehash_t structure itself. This does
require a search to be primed with the proper value in order to restart from
the beginning, of course.
Screenshots no longer pop up the console, so that you can actually take a
screenshot of the console.
The obsolete playersprite variable's processing code has been removed from EDF
(it is still syntactically accepted, but is ignored). The only place using it
has been rewritten to avoid the need for it. This is a necessary change in
preparation for DECORATE state support.
--------------------------------------------------------------------------------
11/10/09
At fraggle's urging I refactored Polyobj_findLines to avoid an unnecessary
goto, and to give the algorithm better documentation.
I have eliminated the old -debugfile and all code related to it, which was
probably added late in id's development cycle while ironing out the network
code. The file was nearly useless to us now, and I have not employed it since
sometime around 2001, making it dead code.
--------------------------------------------------------------------------------
11/09/09
Kate got a confusing error when trying to play a demo named MAP01.LMP - to my
dismay it caused the map it was meant to play with to disappear from visibility,
since demos added with the -playdemo command are linked into the wad directory
as individual file lumps.
I chose to solve this problem (which has existed since vanilla Doom) by placing
such demo lumps into their own new namespace, ns_demos. The lump namespace,
added by BOOM as part of the fix for allowing sprites and flats in PWADs, is
tailor-made to fix this type of problem.
I understand that PrBoom has repaired this by simply not adding demo lumps into
the wad directory, but that requires significantly more code change in g_game.c
than the tweak I came up with, so I am quite happy with it the way it is :)
--------------------------------------------------------------------------------
11/08/09
Unicode input should be supported from ANY key that sends a character within the
ASCII range, not just such keys which also have SDL keysym codes in that same
range. Adding this tweak allows typing numbers with the numeric keypad so long
as the Num Lock key is turned off. This change/fix was requested by SoM ;)
Seems GCC 4.4.2 has added a strndup routine, which causes a conflict with
libConfuse. I renamed libConfuse's routine to cfg_strndup to avoid this.
--------------------------------------------------------------------------------
11/07/09
Essel found that the codepointers A_FadeIn, A_FadeOut, and A_SetTranslucent
were broken due to the argument evaluation rewrite - I accidentally used
E_ArgAsInt where I needed to use E_ArgAsFixed, having forgotten that these
codepointers accepted floating-point argument values to represent alpha levels.
--------------------------------------------------------------------------------
11/05/09
I reformatted mmus2mid.c in the SDL directory after noticing that it was still
a mess from hell. It probably still is even with my changes, but at least it
is readable now.
--------------------------------------------------------------------------------
11/04/09
I have migrated many of Eternity's low-level options to system.cfg so that they
are now universal amongst all supported games, meaning that they do not have to
be set individually for each one.
There may be other options which would be best stored here, and I will continue
to evaluate all of the settings for the best location in which to put them.
I got around to programming a toString method handler function for metadouble
objects. I also added an "isinit" field into the metatype_t structure, which
allowed me to eliminate a couple dozen checks in the source code that were
using "firsttime" guard booleans in order to ensure that metatypes were only
initialized once.
As a result of that change, non-singleton metatypes are now be treated as an
error in MetaRegisterType.
SoM fixed slopes so that they light uniformly in all fields of view.
--------------------------------------------------------------------------------
11/03/09
I made the I_Sleep call on video mode sets exclusive to fullscreen, since it
was bothering SoM in windowed mode.
--------------------------------------------------------------------------------
11/01/09
I finally got rid of that ridiculous extraneous "eternity" directory inside of
our SVN trunk, without managing to screw up the entire repo (I will note it did
cause the untimely death of my principal working copy, however :P )
Took the opportunity created by this change to create tags for all of Eternity's
releases that have occurred since the SVN repository was established.
--------------------------------------------------------------------------------
10/31/09
Using calloc on vertexes array loaded from the map allows me to get rid of some
useless assignments to zero in the dynaseg code when copying vertices from
linedefs into dynasegs.
I have wiped out all remaining traces of seg_t::angle, which is never used in
Cardboard, and a corresponding R_PointToAngle2 call in the dynaseg code which
was now a total waste of CPU cycles.
The skin viewer now has additional safety code when remapping sprites, due to a
bug Kate was able to trigger with an odd barrel skin which caused the skin
viewer to begin showing graphics from all over memory before crashing the game.
Woops!
--------------------------------------------------------------------------------
10/30/09
Kate seems to believe the crash has to do with the conversion of floats to
integers in R_AddLine. Initial testing with a bandaid patch indicates that the
crash is indeed stemming from here. Seems conversion of out-of-range floats to
integers has undefined results.
Once SoM came back, he figured out the problem definitively and worked out the
best way to fix it :) No more crashes!
I made a minor possible fix to the dynaseg system that will hopefully fix the
recently observed problem of automap rendering continuing beyond them.
Basically, the backsector of the dynamic seg should not be set to a sector
unless the dynamic seg's parent linedef is actually two-sided (it really
shouldn't be, but people have found ways to get away with 2S polyobjects in
some circumstances ;)
--------------------------------------------------------------------------------
10/29/09
3DMidTex movers have been broken for a very long time evidently, as I missed a
break statement in P_ChangeSector3D which caused it to fall through into the
default case and malfunction.
Eternity has developed a new extremely random and difficult-to-trigger crashing
error somewhere in the renderer in the neighborhood of R_RenderSegLoop. Time
to call up SoM ;)
--------------------------------------------------------------------------------
10/27/09
Inspired by Chocolate Doom, I added a -noframe command line parameter and a
matching "n" flag character for geom resolution-setting strings that allows
the creation of Eternity's main graphics window without a window border.
Note that frameless windows cannot be moved, at least not in Windows XP.
Example of use:
eternity -geom 1024x768wn
--------------------------------------------------------------------------------
10/25/09
I decided to finally bite the bullet and revamp portions of Eternity's low-level
input system so that we can support the smart Unicode mapping provided by SDL
which properly reflects international keyboards. This mapping must be enabled by
the end user.
The changes made to allow this also allowed me to instantly add support for
binding to the numeric keypad and the delete key, as the DOSisms that were
holding these features back have either been eliminated entirely or confined to
processing of events within the proper ASCII range when using the original
shiftxform keyboard mapping.
Pipes can also now be typed into the console, even when using the shiftxform
mapping. We may never know exactly what prompted the infamous "Oh my GOD
does Watcom suck!" comment in hu_stuff.c, but it's pretty irrelevant now,
seeing how we have never supported compilation on Watcom ;)
--------------------------------------------------------------------------------
10/24/09
Now that Eternity has sector flags, I removed my previously added wassecret
field from sector_t and changed it into a SIF_WASSECRET internal sector flag
value. No use having an entire boolean do the job of a single bit!
By user request, I have modified the behavior of the BOOM "ammo" cheat so that
it gives max ammo if the user is currently not at max, and removes the ammo
otherwise. The behavior in BOOM was to take away the ammo if you had any at all,
which was obviously somewhat confusing and less useful.
--------------------------------------------------------------------------------
10/23/09
Integration of the new argument evaluation system has been completed, along with
a change in the way that EDF keywords are stored and evaluated, and the rewrite
of ALL parameterized codepointers to support the new system (this was NOT a
trivial undertaking either...)
Individual codepointers will require testing to verify that I have not broken
anything, as I most certainly have, since it was 4:30 am by the time I finished
this.
--------------------------------------------------------------------------------
10/22/09
I have begun work on converting EDF to a runtime evaluation model for the
arguments to parameterized codepointers, which will be necessary to properly
support some of the advanced features of DECORATE states. This also simultaneously
solves Eternity's problem of having a limited number of arguments per state.
The argument evaluation engine provides evaluator functions for various data types
including int, fixed, double, thing types, states, EDF string objects, etc. Note
that there is not currently any support for expressions, but such support should
be simple to add once the Aeon scripting system is in place (note that such
expressions will then need to be valid ECMAScript, rather than a bespoke
expression language like the one currently used by ZDoom).
--------------------------------------------------------------------------------
10/16/09
I added an I_Sleep call to video mode setting to allow the screen to settle when
the engine changes into a fullscreen video mode.
Out-of-bounds palette changes in ST_Ticker are no longer performed, and
P_DamageMobj will no longer allow mobj_t damagecounts to become negative. These
together were the cause of the undefined behavior exhibited in vanilla Doom when
doing negative damage to the player via DeHackEd patches. This instead crashes
in Eternity and other modern ports without these modifications.
I also took time to add entryway's new emulation code for the undefined behavior
exhibited by donut linedef actions which have no valid model sector. Under
vanilla Doom they infact read the target movement height from the interrupt
vector as it is stored at memory offset 0. Why does DOS4GW not catch those
invalid accesses? I have never figured out why it sucked so badly as a DPMI
provider.
Of course if it had worked even half-way properly, DOOM would have almost
certainly missed its release date by a couple of months ;)
--------------------------------------------------------------------------------
10/12/09
EDF damagefactors are now implemented, using the new multi-valued property syntax.
This allows monsters to take different amounts of damage from different types of
attacks.
Added a new kind of routine to the metatable - MetaSetDouble. This routine will
add the field to the metatable if it doesn't exist already, but if it does, it
instead changes the value of the existing metadouble object to the new provided
value. This is very useful, and I will be adding matching routines for the other
basic data types as they are needed.
--------------------------------------------------------------------------------
10/06/09
Added a couple of methods to the qstring pseudoclass:
M_QStrReplace
M_QStrReplaceNotOf
These allow replacement of specific characters inside a qstring with other
characters. I have noticed this operation being carried out on some qstrings,
so I'll need to find those and replace them with calls to these nicer functions.
--------------------------------------------------------------------------------
10/03/09
I have added separating header comments to distinct sections in d_main.c to
outline the different regions of functionality. This module, along with several
others in the source code, has become too large and will probably be split up
in the future.
To my dismay, after essel pointed out some mysterious polyobject issues, I found
that I completely failed to define any values for the polyobject flags added
previously. This was causing massive malfunctions throughout the entire
polyobject system, not limited to the constant addition of more and more
dynasegs to the subsector dynaseg lists until the game was running at a grand
1 FPS or less. Woops!
--------------------------------------------------------------------------------
10/01/09
The IWAD picker was firing up when the -game parameter was in use, which
prevented the use of normal IWAD searchinig inside of the specified base/game
directory. Fixed by turning off the IWAD picker under this circumstance.
--------------------------------------------------------------------------------
09/29/09
I removed A_Look recursion detection entirely because it is too difficult to do
in a way that will retain compatibility (Lost Souls can theoretically recurse
through A_Look to any possible depth when colliding with something, although
due to the way the engine works it typically only recurses 1 or 2 times).
However, outside of playing old demos, EE will no longer allow Lost Souls to do
this. They will now call the function P_SetMobjStateNF, which does not call the
state's action routine, when resetting to their spawnstate upon a collision.
This prevents recursion through the non-reentrant P_Move codepath by SKULLFLY
objects. The soul will still call A_Look upon its next state transition, so at
most it might simply sit there for a couple of tics.
--------------------------------------------------------------------------------
09/28/09
After the rewrite of damagetype pain and death states to utilize the new
metastate maintenance code (which stores states into the mobjinfo metatable),
several functions were reduced to single lines of code. I eliminated these
unnecessary middle-man functions and replaced them with direct calls to the
metastate functions. Overall the code still didn't get smaller than it was
originally, but the dividends will pay off when metastate management is employed
for DECORATE state support.
--------------------------------------------------------------------------------
09/27/09
I had to make an accidental megacommit after the finding of yet another serious
issue:
1. I have added the ability to parse "multi-valued properties" to libConfuse.
These are properties in this form:
field [=] <value> [, <value>]*
The equal sign is optional, as per yesterday's extension, and the list of
values is managed by an independent cfg_opt_t array, which functions the
same way internally as a libConfuse section. CFG_MVPROP items are basically
identical to CFG_SEC items, except that they cannot have titles, may only
contain basic options (no sections or lists), and the items must be present
in the same order specified in the MVPROP's cfg_opt_t.
This matches syntax used by DECORATE for items such as damagefactors.
2. Started addition of meta damagefactors, but this has stalled out for the
time being while I worked on other problems.
3. I broke down and reassigned the ownership of metaobject keys to the
metaobjects themselves, due to persistant problems with the copying of keys
between objects when copying the metatable. My initial design was made under
the assumption that metaobject keys would largely be static strings, and that
dynamically allocated ones would be the exception. It is increasingly obvious
that dynamically allocated key strings are going to be the rule after all,
since so many of the features I want to use the metatable for rely on names
generated via concatenation of strings derived from EDF definitions which
may have arbitrary length.
4. I added another incredibly powerful C++-like mechanism to the metatype
system: a toString method. If not defined, the default metatype::toString
method prints out a hex dump representation of the object. Otherwise, the
toString method may be overridden by passing a function to MetaRegisterTypeEx
and it will be called for objects of that type.
Metaobjects without a registered metatype will not be printed; instead a
message is given that the object is not a registered metatype object.
This was added to support the new e_dumpmeta command, which allows me to look
at an mobjinfo's metatable in the console. This is a pretty critical thing to
have for debugging new metatable-reliant EDF features, which are going to be
popping up like weeds for a good while now.
5. Aaaand finally, as alluded to above, a new extremely serious issue abruptly
appeared while I was trying to work on the metatable. EE started crashing
during P_SetupLevel on Heretic E1M3 with approximately 10% reliability. The
crash initially only seemed to work in the release build, so I was forced to
try to reconstruct the call stack from the crash reporter output, and I found
something extremely disturbing:
The game was crashing in P_UpdateThinker.
That such reliable, "finished" code that hasn't changed for years would
suddenly manifest a segmentation violation was very troubling to me, so I
wasn't about to give up on trying to find the root cause of this. I finally
got the bug to occur in the debug build by idclev'ing between E1M2 and E1M3
repeatedly, and captured the crash in the debugger, in the exact same locale
it had crashed before - P_UpdateThinker as called for the addition of a
"strobe light" thinker for a particular sector.
For *some strange reason*, it turned out that this strobe light thinker had
the function P_MobjThinker in its thinker->function field. Why? Because it
turns out that none of the thinker classes were initialized before being
sent to P_UpdateThinker. This wasn't a problem before EE, but became such
because of the addition of the PrBoom fix to MBF's thinkerclass list
corruption bug, which has to test the value of thinker->function.
How this managed to survive so long is troubling, but even more troubling is
why the Windows debug heap was not doing its job. When memory is freed in a
Windows debug program, the system contractually guarantees you that the
memory will be overwritten with the value 0xDEADBEEF before being reused.
This did not happen, however.
Instead, an mobj_t's memory block freed during Z_FreeTags from the previous
level was handed back to me during the request for a strobe light thinker.
Because both structures have the thinker->function field at the exact same
offset, this allowed a "perfect storm" bug to occur that would otherwise be
impossible.
While I am glad that this enabled me to find the problem, which has been
fixed by changing all thinkers to use Z_Calloc instead of Z_Malloc, I am
not pleased in learning that I cannot count on the protections afforded by
use of the Windows debug heap.
--------------------------------------------------------------------------------
09/26/09
I have introduced a relaxation into the libConfuse grammar/state machine which
makes equals signs optional by simply proceeding to the "expect value" state
without reading a new token when the token just read is not an '='. This makes
EDF a bit more DECORATE-ish in that it now has less unnecessary yet mandatory
syntactic sugar.
--------------------------------------------------------------------------------
09/25/09
I defined some additional new flags4 bits:
* NODAMAGE prevents objects from taking damage from normal sources, although
they still behave as if they take the damage, including being thrust and
entering their painstate. I am told that in ZDoom this is completely
unconditional, but I have made it inapplicable to telefrag-level damage in EE
so that nuke, mdk, and telefragging can still destroy such objects.
* LOOKALLAROUND causes a monster to be able to see targets 360 degrees around it
in the same manner as AMBUSH monsters after their sectors' soundtargets have
been activated. Normal monsters only have a 180 degree field of vision.
--------------------------------------------------------------------------------
09/23/09
Added new thingtype properties respawntime and respawnchance, matching ZDoom, to
allow fine-tuned control of Nightmare (or ALWAYSRESPAWN) respawn behavior.
Respawn time is in tics, and is the minimum amount of time that a thing must
stay still in its death state before it can get a chance to respawn. Respawn
chance is a number from 0 to 255 (ideally, at least) representing a chance
similar to the pain chance. If a random number is less than respawnchance, then
the object will respawn provided that it can fit at the location at which it
wants to respawn. Using 0 would mean that it never respawns, whereas using 256
or anything higher would mean that it will always respawn as soon as it gets
the first chance. Defaults are 12*35 tics, and 4, respectively.
Modified P_RadiusAttack to make it reentrant outside of old demos by adding a
bombdata_t structure which is "pushed" and "popped" via the incrementing or
decrementing of a pointer over a fixed-size array of objects. The static limit
is introduced in order to eliminate the possibility of stack faults when
recursive explosions occur. The limit is currently 128, which is actually a lot
lower than what can be theoretically allowed, but which is already a very
lenient number considering the number of barrels this implies going off in a
single event.
While in p_map.c I took the long-neglected opportunity to fix blood effects for
crushers while the particle system is enabled, by getting rid of the code to
draw a particle effect altogether. The effects we have don't even come close to
matching the blood spray which is caused by crushers, so we'll just draw blood
sprites here unconditionally, for now at least.
--------------------------------------------------------------------------------
09/22/09
I refactored the mess that was P_HitFriend to de-Killoughify it (in other words,
to eliminate a single-expression function with tons of side effects in every
branch). I then added a check in this function that makes it return false if the
object in the way has the NOFRIENDDMG flag, since such objects won't take damage
if the caller continues to fire.
--------------------------------------------------------------------------------
09/21/09
Added new flags4 bits NORADIUSDMG and FORCERADIUSDMG (matching ZDoom) to allow
more control over explosions.
Corrected behavior of the NOFRIENDDMG flag so that objects can still cause
damage to themselves if they have this flag.
Also tweaked the behavior of the SUPERFRIEND flag so that superfriend objects
can *never* acquire friends as attack targets at all, since the checks in
A_Chase are evidently insufficient to prevent retaliation in some odd cases.
The BOSSIGNORE flag is now officially deprecated, in anticipation of addition of
fine-tunable infighting control via the mobjinfo metatable. Do not use this
flag, and if you have already used it in an in-progress mod, phase it out now.
--------------------------------------------------------------------------------
09/19/09
The crash report app will no longer be executed for stack faults, since the
stack is still trashed during execution of the exception routine. Somehow there
is enough space to execute writing of the log file, but not enough for the
CreateProcess routine, which apparently generates an exceptionally deep call
stack.
In case you wonder why EE would have a stack fault, I have been experimenting
with the P_RadiusAttack code after finding out that it is both non-reentrant,
and can be forced to recurse to an arbitrary depth simply by modifying barrels
to call A_Explode during their first death frame.
Also added a new A_GenRefire codepointer for Kate, which allows customization of
refiring behavior exhibited by monsters like the chaingunner and Spiderdemon.
--------------------------------------------------------------------------------
09/16/09
Defined a new GameModeInfo flag, GIF_SKILL5RESPAWN, which controls whether or
not a gamemode's fifth skill level has -respawn behavior enabled by default.
This flag is now given to all DOOM gamemodes, but not to Heretic, so that Black
Plague skill level no longer has respawning monsters, as expected. Do note that
at no point had I forgotten about this - it's been on my 3-mile-long TODO list
for an age and half ;)
Added another flag, GIF_SKILL5WARNING, which controls the presence of the
warning message given when starting a game on skill level 5. Only DOOM games
give this message, as apparently Raven felt their games were more fair (I
would beg to differ, however, given how much I suck at Black Plague in Heretic
despite owning much face in Smite-meister).
--------------------------------------------------------------------------------
09/15/09
I have removed the console-menu interlock and made some fixes that enable the
menus to operate over the console with absolute priority, after the scope of
problems caused by the interlock was revealed to me by some users. This stuff
has been a nagging problem since SMMU, so hopefully it is fixed now.
--------------------------------------------------------------------------------
09/13/09
Adjustments for Catoptromancy:
1. The typedef for ehash_t has to be placed before the ECompFunc_t typedef
because apparently GCC cannot understand that "struct foo *" as a function
pointer parameter type is not a declaration of the structure type but is
just an incomplete type. I don't *think* this is correct behavior out of
the compiler, but it wouldn't be the first time I was wrong either. C
seems to be getting harder to write every few months...
2. z_zone.h needs to be made visible through metaapi.h due to use of size_t.
3. The temporary TestDSParser function had to be changed from static to
extern because apparently you also cannot forward prototype a static
function inside a local scope block (this doesn't sound right either but
who am I to argue with a bunch of neckbeards).
After all that unnecessary annoyance was out of the way, I also found time to
work on new features:
The flags4 field is now available. After adding an R_TranslationNumForName
routine and cleaning up some EDF code to use it, I defined the first new
flag for this field, MF4_AUTOTRANSLATE. This flag will kick in if an object
using it is spawned in a non-DOOM gamemode. The thing will then have its
translation (if it does not have a valid one already, anyway) changed to the
gamemode's default palette-conversion translation (for Heretic, this is the
TR_DINH lump provided in eternity.wad). This flag has been given to the MBF
helper dog, and lets me avoid translating its sprites into the Heretic palette
by hand. This flag is only useful for sprites that are in the DOOM palette, and
thus may not find a lot of end-user application.
I added a double floating-point specialization for metaobjects to the metaapi
preemptively, and after observing ZDoom's code a little more closely (note that
I did NOT use its implementation as a reference while creating EE's metatable),
I made an important alteration to the specialization MetaGet* routines that
causes them to take a default value as a parameter. If the metatable does not
contain that field, then the default value is returned, instead of the previous
behavior of returning 0. metaerrno is still set to META_ERR_NOSUCHOBJECT when
this happens, however, in order to enable testing of this if and where it may
be needed.
--------------------------------------------------------------------------------
09/11/09
I have added support for the ZDoom 9303 Polyobject spawn spot, which enables
polyobjects that do damage simply by touching an object rather than by requiring
it to be stuck (ie, crushing damage).
--------------------------------------------------------------------------------
09/10/09
Turns out the oh-so-wonderfully-explicit documentation, if you can call it
documentation, of the DECORATE states format that I got from the ZDoom wiki
was incomplete in some very important ways.
Chiefly, it turns out that consecutive labels are not only allowed, but actually
important in some ways. So apparently I will now have to throw labels onto a
stack of sorts until I see a state, and then create metastates for that first
state for each label that I have put on the stack. A little bit more complicated
but certainly nothing I cannot handle. It just would have been nice to know that
this was possible before I assumed that it wasn't and ended up having to rewrite
a bunch of code and restring portions of my state table.
The other feature it so kindly skipped over mentioning was the namespace
operator, which is apparently only used for goto targeting of super-class
states where the child class has overridden the named state (as this is
unnecessary otherwise, since the child class has inherited all the metastates
from its parent type already). This required addition of another state to the
lexer, since the decision of whether a : is just a : or a :: must be deferred
until the next character has been seen.
I have slightly modified the way that metatable object copying is carried out
through the metatype system in order to reassign responsibility for the copying
of the key from the source object to the destination object to the copy method
of the metatype. This enables us to get around what would otherwise be an
intractible issue for metaobjects with dynamically allocated key strings. I
realized this problem while creating the metastate type earlier, but had to
give it a lot of thought.
--------------------------------------------------------------------------------
09/09/09
I added a case-sensitive specialization of the ehash_t in order to support a
new independent by-type hash within metatables, to fully support hashing and
lookups by either field name or type with full efficiency. This did necessitate
the creation of an independent metatable_t type, though, which now consists of
two ehash_t's - one fixed in size for types, and the other growable for field
names.
--------------------------------------------------------------------------------
09/08/09
Custom damagetype pain and death states have now been converted into
metaproperties, with the addition of code which will manage the addition and
removal of metastates (metaobject wrappers for state_t) to/from mobjinfo.
--------------------------------------------------------------------------------
09/07/09
An initial design of error handling for the DECORATE parser is in place. It
will currently always attempt to keep parsing, usually by discarding the
remainder of the current line and returning to an appropriate state such as
EXPECTLABEL or EXPECTKWORSTATE. We will see how practical this turns out once
real-life examples are being turned into state lists.
--------------------------------------------------------------------------------
09/06/09
The DECORATE state parser has now passed its first successful dry run after
some minor modifications.
--------------------------------------------------------------------------------
09/04/09
The DECORATE state parser's core loop is now functional, so I am prepared to
give it its first dry run. I'm confident that many problems will be revealed,
such is the nature of these things.
--------------------------------------------------------------------------------
09/01/09
I had to tweak my fix to the SWITCHES problem a bit - the lump has to actually
be named SWITCHES before I can decide whether or not it is a SWITCHES lump.
I sort of forgot that the hash chain could have other lumps on it as well ;)
I have had to disable particles in low detail mode until the root cause of their
bizarre undefined behaviors is rooted out. Any attempts to modify the way they
work cause the game to crash hard currently. I also have no idea where exactly
the particle system is getting its clipping information from. To me, it seems
that it is completely incorrect. An overhaul is apparently overdue.
Because they seem to be referenced after return-of-call, and because very
large arrays shouldn't be on the stack anyway, I have moved the clipping
arrays outside of the R_DrawSprite-style functions and made them into static
globals.
--------------------------------------------------------------------------------
08/30/09
I have made an experimental change to R_AddLine to fix a weird Cardboard
incompatibility with fiffy5.wad which causes a gigantic HOM in the first room.
Seems that DOOM treats at least SOME closed self-referencing lines as solid,
so I remove the second side of the line for rendering purposes, in order to
force it to be treated like a 1S line for rendering. This smacks of a hack,
but to implement it lower down might be messy and a lot more dangerous.
Console netcommand variables can now correctly be saved to the configuration
file and modified via the menu again.
I have changed the way that sky is handled in the engine by creating an internal
sector flags field, and creating a handler function that must be called to
safely change a sector's ceiling pic. Any time that pic is changed to or from
sky, the SIF_SKY flag will be set or cleared on the sector.
With this fix, I have finally rewritten and cleaned up the old Hexen lightning
engine so that it is savegame safe (lightning strikes are saved and restored
properly), and so that it affects ALL sky sectors, not just sectors that had
sky at the start of the level. I added some new console commands that can
enable lightning and force a lightning strike in order to easily test this code
without having to make a dummy EMAPINFO lump.
--------------------------------------------------------------------------------
08/29/09
The use_doom_config variable will also now remap the path to the keys.csc file,
which is considered part of the configuration. This was an oversight in the
earlier implementation of this feature - one which I noticed a while ago but put
off fixing ;)
I had to manually repair the VC++ 6.0 project file because I somehow dragged the
P_ folder into the M_ folder 9_9
A round of compatibility fixes have been implemented in order to fix some
interesting problems in the difficult-to-support spcial1b.wad by Jens
Hykkelbjerg, which was a special effects tutorial level. This required a
major modification to the BOOM SWITCHES code to cause it to walk down the wad
lump hash chain for "SWITCHES" lumps until it finds one that looks valid. This
problem was necessitated by spcial1b containing a patch named "SWITCHES".
This also required addition of a new compatibility variable, comp_special, to
allow toggling of the BOOM change to the behavior of one-time-use line specials
at runtime, instead of having it be solely contingent on demo_compatibility.
Turns out spcial1b actually exploits the use of a one-time-use gun action line
that is tagged to a non-existent sector to enable a blastable shield switch
over a switch that is deeper inside. With the BOOM bug fix active, it becomes
impossible to activate either switch.
With these tweaks, EE has restored 100% support for spcial1b.wad, and is to my
knowledge one of the only source ports, along with Choco Doom, that can claim
that currently. This thing is a true litmus test for compatibility, and it's
pretty amazing IMO that Cardboard is able to replicate all of the special
effects with no problems.
--------------------------------------------------------------------------------
08/26/09
Modified the DECORATE state parser to elimiate the difference between "text" and
"quoted string" token types, as either can be accepted in any location. This is
mainly important for allowing frame strings to contain [\] characters, as when
they do, they are required to be contained inside quotation marks.
--------------------------------------------------------------------------------
08/25/09
Added the ability to set the menu background flat and a selection box in which
flats can be picked.
Catoptromancy pointed out that the old problem of snd_card/mus_card being set
to 0 on SDL_mixer initialization failure was still acting up, particularly on
Linux, so I have eliminated that behavior entirely. nosfxparm or nomusicparm
are now set instead, to cause the game to act as if though -nosfx or -nomusic
have been used.
Menu and credit screen backgrounds, as well as the console backgroudn, can now
be overridden in missioninfo structures in order to give HacX appropriate
defaults.
Began a new massive round of GCC warning fixes. That thing is utterly impossible
to please.
--------------------------------------------------------------------------------
08/23/09
The EDF correct_height field will now be removed from certain classes of thing
types when those types are modified by DeHackEd patches, in order to improve
the default presentation of multiple wads, including HacX and Aliens TC, which
put previously grounded objects on the ceiling using the SPAWNCEILING flag.
It was always possible to avoid this by turning on comp_theights, but you
won't know it's a problem until you see things hanging in midair.
Added an overlooked and sorely needed M_QStrCaseCmp routine to clean up some
dodginess in the DECORATE state parser.
Added a "runiswalk" configuration and console variable by popular demand that
makes the shift key cause you to walk if autorun is enabled. Personally, I
can't stand this, since I often hold down shift unnecessarily out of years of
ingrained habit, but some people swear by it.
Menu help and error messages on the bottom will now be centered, because it
looks much better.
--------------------------------------------------------------------------------
08/22/09
A skeletal outline of the DECORATE state parser is now in place, with complete
state transition logic (that is to say, the functions will soon be capable of
doing a "dry run" on a string so that I can watch the state machine run through
the motions without anything important such as state generation happening).
ZONE_NATIVE and HAVE_SCHED_SETAFFINITY have been added to the Linux makefiles by
user request, so that users no longer have to enable these settings after
doing a SVN checkout. If either of them prove to be a problem, users will now
need to disable them, but I don't anticipate this.
I found an unnecessary conversion of int to float for purposes of doing a
comparison against the viewwidth in the core loop of R_DrawMaskedColumn which
was causing some major register motion in the generated assembly. Eliminating
this provided a small boost to sprite rendering, adding approximately 3 or 4
sprites before a noticeable dip in FPS occurs due to overdraw.
--------------------------------------------------------------------------------
08/20/09
Metatypes are now registered for provided metaobject specializations such as
metaint and metastring. I also did an audit of the source for "firsttime"
guard booleans that were missing their "static" keyword, which is the only
thing that makes them work in the first place, but is painfully easy to forget.
While testing HacX 1.2 I also discovered the consequences of an ancient MBF
bug relating to support of DEHACKED lumps in WAD files - \r characters must be
manually skipped over when reading in DEH-style string values, or the count of
characters is messed up completely, causing none of the multiline string matches
to work properly.
The reason this only matters for deh-in-wad support is, obviously, the fact that
WAD file i/o is actually binary, no matter how much the DWFILE system wants to
pretend that it is emulating a FILE pointer. Stripping \r's in the DWFILE
system is neither practical nor particularly safe, so it is better to have any
concerned parsers simply skip over that character.
--------------------------------------------------------------------------------
08/19/09
I have begun work on adding native support for the new HacX 1.2 stand-alone
IWAD as a gamemission under the DOOM II gamemode (similar to Final Doom).
--------------------------------------------------------------------------------
08/18/09
The metatype registry has been completed with addition of the supporting
E_HashTableIterator, MetaRegisterType(Ex), and MetaCopyTable routines. It works
like functional magic, and is almost scary ;) I have converted nukespecials into
a native property of mobjinfo, since I see no current value in making them a
metaproperty.
--------------------------------------------------------------------------------
08/17/09
Began integration of the metatable with mobjinfo. -fast speedsets are the first
code to be rewritten to use them. This rewrite however has revealed the need for
a metatable copying mechanism, in order to achieve the desired streamlining of
mobjinfo inheritance that was envisioned during my initial conception of the
metatable.
The only way I can see of enabling objects in the metatable to be copied
through a single routine is to establish a secondary cache of type information,
which I will call the metatype table. This will be a global hash (a metatable
itself, in fact), and types will be registered with it one time each, before
the first object of that type is added to any metatable. Only objects with a
registered metatype will be capable of being copied between tables in this
automated manner.
If designed properly, this should approach the power of copy constructors in
C++.
--------------------------------------------------------------------------------
08/09/09
I made a long-requested tweak to the A_MissileAttack parameterized codepointer
that enables the function to call P_SpawnMissileAngle when the thing does not
have a valid target. Missiles will simply be fired from the thing's direction
plus or minus the provided offset in this instance.
Another critical long-existing bug was uncovered by use of the native heap -
player mobj pointers are left dangling after Z_FreeTags, and this could cause
a random crash in the sound code.
--------------------------------------------------------------------------------
08/04/09
Building further on top of the generic hash, I have written the previously
planned metatable API which will enable EE to associate arbitrary information
with particular instances of structures such as mobjinfo without having to add
each object as a native structure field.
It is already possible to make use of the metatables by converting over certain
problematic, pre-existing EDF features such as damagetype states, nukespecs,
and fastspeeds to use this instead of their individual list or queue data
structures.
It will additionally almost certainly prove useful in the implementation of
UDMF, and in its subsequent exposure via the Aeon API.
A metatable is just an ehash_t which has been sent into MetaInit, initializing
it as a case-insensitive string hash which functions on metaobject_t structures.
A metaobject_t structure is an mdllistitem_t descendant which additionally
provides RTTI and an object pointer which may be either self-referential, or may
point to a different address in the case of metaobject_t's which are not
contained as the first item in a larger structure, or are not in fact contained
within the structure they link to at all.
Specializations of metaobject_t are provided for integers and strings as
metaint_t and metastring_t. Creating these objects through their respective
APIs lets the memory management of these otherwise intrinsic data types become
more automated, and the manipulation of such fields is easier.
Again, as with the underlying hash table class, specializations can be provided
to store any type of structure or data type in the metatable. But unlike the
hash by itself, the RTTI and higher-level genericity of the metatable allow
objects of multiple types to be in the same table at the same time.
RTTI for metaobjects is implemented with string comparison. Types are passed to
relevant functions using the METATYPE macro, which simply utilizes the C
preprocessor's "stringize" operator, #.
An example of basic metatable usage follows:
MetaInit(info->meta);
MetaAddInt(info->meta, "foo", 10);
...
int foo = MetaGetInt(info->meta, "foo");
This is just illustrative, and very basic. It is capable of a lot more, and
there are several different ways in which it can be used.
--------------------------------------------------------------------------------
08/02/09
I have completed an awesomely powerful generic hash data structure, previously
created for the Experimental Rudimentary Actionscripting system which was
abandoned in preference to forging ahead with Aeon based on ECMAScript.
This hash table is built on top of the genericity provided by mdllistitem_t,
giving it chains which are double-linked lists with **prev pointers, like the
majority of EE's currently innumerable private hash implementations that it
will eventually replace. Any type of object can be stored inside it via use of
callback functions registered with the hash table which retrieve the key,
object, and link pointers and return them as void *.
Specializations are provided for objects keyed to integers, unsigned integers,
and strings. Any other possible specialization imaginable is possible, but those
three are all that I can see us needing in the future.
The second Two-Week-Bug-style problem in the last two months has suddenly come
out of the woodwork. This time it turned out that mobjinfo and states have never
been initialized to zero after being allocated. Almost all structs in EE are
now allocated using calloc, but these were still using Z_Malloc directly.
This suddenly caused E_DisposeDamageTypes to access 0xBAADF00D, the value that
the Win32 heap fills freshly allocated memory with. I guess that until now the
values in this location were always coincidentally set to zero somehow. I really
REALLY hate this kind of stuff, and I hope that it will be the last such error
for quite some time.
I am glad, however, that this stuff is exploding all over me during development,
and not waiting to spring on our users after the next release. It is a pleasant
change of pace, I must say :P
This is at least in part thanks to our direct use of the Win32 heap, and more
importantly, our previous change to eliminate use of the SDLmain library, which
allows us to use the the Debug Multithreaded runtime library, which has a
heap checking system which puts the one implemented in z_zone to utter shame.
--------------------------------------------------------------------------------
08/01/09
64-bit portability work continues with elimination of almost all instances of
the "long" data type in preference to "int" or "int32_t" where appropriate.
Exceptions have had to be made in the Small AMX and in the implementation of
opendir for Visual C++. The former is not going to be portable to 64-bit, as
I have positively identified it writing function pointers into integers, and
the latter is meant to return a "long" in order to be POSIX-compliant. Let us
hope it doesn't matter ;)
Yet another round of GCC warning fixes have been implemented, this time thanks
to hobbs. According to CSonicGo, EE has now compiled for the first time under
GCC with no meaningful warnings (a few remain about unused functions, but these
are being generated by currently unfinished features such as ExtraData sectors,
EDF inventory, and the DECORATE states processor...).
--------------------------------------------------------------------------------
07/30/09 - 07/31/09
After severe problems with pstdint.h on GCC (GCC has non-ISO-compliant
redundant declarations of inttypes.h types in its sys/types.h file.....), I have
been forced to remove that file in preference for a Visual C++-specific stdint.h
and inttypes.h. On GCC, we will use the normal inttypes.h which is always
available now.
However, thanks to the morons in SDL, I must now globally disable VC++ warning
4142, "benign redefinition of type," which is caused by SDL_config_win32.h
unconditionally defining ISO integer types with the same types as those used by
inttypes.h.
Libraries have no business defining standard library types. SDL should be setup
to use proper intermediary types rather than defining the ISO names here, and
at the very least, it should be possible to disable these by defining a symbol
such as HAVE_STDINT_H. But no, that would be good design, and we can't have
that!
I have completed and committed the initial version of eecrashreport, which is
an MFC application that will be launched via the CreateProcess API when
Eternity crashes. This dialog app will read crashlog.txt and offer to copy and
paste it into a new browser tab or window pointing to the Eternity forum on
Doomworld.
Do note that simply opening such a dialog from Eternity itself is considered
off-limits. Opening dialogs from inside exception handlers is beyond bad
practice into the realm of stuff you should never EVER do. Other port authors
should take notice...
--------------------------------------------------------------------------------
07/28/09
More 64-bit portability changes, this time in the configuration file. I have
rewritten the system entirely to be type-safe, so that string pointers are no
longer forced into integer variables. Instead, defaults for various types, which
now include support for float and boolean rather than just ints and strings, are
all stored in separate variables within the default_t structure.
To keep this clean, items are now defined in the defaults array using macros, in
a system extremely similar (and in fact entirely redundant) to that used by
libConfuse.
--------------------------------------------------------------------------------
07/26/09
"byte" has been changed from a typedef of unsigned char to a typedef of uint8_t.
--------------------------------------------------------------------------------
07/24/09
Began a massive round of changes which will hopefully bring us all the way to
64-bit portability. I have begun by adding a portable implementation of stdint.h
so that we can transition to using size-safe integer types everywhere that this
is needed.
The first victims were the types fixed_t and angle_t, which have become typedefs
for int32_t and uint32_t respectively. The latter is especially critical,
because the Binary Angle Measure system relies on the behavior of 32-bit
integers, but was defined previously as "unsigned long." On x64, long has been
promoted to 64-bit, so this would immediately cause a number of severe game-
breaking problems.
The next victims were the Long64 / ULong64 custom typedefs, which were manually
configured to resolve to various system-dependent 64-bit integer types on our
supported platforms (__int64 or long long respectively). These two have been
eliminated entirely in preference for the standard int64_t.
--------------------------------------------------------------------------------
07/23/09
An incredibly improbably-undetected error which has survived in the code for
several years suddenly blew up Two-Week-Bug-style on me today when the skin
code started to access a NULL pointer. What should it be other than the fact
that the savegame code didn't initialize mo->info until after calling
P_SetSkin, which uses the exact same field? I need somebody to explain to me
how this never crashed until now, because it makes no sense 6_9
Finished integration of a proper structured exception handler for Windows, to
replace the totally useless silent SDL parachute. Crash information is now
dumped in to crashlog.txt in the application's directory using routines which
are supposed to be safe to call while in an exception handler.
--------------------------------------------------------------------------------
07/22/09
Cleared out some dead code, tweaked an initialization problem in txt_gui.c found
by yagisan's analysis in order to match the commit in Choco Doom, and finally,
we now clear the channel id in the low-level SDL digital sound channel structs
when a sound is stopped, so that the game engine will see an ID mismatch and
will not attempt to further update that sound channel. This is just a safety
measure, as I've seen nothing to really suggest that this actually causes a
problem.
I do however believe there is still a subtle problem somewhere in the sound
engine, as I swear I am getting occasional phantom cutoffs. We'll see what
happens.
--------------------------------------------------------------------------------
07/20/09
Analysis of the game while stress-testing the new native heap implementation
revealed extreme allocator noise coming from P_SetPSprite. Found out I need to
keep gunaction_t's on a freelist, so that they're not constantly being malloced
and freed, sometimes several times per tic.
yagisan sent me the output of a static code analyzer (like a special compiler,
but it traces through your code and looks for logic errors instead of compiling
it into a binary). This found a lot of iffy stuff which I have knocked out.
--------------------------------------------------------------------------------
07/18/09
After some misguided and unsuccessful attempts to stack a fast-fit system on
top of the zone heap, I have decided to bite the bullet and provide an optional
native heap implementation, which can be enabled by globally defining the
symbol ZONE_NATIVE. This #if's out z_zone.c and compiles z_native.c instead.
During normal play there is virtually no difference. The zone heap and Win32
heaps have similar overhead, believe it or not. But when the game is under
stress, such as during nuts.wad, the difference is unbelievable. The native heap
never dumps cachables unless a system out-of-memory event occurs, so there is no
constant unloading and reloading of resources, and there is no unnecessary
static limit on the amount of memory allocated with a single strategy.
The plain fact of the matter is, BOOM's zone block purging stategy is crap, and
its extreme-case behavior is pure O(N) garbage. It's a shame because the zone
system would otherwise be sustainable.
I may in the future do another attempt at a pure fast-fit heap, but as long as
the system heap offers superior behavior, I question the need for any futher
misadventure. I'd have to be pretty arrogant to believe that I could match the
know-how and manpower that have gone into the heaps used by operating systems
like Windows and Linux.
After attempting to modify one myself, I have nothing but the utmost respect for
the people who created and maintain that stuff. They must be geniuses.
After a lot of fighting with GCC, it was discovered, again by Russell, that EE's
global define values were NOT visible to modules in any directory except the
main source directory. This was due to joe not coding any way for those defines
to be propagated downward during the build process.
This means I am now required to duplicate and maintain the list of global
define symbols in every Makefile.am individually. What a bitch.
--------------------------------------------------------------------------------
07/17/09
After being frustrated with the obviously disproportionately bad performance on
nuts.wad under a 16MB heap, I have been experimenting with the zone system.
My initial idea was to code a load factor over which purgables would stop being
dumped and the block list would not be searched, but rather the game would go
immediately to using so-called "virtual" memory. This did offer a 2-3 FPS
increase, but only once the heap was so saturated with static allocations that
the speed of incoming allocations could completely overwhelm the savings
offered by dumping cachables.
If the rate of allocation could not exceed the rate of reclamation, then the
system would see-saw around the load factor, and performance would plummet as
usual. This heap stuff is exceptionally difficult.
As a result of playing with this stuff, I found not one but TWO places from
BOOM where an extra BLOCK_HEADER was not being subtracted from virtual_memory,
the variable tracking the amount of virtual allocations for instrumentation
purposes. This would cause the program to appear to have a gigantic memory
leak, when in fact it was just some bad math.
--------------------------------------------------------------------------------
07/15/09
v_mode has been eliminated entirely. i_videomode now stores the user's video
mode in the form of a geom string, and the menus have been recoded to present
this string, and the option to set its current value as the default.
To support this separate default setting required fixing a well-hidden bug in
the console whose behavior I had previously noticed but could not explain.
By default the console will not set the default value of a variable when
changing its value unless the source of the command doing the setting is
the console itself. However, the menu system has always set the command type to
c_menu, and yet it was still setting the defaults, which is the expected
behavior anyway.
Well, the reason this worked was a happy accident. In the console code, the
c_typed enumeration value was being compared against the wrong variable - cmdsrc
instead of cmdtype. cmdsrc is the player who issued the command, whereas cmdtype
represents the game system from which the command emanated. Because in a single
player game cmdsrc is always 0, this caused it to always match the enumeration
value c_typed - meaning that the entire console system only worked at all
because of an ACCIDENT!
Fixed by comparing against the proper value, but this in turn then required a
change to the menu system - a new menu widget type is supported which sets the
command type to c_menu, and default widget types set the command type to c_typed
instead so that they cause default values to be set.
Otherwise, no changes made via the menu system would stick.
Tweaked I_ParseGeom to handle termination of the lexer while in STATE_HEIGHT.
If parsing ended here, the height would be set to the default value and you
would end up trying to set a strange video mode such as 1024x480.
With SoM's completion of the automatic field-of-view adjustment, and my
completion of the geom-based video mode setting, widescreen support is now
finished. I have raised the rescap to its new value of 2560x1600, which we
expect to be sufficient until the next generation of ATi cards is released ;)
Turns out dynasegs need to attach to the sector of the subsector in which they
reside, rather than referring to the sector of the original linedef's
location. Not doing this apparently causes some odd side effects, but only in
certain situations.
--------------------------------------------------------------------------------
07/13/09
After getting my butt whipped in a conversation with myk on DWF I fixed the
behavior of the iwadhandle variable in EE's wad system so that it is only ever
set to the FIRST IWAD to pass through W_AddFile, and not to every subsequent
IWAD.
This keeps things like loading DOOM over DOOM II working the way you would
expect, rather than having DOOM II's maps mysteriously start to call themselves
"New Level" even though they exist in the IWAD. This was indeed a bug, inherited
from SMMU, but amplified by EE's further application of the feature.
--------------------------------------------------------------------------------
07/11/09
Empty-string filenames provided to the -file command will now be ignored, to
help abate problems with Doom Builder and other utilities which can sometimes
accidentally insert empty filenames into command lines when they invoke other
programs.
This is kind of a bug on their end, but it's also a design issue on ours, and so
I think correcting it is fine.
--------------------------------------------------------------------------------
07/10/09
Thanks to Russell Rice for helping me eliminate further instances of undefined
sprintf behavior in the console. I eliminated them in this case by employing
qstring_t, which also simultaneously makes this code vastly more secure and
robust. It's still a fuckin' mess though.
Removed support for comments in console scripts, because they conflicted with
the ability to use ; as a command separator and as a literal character inside
strings. I then additionally repaired the keybinding save code to surround ; in
quotation marks when it occurs as the name of a key. Due to these two problems
together, it was impossible to have any action bound to ; to survive from one
play session to the next.
Added graceful handling of errors to I_ParseGeom, in preparation for making it
Eternity's primary means of setting the screen resolution. Any error in parsing
will cause the routine to set the resolution info to a reasonable default, which
is 640x480w.
Added a new deathmatch flag on a whim: INSTAGIB! Try using this in single player
for an entirely new sort of challenge - the honest no-damage run. You could play
DOOM in a no-damage run previously, but you had to watch your health very
carefully. Sometimes it's easy to ignore minor damage in DOOM and not even
realize it happened. With instagib, if you take damage, you die. After initial
testing, I did remove instant death for NULL damage sources, however, since
this makes it impossible to cross nukage sectors or survive minor fall damage,
things that some maps require you to suffer in order for them to be playable.
SoM noted that the Boss Brain object had accidentally been given a correct
height field in EDF. EE's policy on adoption of the correct height data from the
abandoned BOOM OVER_UNDER implementation was to apply it only to decorative
objects - nothing shootable except barrels. As a result of this error, it was
possible for your aim to become distracted by the Boss Brain object on DOOM II
MAP30, and your rockets would strike the Baphomet in the face instead of going
into his brain.
Don't ask me how this went unnoticed for so long. I guess nobody has ever played
MAP30 with comp_overunder off :/
--------------------------------------------------------------------------------
07/07/09
The old legacy -vwidth / -vheight parameters needed to use MAX_SCREENWIDTH and
MAX_SCREENHEIGHT for bounds checking rather than using hardcoded constants.
--------------------------------------------------------------------------------
07/06/09
Added long-overdue save and restore code for acsthinker_t. As a result I had to
change the location of the call to ACS_RunDeferredScripts, because deferred
scripts cannot run from G_DoLoadLevel when loading a savegame. This needs a
serious fix when the level transfer crap is rewritten for hub support later.
I changed the way that video.screens[4] is allocated. It is now a full-sized
screen buffer allocated with Z_SysAlloc, even though only a status-bar-sized
buffer is actually required. This is just the simplest and safest way to ensure
that the buffer is the appropriate size. If we eliminate buffering of the status
bar later, screens[4] can be eliminated entirely.
This change helps open the way for the impending change to Eternity's resolution
system that will enable widescreen support.
Rescap raising has been undertaken: the new cap for testing purposes is
1280x800, which is the fullscreen resolution of my laptop.
--------------------------------------------------------------------------------
07/05/09
Added save and restore code for floorwaggle_t thinkers.
--------------------------------------------------------------------------------
07/03/09 - 07/04/09
Continuing the effort to move more data into GameModeInfo, I have defined a
GIF_HASEXITSOUNDS flag, and broke up the quitsounds array and mapped it through
GameModeInfo.
Added a GIF_WOLFHACK flag, which indicates that the gamemode needs the DOOM II
hack to check for the presence of secret levels. This is technically silly, but
I am being rigorous about not branching on the value of GameModeInfo->id
wherever it is practical to avoid it.
Level exit rules, previously implemented as hard coded functions called through
an array indexed by GameModeInfo->type, are now tablified into arrays of rules
that function in a cascading manner. Each rule is checked until a match is
found, and the last rule in the table is always a catch-all which determines
the default exit destination behavior for that gamemode. The arrays are, of
course, attached to GameModeInfo.
Added a GIF_SETENDGAME flag and teleEndGameFinale field which control the
behavior of the Teleport_EndGame special through GameModeInfo.
Added a GIF_CLASSICMENUS flag which indicates support for the classic menu
emulation feature. There is no current plan to support emulated menus for any
gamemodes except DOOM-based ones, so this flag is considered part of the flags
that define the "DOOM" game type. Fixed a bug as a result of testing this that
resulted in the value of mn_classic_menus being ignored by the menu system
until its value had changed at least once during runtime.
Tablified finale defaults in the same manner as level exit rules, applying the
same design of cascading rule sets.
Fixed another bug: don't try to go to a non-existant E4M2 when exiting E4M1 in
Heretic registered version. Dunno what was up with this, it was probably just a
typo or an off-by-one error caused by the annoying fact that level numbers sent
into the finale code are off-by-one (ROMERO!!!!!!!! >_<)
Music selection routines are now mapped through GameModeInfo, though due to the
complexity of music selection, these are not tablified but instead handled with
a function pointer. The music selection methods are completely different for
each gamemode, so trying to handle them with rulesets would be overly complex.
The particle blood effect has yet again been significantly improved. The
particles are smaller, do not accelerate on the x-y plane, spawn higher up,
fall faster, are more voluminous, can be given greater positive z momentum,
live longer, and most importantly, can use any color between the two normal
blood colors if those colors are separated in the palette by fewer than 16
indices.
--------------------------------------------------------------------------------
07/02/09
I fixed the cast call caused in DOOM 2 by a call to Teleport_EndGame so that
the cast call actually occurs ;)
If you summon boss brain spots via the console, the list of boss brain spots
will now be reinitialized, so that any already active cube spitters will begin
spitting cubes to that spot. Previously, any such spots added would simply be
ignored. I'll admit I mainly did this just so that I could see the Seven Portals
fill up with 5000 DOOM monsters :) What's the point of writing this program if
I'm not allowed to have fun? :P
After worrying some more about the fact that all the special linedefs in Hexen
seemed to give off switch sounds, even when this was totally inappropriate, I
decided to find out what the deal was.
Turns out this was actually a bug in Eternity in general, and a result of the
change to R_TextureNumForName which allows the game to keep running in event of
missing textures. Under the BOOM switches system, if a non-existant texture is
resolved in the table, then any texture will start to match the entry for that
switch, and thus a switch sound will be given off inappropriately.
This was fixed by ignoring texture number 0 in the switch finding routine.
I have begun another massive effort to remap static data in the executable
through the GameModeInfo structure, and to eliminate as many uses of the
gamemode and gamemission id's to branch or select behaviors directly.
This starts with the breakup of the demostates table in d_main.c. Lee
previously assembled this into one table, but with the way we do things now, it
makes more sense to have one self-contained table per gamemode, and to select
them by accessing GameModeInfo->demostates. I think it's actually a lot cleaner
now, since the tables no longer have to be padded out with NULL pointers so that
they're all the same length.
--------------------------------------------------------------------------------
07/01/09
I implemented script wait notification in the ACS interpreter, so that scripts
waiting on other scripts can resume execution. Yet more Hexen areas have come
to life.
I also implemented a "puke" console command (in reference to the Hexen cheat
code of the same name) which can run ACS levelscripts. This has proven useful,
since missing linedef specials keep some of the more interesting scripts in
Hexen from executing currently.
Thing_SpawnNoFog and Teleport_EndGame are now implemented.
--------------------------------------------------------------------------------
06/30/09
I have added a system for specification of gamemode-dependent default overrides,
which alter the values of defaults stored in the configuration file system.
This enables Heretic to specify its proper default values for, amongst other
things, automap and HUD widget colors.
OPEN scripts in the ACS interpreter were being executed too early, leading to
undefined behavior. Repaired by deferring the call until initialization of the
virtual machine is complete.
Floor_Waggle (Hexen line type 138, ED type 397) is now implemented.
Added Thing_Spawn for ACS. Zombiemen now warp in all over Hexen maps.
--------------------------------------------------------------------------------
06/29/09
I finally got around to fixing one of my longest-standing pet peeves left over
from SMMU, and that's the fact that the automap timer was attached to a
redundant variable named levelTime, which is not saved in savegames, instead of
using the normal leveltime variable that has always existed and is savegame and
demo safe.
levelTime no longer exists at all, as the automap was the only thing using it.
--------------------------------------------------------------------------------
06/27/09
I cleaned up a hack in the IWAD picker subsystem which was written to avoid a
double-free due to two of the IWADs using the same titlescreen resource by
creating a W_FreeDirectoryLumps routine in w_wad.c. This routine goes down the
wad directory and frees all loaded lumps, in preparation for the destruction of
a private directory.
--------------------------------------------------------------------------------
06/23/09
While looking over the zone heap I noticed that if a malloc or free occurs on
a gametic ending with the LSB zero, the memory ends up getting memset to 0. This
is exceptionally harmful for catching errors, as zero is the general default
value for intialized pointers as well as most other variables and structures.
I modified this by OR'ing 1 with gametic & 0xff, so that at least one bit is
always set. It's still not fool-proof of course, but it's an improvement.
--------------------------------------------------------------------------------
06/21/09
I have added a "parent" field to mobjinfo_t which will permanently remember
EDF thingtype inheritance, which will be necessary to support the virtual
behavior of the A_Jump pointer defined in DECORATE. EE has never had any such
concept of runtime inheritance behaviors, but with this addition they will now
be possible.
--------------------------------------------------------------------------------
06/19/09
It are my birfday!
I have defined a formal grammar for parsing DECORATE states, and have done
additional work on the parser for them based on it. I have also modified EDF
heredocs once again to require matching opening and closing delimiters (either
@" "@ or @' '@ respectively). This was originally intended, and was to be
documented as a requirement, but now it is a formal fact in the code as well.
This enables heredocs to ignore only the specific type of delimiter with which
they are started.
--------------------------------------------------------------------------------
06/18/09
states is now finally declared as an array of state_t *, so this opens the door
for making states fully dynamic and allowing them to be declared outside of the
root EDF, as most other EDF entities now function. This leaves thingtypes and
sprites as the remaining critical definitions which are still runtime-static.
--------------------------------------------------------------------------------
06/17/09
Work continues on the DECORATE state parser.
--------------------------------------------------------------------------------
06/16/09
I added a call to the I_Sleep routine inside the dedicated screenwipe loop used
when wipewait is set to "always" or "demos". This should make the screenwipe
significantly more friendly toward fraps. As it was, fraps' recording would lag
horribly during any wipe, regardless of the routine being used (though the melt
wipe was by far the worst, as it's already a CPU muncher on its own).
I have begun tenative work on DECORATE state parsing, laying out an enumeration
based on my initial draw-up of the FSA required to parse them. This stuff is
very likely to change as I get deeper into it, though.
--------------------------------------------------------------------------------
06/15/09
I have eliminated all uses of pointer math on the states array in further
preparation for DECORATE states. All such statements like state - states,
states + foo, etc. now either access states directly or use the new state->index
field as is appropriate.
Added static keywords to all private automap routines. Dave Taylor never used
that keyword apparently ;)
--------------------------------------------------------------------------------
06/13/09
The Wu line drawer has been completed, with some modifications demonstrated to
me by SoM - if sine and cosine lookups are used on the amount of error in the
pixel and its inverse, for some interesting reason which probably has a deep
geometric explanation that escapes my grasp, the antialiasing effect becomes
significantly smoother and somewhat brighter than the normal linear treatment
of these numbers used in the stock Wu algorithm.
I have titled this modification the Wu-McGranahan line drawing algorithm, since
I cannot find any literature or hits referring to it, though it seems certain
somebody else would have thought of this before.
TODO is making Wu antialiasing optional. The Bresenham line drawer is still
present, since it must be used for special cases in Wu line drawing (horizontal,
vertical, and perfectly diagonal all collapse to Bresenham), so it should be
possible to use it directly, for those ultrapurists out there who object to
having a better looking automap...
--------------------------------------------------------------------------------
06/12/09
There was a tornado warning, so I threw in a quick commit just in case my house
ended up somewhere in Oz. I have started work on a Wu line drawing routine for
the automap which uses the ZDoom-style translucency table rather than the
traditional method of precalculated color ramps, which wouldn't work efficiently
for EE since we allow customization of automap colors.
I have also added an "index" field to state_t, in preparation for support of
DECORATE state syntax within EDF heredocs. This will require turning states into
an array of state_t pointers rather than an array of state_t objects as it
currently is, and when that modification is made, it will be necessary to get
the index of a state in states[] using state->index, rather than the current
hackish method of doing state - states. state_t::index is filled in sequentially
by E_ProcessStates.
--------------------------------------------------------------------------------
06/10/09
The automap boundary calculation is now sensitive to portal overlays, and thus
the edges of the map will be consistent with the map when overlays are turned
on. Previously, the normal map boundaries were always in effect, and this often
lead to not being able to scroll the map far enough to even see your own
location.
--------------------------------------------------------------------------------
06/09/09
Although it is probably largely pointless now, I have repaired a long-standing
issue with the BOOM (or is it MBF?) tranmap.dat caching code which creates the
tranmap.dat file in EE's base/game folders. Lee was accidentally saving the
8-bit 256-color palette, which is 768 bytes long, into a 256-byte buffer - a
clear and simple logic error. This meant that if a palette was identical to the
palette used to calculate the tranmap within the first 256 bytes (or the first
third of the palette, in other words), it would fail to detect any change and
would use the cached tranmap with an incorrect palette, resulting in weird and
ugly colors showing up on translucent sprites.
Thanks to esselfortium for catching this in his vapor project, or it probably
would have gone unnoticed forever.
I took the opportunity to add versioning information to the header of the cached
tranmap, so that if we ever need to change its format again, EE will know to
discard files in earlier formats rather than trying to use them. This was
critical for making sure that the next release doesn't cause a bevy of
complaints about supposedly broken translucency.
After a complaint was registered on the Doomworld forums and an email was sent
to me personally by the same person about the little-known fact that BOOM
generalized walk-over crusher line types are no-ops, I repaired them for
demo_version >= 335. It is evident enough that the lines are intended to work
if they have been placed in a map.
Jim Flynn must have left the case for them out of the switch in
P_CrossSpecialLine, and this is not very surprising since they were apparently
the very last generalized line category implemented by the BOOM team. God rest
your soul, Jim.
--------------------------------------------------------------------------------
06/08/09
I have finally gotten around to my long-planned rewrite of the automap to use
floating-point arithmetic. This brings it up to par with Cardboard for numeric
stability, so that we can now map overly large levels or overly long linedefs
with the same precision and unerring grace which we have previously had only
when rendering them. Plus on modern architectures, floating point is just as
fast and in some cases (believe it or not) faster than fixed point.
In the process of testing this rewrite, I discovered a hitherto little-known
vanilla DOOM bug - automap grid lines are lost near the left and bottom edges of
the map due to some faulty math - a minimum amount (MAPBLOCKUNITS) is added in
which causes the drawing loop to start beyond the position of a line that is
closer to the edge of the screen than that amount. The proper thing to do is to
start at the closest line to the edge of the screen by subtracting the
remainder, even if that line is offscreen as a result - an offscreen line will
be efficiently rejected by the Cohen-Sutherland clipping algorithm and won't be
drawn. At most one line can be offscreen on a given edge with this method.
--------------------------------------------------------------------------------
06/07/09
Another round of GCC warning fixes.
Ever since the thread-safety rewrite of the SDL digital sound update loop in
3.35.92, I had been hearing sample loss, as if the update loop were being barred
from modifying sounds for an unacceptable amount of time. After thinking about
it for a while, I realized that this must have been coming from the positional
sound update code called via the main loop, which modifies the volume lookup
ramps for each channel to keep them in sync with changes in the player's
relative position.
I realized that a critical section is not needed around this code, since
changing the volume ramp pointers should not in fact affect the sound update
loop in any serious way - at worst, the two stereo channels' volume ramps might
become out of sync with each other for an unnoticably short amount of time.
Removing the semaphore acquisition around the volume ramp modfication code
alleviated all of the sample loss and made the sound engine run smoothly once
again.
--------------------------------------------------------------------------------
06/04/09
Modified EDF heredocs so that they do not require a linebreak immediately before
the terminating sequence. I don't feel this is a good or necessary restriction.
After having followed a similar course with my experimental but ultimately
abandoned "ERA" parser, I rewrote the libConfuse lexer to use a state table.
This reduces the core lexer loop to 3 lines of code, whereas it was previously
several hundred. Each state is now handled on its own, and everything is much
cleaner as a result.
SoM has completed the core integration of slopes.
--------------------------------------------------------------------------------
06/02/09
I completed two new gun codepointers for esselfortium, EjectCasing and
CasingThrust. These are fairly standard object spawn and thrust pointers,
except that they behave in a way that attempts to keep them consistent with the
position of the player's gun psprite on screen, and their parameters allow
extremely fine-tuned adjustment in 32nds of a unit.
These pointers look great in first-person perspective in a single player game,
but they look pretty ridiculous in multiplayer or with the chasecam active.
This is what essel wanted though, so who am I to complain? :P
Dynasegs got another smaller optimization. I do not need to calculate the
sidedness of the polyobject origin point *unless* both of the vertices of the
seg are within epsilon distance of the partition line, since the calculated
value is otherwise unused. This was a silly thing just caused by wanting to get
the code over and done with.
While wrapping up essel's codepointers I also noticed that my previous
modification to allow copying of arguments into the player mobj state from the
player gun state during P_SetPSprite calls was unsafe if P_SetPSprite became
reentrant. In order to safely operate in this manner, I must store the states
and their original argument values on a stack, pushing one gunaction_t struct
for every call, and popping one off the stack afterward. This makes it properly
recursive.
--------------------------------------------------------------------------------
05/30/09
I added the "smart" background drawer routine which I had been thinking about
implementing for a very long time. V_DrawFSBackground will, when given the
pointer to a cached lump and the lump's size, determine if the lump should be
drawn as a tiled 64x64 flat, a 320x200 linear graphic, or a screen patch.
Patch vs. linear determination is not "smart" here - the linear graphic must be
exactly 64000 bytes, which is not a valid size for 320x200 fullscreen patch
graphics (they must be larger than that, or else they contain erroneous
transparent areas and are not really fullscreen graphics).
I rewrote several functions to call this new routine instead of doing their own
private determination of resource types. This added support for some of the
formats where it did not exist before, as well as cutting down on duplicated
code.
--------------------------------------------------------------------------------
05/16/09
Line IDs and line tags are now effectively the same thing as far as DOOM-format
maps are concerned in Eternity now, in order to bring us into compliance with
UDMF semantics and to fix what could have actually become quite an awful mess.
It is good, although humbling, to admit when you've made a design error.
--------------------------------------------------------------------------------
05/13/09
I applied yet another round of GCC warning fixes today.
I also found a massive optimization opportunity in R_IntersectPoint in the
dynaseg code. The expressions I used to calculate the (a2,b2,c2) values for the
node line at the beginning of the function turned out to be, under some minor
mathematical analysis, exactly equivalent to the negation of the node line's
A,B,C general line equation coefficients.
So instead of calucating these values again, I now just read them out of the
node and negate them. Should provide a good speed boost to dynaseg splitting,
since R_IntersectPoint is called every time a seg crosses a node line.
I also added a seg length calculation to dynaseg attachment, to satisfy SoM's
new Cardboard optimization that caches seg lengths instead of computing them
every frame.
--------------------------------------------------------------------------------
05/12/09
Randomized sounds are now supported via EDF sound definitions. A new random
block inside the sound serves to alias that sound definition to any one of the
sounds listed by name within the block. Random and alias resolution are
performed simultaneously, so that random sounds can both refer to sound aliases,
and sounds referred to by random sound lists can both be random sounds, or
aliases of normal sounds, or aliases of random sounds, etc. Yeah, it's kind of
complicated. There's rarely any need to actually mix the two features, though.
--------------------------------------------------------------------------------
05/06/09
I changed the behavior of the Heretic deathmatch intermission somewhat from how
it worked in vanilla because it seemed backward to me.
In vanilla Heretic, if there was a frag leader, the "wpnup" laughing sound would
play and the player(s)' frag count(s) would then flash. However, at the top, all
the players' pics except your own would then fade out to indicate who you were.
I have changed this so that your own row of frag counts is instead indicated by
a pulsing translucent yellow color block beneath the numbers, and when there is
one or more "slaughterboy," the players who were defeated by him or them will
have their pictures fade out to indicate their loserly status. To me this makes
more sense, and looks awesome.
--------------------------------------------------------------------------------
04/26/09
Finished the deathmatch intermission, and added a definition for the Heretic
wpnup sound effect.
--------------------------------------------------------------------------------
04/25/09
Completed the cooperative-mode intermission statistics for Heretic, and started
work on the deathmatch-mode frags table.
--------------------------------------------------------------------------------
04/19/09
I discovered that a subtle problem still existed with my new approach to
button_t sound-playing for switches which could result in the switch sounds
jumping to global position (0,0) - this was due to the fact that button_t's
would be immediately cleared, even while the sound they just started was still
playing.
To fix this, I went ahead and implemented my previous idea of linedef sound
origins, which consist of a degenmobj_t which is positioned at the center point
of every linedef at startup (and yes, polyobjects drag this around too, keeping
it consistent with the linedef's location).
This may open the door in the future for some new linedef-based features, an
area of DOOM-editing which I do not think has been exploited to even a fraction
of its potential by any port so far.
--------------------------------------------------------------------------------
04/17/09
Esselfortium found that Eternity could crash when less than 17 members were a
part of the cast call sequence. A small logical error lead to some very
selective trashing of memory near the castinfo array. This was due to use of
OLDCASTMAX without checking if numcastinfo was less than that value when
populating cast members with the default names defined via BEX strings.
Also, the screen clear routine in the IWAD picker module was always resolving
RGB (0,0,0) to a surface-format color value on every call because I forgot the
keyword "static" before the firsttime variable's declaration. I fixed that up.
--------------------------------------------------------------------------------
03/31/09
While trying to play an interesting PrBoom 2.02 demo posted on Doomworld where
a Pain Elemental is fooled into killing a Boss Brain whilst trying to attack a
Revenant while the player is dead, ending the level, I discovered that a simple
logic error in MBF's comp vector handling prevented such demos from synching
artificially. Basically, if the demo was recorded with "compatibility," the old
BOOM setting, disabled, then MBF's logic would set the entire comp vector to
"false," including any comp vars that didn't exist as of 2.02. This was just
plain wrong.
The proper action is to set all the comp vector values that were included as
part of "compatibility" originally to the state of that variable, and to set any
newer ones to true unconditionally. Eternity should have excellent PrBoom 2.02
demo compatibility so long as the demos in question are not for maps that use
certain BOOM features - friction, mainly.
================================================================================
EOF
================================================================================