===============================================================================
This log contains a summary of changes performed in the edf3-branch to
add fully dynamic EDF, inventory and weapons support, and other miscellaneous
features. edf3-branch spans SVN revisions from 1754 to 1871.
===============================================================================
-------------------------------------------------------------------------------
02/04/12
Merged with trunk to pull in Ladna's new keybindings system, which he brought
in early from C/S branch and modified. I then went back to trunk, made a few
modifications of my own, and merged those in as well, along with some conflict
resolutions.
Due to the amount of feature creep that has occurred on edf3-branch, and the
fact that some of the fundamental APIs of the engine have changed on this
branch in a way that creates serious burdens for merging of changes to trunk,
I am going to declare this branch closed as of r1871.
Finishing of inventory support will start immediately on a new branch, which
I will attempt to keep more focused to task (haha, yeah right :P )
-------------------------------------------------------------------------------
01/31/12
qstring has been modified to have a proper copy constructor so that it can
work with the new MetaAdapter<T> template class. This class inherits from
MetaObject in a way that allows provision of metatable storage to objects
which are *not* MetaObject descendants, and in particular to those that cannot
be made MetaObject descendants without violating Eternity's ban on multiple
inheritance. (An object requiring multiple inheritance of ZoneObject
descendants would break the entire MetaObject heirarchy and force it to use
dynamic_cast everywhere! This is not acceptable.)
MetaQString inherits from MetaAdapter<qstring> and provides a proper class
name as well as a toString() implementation that invokes qstring::constPtr().
The MetaAPI modules, which now number one C++ module and 3 headers, have been
moved into their own folder in the Visual C++ 2008 project.
-------------------------------------------------------------------------------
01/30/12
Modifications were pulled in from trunk to avoid use of non-standard strnlen
in m_strcasestr.cpp, and a requested fix to ExtraData attachment to polyobject
line specials 348 and 349 specified directly within maps.
Sector colormap values should additionally not be overwritten by ExtraData
sector definitions when there was no value specified, in order to agree with
the documented and intended behavior that "@default" implies that any global
or sky map for the level as settable via MapInfo will apply as usual. Instead
what was happening is that such sectors would get reset to the default COLORMAP
lump on account of the mapsectorext_t fields being 0.
-------------------------------------------------------------------------------
01/29/12
ExtraData thing definitions may now specify an EDF thingtype mnemonic in
place of a doomednum without using the "thing:" prefix, as this never should
have been required except that I didn't know how to properly work the char **
of strtol back when I first started work on ExtraData ;) Damn, that was a long
time ago :P
-------------------------------------------------------------------------------
01/28/12
The new hashCode and hashCodeCase methods of qstring could not use checkBuffer,
since it is not a this-const method (it modifies qstring::buffer and
qstring::size).
I went ahead and added static versions of HashCode() and HashCodeCase() to
qstring, so that they can be used on other strings wherever qstrings need to
be hashed without requiring the d_dehtbl.h header file to be included and
effectively exposing the implementation details of qstring's hash code
calculation implementation.
I renamed the lstrip and rstrip methods of qstring for consistency with EE
style guidelines in regard to method names (initial lower camel case for
most ordinary methods, initial capital camel case for statics).
I found a bug in qstring::swapWith which came into being when the code was
transitioned to C++ - use of a temporary qstring instance in this function
would result in one of the two persistent qstring objects' buffers getting
deleted when the function returned. This didn't happen in C since it doesn't
have destructors, obviously.
-------------------------------------------------------------------------------
01/27/12
Added adapter classes EQStrHashKey and ENCQStrHashKey, for using qstring as
the storage type for EHashTable keys. This should be very useful in the future
for avoiding reallocating C string logic and the need to use E_ReplaceString
everywhere.
To test it out, I converted the MUSINFO hash table to use qstrings as keys.
Works flawlessly.
-------------------------------------------------------------------------------
01/25/12
E*HashKey classes all now define a param_type typedef in addition to their
basic_type typedef. This allows the type of parameter passed into functions
such as EHashTable<>::objectForKey to differ from the type used as internal
storage. This allows use of non-POD types for internal storage without
resulting in a requirement for EHashTable to receive them passed by value,
with its ominous invocation of copy constructors.
qstring::hashCode and qstring::hashCodeCase have been added as wrappers around
the global D_HashTableKey functions.
The dead code for the old SMMU translations list has been removed. ptoing also
submitted updated versions of his tomato, blue, and gold player translations
for DOOM, and those have been integrated now. I think they are much improved,
especially blue which I myself use.
Did I ever mention that blue is my favorite color? :P
Numerous modifications by Ladna to trunk have been merged in. Ladna will be
working on trunk a lot more in the future, or at least that's his plan. I
hope to see many of the things he has planned come to full fruition.
-------------------------------------------------------------------------------
01/23/12
I was again without an Internet connection for two days, so I have a huge
amount of work to report on in this revision:
MetaState has been exported from the middle of e_things.cpp into its own
e_metastate.h header. This allows me to reference it elsewhere.
The DOOM II finale will now draw the player using the consoleplayer's
translation setting. I think it used to do this a long time ago in SMMU, and
I removed and/or broke it somehow. It *has* always displayed the player's skin
properly, at least.
The menu responder now eats all events while the menus or a menu widget are
up, as doing otherwise leads to some very poor behaviors, *especially* during
netgames!
Unfortunately this has made it impossible to take screenshots of the menus.
I need to think of a fix to this problem ASAP.
Additional refactoring to VerticalDoorThinker has enabled me to entirely
eliminate numerous redundant door type enumeration values introduced by BOOM
team during their implementation of generalized line specials. Blazing doors
are mostly now treated the same as regular doors as well, with the internal
"turbo" field of VerticalDoorThinker keeping track of whether a door should
act like it is turbo or not.
Added a V_DrawPatchTranslatedLit function so I could...
Add support to the skin viewer for viewing the player's sprite at various
light levels. Just press the up or down key and the light level will raise or
lower respectively. Fullbright frames are properly respected, even.
In addition, I also added support to the skin viewer for any metatable death
or extreme death states in the player class. Pressing "m" will cycle through
these in metatable order (which is arbitrary due to the list of them being
drawn from a hash table). So for example you can now see Corvus's fiery death
sequence in the menu skin viewer.
As a part of the previous change, I modified the handling of sounds inside the
skin viewer's state transition engine so that they are emulated based on the
action in the state_t - this means that A_PlaySoundEx, A_Scream, A_XScream,
and A_PlayerScream are all accurately emulated and the sounds played to
represent them (not always strictly in the same way the pointer selects sounds)
will occur on the same frame they would when killing the player during
gameplay.
Internal copyright dates have been bumped to 2012, and I also remembered to
update the window caption string to be consistent with the current dev version
number.
Another translation bug fix was pointed out by ptoing - since my change the
other day to fix SMMU's glitch of thinking it had one too many translations,
the very last translation that EE *does* have became unselectable via the
console or menus due to additionally incorrect treatment in the console
variable's handler function (it "forgets" that 0 means no translation, and
that the total range of values is thus one greater than the number of
translation table entries).
I finally discovered the cause of monsters sticking to locked door linedef
types 32-34, a bug stemming all the way back to vanilla Doom: the game
deliberately tells monsters in P_UseSpecialLine that they are capable of
opening locked doors.
The subsequent call to EV_VerticalDoor denies the attempt if the opener is not
a player. However, P_UseSpecialLine is braindead and does not return the
result of any of the EV_ functions it calls (EV_VerticalDoor notably DOES
return 0 if rejecting a monster for locked doors!), so the monster believes
it has opened the door successfully.
This leads to the exact same kind of stickiness that door tracks previously had
in vanilla DOOM, which has long been fixed with a comp var. Upon discussing
this in #doom-tech, it turns out RjY also discovered this a few years ago and
had planned to implement a fix to it in "prboom 7" complevel. Unfortnately it
never came to fruition.
It is special cases like these that are beginning to make me feel like a
completely unified linedef specials engine will never be possible for Eternity.
The number of compatibility quirks requiring precise vanilla behavior in order
to avoid desyncing demos or breaking esoteric maps is enormous, and it is far
too simple to break some of these things by making seemingly innocuous changes
to the logic.
Instead I am starting to lean in the same direction that SoM has taken with
his current idea on portal clipping. It may be best to separate out all linedef
special activation logic from the underlying thinkers, and then implement an
inheritable interface - instantiated once per map type/gamemode combination -
for linedef triggering.
This would keep a hard and fast separation between DOOM and Hexen logic, the
exact opposite of what ZDoom did. ZDoom didn't worry about maintaining
compatibility though, so I have entirely different design criteria to consider.
-------------------------------------------------------------------------------
01/21/12
ptoing has returned and is working on various things - the first thing that has
drawn his interest are player translations, and so I've started working on
those internally again.
R_InitTranslationTables has been rewritten so that it uses an array of ZDoom
translation language strings instead of fraggle's numeric range table. This
allowed me to catch a bug in my implementation of the parser, though - when
reaching end-of-string after a single-color destination range, the parser would
back up one character - this works in every case *except* when the parser is
at EOS and in that case it causes a parse failure by backing up more than the
tokenizer has advanced.
After fixing that problem, ptoing's new translations are functional. He has
done his best to create ones that are compatible with the existing SMMU color
set but look much better. A couple of them still require some tweaking though,
as some of the colors fade to gray too fast at lower levels and stick out
badly.
Eventually translations will move out into EDF and become player class
properties, and this will most likely occur on this development branch.
I noticed a SMMU glitch - the function thought there was one more translation
in the array than fraggle had defined - this was due to fraggle having removed
one of the translations by commenting it out (apparently it was too ugly to
grant countenance). This never had an ill effect before because the array
was defined at size TRANSLATIONCOLOURS, and the final entry was zero-inited
and therefore useless.
Added an additional ecalloctag macro and replaced more Z_Calloc calls.
-------------------------------------------------------------------------------
01/19/12
I have finally gotten around to removing the old array-style sprite
declarations from sprites.edf for Heretic gamemode, as they are no longer
required - those sprites are created while parsing the states now.
This does require EDF spritevar and pickup effects processing to occur after
thingtypes and states are fully processed however, the former in particular
since DECORATE states may also create new sprite values.
An inventory classing mechanism has been re-implemented (this was also in the
old aborted inventory attempt, but worked differently there). I started out by
adding one of the simplest inventory types, Health, which give the player
health immediately upon being picked up.
However I am running into a lot of roadblocks, much more than I expected, so it
is going to be slow going from here on out with the principal project for this
branch.
Added an E_MetaStringFromCfgString utility (possibly the longest function name
in Eternity currently) that turns a cfg_opt_t CFG_STR property into a
MetaTable property under the same key.
It seems, upon greater thought, that there might be a use for an entire
framework of functions that do the same thing for various types - int, float,
etc.
For that matter libConfuse itself *could* be rewritten ground-up to be based
on MetaTables, but I am divided on whether that would be a useful diversion at
the present time.
-------------------------------------------------------------------------------
01/18/12
I tried implementing some suggestions by Ryan C. Gordon on the SDL mailing list
that might would address the way that Eternity starts out in SDL GL2D mode
showing a screen that is variously bright white or full of weird "dead" screen
contents depending on how it is started up, but unfortunately they had no
effect.
As a result I have removed the fillPBO method of SDLGL2DVideoDriver, as it is
not having the effect that it seemed to have at first. As part of my various
other experiments, I added a GL_OrthQuadFlat function to gl_primitives. It is
not currently used for anything.
-------------------------------------------------------------------------------
01/17/12
All Thinker descendants have been adjusted to use their new Super typedef to
call up to their parent serialize method. The advantage here is obvious: if
the parent type changes, the code will continue to work without intervention.
I have simplified the new e* allocation macros to eliminate their use of a
template function as an intermediary, as it's simply totally unnecessary. The
type parameter can be placed into the angle brackets of static_cast directly by
the macro.
I also added estructalloctag, so that I can use estructalloc in situations
where the allocation tag isn't PU_STATIC.
estructalloc is now used pretty much everywhere, in preference to direct
Z_Calloc calls to allocate structs or arrays of structs.
Sector sound sequence functions' origin type parameter has always bugged me in
that I wrote them to take that parameter as an arbitrary boolean that is
false to mean floor and true to mean ceiling. I have rewritten them to take
the sequence origin type enumeration values that are defined in the s_sndseq
header file instead.
-------------------------------------------------------------------------------
01/16/12
EEMusicChanger was using the wrong doomednum courtesy of me misreading the
ZDoom wiki page about it.
However, I am going to put sector actions and everything they entail back on
hold until udmf-branch (the next scheduled major project after this one),
because they run into problems with Eternity's current gigantic mess of a
linedef special system.
MobjCollection has been altered so that the mobjType is stored as a name
instead of a type number. This ensures that it will be looked up every time
the collectThings method is invoked instead of only once at startup, meaning
that the thingtype they depend on can change via runtime EDF parsing once
EDF collections are functional.
MapInfo level type enumeration values were added for Strife and Hexen (just a
bit of early prep work ;)
In advance of udmf-branch I am also looking at various aspects of the linedef
special system and seeing if they cannot be cleaned up at least a little bit
right now. I started, as always, with doors, and managed to merge several
special cases in VerticalDoorThinker::Think whereby multiple types of doors
always act the same way.
-------------------------------------------------------------------------------
01/15/12
DECORATE states did not set the state_t::oldaction member which is critical for
the functioning of DeHackEd. This caused a significant game-breaking regression
for HACX which was stumbled upon by DavidPH while testing the network code.
When calculating the network packet checksum, do not write it back into the
netbuffer structure until it is verified to be valid. This might improve
behavior somehow, though it should really have no effect if everything is
otherwise working the way it ought.
-------------------------------------------------------------------------------
01/14/12
I am working on adding support for Risen3D's MUSINFO lump, now that it is
supported by both ZDoom and prboom-plus, and is used in the new Memorial wad
for DOOM II which is otherwise completely Eternity-compatible.
In advance of this, I need to support sector actions like ZDoom has, since the
MusicChanger object is a special case of a "player enters sector" action. I
will be allowing type 9998 as "EESectorActionEnter" at the same time.
Some problems have been identified in EE's peer-to-peer networking code and I
am trying to iron them out. Two fixes so far:
* When sending the startup packet, the size of the game options should be
added in, not assigned to packetsize.
* -net should not crash the game if given less than two parameters
(anarkavre!!!! >:O )
-------------------------------------------------------------------------------
01/13/12
Advanced menu filebox selection widgets should not be launched when the file
list is empty. This avoids the crash mentioned in the 1/12 entry below.
-------------------------------------------------------------------------------
01/12/12
Most calls to EHashTable<>::initialize have been eliminated throughout the
engine in preference to use of the new parameterized constructor.
I found a bug whereby the mn_selectmusic menu could start out empty, and if you
tried to use it, the game would crash. I have to make sure that the static
music definitions are added to the music hash table at startup always, and
cannot be missed due to a music-playing error.
The fact the menu crashes is a separate issue that I will look into later.
MobjCollection objects can now be managed by a global MobjCollectionSet class.
I intend to use this to make the definition of mapspot collections, such as
those used by the Boss Brain in DOOM II and D'Sparil in Heretic, defined via
EDF and extensible to additional object types. This will allow me to add the
Mace Spots from Heretic, which are the last missing thingtype with a doomednum
in that game.
-------------------------------------------------------------------------------
01/11/12
EHashTable will now self-initialize, in much the same fashion as qstring, if
objects are added to the table before it has been explicitly initialized. An
additional constructor has also been added which accepts the number of chains
to start the table out with, and this constructor also results in immediate
initialization.
-------------------------------------------------------------------------------
01/08/12
Two additional steps were needed for fully dynamic thingtypes:
* Caching of thingtype lookups throughout the engine, mostly in action
routines, had to be eliminated, as was done previously for states. (Note
this does not include the caching mechanism of argument lookups, which
has a mechanism for resetting all evaluations when EDF is processed at
runtime).
* GameModeInfo->teleFogType is resolved at runtime instead of at startup.
A bug was additionally fixed which caused a regression in G_CheckDemoStatus
whereby the demo loop would continue after playing a single demo from the
command line: the function G_ReloadDefaults changes the value of the singledemo
variable, and so it needs to be cached in a local across the call to that
function in G_CheckDemoStatus, or else it "forgets" that it was in singledemo
mode.
-------------------------------------------------------------------------------
01/04/12
I adjusted the MetaObject copy constructor to be safe if it is trying to copy
over itself.
-------------------------------------------------------------------------------
01/03/12
My Internet was down all weekend due to my DSL apparently shorting out on a bad
terminal contact, so I was able to complete a large amount of work:
* const_cast is no longer used anywhere to get around passing const char *'s
which actually point to heap-allocated values into efree() to free them.
Instead, structures with this problem have additional members that point
to the heap-allocated value and serve to track it for purposes of freeing
and reallocation separately from the const char * that is used to access
it otherwise.
* Added an e_dumpstate command so you can look at states as defined by EDF in
the console.
* DECORATE state blocks may now be used to populate a set of consecutively
pre-defined global states in EDF. A firstdecoratestate member has been
added to the EDF thingtype block to specify what the first global state a
thingtype should start filling in with its DECORATE states is. Note that
this property is "immediate" - it is not stored in the thingtype, and it
is NOT inherited by child types, as that simply wouldn't make sense.
* libConfuse now keeps track of the displaced copy/copies of a section when
it overwrites a section with a newer section of the same name, instead of
freeing and destroying the previous definitions.
* The previous change allows for displaced thingtypes' DECORATE state blocks
to still be processed. This is required when an EDF thingtype overrides one
of the default thingtypes, since if they didn't get their DECORATE state
block processed, they would never fill in the global state blocks either
and DeHackEd and older EDFs would stop working as a result.
* Added a +decorate flag to global EDF frame definitions, which is required
to be present in order for DECORATE states to overwrite that frame. If a
frame is modified first by a global EDF frame definition, it will clear
this flag.
This is required to keep the ability to override global frames by
redefintion. Otherwise, they'd always get overwritten by the original data
when DECORATE-format states are parsed into them.
* doom/things.edf and doom/frames.edf have been refactored to use DECORATE
state format. Heretic will follow soon. Hexen and Strife definitions will
start out in this format eventually.
* Inventory inheritance and basic field processing are now functional.
* Added an E_ReplaceString utility, which frees the current value of a char *
if it is valid and then assigns it a new value. This can be used like the
following example:
E_ReplaceString(foo, estrdup("bar")); // if duplicating a non-dynamic value
E_ReplaceString(baz, qux); // if 2nd param is already dynamic
* The development version has been bumped up to 3.40.21 and the name of the
next version will be "Midgard," which is the Norse mythological name for
the Earth, meaning literally "middle earth" (familiar?).
Middle Earth is where humans live, and it is supported by the trunk of the
great world tree, Yggdrasil. It is "middle" on account of being between
the various other nine realms of existence, amongst them Asgard, the home
of the gods, and Jötunheimr, the home of the giants. Being in the middle,
it is naturally where all the other beings meet to fight and end up making
trouble for humans rather than minding their own business :P
-------------------------------------------------------------------------------
12/31/11
Merged changes from trunk to pull in fix to white startup screen in GL2D mode.
-------------------------------------------------------------------------------
12/29/11
Merged changes from trunk to pull in support for hi-def music.
-------------------------------------------------------------------------------
12/23/11
The ThinkerType hierarchy has again been altered to eliminate the need to
specify a Thinker's parent class inside the IMPLEMENT_THINKER_TYPE macro,
which is away from the class definition inside a source file.
Instead, the parent class is now specified in the DECLARE_THINKER_TYPE macro,
inside the Thinker descendant class definition in the header file.
The following typedefs are created as part of this change:
* All Thinker descendants define a Super typedef that refers to their
immediate parent class (ie. Mobj::Super is equivalent to PointThinker,
and PointThinker::Super is equivalent to Thinker).
* All ThinkerType descendants define a Class typedef, which refers to the
Thinker descendant class for which they serve as an RTTI "whipping boy."
(ie. MobjType::Class is Mobj, and PointThinkerType::Class is PointThinker).
As a result of these changes, there is no longer a need for startup code to
dynamically lookup the parent type of each class in the ThinkerType hash table.
Instead, a virtual ThinkerType *getParentType() method in ThinkerType classes
is able to simply return Class::Super::StaticType. The StaticType member of
each Thinker class was already used to refer to its ThinkerType instance and
is initialized at startup as the ThinkerType objects construct themselves.
-------------------------------------------------------------------------------
12/22/11
Added a call to E_ProcessInventoryDeltas to main EDF processing.
Merged code from trunk to process SNDINFO music definitions, and altered it
accordingly for edf3-branch changes to the EHashTable API.
-------------------------------------------------------------------------------
12/20/11
metaapi.h must include DLListItem; I have no idea why only certain compilers
could get away without it while others object.
MetaTable instantiation and delta section processing have rendered the basic
inventory definition section of EDF parsable and processable (still doesn't
really DO anything, of course), and so I have added call downs to the new
module's routines in e_edf.cpp.
configure.in changes were merged in from trunk.
-------------------------------------------------------------------------------
12/18/11
Since inventory definitions are completely dynamic from the get-go, I have
determined that the inventoryDefs array is completely unnecessary and thus I
have eliminated it and all "GetNumForName" style routines, which now return
a pointer to an inventory_t structure instead of an index into inventoryDefs.
ThinkerType RTTI has been altered to use case sensitive name hashing, since the
identifiers it is based on are also case sensitive at compile time.
-------------------------------------------------------------------------------
12/17/11
Merged changes from trunk to pull in the fix to "goto" being treated as case
sensitive inside DECORATE state blocks.
-------------------------------------------------------------------------------
12/12/11
After additional thought about the way that inheritance should work in the
presence of dynamically changing thingtypes, I determined that the proper
course of action to take is to always create a new mobjinfo and let it override
the one previously defined through the "last-added-is-first-found" semantics of
EHashTable, which were precisely created originally to allow this behavior.
This allows thingtypes that inherited from a type altered later in dynamically
loaded EDFs to retain their proper behavior, as they continue to point to the
old, displaced mobjinfo as their parent type, including any DECORATE states it
may possess in its MetaTable.
The rest of the game engine will see the newly defined thingtype when it is
looked up via E_ThingNumForName and friends.
DeHackEd number hashing works again (it was temporarily broken due to some
mistakes).
Inventory reallocation routines E_CollectInventory and E_ReallocInventory have
been added, matching like-named routines for thingtypes.
-------------------------------------------------------------------------------
12/03/11
Elimination of the assembly version of D_abs was merged from trunk.
-------------------------------------------------------------------------------
11/28/11
Additional changes have been merged from trunk, with some following changes
made to hash tables defined in this branch that don't exist in trunk yet
necessitated by more changes to EHashTable, as suggested by DavidPH, which
resulted in elimination of passing the pointer-to-members to the class
constructor in preference of passing them into the template as parameters.
Another -timedemo increase was noted as a result of this change, since it
enables the templated EHashTable to incorporate the pointers to members as
constants at compile time and bake them into the code itself, with full
optimizations, instead of accessing them at runtime as if they're variables.
-------------------------------------------------------------------------------
11/26/11
The change to the semantics of the E*HashKey classes has made the inclusion of
the e_hashkeys.h header file unnecessary in many locations.
I have merged changes from trunk, including the addition of hal/i_platform
defines.
-------------------------------------------------------------------------------
11/23/11
DavidPH suggested another modification to EHashTable which would have the
E*HashKey interface classes use static methods.
As a result of this change, which did result in an extremely small but
consistent boost in -timedemo mode, I needed to make the names of states and
mobjinfo definitions fully dynamic, so there is no longer any mnemonic length
limit for these sections in EDF.
-------------------------------------------------------------------------------
11/22/11
A functional inventory system is the next major milestone for edf3-branch.
I have added hashing for inventory definitions.
-------------------------------------------------------------------------------
11/21/11
With the following list of changes, EDF has become fully dynamic:
* sprite variables can be set more than once
* pickup items are additive
* castinfo can be defined more than once (the new cast entirely replaces any
existing one, however, since castinfo is a singular resource)
* boss spawn types can be defined more than once, with the same caveat as
for castinfo
* EDF processing for the initial root parsing and for runtime additive
parsing has been folded together into a single routine, as earlier planned
for once all sections became properly dynamic.
I also found and fixed yet another problem with castinfo whereby F_CastPrint
could cause an access violation if any of the castinfo entries after the first
17 had an undefined name.
-------------------------------------------------------------------------------
11/20/11
ThinkerType RTTI is now fully functional. I have observed an average gain of
1.2 FPS in -timedemo ran on Sunder MAP11 allowing the default DEMO1 lump to
load and play - which is the DEMO1 from the IWAD, and features the player
wandering around in the first room shooting at walls while thousands of demons
grunt and growl outside - every one of them generating at LEAST one and
possibly several dynamic_cast calls per frame.
-------------------------------------------------------------------------------
11/14/11
I have completed a previously planned revamp of the ThinkerType class hierarchy
so that it supports inheritance information. With some additional minor
changes, this will enable EE to use these classes as its own form of RTTI in
preference to C++'s built-in mechanisms. With a simple test program, I found
that employing your own custom RTTI solution to implement a dynamic_cast
equivalent operation results in up to ten times faster code in Visual C++.
It is unfortunate that the C++ object model is not only so minimal, but also
at the same time so broken on account of having to support hateful features
such as multiple inheritance and virtual inheritance. These make dynamic_cast
and the obtaining & using of pointers to members significantly more complicated
than any basic, clean, and properly designed code needs.
-------------------------------------------------------------------------------
11/12/11
A bug was found in EE where if you started out in a position where you could
only see HOM, a segmentation violation would occur. This happened because the
drawsegs_xrange table would not be allocated until at least one drawseg was
added to it. If there were NO drawsegs, the pointer stayed NULL unexpectedly
and the renderer would crash trying to access it without checking.
This was fixed by making sure drawsegs_xrange starts out with room for at least
one drawseg allocated, instead of starting out NULL. This resolves the crash in
sargezar.wad MAP10 when -warp is used to go to that map directly.
This ideally should have been done on trunk, but I have decided that
edf3-branch will, to the greatest extent reasonable, serve as the development
branch for the next version of EE, to ensure that it does not end up back-
burnered and ignored in favor of other projects. This means that most fixes
and other changes will be occuring in this branch for a while, instead of on
trunk.
-------------------------------------------------------------------------------
11/06/11
Dynamic mobjinfo has been largely completed and tested to work at a basic
level, so the first major milestone is complete.
* thingtype definitions are now recognized during runtime EDF parsing.
* inheritance is now more friendly to dynamic definition, with generation
tracking (a generation being each individual EDF parsing process).
-------------------------------------------------------------------------------
11/05/11
E_IsMobjInfoDescendantOf has been optimized by altering it to return a pointer
to the parent type (or NULL when the type in question does not inherit from the
indicated type). This prevents the need to immediately lookup the parent type
using E_ThingNumForName when it was already iterated over in the process of
answering the inheritance question.
-------------------------------------------------------------------------------
11/04/11
Again exactly as was previously carried out for states, I have made mobjinfo
hashing fully dynamic via use of EHashTable, and I have changed the return
value from mobjinfo lookup functions that means "no such object" from
NUMMOBJTYPES to -1, so that its value need not change at runtime.
-------------------------------------------------------------------------------
11/03/11
I have created a new branch to continue development of EDF 3.0 features, which
may require significant refactoring that will render the branch unbuildable or
create temporary regressions and functionality gaps that wouldn't be acceptable
on the trunk build.
The first significant project for edf3-branch is the goal of fully dynamic EDF,
so that no definition types are treated non-additively or are ignored when
loading EDFs at runtime.
The first step on that road is to make mobjinfo fully dynamic. As with states
during edf-branch (which resulted in EDF 2.0), I have begun by turning mobjinfo
into an array of pointers to mobjinfo_t structures, instead of a solid array.
For cache coherency, the actual mobjinfo structures are still allocated as
contiguous blocks during each EDF parsing.
===============================================================================
EOF
===============================================================================