This should be less of a drag on the playsim than having each light a separate actor. A quick check with ZDCMP2 showed that the light processing time was reduced to 1/3rd from 0.5 ms to 0.17 ms per tic.
It's also one native actor class less.
# Conflicts:
# src/g_shared/a_dynlight.cpp
# src/g_shared/a_dynlight.h
# src/hwrenderer/dynlights/hw_dynlightdata.cpp
# src/hwrenderer/dynlights/hw_dynlightdata.h
# src/hwrenderer/scene/hw_renderhacks.cpp
# src/namedef.h
# src/scripting/thingdef_data.cpp
# src/swrenderer/line/r_walldraw.cpp
# Conflicts:
# src/d_main.cpp
# src/g_levellocals.h
# src/g_shared/a_dynlight.cpp
# src/g_shared/a_dynlight.h
# src/gl/dynlights/gl_dynlight.h
# src/gl/dynlights/gl_dynlight1.cpp
# src/gl/scene/gl_spritelight.cpp
# src/gl/scene/gl_walls.cpp
# src/hwrenderer/dynlights/hw_shadowmap.cpp
# src/hwrenderer/dynlights/hw_shadowmap.h
# src/hwrenderer/scene/hw_flats.cpp
# src/p_setup.cpp
Sorting modes are
1 - by name, from A to Z
2 - by name, from Z to A
3 - number of calls, ascending
4 - number of calls, descending
5 - total time, ascending
anything else - total time, descending
This is to ensure that the Class pointer can be set right on creation. ZDoom had always depended on handling this lazily which poses some problems for the VM.
So now there is a variadic Create<classtype> function taking care of that, but to ensure that it gets used, direct access to the new operator has been blocked.
This also neccessitated making DArgs a regular object because they get created before the type system is up. Since the few uses of DArgs are easily controllable this wasn't a big issue.
- did a bit of optimization on the bots' decision making whether to pick up a health item or not.
(Is there anyway to tone down GCC's warning level? It outputs too many false positives for potentially uninitialized variables in which the genuine errors get drowned.)
- block creation of actors with the 'new' instruction. Unlike the above these cannot be made abstract because without ConstructNative they cannot be serialized.
This was done to ensure it can be properly overridden in scripts without causing problems when called during engine shutdown for the type and symbol objects the VM needs to work and to have the scripted version always run first.
Since the scripted OnDestroy method never calls the native version - the native one is run after the scripted one - this can be simply skipped over during shutdown.
It can actually happen that the thinker chain gets broken if an actor being iterated happens to destroy the immediately following actor in the chain as well. In that case both actors lose their chain links and the iterator cannot advance any further, the only solution to avoid a crash is to terminate the iteration of the current list.
It is utterly pointless to require every function that wants to make a VM call to allocate a new stack first. The allocation overhead doubles the time to set up the call.
With one stack, previously allocated memory can be reused. The only important thing is, if this ever gets used in a multithreaded environment to have the stack being declared as thread_local, although for ZDoom this is of no consequence.
- eliminated all cases where native code was calling other native code through the VM interface. After scriptifying the game code, only 5 places were left which were quickly eliminated. This was mostly to ensure that the native VM function parameters do not need to be propagated further than absolutely necessary.
Needless to say, this is simply too volatile and would require constant active maintenance, not to mention a huge amount of work up front to get going.
It also hid a nasty problem with the Destroy method. Due to the way the garbage collector works, Destroy cannot be exposed to scripts as-is. It may be called from scripts but it may not be overridden from scripts because the garbage collector can call this function after all data needed for calling a scripted override has already been destroyed because if that data is also being collected there is no guarantee that proper order of destruction is observed. So for now Destroy is just a normal native method to scripted classes
- added an 'exact' parameter to FThinkerIterator's Next function. This is mainly for scripting which allows to do a lot more checks natively when running the iterator while looking for one specific class.
- exported thinker iterator and drop item chain to scripting. Unlike its native counterpart the script-side iterator is wrapped into a DObject to allow proper handling for memory management.
- fixed: The VMFunctionBuilder only distinguished between member and action functions but failed on static ones.
- fixed: FxAssign did not add all needed type casts. Except for purely numeric types it will now wrap the expression in an FxTypeCast. Numeric handling remains unchanged for both performance reasons and not altering semantics for DECORATE.
- exported all internal flags as variables to scripting. They still cannot be used in an actor definition.
- make ATAG_STATE the same as ATAG_GENERIC. Since state pointers exist as actual variables they can take both values which on occasion can trigger some asserts.
- gave PClass a bExported flag, so that scripts cannot see purely internal classes. Especially the types like PInt can cause problems.
Todo: we need readonly references to safely expose the actor defaults. Right now some badly behaving code could overwrite them.
I wish I had realized this the last time it came up - it would have saved me a lot of trouble.
But as it turns out, the more recent travelling code makes all of this completely unnecessary, working perfectly fine with deleting the player pawns along with the rest of the thinkers before loading the stored ones from the savegame (and getting rid of those in G_FinishTravel.)
And with a sane savegame format that does not depend on side effects from how the thinker serializing handled linking into the lists the old code was even harmful, leaving voodoo dolls behind.
I had the exact same effect when I tried to reshuffle some things for reliably restoring portals, but did not make the connection to interference between two mutually incompatible player travelling mechanisms that just worked by sheer happenstance with the original order of things.
After testing with a savegame on ZDCMP2 which is probably the largest map in existence, timing both methods resulted in a speed difference of less than 40 ms (70 vs 110 ms for reading all sectory, linedefs, sidedefs and objects).
This compares to an overall restoration time, including reloading the level, precaching all textures and setting everything up, of approx. 1.2 s, meaning an increase of 3% of the entire reloading time.
That's simply not worth all the negative side effects that may happen with a method that highly depends on proper code construction.
On the other hand, using random access means that a savegame version change is only needed now when the semantics of a field change, but not if some get added or deleted.
- do not I_Error out in the serializer unless caused by a programming error.
It is better to let the serializer finish, collect all the errors and I_Error out when the game is known to be in a stable enough state to allow unwinding.
Note that even with this change it is still not possible to unarchive any thinker pointers before the thinker list has been loaded as it would create broken lists.
This was to resolve some circular dependencies with the portal code.
The most notable changees:
* FTextureID was moved from textures.h to doomtype.h because it is frequently needed in files that don't want to do anything with actual textures.
* split off the parts from p_maputl into a separate header.
* consolidated all blockmap related data into p_blockmap.h
* split off the polyobject parts into po_man.h
- Fixed: DThinker::Destroy(Most)ThinkersInList() were unreliable when
destroyed thinkers destroyed more thinkers in the same list.
Specifically, if the thinker it destroyed caused the very next thinker
in the list to also be destroyed, it would get lost in the thinker list
and end up with a NULL node. So just keep iterating through the first
thinker in the list until there are none left. Since destroying a
thinker causes it to remove itself from its list, the first thinker will
always be changing as long as there's something to destroy.
- Instead of tying NoDelay behavior to OF_JustSpawned, use a new actor
flag, MF7_HANDLENODELAY. This only gets cleared once it has actually
been checked by Tick(). This is necessary because freeze mode delays the
initial run of Tick() past the initial spawn, so OF_JustSpawned will no
longer be set when it does the initial tick.
- Delay NoDelay processing if an actor is spawned dormant. Actors spawned
dormant have Deactivate() called before they tick, so MF7_HANDLENODELAY
will remain set as long as an actor is dormant. This allows the NoDelay
handling to occur as expected once it is activated.
When set, the actor will run this state during its first tick. This means Spawn
states may now run an action function if you set this flag. Note that this action function
is executed during the actor's first tick, which is not the same as when it is spawned.
SVN r4240 (trunk)
about this now. I spent three days trying to figure out why the VC++ debug version was so slow.
It turns out I had a conditional breakpoint set in a high-traffic area. D'oh!
The rest of this stuff should get merged into trunk:
- Fixed: Writing to debugfile uses the standard fprintf, which does not understand %td on VC++.
- Fixed: Instead of crashing when a sprite has been scaled to 0, just don't draw it.
SVN r3896 (scripting)
