Previously it tried to copy all patches of composite sub-images directly onto the main image.
This caused massive complications throughout the entire true color texture code and made any attempt of caching the source data for composition next to impossible because the entire composition process operated on the raw data read from the texture and not some cacheable image. While this may cause more pixel data to be processed, this will be easily offset by being able to reuse patches for multiple textures, once a caching system is in place, which even for the IWADs happens quite frequently.
Removing the now unneeded arguments from the implementation also makes things a lot easier to handle.
On Windows none of this is needed, because we can generate a proper unwind frame for the JITed functions, but even on Linux, it would require manual additions to each single piece of native code that ever gets called from inside a JIT compiled function.
This is an utterly prohibitive proposition because it makes direct native calls a virtual impossibility
So, in order to get the thrown error properly presented both I_Error and ThrowAbortException will now forward to I_FatalError if it is called from inside a JIT context.
src/doomerrors.h:74:14: error: exception specification of overriding function is more lax than base version
src/posix/sdl/i_main.cpp:272:28: error: 'class std::exception' has no member named 'GetMessage'
Like Linux and macOS this will only support borderless fullscreen in the active desktop resolution now, which is what modern systems need.
The list of discrete resolutions has been removed as it makes no sense anymore with a fixed video mode - all the other scaling options remain active, though.
- with renderers freely switchable, some shortcuts in the 3D floor code had to be removed, because now the hardware renderer can get FF_THISINSIDE-flagged 3D floors.
- changed handling of attenuated lights in the legacy renderer to be adjusted when being rendered instead of when being spawned. For the software renderer the light needs to retain its original values.
This does not work with a setup where the same backend is driving both renderers.
Most of this is now routed through 'screen', and the decision between renderers has to be made inside the actual render functions.
The software renderer is still driven by a thin opaque interface to keep it mostly an isolated module.
This was originally invented to fix the sprite offsets for the hardware renderer.
Changed it so that it doesn't override the original offsets but acts as a second set.
A new CVAR has been added to allow controlling the behavior per renderer.
This was done mainly to reduce the amount of occurences of the word FTexture but it immediately helped detect two small and mostly harmless bugs that were found due to the stricter type checks.
Since this is a non-standard function it's better kept to as few places as possible, so now DirEntryExists returns an additional flag to say what type an entry is and is being used nearly everywhere where stat was used, excluding a few low level parts in the POSIX code.
- moved timer definitions into their own header/source files. d_main is not the right place for this.
- removed some leftover cruft from the old timer code.
This fixes two issues:
* timer related texture animations are not being recreated multiple times if a scene renders multiple viewpoints (e.g. camera textures or portals.)
* interpolation is smoother when maps have a high think time of multiple milliseconds. A good map to see the difference would be ZDCMP2 which has a think time of 4-5 milliseconds. With the timer taken in real time after the thinkers have run and VSync on this resulted in alternating time slices of 11 and 21 ms between frame interpolations instead of an even 16 as should be done for smooth 60 fps because roughly every second frame was offset by those 5 ms.
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.
The global variable holding a pointer to this thinker should be a weak reference to the instance in the thinker chain, there is no need to mark this global variable, as the thinker's lifetime is only determined by the thinker chain.
- committed a few Posix related file the last commit missed.
I have no idea why they were even in there, as they intentionally circumvented all GC related features - they declared themselves fixed if prone to getting collected, they all used OF_YesReallyDelete when destroying themselves and they never used any of the object creation or RTTI features, aside from a single assert in V_Init2.
Essentially they were a drag on the system and OF_YesReallyDelete was effectively added just to deal with the canvases which were DObjects but not supposed to behave like them in the first place.
