There might be various error flags and we need to clear all of them to get
precise error results. Thus, we need to call glGetError in a loop to achieve
that.
Formerly the taskbar support was *always* enabled except when the backend or
user specified that it should be disabled. This causes nasty crashes for
backends which do not have any taskbar support (like DC, Tizen and probably
more which simply did not disable it so far) when defaultErrorHandler was
called for example (Mass Add is also broken for those).
The SDL (and derived backends) worked around missing taskbar support by
simply faking a dummy taskbar implementation (but still claiming in configure
that we feature taskbar integration, ouch).
To avoid all non-SDL backends from manually specifying _taskbar=no I added
some auto detection code which simply only enables taskbar support in case
ScummVM is built on Win32 or libunity is present.
This changes makes ScummVM compilable with newer versions of DevKitPPC. ScummVM can be linked against the original libogc and libfat. That makes some newer WiiMotes work, improves audio-/video-playback and contains various improvements.
copyFrom frees and reallocates the surface for every update, as long
as the dimensions and format stay the same, we can do with just a
memcpy.
This gives a tiny improvement in the update-part of the Theora-player
(on the order of a bit more than 1 second saved total in the 1:28 long
J.U.L.I.A.-intro)
If we have only one thing being drawn, and that is opaque, we can
skip filling the render surface with background color.
This shaves another few wasted cycles of the FMV playback. (Since we
now don’t have to write the entire render surface TWICE).
This reduces the time spent in drawTickets() to ~60% of what it was before.
This caused an exception if a error() call occurs, rather than a
clean exit. This occurred in the defaultErrorHandler() function of
engines/engine.cpp, probably due to g_system->getTaskbarManager()
returning a null pointer.
Untangled the audio and video track from each other, and the parsing
of the stream from the decoding of its data. Also fixed a memory leak
as it turns out deleting a Surface doesn't free its data. You have to
call free() in it.
I have only checked the intro, not every cutscene, but that seems to
work fine at least.
As an alternative to using the Screen class's functions, we can let
the VQA decoder draw directly to the backend. This won't work if the
game uses "hi-res mode", but I don't think that's ever the case for
Malcolm's Revenge. I believe the KyraEngine_MR::playVQA() function
ensures that the screen is properly updated after the movie has
finished.
This almost limits the VQA rewrite to vqa.cpp and vqa.h. Whether it's
better this way than changing the Screen functions to take a 'pitch'
parameter...? I don't know. But it's an alternative.
