This mode ensure that an integral scaling is used both for the width
and for the height. Compared to the old pixel perfect stretch mode
it thus differs in the way it handles the aspect ratio correction.
It may stretch or squeeze the vertical direction to snap it to a
multiple of the original game height.
This reverts commit 8af0b8baa5.
That commit introduced a regression causing toggling the
aspect ratio correction using the hotkey to be ignored. And
I checked that reverting the commit did not reintroduced the
bug that it was supposed to fix. However since commit ceca0b5ae
the window size in OpenGL mode no longer depends on whether
the Aspect Ratio correction is enabled or not.
Force the creation of a new screen buffer for the scaler after leaving the overlay. The whole "old source" code looks a bit unfinished. Anyway, this fix should only kick in for one particular scaler and one particular situation and one particular backend. So I guess it is safe enough...
This fixes an assert with the new GUI (bug #12578).
This also required changing the logic to handle the cursor key color
as the new GUI does not use that key color but use the alpha channel.
The logic in the iOS backend was overwriting the alpha channel
by checking the key color, removing all the transparency. Now the
two are combined (hopefully there is no case were it uses the key
color while also using a pixel format with alpha, but not setting
this alpha, because if there is, this will now result in a fully
transparent cursor).
Supported graphics modes are grouped in an merged array in which
the first part contains modes from the non-OpenGL SDL manager,
obtained by a copy dynamically allocated.
When the array was rebuilt or aimed to be destroyed, it was only
cleared (removing elements) but not freeing duplicated fields
name and description.
An additional leak came from the temporary array (srcModes)
not freed.
OSystem now just returns a nullptr if there is no text to speech manager instance
(because none is compiled into the binary, or the system doesn't provide support
for it). This removed the need for the engine authors to add scummvm osystem compile
time options checks into their engine code
- Cleaned up comments
- Debug messages enhanced
- Fix some leftover spaces instead of TABs
- Removed the newlib workaround: A new version has been released which fixed the cause
- Removed an old workaround dealing with a buffer set too low due to an ill combination of newlib build and disk filesystem.
Even forcing the described behaviour i was not able to see slowdowns in videos.
Also, since "slow video playback" is rather vague and no specific case were given in the comment, i wasn´t able to do specific tesing.
Works for me...
