After fad65f2f5d, the viewport size
is always same as the framebuffer size. The 'screen' image size
was the actual rendering region size, and padding was not
considered in the size. The padding is added on the fullscreen
mode.
This fix adds the extra padding values to the 'screen' image size
so that the 'screen' framebuffer will be same size as the monitor
size.
Fixes#693
EBITEN_INTERNAL_IMAGES_KEY is now disabled by default, and enabled
only when a build tag 'ebitendebug' is specified. It is because
game developers might want to disable EBITEN_INTERNAL_IMAGES_KEY
when they release their games, or anyone can dump the internal
images.
Fixes#632.
Actually, at least 704d4cf464,
examples/moire doesn't work well without roundTexel on MacBook Pro.
However, now other parts are fixed (e.g. highp is used), it looks
like roundTexel is no longer needed.
Some pretty old machines don't support OpenGL 3.x or later, and in
such environment, some framebuffer functions like glGenFramebuffers
are not available. Instead, EXT versions can be used even on
machines that don't support OpenGL 3.x.
After this change, Ebiten always tries to use EXT version of
framebuffer functions. I believe EXT version is always available
when non-EXT version is available, so this chang eshould be safe.
Fixes#602
nosync package for GopherJS is problematic since nosync's mutex
assumes that Lock call is never duplicated, which is not true in
actual applications.
See also #603.
nosync package is good in terms of performance, but this assumes
that duplicated lock never happens. As audio package runs multiple
goroutines, theoretically duplicated lock can happen, and it looks
like this is an actual case (#603).
This change replaces nosync usages with regular sync usages.
Probably I'll deprecate nosync usages via internal/sync package
everywhere in Ebiten.
This might fix#603.