The attached patch fixes a few bugs in SDL related to the mouse position
in windowed mode on Mac OS X, reproduced using the attached minimal test
program - at least here on 10.3.9, with SDL CVS from today. Could anyone
test whether the bugs exist and are fixed by the patch on 10.2 and 10.4?
When using OpenGL, the vertical mouse positions obtained through
events or SDL_GetMouseState() are off by one.
When using OpenGL, SDL_WarpMouse() inverts the y coordinate.
Clicks on the topmost pixel row of the window are not recognized.
1 and 2 do not occur in non-OpenGL mode, while 3 does. All three only
occur in windowed mode, not in fullscreen.
The cause for 1 and 3 is that in Cocoa, "the location of the mouse"
seems to be defined as “the location of the top left corner of the mouse
pointer’s hot pixel” (this is not documented, it’s just what I found out
here), which together with the fact that Cocoa’s usual y coordinates
start at the bottom and increase upwards means that the y coordinate of
the mouse runs from 1 to h, not from 0 to h-1, in a window of height h.
If it does work on 10.2 and 10.4 (I’ll try to test it as soon as I can,
but at the moment all I have at hand is 10.3.9), can this be applied to
the CVS?
If it does work on 10.2 and 10.4 (I’ll try to test it as soon as I can,
but at the moment all I have at hand is 10.3.9), can this be applied to
the CVS?
We had some pain with vertical mouse coordinates getting flipped when
Panther (or was it Jaguar?) first shipped.
Your patch is probably good, but I’ll run it through various OS versions
on my test box after Thanksgiving.
We had some pain with vertical mouse coordinates getting flipped when
Panther (or was it Jaguar?) first shipped.
I remember that, and it’s why I agree the patch needs testing before
being incorporated.
Your patch is probably good, but I’ll run it through various OS versions
on my test box after Thanksgiving.
Thanks. I took my 10.2 CDs with me when I was at home this weekend, only
to discover that I can’t install from them because they are older than
my powerbook. So testing on 10.2 (and maybe 10.1 - does SDL even run on
that?) will have to wait until next weekend when I can get my hands on
the powerbook’s own software restore DVD.
I did test it on 10.4.0, and the results are the same as on 10.3.9 - the
bugs exist and are fixed by my patch.
When using OpenGL, SDL_WarpMouse() inverts the y coordinate.
The mouse inversion problem for windowed-OpenGL appeared going from
Jaguar to Panther. I don’t remember if we ever resolved this in an
elegant way. At one point I thought I did, but I remembered I ended up
breaking more stuff and I can’t remember if I fixed this or went to an
OS version check to case the two different situations.
So be careful about testing this one between the Jaguar/Panther
barrier. If not done correctly, it will probably (incorrectly) invert
pre-10.3 systems. I would also advise testing the non-OpenGL modes and
maybe even the fullscreen modes.
When using OpenGL, SDL_WarpMouse() inverts the y coordinate.
The mouse inversion problem for windowed-OpenGL appeared going from
Jaguar to Panther. I don’t remember if we ever resolved this in an
elegant way.
Currently, there is an inversion in QZ_PrivateCocoaToSDL(), which makes
getting mouse coordinates work correctly (apart from the off-by-one
bug). The bug is that the same inversion is also needed in
QZ_PrivateSDLToCocoa(), which is used when setting mouse coordinates,
i.e. SDL_WarpMouse(). My patch fixes that.
At one point I thought I did, but I remembered I ended up
breaking more stuff and I can’t remember if I fixed this or went to an
OS version check to case the two different situations.
So be careful about testing this one between the Jaguar/Panther
barrier. If not done correctly, it will probably (incorrectly) invert
pre-10.3 systems. I would also advise testing the non-OpenGL modes and
maybe even the fullscreen modes.
Yup, my test program does all combinations of GL/non-GL and
windowed/fullscreen, and I intend to test it at least on 10.2 and maybe
10.1. I already did on 10.4.0, and behavior there is identical to 10.3.9.
It sounds like you covered everything. There isn’t any behavior change
for the mouse coordinates between Panther and Tiger. It’s just the
Jaguar/Panther boundary that I’m aware of. (Though we never actually
tested the fix on 10.0-10.1.)
Thanks,
Eric> From: Christian Walther
Currently, there is an inversion in QZ_PrivateCocoaToSDL(), which makes
getting mouse coordinates work correctly (apart from the off-by-one
bug). The bug is that the same inversion is also needed in
QZ_PrivateSDLToCocoa(), which is used when setting mouse coordinates,
i.e. SDL_WarpMouse(). My patch fixes that.
At one point I thought I did, but I remembered I ended up
breaking more stuff and I can’t remember if I fixed this or went to an
OS version check to case the two different situations.
it works everywhere. And if that’s true, I guess my patch should also be
correct everywhere, but I’ll still test it to be sure.
So be careful about testing this one between the Jaguar/Panther
barrier. If not done correctly, it will probably (incorrectly) invert
pre-10.3 systems. I would also advise testing the non-OpenGL modes and
maybe even the fullscreen modes.
Yup, my test program does all combinations of GL/non-GL and
windowed/fullscreen, and I intend to test it at least on 10.2 and maybe
10.1. I already did on 10.4.0, and behavior there is identical to 10.3.9.
I’ll try to test it as soon as I can, but at the moment all I have at hand is 10.3.9
So, here are the results of my tests (with patched and unpatched
frameworks compiled with Xcode 1.5 (gcc 3.3) on 10.3.9):
On 10.1.5, my test program doesn’t run because of “Undefined symbols:
SDL undefined reference to _CGMainDisplayID expected to be defined in
Carbon”. I guess not supporting 10.1 was a deliberate decision then and
that’s OK with me.
On 10.2.8, 10.3.9, and 10.4.0, the bugs exist as described in my
original post and are fixed by my patch. That is, there is no difference
between pre/post 10.3 and the patched version works correctly in all
combinations of GL/non-GL and windowed/fullscreen.