On 23/05/12 23:46, Chad Versace wrote:
On 05/23/2012 07:55 AM, Pauli Nieminen wrote:
On Tue, May 22, 2012 at 03:43:46PM -0700, Chad Versace wrote:
On 05/21/2012 11:09 AM, Pauli Nieminen wrote:
Protable code should cast to EGL native types always before passing
variables to functions taking a native parameter.
Signed-off-by: Pauli Nieminen<[email protected]>
---
tests/egl/egl-util.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/tests/egl/egl-util.c b/tests/egl/egl-util.c
index 41fc31f..639515e 100644
--- a/tests/egl/egl-util.c
+++ b/tests/egl/egl-util.c
@@ -179,7 +179,7 @@ egl_util_run(const struct egl_test *test, int argc, char
*argv[])
}
- state.egl_dpy = eglGetDisplay(state.dpy);
+ state.egl_dpy = eglGetDisplay((EGLNativeDisplayType)state.dpy);
if (state.egl_dpy == EGL_NO_DISPLAY) {
fprintf(stderr, "eglGetDisplay() failed\n");
piglit_report_result(PIGLIT_FAIL);
I think that this indicates that there is a type mismatch somewhere.
I don't believe such casting is ever necessary. In fact, I believe such
casts will mask real bugs.
The bugs like eglplatform.h having matching integer types instead of
pointers or vice-versa. But that probably should be considred in
eglplatform.h.
Waffle never makes casts for EGLNative types, and it compiles without any
warnings.
For example, in the x11_egl backend, Waffle passes a Display* to eglGetDisplay
without any typecast;
see the two lines below:
line 54,
http://cgit.freedesktop.org/~chadversary/waffle/tree/src/waffle/x11_egl/xegl_display.c?id=0.3#n54
line 29,
http://cgit.freedesktop.org/~chadversary/waffle/tree/src/waffle/egl/egl_native_template.c?id=0.3
What problem did you encounter that prompted this patch?
There is problem that EGLNative* types vary between platform but are
defined in dispatch. Including EGL/eglplatform.h did feel like a
working solution to provide platform independ interface.
I had a idea that dispatch could defined native types to uintptr_t
and generate platform specific wrappers to truncate types back to
platform specific types behind dispatch.
Another option would be filtering out native types from dispatch. Too
bad that prevents user of dispatch from using egl native types in same
C file as egl functions use through dispatch. Or at least makes it a
lot more complex than it should be.
This is a difficult thing to solve without introducing lots of complexity.
I think I need to better understand your dispatch patches before discussing
this more.
Now that I remember to check the system V 64bit ABI I noticed that we
can simple make all native types 64bit wide. They will be handled
correctly because parameters are passed in registers. That probably
isn't true for big-endian system if there is some byte swapping magic
required to get lowest 32bits from a 64bit integer register to correct
place for 32bit access.
32bit systems won't have problems because everything is 32bit pointers
or integers.
Pauli
_______________________________________________
Piglit mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/piglit
