[Mesa-dev] [PATCH] wayland/egl: initialize window surface size to window size

[Juan A. Suarez Romero]

On Mon, 2018-08-06 at 16:02 +0100, Daniel Stone wrote:
> Hi Juan,
> 
> On Tue, 5 Jun 2018 at 18:23, Juan A. Suarez Romero <jasuarez at igalia.com> wrote:
> > On Tue, 2018-06-05 at 12:41 +0100, Daniel Stone wrote:
> > > > - In Wayland/Mesa, the surface is not resized at the time the native window is
> > > > resized; this is done in *eglSwapBuffers*, following 3.10.1.1.
> > > 
> > > Disagree, given that we are in charge of exactly when the native
> > > window is 'resized'. It's unclear to me what 'resize the surface prior
> > > to copying its pixels to the native window' even means: does this
> > > require we do a scale blit, or a crop blit, or?
> > > 
> > > The semantics Mesa has interpreted until recently is that
> > > wl_egl_resize_window() constitutes a resize _request_, and that the
> > > last submitted request is acted upon and latched at the first draw
> > > call. This then constitutes the native surface size up until the next
> > > first draw call. I think these are good semantics, and we have to be
> > > quite picky about what exactly we implement, since clients rely on
> > > them. Not just cosmetically (atomic resizing of subsurface chains),
> > > but also because if you get your window sizing wrong on the desktop,
> > > you'll catch a fatal error from the compositor (X11 just shrugs and
> > > does nothing).
> > > 
> > > I think these are good semantics (as to exactly when resize takes
> > > effect), but for the moment we only update the EGL_{WIDTH,HEIGHT}
> > > surface query results after the first draw call. I would suggest that
> > > in the period after eglSwapBuffers but before the first draw call, we
> > > also update those values, and that we update them after eglSwapBuffers
> > > has executed as well. This makes the semantics that any native window
> > > resize requests take effect immediately after eglSwapBuffers (or just
> > > immediately, if no swap has ever been executed) and before the first
> > > draw call (or buffer_age query ... cf. 9ca6711faa03) for a given
> > > frame; in between the first draw call and eglSwapBuffers, the native
> > > resize requests will be queued until after eglSwapBuffers.
> > 
> > 
> > I see. The surface size should be updated inmediately, after a window resize is
> > done, as you said.
> > 
> > I'll send then a new patch version. Actually I'll split this patch in two: one
> > to correctly initialize the surface size (there is no doubt on this, and this
> > patch alone will fix the test), and another patch to correctly update the
> > surface size when the window is resized.
> 
> I think the v3 2/2 patch doesn't actually conform to this behaviour,
> as it unconditionally sets the surface width/height from
> wl_egl_window_resize() rather than waiting until the resized buffers
> are actually going to take effect.
> 

Yes, the idea is that any window resize take effect immediately, without
requiring to run eglSwapBuffers() (that is, if window is resized and the surface
size is immediately queried, the new size is returned; only when the buffers are
updated a new buffer with is created).


	J.A.




> Cheers,
> Daniel
>