[RFC weston] libweston: Do not include subsurfaces with NULL buffer in bounding box

[Pekka Paalanen]

On Mon, 05 Mar 2018 13:27:27 +0100
Philipp Kerling <pkerling at casix.org> wrote:

> Hi,
> 
> thanks for resurrecting this :-) reply below
> 
> 2017-12-04 (月) の 21:37 +0000 に Daniel Stone さんは書きました:
> > Hi Philipp,
> > 
> > On 28 July 2017 at 15:41, Philipp Kerling <pkerling at casix.org> wrote:  
> > > I was pondering how to remove the window decorations of my
> > > application
> > > (which live in subsurfaces) when going full screen without
> > > flickering.
> > > 
> > > At first I just destroyed the surfaces, but that lead to flicker
> > > (window without decoration will appear for one frame before the
> > > full
> > > screen application is displayed) since subsurface destruction is
> > > not
> > > specified to be synchronized to the commit of the main surface.
> > > 
> > > Then I tried attaching a NULL wl_buffer to the surfaces. This works
> > > and
> > > does not flicker, but it has the problem that weston will
> > > disconnect my
> > > application when going to full screen for violating the configured
> > > xdg_shell buffer size. The reason seems to be that attaching a NULL
> > > wl_buffer does not reset the surface size and
> > > weston_surface_get_bounding_box still thinks that the surface has
> > > the
> > > prior size.
> > > 
> > > The patch fixes this by checking explicitly for the buffer and is
> > > confirmed to fix the issue, but I am not sure whether this is the
> > > right
> > > solution. Maybe surface->width and surface->height should be set to
> > > 0
> > > anyway?  
> > 
> > Hm. Yes, it does seem like the width/height should be 0, which is
> > undefined but not disallowed by the subsurface spec. I have no idea
> > if
> > this would work with Mutter in particular, which had problems
> > implementing that for top-levels.
> > 
> > In any case, perhaps one way you could do it would be to stack the
> > decoration subsurfaces below your primary surface before committing
> > your fullscreen buffer. It's ugly, but does seem like it should work
> > ...  
> I already did solve the problem for my application by attaching a NULL
> buffer and explicitly setting the window geometry on my xdg toplevel -
> then the bounding box calculation code in weston is not executed (think
> that was discussed in IRC).
> Still, it would be nice to get this right in weston, which is what the
> patch was about.
> 
> With "setting width/height to 0" I was referring to the variables in
> the weston code (subsurface->surface->width and ...->height), which I
> think may be better set to 0 if no buffer is attached to the
> subsurface. Currently, they seem to retain the size of the last
> attached buffer if you attach a NULL buffer.

Hi,

I didn't actually look at the code, so this comment is a vague
recollection from years ago: if the role interface (here wl_subsurface)
allows a NULL buffer to be committed, then by the general rule it will
cause that wl_surface to be unmapped, and it should also reset the
weston_surface size to zero. I do not recall any reason for not doing
that.

I actually have some hazy recollection of designing the hiding of a
sub-surface to intentionally need attaching a NULL wl_buffer to the
wl_surface, exactly for the synchronization reasons.


Thanks,
pq
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/wayland-devel/attachments/20180306/6379fa3a/attachment.sig>