Re: [PATCH weston 08/11] compositor: Untangle mapedness from outputs
On Thu, 23 Jun 2016 14:41:57 +0200
Giulio Camuffo wrote:
> 2016-06-18 19:15 GMT+02:00 Armin Krezović :
> > Currently, weston assumes a surface/view is mapped if
> > it has an output assigned. In a zero outputs scenario,
> > this isn't really desirable.
> >
> > This patch introduces a new flag to weston_surface and
> > weston_view, which has to be set manually to indicate
> > that a surface/view is mapped.
> >
> > Signed-off-by: Armin Krezović
> > ---
> > src/compositor.c | 29 ++---
> > src/compositor.h | 4
> > src/data-device.c | 2 ++
> > src/input.c | 2 ++
> > 4 files changed, 18 insertions(+), 19 deletions(-)
> >
> > diff --git a/src/compositor.c b/src/compositor.c
> > index d246046..b25c671 100644
> > --- a/src/compositor.c
> > +++ b/src/compositor.c
> > @@ -1549,19 +1549,13 @@ weston_view_set_mask_infinite(struct weston_view
> > *view)
> > WL_EXPORT bool
> > weston_view_is_mapped(struct weston_view *view)
> > {
> > - if (view->output)
> > - return true;
> > - else
> > - return false;
> > + return view->is_mapped;
> > }
> >
> > WL_EXPORT bool
> > weston_surface_is_mapped(struct weston_surface *surface)
> > {
> > - if (surface->output)
> > - return true;
> > - else
> > - return false;
> > + return surface->is_mapped;
> > }
>
> I don't understand what this function means. It's here from before the
> surface/view split happened, but right now i am not sure what the
> purpose of it is, and i would argue it shuold go. How can a surface be
> mapped if no views of it are mapped? Maybe you have a different
> definition of "mapped" than me, in which case i would like to see it
> as part of the documentation for these functions.
> For the use-case Pekka mentioned, remembering if a surface was mapped
> earlier and so retaining information like the position, all that can
> be done by having custom logic in the shell_surface, which knows all
> it needs to know about the surface and the views. We don't need the
> surface mappedness for that.
It can be done, yes, for shell surfaces.
You can check all the calls to weston_surface_is_mapped() with 'git
grep -p weston_surface_is_mapped'. Ignoring all shells, we are left
with the following notable places:
weston_view_unmap():
Skips checking and fixing input device foci if the surface
remains mapped after the view is unmapped. This could be
regarded as a bug though, at least for the foci that are
tracked by the view rather than the surface.
weston_surface_attach(), called from weston_surface_commit_state():
Avoids a redundant call to weston_surface_unmap().
view_list_add_subsurface_view(), part of
weston_compositor_build_view_list() machinery:
Only if surface is mapped, creates a new view for it
dynamically, as necessary, or re-uses a stashed view for it.
This is crucial to be done only for mapped surfaces, because
there are conditions that prevent a sub-surface from being
mapped even if it has both the role and the content. This can
also be the first time a view is being created, so there may
not be a view to check for. Furthermore, sub-surface views are
to be created for *each* parent surface view. If your shell
makes multiple views for the parent, the sub-surfaces will come
along automatically.
There are also other places, but those are either avoiding redundant
calls to weston_surface_unmap() or protecting the layer setup, i.e.
mapping.
You're right, weston_surface_is_mapped() looks like it could go. Each
shell needs to track it's own surface mappedness state, struct
weston_subsurface needs to add that too, weston_surface_unmap() can
probably cope with redundant calls, and weston_view_unmap()'s focus
handling would need fixing anyway.
Changing weston_surface_is_mapped() to look at view assignments to
layerr would not work for sub-surfaces.
IOW, it's not in scope for Armin's project, I don't think.
The immediate goal is to get outputs separated mappedness, so that
Armin can continue his project on outputs.
>
> >
> > static void
> > @@ -1738,6 +1732,7 @@ weston_view_unmap(struct weston_view *view)
> > weston_view_damage_below(view);
> > view->output = NULL;
> > view->plane = NULL;
> > + view->is_mapped = false;
> > weston_layer_entry_remove(&view->layer_link);
> > wl_list_remove(&view->link);
> > wl_list_init(&view->link);
> > @@ -1769,6 +1764,7 @@ weston_surface_unmap(struct weston_surface *surface)
> >
> > wl_list_for_each(view, &surface->views, surface_link)
> > weston_view_unmap(view);
> > + surface->is_mapped = false;
> > surface->output = NULL;
> > }
> >
> > @@ -2131,6 +2127,7 @@ view_list_add_subsurface_view(struct
> > weston_compositor *compositor,
> >
> > view->parent_view = par
Re: [PATCH weston 08/11] compositor: Untangle mapedness from outputs
2016-06-18 19:15 GMT+02:00 Armin Krezović :
> Currently, weston assumes a surface/view is mapped if
> it has an output assigned. In a zero outputs scenario,
> this isn't really desirable.
>
> This patch introduces a new flag to weston_surface and
> weston_view, which has to be set manually to indicate
> that a surface/view is mapped.
>
> Signed-off-by: Armin Krezović
> ---
> src/compositor.c | 29 ++---
> src/compositor.h | 4
> src/data-device.c | 2 ++
> src/input.c | 2 ++
> 4 files changed, 18 insertions(+), 19 deletions(-)
>
> diff --git a/src/compositor.c b/src/compositor.c
> index d246046..b25c671 100644
> --- a/src/compositor.c
> +++ b/src/compositor.c
> @@ -1549,19 +1549,13 @@ weston_view_set_mask_infinite(struct weston_view
> *view)
> WL_EXPORT bool
> weston_view_is_mapped(struct weston_view *view)
> {
> - if (view->output)
> - return true;
> - else
> - return false;
> + return view->is_mapped;
> }
>
> WL_EXPORT bool
> weston_surface_is_mapped(struct weston_surface *surface)
> {
> - if (surface->output)
> - return true;
> - else
> - return false;
> + return surface->is_mapped;
> }
I don't understand what this function means. It's here from before the
surface/view split happened, but right now i am not sure what the
purpose of it is, and i would argue it shuold go. How can a surface be
mapped if no views of it are mapped? Maybe you have a different
definition of "mapped" than me, in which case i would like to see it
as part of the documentation for these functions.
For the use-case Pekka mentioned, remembering if a surface was mapped
earlier and so retaining information like the position, all that can
be done by having custom logic in the shell_surface, which knows all
it needs to know about the surface and the views. We don't need the
surface mappedness for that.
>
> static void
> @@ -1738,6 +1732,7 @@ weston_view_unmap(struct weston_view *view)
> weston_view_damage_below(view);
> view->output = NULL;
> view->plane = NULL;
> + view->is_mapped = false;
> weston_layer_entry_remove(&view->layer_link);
> wl_list_remove(&view->link);
> wl_list_init(&view->link);
> @@ -1769,6 +1764,7 @@ weston_surface_unmap(struct weston_surface *surface)
>
> wl_list_for_each(view, &surface->views, surface_link)
> weston_view_unmap(view);
> + surface->is_mapped = false;
> surface->output = NULL;
> }
>
> @@ -2131,6 +2127,7 @@ view_list_add_subsurface_view(struct weston_compositor
> *compositor,
>
> view->parent_view = parent;
> weston_view_update_transform(view);
> + view->is_mapped = true;
>
> if (wl_list_empty(&sub->surface->subsurface_list)) {
> wl_list_insert(compositor->view_list.prev, &view->link);
> @@ -2152,6 +2149,7 @@ view_list_add(struct weston_compositor *compositor,
> struct weston_subsurface *sub;
>
> weston_view_update_transform(view);
> + view->is_mapped = true;
If you always set is_mapped to true here, what is the difference to
just checking if the view is in the view_list?
>
> if (wl_list_empty(&view->surface->subsurface_list)) {
> wl_list_insert(compositor->view_list.prev, &view->link);
> @@ -3244,7 +3242,6 @@ subsurface_get_label(struct weston_surface *surface,
> char *buf, size_t len)
> static void
> subsurface_configure(struct weston_surface *surface, int32_t dx, int32_t dy)
> {
> - struct weston_compositor *compositor = surface->compositor;
> struct weston_view *view;
>
> wl_list_for_each(view, &surface->views, surface_link)
> @@ -3256,8 +3253,9 @@ subsurface_configure(struct weston_surface *surface,
> int32_t dx, int32_t dy)
> * mapped, parent is not in a visible layer, so this sub-surface
> * will not be drawn either.
> */
> - if (!weston_surface_is_mapped(surface)) {
> - struct weston_output *output;
> +
> + if (!weston_surface_is_mapped(surface))
> + surface->is_mapped = true;
>
> /* Cannot call weston_view_update_transform(),
> * because that would call it also for the parent surface,
> @@ -3265,18 +3263,11 @@ subsurface_configure(struct weston_surface *surface,
> int32_t dx, int32_t dy)
> * inconsistent state, where the window could never be
> * mapped.
> *
> -* Instead just assign any output, to make
> +* Instead just force the is_mapped flag on, to make
> * weston_surface_is_mapped() return true, so that when the
> * parent surface does get mapped, this one will get
> * included, too. See view_list_add().
> */
> - assert(!wl_list_empty(&compositor->output_list));
> -
Re: [PATCH weston 08/11] compositor: Untangle mapedness from outputs
On Wed, 22 Jun 2016 12:16:31 +0200
Armin Krezović wrote:
> On 21.06.2016 14:05, Pekka Paalanen wrote:
> > On Sat, 18 Jun 2016 19:15:23 +0200
> > Armin Krezović wrote:
> >
> >> Currently, weston assumes a surface/view is mapped if
> >> it has an output assigned. In a zero outputs scenario,
> >> this isn't really desirable.
> >>
> >> This patch introduces a new flag to weston_surface and
> >> weston_view, which has to be set manually to indicate
> >> that a surface/view is mapped.
> >>
> >> Signed-off-by: Armin Krezović
> >
> > Hi Armin,
> >
> > there is a slight problem with the split and ordering in patches 8 -
> > 11. Patch 8 removes the old conditions, which breaks "everything" and
> > the follow-up patches fix things up piece by piece. The problem is that
> > we now have known-broken commits from the start. Those will make
> > bisecting unrelated issues harder.
> >
>
> Ugh, that's a valid concern.
>
> > There are two possible solutions:
> >
> > - split patch 8 to to only add the new fields, and make a new patch as
> > the final one in the series that switches the
> > weston_{view,surface}_is_mapped() to use the new flags
> >
> > - squash all patches 8 - 11 into one
> >
> > Because the changes in patches 9 - 11 are fairly trivial and unlikely
> > to break anything in themselves, the patch that switches the
> > *_is_mapped() functions will be the one causing the regressions if any.
> > Hence it would be also ok to just squash these into one patch. It
> > doesn't make too much difference for reviews, and no difference for
> > bisecting.
> >
>
> I wanted to split the shells stuff mainly because different people work(ed)
> on different shells. I like the first approach.
Hi,
ok, that is the way I would consider more proper too.
> >> ---
> >> src/compositor.c | 29 ++---
> >> src/compositor.h | 4
> >> src/data-device.c | 2 ++
> >> src/input.c | 2 ++
> >> 4 files changed, 18 insertions(+), 19 deletions(-)
> >
> > There is also a more subtle change of semantics in this patch, which
> > would be good to underline in the commit message.
> >
> > Previously when mappedness was determined by outputs, unmapping the
> > last view unmapped also the surface, or so I believe. That no longer
> > happens, which is actually a good thing. This allows a shell to unmap
> > all views without losing the mappedness information of the surface.
> >
> > A practical example of this is hiding a desktop-shell top-level window.
> > When the window is shown again, it should appear on the position where
> > it was hidden, but that will not happen if desktop-shell thinks the
> > surface is not mapped. Instead, desktop-shell will randomize the window
> > position again. I haven't looked if this actually happens or if there
> > is code to work around that, but I think the change by this patch is
> > for the better.
> >
> > The mappedness states of a surface and its views are no longer
> > implicitly connected.
> >
>
> I'm not sure I'm getting all this. We can discuss about it more on
> IRC or I can just copy/paste most of the text above :).
> >> @@ -2152,6 +2149,7 @@ view_list_add(struct weston_compositor *compositor,
> >>struct weston_subsurface *sub;
> >>
> >>weston_view_update_transform(view);
> >> + view->is_mapped = true;
> >
> > This I am not so sure about. Isn't it the responsibility of whoever
> > puts a view on a layer to set is_mapped to true?
> >
> > Doing it here might paper over some overlooked places.
> >
>
> Again, I was trying to mimic the previous behavior.
> weston_view_update_transform assigns an output to the view, and when
> the output got assigned it was assumed the view was mapped.
Mimicing the old behaviour would require you to make
weston_surface_assing_output() set the surface->is_mapped based on the
various view->is_mapped. You don't do that, and I don't think we should
either because of what I explained above.
That makes us change the behaviour more than strictly necessary, but I
think it's worth it.
Thanks,
pq
pgpLIsx3Fn_WN.pgp
Description: OpenPGP digital signature
___
wayland-devel mailing list
wayland-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/wayland-devel
Re: [PATCH weston 08/11] compositor: Untangle mapedness from outputs
On 21.06.2016 14:05, Pekka Paalanen wrote:
> On Sat, 18 Jun 2016 19:15:23 +0200
> Armin Krezović wrote:
>
>> Currently, weston assumes a surface/view is mapped if
>> it has an output assigned. In a zero outputs scenario,
>> this isn't really desirable.
>>
>> This patch introduces a new flag to weston_surface and
>> weston_view, which has to be set manually to indicate
>> that a surface/view is mapped.
>>
>> Signed-off-by: Armin Krezović
>
> Hi Armin,
>
> there is a slight problem with the split and ordering in patches 8 -
> 11. Patch 8 removes the old conditions, which breaks "everything" and
> the follow-up patches fix things up piece by piece. The problem is that
> we now have known-broken commits from the start. Those will make
> bisecting unrelated issues harder.
>
Ugh, that's a valid concern.
> There are two possible solutions:
>
> - split patch 8 to to only add the new fields, and make a new patch as
> the final one in the series that switches the
> weston_{view,surface}_is_mapped() to use the new flags
>
> - squash all patches 8 - 11 into one
>
> Because the changes in patches 9 - 11 are fairly trivial and unlikely
> to break anything in themselves, the patch that switches the
> *_is_mapped() functions will be the one causing the regressions if any.
> Hence it would be also ok to just squash these into one patch. It
> doesn't make too much difference for reviews, and no difference for
> bisecting.
>
I wanted to split the shells stuff mainly because different people work(ed)
on different shells. I like the first approach.
>> ---
>> src/compositor.c | 29 ++---
>> src/compositor.h | 4
>> src/data-device.c | 2 ++
>> src/input.c | 2 ++
>> 4 files changed, 18 insertions(+), 19 deletions(-)
>
> There is also a more subtle change of semantics in this patch, which
> would be good to underline in the commit message.
>
> Previously when mappedness was determined by outputs, unmapping the
> last view unmapped also the surface, or so I believe. That no longer
> happens, which is actually a good thing. This allows a shell to unmap
> all views without losing the mappedness information of the surface.
>
> A practical example of this is hiding a desktop-shell top-level window.
> When the window is shown again, it should appear on the position where
> it was hidden, but that will not happen if desktop-shell thinks the
> surface is not mapped. Instead, desktop-shell will randomize the window
> position again. I haven't looked if this actually happens or if there
> is code to work around that, but I think the change by this patch is
> for the better.
>
> The mappedness states of a surface and its views are no longer
> implicitly connected.
>
I'm not sure I'm getting all this. We can discuss about it more on
IRC or I can just copy/paste most of the text above :).
>>
>> diff --git a/src/compositor.c b/src/compositor.c
>> index d246046..b25c671 100644
>> --- a/src/compositor.c
>> +++ b/src/compositor.c
>> @@ -1549,19 +1549,13 @@ weston_view_set_mask_infinite(struct weston_view
>> *view)
>> WL_EXPORT bool
>> weston_view_is_mapped(struct weston_view *view)
>> {
>> -if (view->output)
>> -return true;
>> -else
>> -return false;
>> +return view->is_mapped;
>> }
>>
>> WL_EXPORT bool
>> weston_surface_is_mapped(struct weston_surface *surface)
>> {
>> -if (surface->output)
>> -return true;
>> -else
>> -return false;
>> +return surface->is_mapped;
>> }
>>
>> static void
>> @@ -1738,6 +1732,7 @@ weston_view_unmap(struct weston_view *view)
>> weston_view_damage_below(view);
>> view->output = NULL;
>> view->plane = NULL;
>> +view->is_mapped = false;
>> weston_layer_entry_remove(&view->layer_link);
>> wl_list_remove(&view->link);
>> wl_list_init(&view->link);
>
> Looks like weston_view_unmap() has a problem. It won't fix up the input
> device foci unless the surface is not mapped. So even before this
> patch, if a shell unmaps a view that had pointer focus, the pointer
> focus is not removed. The same for touch. But this does not apply to
> kbd focus, because kbd focus is tracked by the surface, not by the view.
>
> Then I wonder if the sloppy handling of pointer focus is actually
> relied on somewhere, or if it's just accidental. Or both.
>
> Giulio, do you have any recommendations on what to do here?
> Since this ties in with the issue below...
>
> Fixing this is not a matter for this patch, though.
>
>> @@ -1769,6 +1764,7 @@ weston_surface_unmap(struct weston_surface *surface)
>>
>> wl_list_for_each(view, &surface->views, surface_link)
>> weston_view_unmap(view);
>> +surface->is_mapped = false;
>> surface->output = NULL;
>
> Because weston_view_unmap() is called first, and surface->is_mapped is
> set only afterwards, the input foci clean-up never runs.
>
> I think surface->is_mapped = false should be before th
Re: [PATCH weston 08/11] compositor: Untangle mapedness from outputs
On Sat, 18 Jun 2016 19:15:23 +0200
Armin Krezović wrote:
> Currently, weston assumes a surface/view is mapped if
> it has an output assigned. In a zero outputs scenario,
> this isn't really desirable.
>
> This patch introduces a new flag to weston_surface and
> weston_view, which has to be set manually to indicate
> that a surface/view is mapped.
>
> Signed-off-by: Armin Krezović
Hi Armin,
there is a slight problem with the split and ordering in patches 8 -
11. Patch 8 removes the old conditions, which breaks "everything" and
the follow-up patches fix things up piece by piece. The problem is that
we now have known-broken commits from the start. Those will make
bisecting unrelated issues harder.
There are two possible solutions:
- split patch 8 to to only add the new fields, and make a new patch as
the final one in the series that switches the
weston_{view,surface}_is_mapped() to use the new flags
- squash all patches 8 - 11 into one
Because the changes in patches 9 - 11 are fairly trivial and unlikely
to break anything in themselves, the patch that switches the
*_is_mapped() functions will be the one causing the regressions if any.
Hence it would be also ok to just squash these into one patch. It
doesn't make too much difference for reviews, and no difference for
bisecting.
> ---
> src/compositor.c | 29 ++---
> src/compositor.h | 4
> src/data-device.c | 2 ++
> src/input.c | 2 ++
> 4 files changed, 18 insertions(+), 19 deletions(-)
There is also a more subtle change of semantics in this patch, which
would be good to underline in the commit message.
Previously when mappedness was determined by outputs, unmapping the
last view unmapped also the surface, or so I believe. That no longer
happens, which is actually a good thing. This allows a shell to unmap
all views without losing the mappedness information of the surface.
A practical example of this is hiding a desktop-shell top-level window.
When the window is shown again, it should appear on the position where
it was hidden, but that will not happen if desktop-shell thinks the
surface is not mapped. Instead, desktop-shell will randomize the window
position again. I haven't looked if this actually happens or if there
is code to work around that, but I think the change by this patch is
for the better.
The mappedness states of a surface and its views are no longer
implicitly connected.
>
> diff --git a/src/compositor.c b/src/compositor.c
> index d246046..b25c671 100644
> --- a/src/compositor.c
> +++ b/src/compositor.c
> @@ -1549,19 +1549,13 @@ weston_view_set_mask_infinite(struct weston_view
> *view)
> WL_EXPORT bool
> weston_view_is_mapped(struct weston_view *view)
> {
> - if (view->output)
> - return true;
> - else
> - return false;
> + return view->is_mapped;
> }
>
> WL_EXPORT bool
> weston_surface_is_mapped(struct weston_surface *surface)
> {
> - if (surface->output)
> - return true;
> - else
> - return false;
> + return surface->is_mapped;
> }
>
> static void
> @@ -1738,6 +1732,7 @@ weston_view_unmap(struct weston_view *view)
> weston_view_damage_below(view);
> view->output = NULL;
> view->plane = NULL;
> + view->is_mapped = false;
> weston_layer_entry_remove(&view->layer_link);
> wl_list_remove(&view->link);
> wl_list_init(&view->link);
Looks like weston_view_unmap() has a problem. It won't fix up the input
device foci unless the surface is not mapped. So even before this
patch, if a shell unmaps a view that had pointer focus, the pointer
focus is not removed. The same for touch. But this does not apply to
kbd focus, because kbd focus is tracked by the surface, not by the view.
Then I wonder if the sloppy handling of pointer focus is actually
relied on somewhere, or if it's just accidental. Or both.
Giulio, do you have any recommendations on what to do here?
Since this ties in with the issue below...
Fixing this is not a matter for this patch, though.
> @@ -1769,6 +1764,7 @@ weston_surface_unmap(struct weston_surface *surface)
>
> wl_list_for_each(view, &surface->views, surface_link)
> weston_view_unmap(view);
> + surface->is_mapped = false;
> surface->output = NULL;
Because weston_view_unmap() is called first, and surface->is_mapped is
set only afterwards, the input foci clean-up never runs.
I think surface->is_mapped = false should be before the
weston_view_unmap(). That way at least weston_surface_unmap() will work
as expected. Previously, it relied on the last call to
weston_view_unmap() -> weston_surface_assign_output() to mark the
surface as not mapped via output assignment, so that the input foci
clean-up was executed once all views were unmapped.
> }
>
> @@ -2131,6 +2127,7 @@ view_list_add_subsurface_view(struct weston_compositor
> *compositor,
>
> view->parent_view = parent;
> weston_view_update_trans
[PATCH weston 08/11] compositor: Untangle mapedness from outputs
Currently, weston assumes a surface/view is mapped if
it has an output assigned. In a zero outputs scenario,
this isn't really desirable.
This patch introduces a new flag to weston_surface and
weston_view, which has to be set manually to indicate
that a surface/view is mapped.
Signed-off-by: Armin Krezović
---
src/compositor.c | 29 ++---
src/compositor.h | 4
src/data-device.c | 2 ++
src/input.c | 2 ++
4 files changed, 18 insertions(+), 19 deletions(-)
diff --git a/src/compositor.c b/src/compositor.c
index d246046..b25c671 100644
--- a/src/compositor.c
+++ b/src/compositor.c
@@ -1549,19 +1549,13 @@ weston_view_set_mask_infinite(struct weston_view *view)
WL_EXPORT bool
weston_view_is_mapped(struct weston_view *view)
{
- if (view->output)
- return true;
- else
- return false;
+ return view->is_mapped;
}
WL_EXPORT bool
weston_surface_is_mapped(struct weston_surface *surface)
{
- if (surface->output)
- return true;
- else
- return false;
+ return surface->is_mapped;
}
static void
@@ -1738,6 +1732,7 @@ weston_view_unmap(struct weston_view *view)
weston_view_damage_below(view);
view->output = NULL;
view->plane = NULL;
+ view->is_mapped = false;
weston_layer_entry_remove(&view->layer_link);
wl_list_remove(&view->link);
wl_list_init(&view->link);
@@ -1769,6 +1764,7 @@ weston_surface_unmap(struct weston_surface *surface)
wl_list_for_each(view, &surface->views, surface_link)
weston_view_unmap(view);
+ surface->is_mapped = false;
surface->output = NULL;
}
@@ -2131,6 +2127,7 @@ view_list_add_subsurface_view(struct weston_compositor
*compositor,
view->parent_view = parent;
weston_view_update_transform(view);
+ view->is_mapped = true;
if (wl_list_empty(&sub->surface->subsurface_list)) {
wl_list_insert(compositor->view_list.prev, &view->link);
@@ -2152,6 +2149,7 @@ view_list_add(struct weston_compositor *compositor,
struct weston_subsurface *sub;
weston_view_update_transform(view);
+ view->is_mapped = true;
if (wl_list_empty(&view->surface->subsurface_list)) {
wl_list_insert(compositor->view_list.prev, &view->link);
@@ -3244,7 +3242,6 @@ subsurface_get_label(struct weston_surface *surface, char
*buf, size_t len)
static void
subsurface_configure(struct weston_surface *surface, int32_t dx, int32_t dy)
{
- struct weston_compositor *compositor = surface->compositor;
struct weston_view *view;
wl_list_for_each(view, &surface->views, surface_link)
@@ -3256,8 +3253,9 @@ subsurface_configure(struct weston_surface *surface,
int32_t dx, int32_t dy)
* mapped, parent is not in a visible layer, so this sub-surface
* will not be drawn either.
*/
- if (!weston_surface_is_mapped(surface)) {
- struct weston_output *output;
+
+ if (!weston_surface_is_mapped(surface))
+ surface->is_mapped = true;
/* Cannot call weston_view_update_transform(),
* because that would call it also for the parent surface,
@@ -3265,18 +3263,11 @@ subsurface_configure(struct weston_surface *surface,
int32_t dx, int32_t dy)
* inconsistent state, where the window could never be
* mapped.
*
-* Instead just assign any output, to make
+* Instead just force the is_mapped flag on, to make
* weston_surface_is_mapped() return true, so that when the
* parent surface does get mapped, this one will get
* included, too. See view_list_add().
*/
- assert(!wl_list_empty(&compositor->output_list));
- output = container_of(compositor->output_list.next,
- struct weston_output, link);
-
- surface->output = output;
- weston_surface_update_output_mask(surface, 1u << output->id);
- }
}
static struct weston_subsurface *
diff --git a/src/compositor.h b/src/compositor.h
index 9e5155c..8770982 100644
--- a/src/compositor.h
+++ b/src/compositor.h
@@ -964,6 +964,8 @@ struct weston_view {
/* Per-surface Presentation feedback flags, controlled by backend. */
uint32_t psf_flags;
+
+ bool is_mapped;
};
struct weston_surface_state {
@@ -1082,6 +1084,8 @@ struct weston_surface {
const char *role_name;
struct weston_timeline_object timeline;
+
+ bool is_mapped;
};
struct weston_subsurface {
diff --git a/src/data-device.c b/src/data-device.c
index f04f030..44a08f9 100644
--- a/src/data-device.c
+++ b/src/data-device.c
@@ -421,6 +421,8 @@ drag_surface_configure(struct weston_drag *drag,
weston_layer_entry_inse
