profile/ivi/weston.git
12 years agoMerge remote-tracking branch 'pq/transform-v1'
Kristian Høgsberg [Fri, 27 Jan 2012 16:42:18 +0000 (11:42 -0500)]
Merge remote-tracking branch 'pq/transform-v1'

12 years agodesktop-shell: print what failed on exec
Pekka Paalanen [Fri, 27 Jan 2012 07:50:02 +0000 (09:50 +0200)]
desktop-shell: print what failed on exec

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agowindow: fix resizing of windows backed by shm buffers
Ander Conselvan de Oliveira [Fri, 27 Jan 2012 15:17:39 +0000 (17:17 +0200)]
window: fix resizing of windows backed by shm buffers

When window_attach_surface() calls window_get_resize_dx_dy(),
window->resize_edges is cleared. However if there is already a pending
surface to be attached, the resize won't be done until the following
call to window_attach_surface(). In this next call, since resize_edges
is now zero, the top-left corner of the window will be unchanged. If
the user is resizing from the top or left border, this causes the
resize to happen in the wrong direction.

This patch changes window_attach_surface() to call
window_get_resize_dx_dy() only if an attach will actually happen.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
12 years agowindow: Handle data_device.selection events with offer=NULL
Kristian Høgsberg [Fri, 27 Jan 2012 16:04:18 +0000 (11:04 -0500)]
window: Handle data_device.selection events with offer=NULL

This happens when the current selection data_offer is removed.

12 years agoCollabora copyright updates
Pekka Paalanen [Fri, 27 Jan 2012 14:25:16 +0000 (16:25 +0200)]
Collabora copyright updates

which I forgot to add while working on these files.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: make overlap computation use the boundingbox
Pekka Paalanen [Fri, 27 Jan 2012 12:38:33 +0000 (14:38 +0200)]
compositor: make overlap computation use the boundingbox

This makes the overlap to account for surface transformations.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: use bounding box for damage regions
Pekka Paalanen [Thu, 26 Jan 2012 11:12:45 +0000 (13:12 +0200)]
compositor: use bounding box for damage regions

Change weston_surface_damage*() functions to use the full surface
bounding box or call surface_compute_bbox() to find the bounding box for
an arbitrary rectangle.

This should fix all rendering artifacts for non-opaque (i.e. ARGB)
transformed surfaces.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: weston_surface_assign_output() to use bounding box
Pekka Paalanen [Thu, 26 Jan 2012 09:34:16 +0000 (11:34 +0200)]
compositor: weston_surface_assign_output() to use bounding box

Use the bounding box to compute an approximation of which output
contains most of the surface.

Move the region32 init outside the loop, and fini it, too.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: weston_output_set_cursor() to use bounding box
Pekka Paalanen [Thu, 26 Jan 2012 09:31:01 +0000 (11:31 +0200)]
compositor: weston_output_set_cursor() to use bounding box

If we ever have transformed cursor surfaces, we would better use the
bounding box to check if it is on the given output.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: weston_surface_draw() to use bounding box
Pekka Paalanen [Thu, 26 Jan 2012 09:28:08 +0000 (11:28 +0200)]
compositor: weston_surface_draw() to use bounding box

Use the proper bounding box in clipping the surface repaint area. Fixes
excessive clipping for transformed surfaces.

Also don't leak the region32 on early return.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: move weston_surface::width,height into geometry
Pekka Paalanen [Wed, 25 Jan 2012 13:55:43 +0000 (15:55 +0200)]
compositor: move weston_surface::width,height into geometry

weston_surface::transform.boundingbox depends on width and height, and
therefore geometry.dirty flag, so move width and height into geometry.

Fix all users and check that the dirty flag is set.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: add weston_surface bounding box
Pekka Paalanen [Wed, 25 Jan 2012 13:17:40 +0000 (15:17 +0200)]
compositor: add weston_surface bounding box

Compute a surface bounding box, especially for transformed surfaces, for
which one cannot simply use x,y,width,height.

The bounding box depends on width and height, so these are now under the
geometry.dirty flag.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: fix zoom origin
Pekka Paalanen [Wed, 25 Jan 2012 12:45:18 +0000 (14:45 +0200)]
compositor: fix zoom origin

Now that we can insert a transformation before the surface position
translation, we can drop geometry.x,y from the zoom transformation. That
was just undoing and redoing the position translation.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor-drm: do not scan out transformed surfaces
Pekka Paalanen [Wed, 25 Jan 2012 12:33:33 +0000 (14:33 +0200)]
compositor-drm: do not scan out transformed surfaces

Not sure this check belongs here, but as the position checks are here
too, I added this. Just so we don't forget.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: move weston_surface::x,y into geometry
Pekka Paalanen [Wed, 25 Jan 2012 14:22:05 +0000 (16:22 +0200)]
compositor: move weston_surface::x,y into geometry

weston_surface::transform.position depends on x,y, and therefore the
dirty flag, so move x and y into geometry.

Also add the missing dirty flags.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: make position a surface transformation
Pekka Paalanen [Wed, 25 Jan 2012 11:37:39 +0000 (13:37 +0200)]
compositor: make position a surface transformation

Put the surface translation (absolute position) into the surface
transformations list. This allows to set additional transformations
before and after the global translation.

Having the translation cached, changing the surface x,y now requires to
set the geometry.dirty flag.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: honour repaint regions with transformed surfaces
Pekka Paalanen [Tue, 24 Jan 2012 12:47:37 +0000 (14:47 +0200)]
compositor: honour repaint regions with transformed surfaces

Previously, if a surface was transformed, it repainted as a whole,
regardless of the computed repaint region. As damage regions determine
repaint regions and whether a surface is considered for drawing at all,
this lead to disappearing surfaces if all surfaces were considered
transformed. Also transparent transformed surfaces were redrawn without
the surfaces below being properly redrawn, leading to alpha-saturation.

Fix that by making texture_region() use the proper global-to-surface
coordinate transformation for texture coordinates. This makes it
possible to call texture_region() also for transformed surfaces.

As texture coordinates may now lie outside the valid texture image, the
fragment shader is modified to check the fragment texture coordinates.

The special path texture_transformed_surface() is no longer used and is
removed.

This change fixes many of the rendering artifacts related to transformed
surfaces.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: disable attrib arrays after use
Pekka Paalanen [Tue, 24 Jan 2012 13:53:35 +0000 (15:53 +0200)]
compositor: disable attrib arrays after use

Not strictly necessary right now, but nice to clean up.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: remove unused *_uniform members
Pekka Paalanen [Tue, 24 Jan 2012 12:50:14 +0000 (14:50 +0200)]
compositor: remove unused *_uniform members

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: rewrite draw and input coordinate transformations
Pekka Paalanen [Tue, 24 Jan 2012 07:59:29 +0000 (09:59 +0200)]
compositor: rewrite draw and input coordinate transformations

For unifying the coordinate system handling, introduce functions for
converting explicitly between the global and the surface local
coordinate systems.

Use these functions in the input path, replacing
weston_surface_transform().

In the draw path, rewrite transform_vertex() to take in surface local
coordinates.

As shell now uses the new functions, the rotation origin is properly
placed in the middle of the surface.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: restructure weston_surface::transform
Pekka Paalanen [Tue, 24 Jan 2012 07:53:37 +0000 (09:53 +0200)]
compositor: restructure weston_surface::transform

Separate mutable data and cached immutable data in struct
weston_surface.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: handle non-invertible surface transformations
Pekka Paalanen [Mon, 23 Jan 2012 08:02:47 +0000 (10:02 +0200)]
compositor: handle non-invertible surface transformations

Detect a non-invertible surface total transformation, disable it, and
warn about it.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoshell: add key binding for rotating a surface
Pekka Paalanen [Fri, 20 Jan 2012 14:48:25 +0000 (16:48 +0200)]
shell: add key binding for rotating a surface

Add the key binding Super+Alt+MouseLeftButton to start rotating a
surface by dragging. The rotation is removed, when the drag is near the
rotation origin.

Rotated surface are a stress test for input event coordinate
transformations, damage region tracking, draw transformations, and
window move and resize orientation.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: apply full transformation to input coordinates
Pekka Paalanen [Fri, 20 Jan 2012 12:24:25 +0000 (14:24 +0200)]
compositor: apply full transformation to input coordinates

When converting input coordinates from global to surface-local system,
apply the full inverse surface transformation instead of just
translation.

Move weston_surface_update_transform() implementation realier in the
file, no changes.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoclickdot: implement the purpose
Pekka Paalanen [Fri, 20 Jan 2012 11:04:56 +0000 (13:04 +0200)]
clickdot: implement the purpose

Remove all unneeded resizor features, and add the feature why clickdot
exists: put a visible marker to exactly where mouse was clicked.

This app can be used to check input coordinate transformations in a
compositor.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoclickdot: a copy of resizor
Pekka Paalanen [Fri, 20 Jan 2012 09:09:13 +0000 (11:09 +0200)]
clickdot: a copy of resizor

Start a new application clickdot as a copy of resizor, with the name
changed.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: simplify the matrix inversion API
Pekka Paalanen [Fri, 20 Jan 2012 08:47:57 +0000 (10:47 +0200)]
compositor: simplify the matrix inversion API

The compositor will likely do an order of magnitude less matrix
inversions than point transformations with an inverse, hence we do not
really need the optimised path for single-shot invert-and-transform.

Expose only the computing of the explicit inverse matrix in the API.

However, the matrix inversion tests need access to the internal
functions. Designate a unit test build by #defining UNIT_TEST, and
export the internal functions in that case.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoupdate git ignores
Pekka Paalanen [Mon, 16 Jan 2012 13:38:17 +0000 (15:38 +0200)]
update git ignores

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agotests: add matrix-test
Pekka Paalanen [Mon, 16 Jan 2012 13:04:28 +0000 (15:04 +0200)]
tests: add matrix-test

Add a new directory tests/ for unit test applications. This directory
will be built only if --enable-tests is given to ./configure.

Add matrix-test application. It excercises especially the
weston_matrix_invert() and weston_matrix_inverse_transform() functions.
It has one test for correctness and precision, and other tests for
measuring the speed of various matrix operations.

For the record, the correctness test prints:

a random matrix:
   1.112418e-02   2.628150e+00   8.205844e+02  -1.147526e-04
   4.943677e-04  -1.117819e-04  -9.158849e-06   3.678122e-02
   7.915063e-03  -3.093254e-04  -4.376583e+02   3.424706e-02
  -2.504038e+02   2.481788e+03  -7.545445e+01   1.752909e-03

The matrix multiplied by its inverse, error:
   0.000000e+00  -0.000000e+00  -0.000000e+00  -0.000000e+00
   0.000000e+00   0.000000e+00   0.000000e+00   0.000000e+00
  -0.000000e+00  -0.000000e+00   0.000000e+00  -0.000000e+00
   0.000000e+00   0.000000e+00   0.000000e+00   0.000000e+00
max abs error: 0, original determinant 11595.2

Running a test loop for 10 seconds...
test fail, det: -0.00464805, error sup: inf
test fail, det: -0.0424053, error sup: 1.30787e-06
test fail, det: 5.15191, error sup: 1.15956e-06
tests: 6791767 ok, 1 not invertible but ok, 3 failed.
Total: 6791771 iterations.

These results are expected with the current precision thresholds in
src/matrix.c and tests/matrix-test.c. The random number generator is
seeded with a constant, so the random numbers should be the same on
every run. Machine speed and scheduling affect how many iterations are
run.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: implement inverse matrix transformation
Pekka Paalanen [Mon, 16 Jan 2012 12:27:00 +0000 (14:27 +0200)]
compositor: implement inverse matrix transformation

Implement 4x4 matrix inversion based on LU-decomposition with partial
pivoting.

Instead of simply computing the inverse matrix explicitly, introduce the
type struct weston_inverse_matrix for storing the LU-decomposition and
the permutation from pivoting. Using doubles, this struct has greater
precision than struct weston_matrix.

If you need only few (less than 5, presumably) multiplications with the
inverse matrix, is it cheaper to use weston_inverse_matrix, and not
compute the inverse matrix explicitly into a weston_matrix.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: drop inverse matrix from weston_transform
Pekka Paalanen [Thu, 12 Jan 2012 13:00:57 +0000 (15:00 +0200)]
compositor: drop inverse matrix from weston_transform

Remove the inverse matrix member from struct weston_transform. It is
easier (and probably faster, too) to create and store only forward
transformation matrices in a list, multiply them once, and then invert
the final matrix, rather than creating both forward and inverse
matrices, and multiplying both.

Add a stub for the 4x4 matrix inversion function.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: move matrix code to separate files
Pekka Paalanen [Thu, 12 Jan 2012 12:30:47 +0000 (14:30 +0200)]
compositor: move matrix code to separate files

Move matrix code to separate files to allow writing unit tests for it.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoutil: document matrices
Pekka Paalanen [Tue, 10 Jan 2012 13:40:18 +0000 (15:40 +0200)]
util: document matrices

Add comments explaining the matrix storage and multiplication, so that
no-one else needs to decipher them again.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: implement a stack of surface transformations
Pekka Paalanen [Fri, 6 Jan 2012 12:10:06 +0000 (14:10 +0200)]
compositor: implement a stack of surface transformations

Having at most one transformation object attached to a surface is not
enough anymore. If we have a surface that needs to be scaled to
fullscreen, and then we have the zoom animation, we already need two
transformations combined.

Implement support for multiple transformations by adding a transformation
list. The final transformation is the ordered composite of those in the
list. To avoid traversing the list every single time, add a dirty flag,
and cache the final transformation.

The existing transformation users (only zoom) are converted.

Note: surface drawing should honour all kinds of transformations, but
not damage region code nor input event translating code take
transformations into account, AFAICT. Therefore anything but translation
will probably behave badly until they are fixed.

Cc: Juan Zhao <juan.j.zhao@linux.intel.com>
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocairo-util: Don't anger png.h by including setjmp.h
Kristian Høgsberg [Thu, 26 Jan 2012 16:32:30 +0000 (11:32 -0500)]
cairo-util: Don't anger png.h by including setjmp.h

Older libpngs are a bit fussy about this, apparently.

12 years agowindow: Don't crash when window with focus_widget is destroyed
Kristian Høgsberg [Thu, 26 Jan 2012 16:09:20 +0000 (11:09 -0500)]
window: Don't crash when window with focus_widget is destroyed

We would try to deliver a leave event to a widget in a destroyed window.

12 years agocompositor: Keep fade surface around while we're fading
Kristian Høgsberg [Thu, 26 Jan 2012 15:55:10 +0000 (10:55 -0500)]
compositor: Keep fade surface around while we're fading

This removes more special cases from weston_output_repaint() and we
avoid creating and destroying the surface for each animation frame.
We gain another special case in weston_compositor_top(), but that's
less of a problem, and we'll fix that later.

12 years agocompositor: Use weston_surface_create for creating solid color fade surface
Kristian Høgsberg [Thu, 26 Jan 2012 15:03:45 +0000 (10:03 -0500)]
compositor: Use weston_surface_create for creating solid color fade surface

12 years agocompositor: Create surfaces neutral, select shader later
Kristian Høgsberg [Thu, 26 Jan 2012 15:00:23 +0000 (10:00 -0500)]
compositor: Create surfaces neutral, select shader later

We only setup the shader and create the texture once we attach a buffer
or set a color.

12 years agocompositor: Drop unused saved_texture surface field
Kristian Høgsberg [Thu, 26 Jan 2012 14:28:42 +0000 (09:28 -0500)]
compositor: Drop unused saved_texture surface field

12 years agocompositor: Clear repaint_scheduled when an idle repaint is cancelled
Kristian Høgsberg [Thu, 26 Jan 2012 13:47:04 +0000 (08:47 -0500)]
compositor: Clear repaint_scheduled when an idle repaint is cancelled

We don't hit this case often, but if we did, the compositor would get stuck
thinking a repaint was already scheduled.

12 years agocompositor: Combine repaint and weston_output_repaint
Kristian Høgsberg [Thu, 26 Jan 2012 13:40:37 +0000 (08:40 -0500)]
compositor: Combine repaint and weston_output_repaint

Fewer repaint functions is better.

12 years agocompositor: Compute overlapped early and base hw cursor decision on that
Kristian Høgsberg [Thu, 26 Jan 2012 06:03:58 +0000 (01:03 -0500)]
compositor: Compute overlapped early and base hw cursor decision on that

12 years agocompositor: Move tracking of scanout buffers to compositor-drm
Kristian Høgsberg [Thu, 26 Jan 2012 05:11:01 +0000 (00:11 -0500)]
compositor: Move tracking of scanout buffers to compositor-drm

12 years agocompositor: Remove prepare_scanout_surface callout
Kristian Høgsberg [Thu, 26 Jan 2012 04:59:42 +0000 (23:59 -0500)]
compositor: Remove prepare_scanout_surface callout

For the drm backend, we just refactor setup_scanout_surface into
drm_output_prepare_scanout_surface and call it from drm_output_repaint.

12 years agocompositor: Pull prepare_render and present callouts into repaint
Kristian Høgsberg [Thu, 26 Jan 2012 04:47:45 +0000 (23:47 -0500)]
compositor: Pull prepare_render and present callouts into repaint

This dramatically simplifies the backend repaint abstractions and paves
the way for moving overlay and cursor setup into the backend.

12 years agocompositor: Move repaint loop into a backend function
Kristian Høgsberg [Thu, 26 Jan 2012 04:32:28 +0000 (23:32 -0500)]
compositor: Move repaint loop into a backend function

We've trimmed down the actual repaint loop to just iterating through the
surface list and calling weston_surface_draw(), so we push that to the
backend without too much code duplication.

12 years agocompositor: Insert the fade surface in the surface list
Kristian Høgsberg [Thu, 26 Jan 2012 04:16:29 +0000 (23:16 -0500)]
compositor: Insert the fade surface in the surface list

This lets us repaint it in the repaint loop instead of the fade_output()
special case.

12 years agocompositor: Move shader setup to weston_surface_draw()
Kristian Høgsberg [Thu, 26 Jan 2012 04:02:06 +0000 (23:02 -0500)]
compositor: Move shader setup to weston_surface_draw()

12 years agocompositor: Compute whether or not a surface is overlapped
Kristian Høgsberg [Thu, 26 Jan 2012 03:20:30 +0000 (22:20 -0500)]
compositor: Compute whether or not a surface is overlapped

12 years agocompositor: Remove special casing of fullscreen surfaces
Kristian Høgsberg [Thu, 26 Jan 2012 03:09:10 +0000 (22:09 -0500)]
compositor: Remove special casing of fullscreen surfaces

This was supposed to draw black borders around a fullscreen surface that
was smaller than the output.  We don't want to special case that in the
repaint loop, but may use a different shader or such.  And we want the
surface to have an opaque region that covers the output so that that
will eliminate overdraw of lower surfaces.

12 years agocompositor: Do the per surface clip region math in weston_surface_draw()
Kristian Høgsberg [Thu, 26 Jan 2012 03:02:00 +0000 (22:02 -0500)]
compositor: Do the per surface clip region math in weston_surface_draw()

12 years agocompositor: Assign surface damage before setting up scanout buffers
Kristian Høgsberg [Thu, 26 Jan 2012 02:48:26 +0000 (21:48 -0500)]
compositor: Assign surface damage before setting up scanout buffers

12 years agodesktop-shell: Build in sensible defaults
Kristian Høgsberg [Wed, 25 Jan 2012 21:57:11 +0000 (16:57 -0500)]
desktop-shell: Build in sensible defaults

This adds a default background pattern and a terminal launcher in case
we don't have a config file.

12 years agodesktop-shell: Support tiling of background images
Kristian Høgsberg [Wed, 25 Jan 2012 21:34:36 +0000 (16:34 -0500)]
desktop-shell: Support tiling of background images

12 years agocairo-util: Add png loader and image type sniffing
Kristian Høgsberg [Wed, 25 Jan 2012 20:43:48 +0000 (15:43 -0500)]
cairo-util: Add png loader and image type sniffing

12 years agodesktop-shell: Fall back to solid color if there's no background image
Kristian Høgsberg [Wed, 25 Jan 2012 19:55:33 +0000 (14:55 -0500)]
desktop-shell: Fall back to solid color if there's no background image

12 years agocairo-util: Return NULL instead of exit() on jpeg load failure
Kristian Høgsberg [Wed, 25 Jan 2012 19:54:26 +0000 (14:54 -0500)]
cairo-util: Return NULL instead of exit() on jpeg load failure

12 years agoInstall terminal as weston-terminal
Kristian Høgsberg [Wed, 25 Jan 2012 19:02:05 +0000 (14:02 -0500)]
Install terminal as weston-terminal

12 years agoRevert "weston: Drop priviledges early, and seteuid when needed"
Kristian Høgsberg [Tue, 24 Jan 2012 17:37:17 +0000 (12:37 -0500)]
Revert "weston: Drop priviledges early, and seteuid when needed"

This reverts commit fc6ccb868fa735ee9c6592806f381aa1262bf0b2.

We still need root permissions for drmDrop/SetMaster.  Without
integration with ConsoleKit or systemd we also don't have access
to /dev/dri/cardX in the case where we open a new VT.

12 years agoweston: Drop priviledges early, and seteuid when needed
Benjamin Franzke [Tue, 24 Jan 2012 15:37:15 +0000 (16:37 +0100)]
weston: Drop priviledges early, and seteuid when needed

12 years agosetuid: Operate relative to $(DESTDIR)
Benjamin Franzke [Tue, 24 Jan 2012 09:29:44 +0000 (10:29 +0100)]
setuid: Operate relative to $(DESTDIR)

12 years agowindow: Fix sending button events when there's no grab
Neil Roberts [Mon, 23 Jan 2012 19:11:18 +0000 (19:11 +0000)]
window: Fix sending button events when there's no grab

The code which sends the button events was checking whether there is a
focus widget with a button handler but then always sending the button
event to the grab widget. If the grab widget is different from the
focus widget at this point then it will check the wrong widget for a
button handler and potentially crash. It is also possible for there to
be no grab widget here in the following situation:

1. Press and hold down the left mouse button
2. Press and hold down the right mouse button
3. Release the left mouse button
4. Release the right mouse button

In this case the grab will be released at step 3 because the code only
keeps track of the grab for one button. Then it will try to send the
release event for the right mouse button to a NULL widget so it will
crash.

12 years agoevdev: fix uninitialised pointer value
Tiago Vignatti [Fri, 20 Jan 2012 16:47:46 +0000 (18:47 +0200)]
evdev: fix uninitialised pointer value

==30224== Conditional jump or move depends on uninitialised value(s)
==30224==    at 0x40EE3A0: evdev_flush_motion (evdev.c:284)
==30224==    by 0x40EE6DC: evdev_input_device_data (evdev.c:352)
==30224==    by 0x4034710: wl_event_source_fd_dispatch (event-loop.c:76)
==30224==    by 0x4035171: wl_event_loop_dispatch (event-loop.c:462)
==30224==    by 0x4032F76: wl_display_run (wayland-server.c:785)
==30224==    by 0x8050972: main (compositor.c:2183)

Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
12 years agogears: Make window transparent
Kristian Høgsberg [Thu, 19 Jan 2012 19:05:21 +0000 (14:05 -0500)]
gears: Make window transparent

We need this for the window decorations.

12 years agowindow: Use CAIRO_OPERATOR_OVER for rendering the menu
Kristian Høgsberg [Thu, 19 Jan 2012 18:54:09 +0000 (13:54 -0500)]
window: Use CAIRO_OPERATOR_OVER for rendering the menu

We forgot to switch back from CAIRO_OPERATOR_SOURCE, which is broken,
but also hits a broken fallback in cairo-gl.

12 years agoconfigure: Add --enable-setuid-install
Kristian Høgsberg [Thu, 19 Jan 2012 18:32:17 +0000 (13:32 -0500)]
configure: Add --enable-setuid-install

This adds support for installing the compositor setuid.

12 years agocompositor: initialise segv_action.sa_mask
Pekka Paalanen [Thu, 19 Jan 2012 14:48:37 +0000 (16:48 +0200)]
compositor: initialise segv_action.sa_mask

Valgrind complained sa_mask member is not initialised, fix that.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agodesktop-shell: don't crash on menu
Pekka Paalanen [Thu, 19 Jan 2012 14:40:28 +0000 (16:40 +0200)]
desktop-shell: don't crash on menu

Add dummy menu callback function, so we don't crash after every time
someone opens the menu.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoshell: deal with weston_compositor_pick_surface() = NULL
Pekka Paalanen [Thu, 19 Jan 2012 14:25:40 +0000 (16:25 +0200)]
shell: deal with weston_compositor_pick_surface() = NULL

I could crash Weston by trying to open another menu from a panel while
one menu from it was already showing.

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff40a9872 in popup_grab_focus (grab=0x761968, time=4130706528, surface=0x0, x=-227, y=15) at shell.c:440
440 if (surface->resource.client == client) {
(gdb) bt
 0  0x00007ffff40a9872 in popup_grab_focus (grab=0x761968, time=4130706528, surface=0x0, x=-227, y=15) at shell.c:440
 1  0x0000000000406977 in weston_device_repick (device=0x70b4e0, time=4130706528) at compositor.c:360
 2  0x0000000000406a36 in weston_compositor_repick (compositor=0x619960) at compositor.c:382
 3  0x0000000000406ac8 in destroy_surface (resource=0x6fc6f0) at compositor.c:397
 4  0x00007ffff7bd33d8 in destroy_resource (element=0x6fc6f0, data=0x7fffffffd9fc) at wayland-server.c:355
 5  0x00007ffff7bd8d98 in for_each_helper (entries=0x757808, func=0x7ffff7bd332c <destroy_resource>, data=0x7fffffffd9fc)
    at wayland-util.c:264
 6  0x00007ffff7bd8dd4 in wl_map_for_each (map=0x757808, func=0x7ffff7bd332c <destroy_resource>, data=0x7fffffffd9fc)
    at wayland-util.c:270
 7  0x00007ffff7bd34dc in wl_client_destroy (client=0x7577d0) at wayland-server.c:385
 8  0x00007ffff7bd2e36 in wl_client_connection_data (fd=17, mask=1, data=0x7577d0) at wayland-server.c:187
 9  0x00007ffff7bd5bde in wl_event_source_fd_dispatch (source=0x74cda0, ep=0x7fffffffdae0) at event-loop.c:76
 10 0x00007ffff7bd665b in wl_event_loop_dispatch (loop=0x618900, timeout=-1) at event-loop.c:462
 11 0x00007ffff7bd42a9 in wl_display_run (display=0x6188b0) at wayland-server.c:785
 12 0x000000000040b1e1 in main (argc=1, argv=0x7fffffffdef8) at compositor.c:2182

Modify popup_grab_focus() to deal with a NULL surface.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agowindow: menu leak fixes
Pekka Paalanen [Thu, 19 Jan 2012 13:17:59 +0000 (15:17 +0200)]
window: menu leak fixes

When a menu self-destructs, free also the widget and struct menu.

As menus are self-destructing, it does not make sense to store the
window pointer, since we cannot clear it automatically. Therefore,
rename window_create_menu() to window_show_menu() that does not return
the window pointer. It also calls window_schedule_redraw() internally.

Fixes Valgrind reported memory leaks.

The alternative would be to explicitly destroy the menu in application's
menu callback.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoresizor: free the widget on destroy
Pekka Paalanen [Thu, 19 Jan 2012 12:21:35 +0000 (14:21 +0200)]
resizor: free the widget on destroy

Plugs a memory leak.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agowindow: remove pointers to widget on destroy
Pekka Paalanen [Thu, 19 Jan 2012 11:51:38 +0000 (13:51 +0200)]
window: remove pointers to widget on destroy

Input devices may hold a pointer to the widget being destroyed. Reset
such pointers in widget_destroy().

This fixes a use-after-free in window_destroy(), if an application
destroys its widgets before the window.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agowindow: destroy frame in window_destroy()
Pekka Paalanen [Thu, 19 Jan 2012 11:33:50 +0000 (13:33 +0200)]
window: destroy frame in window_destroy()

Fix a memory leak reported by Valgrind, by destroying the window
decorations widget, if it exists.

All widget pointers returned from toytoolkit to the application should
be destroyed by the application explicitly.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: Flush surface damage to surface below on destroy
Kristian Høgsberg [Thu, 19 Jan 2012 02:41:37 +0000 (21:41 -0500)]
compositor: Flush surface damage to surface below on destroy

When we destroy a surface, we damage the surface below so that the area
exposed by the disappearing surface will be repainted.  However, if that
surface also is destroyed, the damage information is lost and we fail to
repaint that area.

This commit introduces weston_surface_flush_damage(), which flushes the
surface damage the the surface below when a surface is destroyed.  When
multiple surfaces are destroyed at the same time, the damage now accumulates
and sinks down through the surface stack as it should.

12 years agodrm: Don't print warning if cursor stride isn't right
Kristian Høgsberg [Wed, 18 Jan 2012 20:11:07 +0000 (15:11 -0500)]
drm: Don't print warning if cursor stride isn't right

gbm_bo_create_from_egl_image() should catch this based on the
GBM_BO_USE_CURSOR_64X64 flag.  It currently doesn't so we end up with
a cursor buffer with invalid stride.

12 years agoclients: Allow compiling with the cairo glesv2 backend
Kristian Høgsberg [Wed, 18 Jan 2012 19:50:58 +0000 (14:50 -0500)]
clients: Allow compiling with the cairo glesv2 backend

This disables gears and wscreensaver, which use full GL.

12 years agodesktop-shell: fix the unlock dialog decorations
Pekka Paalanen [Wed, 18 Jan 2012 14:51:30 +0000 (16:51 +0200)]
desktop-shell: fix the unlock dialog decorations

The decorations were not responding, could not move or resize the
dialog, and it painted with a black shade.

When the dialog is created, schedule a resize instead of a repaint. The
resize will initialise the widget tree, and let everything draw and work
as it should.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agotty: Use SIGUSR1 for both enter and leave signals
Kristian Høgsberg [Wed, 18 Jan 2012 17:42:16 +0000 (12:42 -0500)]
tty: Use SIGUSR1 for both enter and leave signals

Now that we track has_vt, we can use the same handler for both enter and leave.

12 years agotty: Switch back to launch tty if we switched away
Kristian Høgsberg [Wed, 18 Jan 2012 17:38:33 +0000 (12:38 -0500)]
tty: Switch back to launch tty if we switched away

12 years agodrm: Cancel any scheduled repaints in the leave vt handler
Kristian Høgsberg [Wed, 18 Jan 2012 16:50:31 +0000 (11:50 -0500)]
drm: Cancel any scheduled repaints in the leave vt handler

If we don't cancel the repaint, we end up pointlessly redrawing the output.
What's worse is that pageflipping to the new buffer eventually fails and
we miss the finish_frame callback, leaving the compositor stuck when we
re-enter the vt.

12 years agodrm: Reset kms mode on vt enter
Kristian Høgsberg [Wed, 18 Jan 2012 16:48:46 +0000 (11:48 -0500)]
drm: Reset kms mode on vt enter

Normally the repaint will trigger a pageflip, which flips back to our
fb, but that doesn't work if the kms output has been turned off or
set to a different mode.

12 years agodrm: Complain if pageflip fails
Kristian Høgsberg [Wed, 18 Jan 2012 16:47:41 +0000 (11:47 -0500)]
drm: Complain if pageflip fails

We still don't handle the error in any way, but there's not much we can do.
It's typically a case of not having drm master."

12 years agoshell: Set desktop-shell deathstamp at initial launch
Kristian Høgsberg [Tue, 17 Jan 2012 16:07:42 +0000 (11:07 -0500)]
shell: Set desktop-shell deathstamp at initial launch

Timestamp 0 is a valid timestamp and we can't use that as an uninitialized
timestamp.

12 years agowscreensaver: get decorations back
Pekka Paalanen [Tue, 17 Jan 2012 14:10:41 +0000 (16:10 +0200)]
wscreensaver: get decorations back

In the great toytoolkit widget revolution, wscreensaver demo mode lost
its window decorations. Bring them back.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agoshell: respawn desktop-shell if it dies
Pekka Paalanen [Tue, 17 Jan 2012 12:36:27 +0000 (14:36 +0200)]
shell: respawn desktop-shell if it dies

If the desktop-shell client goes away for any reason, respawn it. To
avoid harmful looping, limit the respawning to 5 times within 30
seconds, and then give up.

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agotty: Open a new vt if not running on a VT
Kristian Høgsberg [Tue, 17 Jan 2012 03:42:22 +0000 (22:42 -0500)]
tty: Open a new vt if not running on a VT

This is typically when launcing from a pty such as an X (or Wayland)
terminal or from an ssh session.  Opening a new vt typically requires root
priviledges, so weston must be setuid root or laucnhed as root for this
to work.

12 years agowindow: remove duplicate widget_resize_handler_t
Pekka Paalanen [Thu, 12 Jan 2012 12:52:09 +0000 (14:52 +0200)]
window: remove duplicate widget_resize_handler_t

Fixes the following build failure:

In file included from window.c:58:0:
window.h:194:16: error: redefinition of typedef 'widget_resize_handler_t'
window.h:178:16: note: previous declaration of 'widget_resize_handler_t' was here

Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
12 years agocompositor: Install a SEGV handler to try and clean up the VT if we crash
Kristian Høgsberg [Mon, 16 Jan 2012 16:53:54 +0000 (11:53 -0500)]
compositor: Install a SEGV handler to try and clean up the VT if we crash

12 years agotty: Fix order of VT ack/release ioctls and the vt handler
Kristian Høgsberg [Mon, 16 Jan 2012 15:57:14 +0000 (10:57 -0500)]
tty: Fix order of VT ack/release ioctls and the vt handler

We can only set up once we've acquired the VT and we shouldn't release the
VT until we've cleaned up.  Before we would release the VT first, and then
race to drop drm master as X tried to get drm master.  Which would kill X.

12 years agoevdev: Drop noisy debug messages on vt swtich
Kristian Høgsberg [Mon, 16 Jan 2012 15:54:07 +0000 (10:54 -0500)]
evdev: Drop noisy debug messages on vt swtich

12 years agocompositor-drm: Exit if we failed to get drm master
Kristian Høgsberg [Mon, 16 Jan 2012 15:52:52 +0000 (10:52 -0500)]
compositor-drm: Exit if we failed to get drm master

We can do anything without drm master.  Also log an error in case we
fail to drop it, which will be trouble for whoever we're switching to.

12 years agoxserver: Create WM window, handle client messages
Kristian Høgsberg [Mon, 16 Jan 2012 03:16:11 +0000 (22:16 -0500)]
xserver: Create WM window, handle client messages

12 years agoevdev: Correct warning on missing input device
Olivier Le Thanh Duong [Sat, 14 Jan 2012 14:57:34 +0000 (15:57 +0100)]
evdev: Correct warning on missing input device

12 years agoxserver: Just dup fd instead of two fcntls to get a non-O_CLOEXEC fd
Kristian Høgsberg [Sun, 15 Jan 2012 21:05:28 +0000 (16:05 -0500)]
xserver: Just dup fd instead of two fcntls to get a non-O_CLOEXEC fd

Yeah, it's the same number of lines of code, but this actually checks
the error return.

12 years agoxserver: Set DISPLAY for clients launched by the compositor
Kristian Høgsberg [Sun, 15 Jan 2012 21:04:33 +0000 (16:04 -0500)]
xserver: Set DISPLAY for clients launched by the compositor

12 years agoxserver: Don't set -logfile option, only available to root user
Kristian Høgsberg [Sun, 15 Jan 2012 20:51:12 +0000 (15:51 -0500)]
xserver: Don't set -logfile option, only available to root user

12 years agotty: Leave tty in KD_GRAPHICS mode during vt switches
Kristian Høgsberg [Sun, 15 Jan 2012 20:17:58 +0000 (15:17 -0500)]
tty: Leave tty in KD_GRAPHICS mode during vt switches

Part of the point of KD_GRAPHICS mode is that the kernel doesn't try
to restore the VT contents when we switch bach, but leaves that
to the user mode process.  This avoids ugly flicker of text mode contents
before the compositor takes over.

12 years agocompositor-drm: Fail, and fail early if we can't set up the tty
Kristian Høgsberg [Sun, 15 Jan 2012 19:29:09 +0000 (14:29 -0500)]
compositor-drm: Fail, and fail early if we can't set up the tty

Typically a permissions problem.  Weston should be setuid root or
be launched by a setuid helper.

12 years agotty: Close tty fd on exit
Kristian Høgsberg [Sun, 15 Jan 2012 20:20:53 +0000 (15:20 -0500)]
tty: Close tty fd on exit