Improve bufferpool handling in ximagesink.
Implement bufferpool handling on xvimagesink.
Based on patches from benjamin gaignard <benjamin.gaignard@linaro.org>
And deprecate the gulong versions. This is to support platforms
where sizeof(unsigned long) < sizeof(void *). Fixes#627565.
API: Add gst_x_overlay_set_window_handle()
API: Deprecate: gst_x_overlay_set_xwindow_id()
API: Add gst_x_overlay_got_window_handle()
API: Deprecate: gst_x_overlay_got_xwindow_id()
API: Add GstXOverlay::set_window_handle()
API: Deprecate: GstXOverlay::set_xwindow_id()
In buffer_alloc, if the buffer caps are new, call
gst_pad_peer_accept_caps once only, it's useless to call it in the
cases where we know it will always fail.
Fixes bug #621190
We can update the geometry in ConfigureNotify (unless we disable event-
handling). If event handling is disabled, one should use _expose() to trigger a
redraw and update the geometry.
The event thread is doing 20 wakeups per second to poll the events. If one
runs ximagesink with handle-events=false and handle-expose=false then we can
avoid the extra thread.
Remove some useless debug info that reported wrong image sizes.
When upstream does not accept out suggested size, fall back to allocating an
image of the requested width/height instead of the currently configured size.
The problem is that an image is reused from the pool because the width/height
match but the caps on the new buffer are the requested caps with possibly
different height/width resulting in errors.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_buffer_alloc):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_buffer_alloc):
Clear all flags on buffers returned from the image pool.
Fixes#563143
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_check_xshm_calls),
(gst_ximagesink_ximage_new):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_check_xshm_calls),
(gst_xvimagesink_xvimage_new):
Apparently on Solaris and OS/X (at least), it's not legal to ask the
X server to attach to a shared memory segment after we've deleted it,
with the result that MIT-SHM is disabled. Instead, remove it only after
X succeeds in attaching too.
Original commit message from CVS:
* sys/ximage/ximagesink.c (gst_ximagesink_ximage_put),
(gst_ximagesink_setcaps):
* sys/ximage/ximagesink.h:
When the caps change, make sure to re-draw borders in
force-aspect-ratio=true mode.
* sys/xvimage/xvimagesink.c (gst_xvimagesink_xvimage_put):
Don't clear the border_draw flag until we actually draw the border.
* tests/check/Makefile.am:
Ignore alsasink/src during the states test too, so it doesn't fail
when running without access to the sound device.
Original commit message from CVS:
Patch by: Tommi Myöhänen <ext-tommi dot myohanen at nokia dot com>
* sys/ximage/ximagesink.c: (gst_ximagesink_xcontext_get),
(gst_ximagesink_change_state):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_xcontext_get):
Don't leak the PAR on errors. Fixes#496731.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_event_thread),
(gst_ximagesink_xcontext_get), (gst_ximagesink_xcontext_clear),
(gst_ximagesink_change_state), (gst_ximagesink_reset):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_event_thread),
(gst_xvimagesink_xcontext_get), (gst_xvimagesink_xcontext_clear),
(gst_xvimagesink_change_state), (gst_xvimagesink_reset):
Make sure that before we clean up the X resources, we shutdown and join
the event thread.
Also make sure the event thread does not shut down immediatly after
startup because the running variable is not yet correctly set.
Fixes#378770.
Original commit message from CVS:
Patch by: René Stadler <mail at renestadler dot de>
* sys/ximage/ximagesink.c: (gst_ximage_buffer_finalize),
(gst_ximage_buffer_class_init):
* sys/xvimage/xvimagesink.c: (gst_xvimage_buffer_destroy),
(gst_xvimage_buffer_class_init):
Correctly chain up finalize with the parent class to prevent
memory leaks. Fixes#474880.
Original commit message from CVS:
Patch by: Jorn Baayen <jorn at openedhand dot com>
* sys/ximage/ximagesink.c: (gst_ximagesink_handle_xevents),
(gst_ximagesink_set_property), (gst_ximagesink_get_property),
(gst_ximagesink_init), (gst_ximagesink_class_init):
* sys/ximage/ximagesink.h:
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_handle_xevents),
(gst_xvimagesink_set_property), (gst_xvimagesink_get_property),
(gst_xvimagesink_init), (gst_xvimagesink_class_init):
* sys/xvimage/xvimagesink.h:
add 'handle-expose' property. Useful for video widgets which may want to
be in control of Expose behaviour. Fixes#380625
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_handle_xevents):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_handle_xevents):
Fix a locking-order bug I introduced with my changes the other day.
Patch by Mike Smith.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_put),
(gst_ximagesink_xwindow_new), (gst_ximagesink_handle_xevents),
(gst_ximagesink_show_frame):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_xvimage_put),
(gst_xvimagesink_xwindow_new), (gst_xvimagesink_handle_xevents),
(gst_xvimagesink_show_frame):
When we create our own window, indicate that we handle the
WM_DELETE client message from the window manager, so that it won't
kill our window (and our app) along with it. Handle ClientMessage,
post an error on the bus, and close the window. Further buffers
arriving will result in a FlowError because the window has been
destroyed.
Fixes: #393975
Clean up the X event handling loop and make them the same for
both xvimagesink and ximagesink while I'm at it.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_new):
When XShm is not available, we might get row strides that are not
rounded up to multiples of four; this is bad, because virtually
every RGB-processing element in GStreamer assumes rowstrides are
rounded up to multiples of four, so let's allocate at least enough
memory to avoid crashes in this case. The image will still be
displayed distorted though if this happens, so that still needs
fixing (maybe by allocating a bigger image with an 'even' width
and then clipping it appropriately when rendering - something for
Xlib aficionados in any case).
Original commit message from CVS:
Patch by: Christian Kirbach <Christian dot Kirbach at googlemail com>
* sys/ximage/ximagesink.c:
Fix build if XShm is not available (#432362).
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_xcontext_clear),
(gst_ximagesink_change_state), (gst_ximagesink_reset),
(gst_ximagesink_finalize):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_change_state),
(gst_xvimagesink_reset), (gst_xvimagesink_finalize):
Move some cleanup stuff from the state change handler into a _reset()
function that can be called from _finalize(). This ensures that things
get freed even if (for some reason) the NULL->READY state transition
fails in the parent class.
Even if a parent state change fails, process our downward state change
logic instead of bailing out early.
Free the correct xcontext pointer in ximagesink's xcontext_clear.
Original commit message from CVS:
* sys/ximage/ximagesink.c:
(gst_ximagesink_calculate_pixel_aspect_ratio):
* sys/xvimage/xvimagesink.c:
(gst_xvimagesink_calculate_pixel_aspect_ratio):
Small constifications.
Original commit message from CVS:
2007-01-23 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximagesink_handle_xevents):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_handle_xevents):
Don't lock on navigation event push, just on keysym to string.
Fixes#397673 again.
Original commit message from CVS:
2007-01-22 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximagesink_handle_xevents):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_handle_xevents):
Looking at Xorg code i can't figure out if that XKeysymToString
function is thread sensible or not. Lock it just in case as
recommended by Radek Doulik <rodo at ximian dot com>.
Original commit message from CVS:
2007-01-22 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximagesink_handle_xevents):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_handle_xevents):
Lock that X Call as well. Fixes#397673.
Original commit message from CVS:
2007-01-07 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximage_buffer_finalize),
(gst_ximagesink_handle_xerror), (gst_ximagesink_ximage_new),
(gst_ximagesink_ximage_destroy), (gst_ximagesink_ximage_put),
(gst_ximagesink_handle_xevents), (gst_ximagesink_setcaps),
(gst_ximagesink_change_state), (gst_ximagesink_set_xwindow_id),
(gst_ximagesink_expose), (gst_ximagesink_set_event_handling):
* sys/xvimage/xvimagesink.c: (gst_xvimage_buffer_destroy),
(gst_xvimage_buffer_finalize), (gst_xvimagesink_handle_xerror),
(gst_xvimagesink_xvimage_new), (gst_xvimagesink_xvimage_put),
(gst_xvimagesink_handle_xevents), (gst_xvimagesink_setcaps),
(gst_xvimagesink_change_state),
(gst_xvimagesink_set_xwindow_id),
(gst_xvimagesink_expose), (gst_xvimagesink_set_event_handling):
Use flow_lock much more to protect every access to xwindow.
Try to catch erros while creating images in case some drivers
are
just generating an XError when the requested image is too big.
Should fix : #354698, #384008, #384060.
* tests/icles/stress-xoverlay.c: (cycle_window),
(create_window):
Implement some stress testing of setting window xid.
Original commit message from CVS:
2007-01-04 Julien MOUTTE <julien@moutte.net>
* gst-libs/gst/interfaces/xoverlay.c:
(gst_x_overlay_handle_events):
* gst-libs/gst/interfaces/xoverlay.h:
* sys/ximage/ximagesink.c: (gst_ximagesink_xwindow_new),
(gst_ximagesink_set_xwindow_id),
(gst_ximagesink_set_event_handling),
(gst_ximagesink_xoverlay_init), (gst_ximagesink_set_property),
(gst_ximagesink_get_property), (gst_ximagesink_init),
(gst_ximagesink_class_init):
* sys/ximage/ximagesink.h:
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_xwindow_new),
(gst_xvimagesink_set_xwindow_id),
(gst_xvimagesink_set_event_handling),
(gst_xvimagesink_xoverlay_init), (gst_xvimagesink_set_property),
(gst_xvimagesink_get_property), (gst_xvimagesink_init),
(gst_xvimagesink_class_init):
* sys/xvimage/xvimagesink.h:
* tests/icles/stress-xoverlay.c: (toggle_events),
(create_window):
Add a method to the XOverlay interface to allow disabling of
event handling in x[v]imagesink elements. This will let X events
propagate to parent windows which can be usefull in some cases.
Be carefull that the application is then responsible of pushing
navigation events and expose events to the video sink.
Fixes: #387138.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_get_type):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_get_type):
Explicitly create our custom buffer classes at a thread-safe
location as well, since g_type_class_ref() doesn't seem to be
entirely thread-safe either (#365501; also see #349410).
Original commit message from CVS:
2006-10-20 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_put):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_xvimage_put):
Patch by: Stefan Kost <ensonic@users.sf.net>
Try to redraw borders only when needed. Apparently this consumes
resources on small devices... :-O (#363607)
Original commit message from CVS:
* gst/videotestsrc/gstvideotestsrc.c:
Use G_UNLIKELY in _create and log one more detail.
(gst_video_test_src_get_times), (gst_video_test_src_create):
* sys/ximage/ximagesink.c: (gst_ximagesink_get_times):
Use gst_util_uint64_scale_int in _get_times().
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_new),
(gst_ximagesink_change_state):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_xvimage_new),
(gst_xvimagesink_change_state):
Implement a locking order to ensure we always take the object lock
before the x_lock and never vice-versa.
Original commit message from CVS:
* gst/playback/gstdecodebin.c: (find_compatibles):
Fix a caps leak when linking (#347304)
* sys/ximage/ximagesink.c: (gst_ximage_buffer_finalize),
(gst_ximagesink_ximage_destroy), (gst_ximagesink_xcontext_clear),
(gst_ximagesink_change_state):
* sys/xvimage/xvimagesink.c: (gst_xvimage_buffer_destroy),
(gst_xvimage_buffer_finalize), (gst_xvimagesink_check_xshm_calls),
(gst_xvimagesink_xvimage_new), (gst_xvimagesink_xvimage_put),
(gst_xvimagesink_xcontext_clear), (gst_xvimagesink_change_state):
Don't leak shared memory resources. Use the object lock to protect
against the xcontext disappearing while returning a buffer from the
pipeline. (#347304)
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximage_buffer_finalize),
(gst_ximage_buffer_free), (gst_ximagesink_ximage_put),
(gst_ximagesink_setcaps), (gst_ximagesink_buffer_alloc):
Avoid type checking in buffer casts.
Avoid caps copy in buffer_alloc when we can.
Use pad_peer_accept.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_new),
(gst_ximagesink_xcontext_get), (gst_ximagesink_show_frame):
Improve the errors produced on bad output, including some human
readable description strings.
Handle the (theoretical for ximagesink) case where the XServer
has a different idea about the size required for a particular
frame and gives us too small a memory allocation.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_get_type):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_get_type):
Register special buffer types along with the objects so
that they are not registered at runtime from N different
streaming threads since they are not threadsafe.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_check_xshm_calls),
(gst_ximagesink_ximage_new), (gst_ximagesink_ximage_destroy):
* sys/xvimage/xvimagesink.c: (gst_xvimage_buffer_destroy),
(gst_xvimagesink_check_xshm_calls), (gst_xvimagesink_xvimage_new):
Set shared memory segments to be deleted as soon as we have attached,
that way they get cleaned up automatically if we crash.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximagesink_check_xshm_calls):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_check_xshm_calls):
Fix up the XShm call testing so that we catch errors, and don't
cause new ones by attempting to detach from a segment we failed
to attach to. Fixes#312439.
Original commit message from CVS:
2006-01-26 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c:
(gst_ximagesink_calculate_pixel_aspect_ratio):
* sys/xvimage/xvimagesink.c:
(gst_xvimagesink_calculate_pixel_aspect_ratio): Handle some
more cases of pixel aspect ratio.
Original commit message from CVS:
2006-01-22 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_destroy):
* sys/xvimage/xvimagesink.c: (gst_xvimage_buffer_destroy),
(gst_xvimagesink_buffer_alloc): Playbin keeps some ref to some
frames. We might get a frame destroyed after changing state to
NULL, adding a safety check on xcontext.
Original commit message from CVS:
* sys/ximage/ximagesink.c: (gst_ximage_buffer_finalize),
(gst_ximagesink_ximage_new), (gst_ximagesink_ximage_put),
(gst_ximagesink_buffer_alloc):
* sys/xvimage/xvimagesink.c: (gst_xvimage_buffer_finalize),
(gst_xvimagesink_xvimage_put), (gst_xvimagesink_show_frame),
(gst_xvimagesink_buffer_alloc):
move all regularly occurring messages to GST_LOG level
add some more object logs
Original commit message from CVS:
2005-11-28 Julien MOUTTE <julien@moutte.net>
* sys/ximage/ximagesink.c: (gst_ximagesink_ximage_put),
(gst_ximagesink_setcaps):
* sys/xvimage/xvimagesink.c: (gst_xvimagesink_xvimage_put),
(gst_xvimagesink_setcaps): Fixed a tricky bug. When caps
renegotiation
happens (only visible on ximagesink but bug is in xv too)
set_caps was
destroying the internal x[v]image used to memcpy non locally
alloced
buffers so that it got renewed on next _chain. The issue is that
_expose will try to put that image as it reffed it in _put.
Using gst_buffer_unref instead of destroy fixes it !
