Original commit message from CVS:
Patch by: Håvard Graff <havard dot graff at tandberg dot com>
* gst/rtpmanager/gstrtpbin.c:
(gst_rtp_bin_propagate_property_to_jitterbuffer),
(gst_rtp_bin_set_property):
Propagate the do-lost and latency properties to the jitterbuffers when
they are changed on rtpbin.
Original commit message from CVS:
* gst/interleave/deinterleave.c: (gst_deinterleave_base_init),
(gst_deinterleave_class_init), (gst_deinterleave_init),
(gst_deinterleave_add_new_pads), (gst_deinterleave_sink_getcaps):
* gst/interleave/deinterleave.h:
Don't set a getcaps() function on the src pads as it's not required
and the default getcaps() function returns the correct results for
our src pads.
Complete documentation and add myself to the authors of the element.
Original commit message from CVS:
Patch by: Sjoerd Simons <sjoerd at luon dot net>
* gst/mpeg4videoparse/mpeg4videoparse.c: (gst_mpeg4vparse_push),
(gst_mpeg4vparse_drain), (gst_mpeg4vparse_chain),
(gst_mpeg4vparse_change_state):
Move some code around to integrate the startcode searching with the
other bits of parsing, avoid a whole bunch of peeks.
Get rid of invalid data that should not happen according to the specs.
Fixes#533559.
Original commit message from CVS:
* gst/interleave/deinterleave.c: (gst_deinterleave_class_init),
(gst_deinterleave_init), (gst_deinterleave_add_new_pads),
(gst_deinterleave_set_pads_caps), (gst_deinterleave_set_property),
(gst_deinterleave_get_property):
* gst/interleave/deinterleave.h:
Add a property to select whether channel positions should be kept on
the mono output buffers or should be dropped.
Original commit message from CVS:
* gst/interleave/deinterleave.c: (gst_deinterleave_finalize),
(gst_deinterleave_init), (gst_deinterleave_sink_event),
(gst_deinterleave_process), (gst_deinterleave_sink_activate_push):
* gst/interleave/deinterleave.h:
Queue events until src pads were added and they can be sent. Otherwise
downstream will never get the first newsegment event.
Original commit message from CVS:
* gst/interleave/deinterleave.c: (gst_deinterleave_sink_setcaps),
(gst_deinterleave_getcaps):
Always set the channel positions when gst_audio_get_channel_positions()
returns something, even if they're not set in the caps. This makes
sure that the output channels can be interleaved again correctly
in the mono/stereo cases too.
Don't ask for the peercaps of the current pad in getcaps() as this
might call getcaps() again and deadlock.
Original commit message from CVS:
* gst/interleave/Makefile.am:
* gst/interleave/deinterleave.c: (deinterleave_24),
(gst_deinterleave_finalize), (gst_deinterleave_base_init),
(gst_deinterleave_class_init), (gst_deinterleave_init),
(gst_deinterleave_add_new_pads), (gst_deinterleave_set_pads_caps),
(gst_deinterleave_set_process_function),
(gst_deinterleave_sink_setcaps), (__remove_channels),
(__set_channels), (gst_deinterleave_getcaps),
(gst_deinterleave_process), (gst_deinterleave_chain),
(gst_deinterleave_sink_activate_push):
* gst/interleave/deinterleave.h:
Add support for all raw audio formats and provide better negotiation
if the caps are changing.
Don't allow changes of the channel positions and set the position of
the corresponding channel on the src pad caps.
General cleanup and smaller bugfixes.
* tests/check/elements/deinterleave.c: (float_buffer_check_probe):
Check the channel positions on the output buffer caps.
Original commit message from CVS:
* docs/Makefile.am:
Don't attempt to build plugin docs when they're disabled.
* gst/bayer/Makefile.am:
Add libgstvideo to the link.
* gst/rtpmanager/Makefile.am:
Fix link order, and move LIBS things to _LIBS
Original commit message from CVS:
* gst/rtpmanager/gstrtpjitterbuffer.c:
(gst_rtp_jitter_buffer_chain):
Simply drop bad RTP packets with a warning instead of just posting an
error and stopping. This is a perfectly recoverable event and we don't
force people to use an rtpbin to filter out bad packets first.
Original commit message from CVS:
* gst/mpeg4videoparse/mpeg4videoparse.c: (gst_mpeg4vparse_init):
Set fixed caps on the srcpad after we created the pad...
Original commit message from CVS:
* gst/audioresample/gstaudioresample.c: (gst_audioresample_init):
Revert previous change which made basetransform handle buffer_alloc
and which breaks things badly in the non-passthrough case since it
returned buffers with a different (ie. sometimes smaller) size than
the size requested.
Original commit message from CVS:
* gst/interleave/Makefile.am:
* gst/interleave/deinterleave.h:
* gst/interleave/interleave.h:
* gst/interleave/plugin.h:
Split definitions into separate header files for better documentation
generation.
* gst/interleave/deinterleave.c: (gst_deinterleave_base_init),
(gst_deinterleave_class_init), (gst_deinterleave_sink_setcaps),
(gst_deinterleave_process):
Don't use alloca, allow caps changes as long as the number of channels
does not change, don't use g_warning, return NOT_NEGOTIATED as early
as possible and some other cleanup.
* gst/interleave/interleave.c: (gst_interleave_base_init),
(gst_interleave_class_init):
Do some random cleanup.
* tests/check/Makefile.am:
* tests/check/elements/deinterleave.c: (GST_START_TEST),
(deinterleave_chain_func), (deinterleave_pad_added),
(deinterleave_suite):
Add unit tests for the deinterleave element.
Original commit message from CVS:
Patch by: Sjoerd Simons <sjoerd at luon dot net>
* gst/mpeg4videoparse/mpeg4videoparse.c:
(gst_mpeg4vparse_set_new_caps), (gst_mpeg4vparse_align),
(get_bits), (next_start_code), (gst_mpeg4vparse_handle_vos),
(gst_mpeg4vparse_push), (gst_mpeg4vparse_drain),
(gst_mpeg4vparse_chain), (gst_mpeg4vparse_sink_setcaps),
(gst_mpeg4vparse_sink_event), (gst_mpeg4vparse_src_query),
(gst_mpeg4vparse_set_property), (gst_mpeg4vparse_get_property),
(gst_mpeg4vparse_class_init), (gst_mpeg4vparse_init):
* gst/mpeg4videoparse/mpeg4videoparse.h:
Parse the config data (either outbound or in the stream) to set
width/height, apect ration, framerate in the caps if applicable.
Mark frames as GST_BUFFER_FLAG_DELTA_UNIT when they are not
intra frames
Set the timestamps of outgoing buffers to the buffer in
which the VOP header was found.
Drop incoming data untill configuration is found (by default,
configurable using a property).
Report a 1 frame latency. Fixes#532723.
Original commit message from CVS:
* gst/rtpmanager/gstrtpjitterbuffer.c:
(gst_rtp_jitter_buffer_loop):
Avoid waiting for a negative (huge) duration when the last packet has a
lower timestamp than the current packet.
Original commit message from CVS:
* gst/rtpmanager/gstrtpsession.c: (gst_rtp_session_query_send_rtcp_src):
Make sure to unref the rtpsession returned by gst_pad_get_parent() to
prevent a memory leak.
Original commit message from CVS:
* gst/rtpmanager/gstrtpjitterbuffer.c:
(gst_rtp_jitter_buffer_loop):
Initialise with GST_CLOCK_TIME_NONE to avoid compiler warning.
Original commit message from CVS:
* docs/plugins/Makefile.am:
* docs/plugins/gst-plugins-bad-plugins-docs.sgml:
* docs/plugins/gst-plugins-bad-plugins-sections.txt:
* docs/plugins/inspect/plugin-gstinterlace.xml:
* gst/deinterlace/gstdeinterlace.c:
* gst/deinterlace/gstdeinterlace.h:
Random doc of the day: the deinterlace element.
Original commit message from CVS:
* gst/mpegtsparse/mpegtspacketizer.c:
* gst/mpegtsparse/mpegtsparse.c:
Make sure all schedule EIT and non-actual transport stream
EITs are parsed. Also add present-following flag and
actual-transport-stream flag to eit bus message.
Original commit message from CVS:
* gst/rtpmanager/rtpsource.c: (rtp_source_finalize):
Make sure to unref the caps used by RTPSource to prevent a memory leak.
Original commit message from CVS:
Patch by: Olivier Crete <tester at tester dot ca>
* gst/rtpmanager/rtpsession.c: (source_clock_rate),
(rtp_session_process_bye), (rtp_session_send_bye_locked):
Unlock the session lock when calling one of our callbacks.
Fixes#532011.
Original commit message from CVS:
Patch by: Sjoerd Simons <sjoerd at luon dot net>
* gst/audioresample/gstaudioresample.c: (gst_audioresample_init):
Let audioresample use the buffer allocation of basetransform instead
of it's own stuff.
* tests/check/elements/audioresample.c: (alloc_only_48000),
(GST_START_TEST), (audioresample_suite):
Add unit test for the recent basetransform bugfix, where upstream
changes caps to something that can't be passed through anymore.
Original commit message from CVS:
Patch by: Jens Granseuer <jensgr at gmx dot net>
* gst/subenc/gstsrtenc.c: (gst_srt_enc_timestamp_to_string):
Declare variables at the beginning of blocks. Fixes compilation with
gcc 2.x and other compilers. Fixes bug #530611.
Original commit message from CVS:
* gst/mpegtsparse/mpegtspacketizer.c:
* gst/mpegtsparse/mpegtspacketizer.h:
* gst/mpegtsparse/mpegtsparse.c:
Detect SI pids (NIT, SDT, EIT etc.) based on table id and not
by pid number. This allows for example the EPG data from UK's
freesat to be picked up.
Original commit message from CVS:
* gst/rtpmanager/gstrtpbin.c: (create_stream), (gst_rtp_bin_init),
(gst_rtp_bin_set_property), (gst_rtp_bin_get_property):
* gst/rtpmanager/gstrtpbin.h:
Expose new jitterbuffer property in rtpbin too.
Original commit message from CVS:
* gst/rtpmanager/gstrtpjitterbuffer.c:
(gst_rtp_jitter_buffer_class_init), (gst_rtp_jitter_buffer_init),
(gst_rtp_jitter_buffer_loop), (gst_rtp_jitter_buffer_set_property),
(gst_rtp_jitter_buffer_get_property):
Disable sending out rtp packet lost events by default and make a
property to enabe it. We will likely enable it by default when the base
depayloaders have a default handler for them so that we don't send these
events all through the pipeline for now.
Original commit message from CVS:
* gst/rtpmanager/gstrtpjitterbuffer.c:
(gst_rtp_jitter_buffer_init), (gst_rtp_jitter_buffer_flush_stop),
(gst_rtp_jitter_buffer_src_event), (gst_rtp_jitter_buffer_chain),
(gst_rtp_jitter_buffer_loop):
Remove private version of a function that is in -base now.
Add src event handler.
Rework the jitterbuffer pushing loop so that it can quickly react to
lost packets and instruct the depayloader of them. This can then be used
to implement error concealment data.
Original commit message from CVS:
* gst/rtpmanager/gstrtpsession.c:
(gst_rtp_session_query_send_rtcp_src), (create_recv_rtcp_sink),
(create_send_rtcp_src):
Set up some internal links functions for the RTCP and sync pads because
the defaults are really not correct.
Implement a query handler for the RTCP src pad, mostly to correctly
report about the latency.
Original commit message from CVS:
* gst/rtpmanager/gstrtpbin.c: (gst_rtp_bin_associate),
(gst_rtp_bin_sync_chain):
* gst/rtpmanager/rtpsession.c: (update_arrival_stats),
(rtp_session_process_sr), (rtp_session_on_timeout):
* gst/rtpmanager/rtpsource.c: (rtp_source_init),
(calculate_jitter):
* gst/rtpmanager/rtpsource.h:
* gst/rtpmanager/rtpstats.h:
Also keep track of the first buffer timestamp together with the first
RTP timestamp as they both are needed to construct the timing of
outgoing packets in the jitterbuffer and are therefore also needed to
manage lip-sync. This fixes lip-sync if the first RTP packets arrive
with a wildly different gap.
Original commit message from CVS:
* gst/flv/gstflvparse.c: (gst_flv_parse_metadata_item),
(gst_flv_parse_tag_script):
Handle NULL returns from FLV_GET_STRING() more gracefully. Fixes
crash caused by a strlen on a NULL string (#527622).
Original commit message from CVS:
* configure.ac:
Bump core/base requirements to released versions to avoid
confusion.
* gst/deinterlace/gstdeinterlace.c: (deinterlace_debug),
(GST_CAT_DEFAULT), (gst_deinterlace_base_init),
(gst_deinterlace_set_caps), (plugin_init):
Add debug category, use _set_element_details_simple and
remove special code path for Y42B to calculate offsets and
strides; libgstvideo knows how to handle this format now.
Original commit message from CVS:
* gst/cdxaparse/Makefile.am:
* gst/cdxaparse/gstcdxaparse.c:
* gst/cdxaparse/gstcdxastrip.c:
* gst/cdxaparse/gstcdxastrip.h:
* gst/cdxaparse/gstvcdparse.c:
* gst/cdxaparse/gstvcdparse.h:
Port VCD parser (formerly cdxastrip) from 0.8 to 0.10. Doesn't do
anything the 0.8 version didn't do though.
Original commit message from CVS:
* gst/rtpmanager/gstrtpbin.c: (free_session):
* gst/rtpmanager/gstrtpsession.c: (gst_rtp_session_finalize):
Avoid leaking pads in the RTP manager.
Original commit message from CVS:
* gst/mpegtsparse/mpegtspacketizer.c:
Cable delivery subsystem descriptors' frequency's bcd
is measured in 100Hz units so adjust multiplier accordingly.
Original commit message from CVS:
* gst/nsf/Makefile.am:
* gst/nsf/fds_snd.c:
* gst/nsf/mmc5_snd.c:
* gst/nsf/nsf.c:
* gst/nsf/types.h:
* gst/nsf/vrc7_snd.c:
* gst/nsf/vrcvisnd.c:
* gst/nsf/memguard.c:
* gst/nsf/memguard.h:
Remove memguard again and apply hopefully all previously dropped
local patches. Should be really better than the old version now.
Original commit message from CVS:
* gst/nsf/memguard.c: (_my_free):
* gst/nsf/types.h:
Unbreak compilation by disabling memguard and doing some dirty hack
fixes to make it compile on 64bits.
Original commit message from CVS:
* gst/selector/gstinputselector.c:
(gst_input_selector_set_active_pad), (gst_input_selector_switch):
Do g_object_notify() only when not holding the lock to get the property
because otherwise we run into a deadlock with the deep-notify handlers
that are possibly installed.
Original commit message from CVS:
* gst/selector/gstinputselector.c: (gst_selector_pad_class_init),
(gst_selector_pad_event), (gst_selector_pad_bufferalloc),
(gst_selector_pad_chain), (gst_input_selector_set_active_pad):
Release the selector lock when pad alloc happens on a non selected pad.
Original commit message from CVS:
* gst/selector/gstinputselector.c: (gst_selector_pad_class_init),
(gst_selector_pad_init), (gst_selector_pad_set_property),
(gst_selector_pad_get_property), (gst_selector_pad_event),
(gst_selector_pad_bufferalloc), (gst_selector_pad_chain),
(gst_input_selector_set_active_pad):
Add pad property to configure behaviour of the unselected pad, it can
return OK or NOT_LINKED, based on the use case.
Original commit message from CVS:
* gst/selector/gstinputselector.c:
(gst_selector_pad_get_running_time), (gst_selector_pad_reset),
(gst_selector_pad_event), (gst_selector_pad_bufferalloc),
(gst_input_selector_wait), (gst_selector_pad_chain),
(gst_input_selector_class_init), (gst_input_selector_init),
(gst_input_selector_dispose), (gst_segment_set_start),
(gst_input_selector_set_active_pad),
(gst_input_selector_set_property),
(gst_input_selector_get_property),
(gst_input_selector_get_linked_pad),
(gst_input_selector_is_active_sinkpad),
(gst_input_selector_activate_sinkpad),
(gst_input_selector_request_new_pad),
(gst_input_selector_release_pad),
(gst_input_selector_change_state), (gst_input_selector_block),
(gst_input_selector_switch):
* gst/selector/gstinputselector.h:
Figure out the locking a bit more.
Mark buffers with discont after switching.
Fix initial segment forwarding, make sure to only forward one segment
regardless of what the sequence of buffers/segments is. See #522203.
Improve flushing when blocked.
Return NOT_LINKED when a stream is not selected.
Not API change for the switch signal in the docs.
Fix start/time/accum values of the new segment.
Correctly unlock and flush a blocking selector when going to READY.
Original commit message from CVS:
* gst/freeze/FAQ:
* gst/freeze/Makefile.am:
* gst/freeze/gstfreeze.c:
Add example to source code documentation blob and remove the 3 line
FAQ.
* gst/interleave/interleave.c:
Add a source code documentation blob.
Original commit message from CVS:
* gst/selector/gstinputselector.c: (gst_selector_pad_event),
(gst_selector_pad_bufferalloc), (gst_selector_pad_chain),
(gst_input_selector_class_init),
(gst_input_selector_set_active_pad),
(gst_input_selector_set_property),
(gst_input_selector_push_pending_stop):
Add lots of debugging.
Fix time member in the newsegment event.
Original commit message from CVS:
* gst/selector/gstinputselector.c: (gst_selector_pad_class_init),
(gst_selector_pad_finalize), (gst_selector_pad_get_property),
(gst_selector_pad_event), (gst_input_selector_class_init),
(gst_input_selector_init), (gst_input_selector_set_active_pad),
(gst_input_selector_set_property),
(gst_input_selector_get_property),
(gst_input_selector_request_new_pad),
(gst_input_selector_release_pad),
(gst_input_selector_push_pending_stop),
(gst_input_selector_switch):
* gst/selector/gstinputselector.h:
Various cleanups.
Added tags to the pads.
Select active pad based on the pad object instead of its name.
Fix refcount in set_active_pad.
Add property to get the number of pads.
* gst/selector/gstoutputselector.c:
(gst_output_selector_class_init),
(gst_output_selector_set_property),
(gst_output_selector_get_property):
Various cleanups.
Select the active pad based on the pad object instead of its name.
Fix locking when setting the active pad.
* gst/selector/gstselector-marshal.list:
* tests/check/elements/selector.c: (cleanup_pad),
(selector_set_active_pad), (run_input_selector_buffer_count):
Fixes for pad instead of padname for pad selection.
Original commit message from CVS:
Based on patch by: Ole André Vadla Ravnås <ole.andre.ravnas@tandberg.com>
* gst/rtpmanager/gstrtpsession.c: (gst_rtp_session_init),
(rtcp_thread), (start_rtcp_thread), (stop_rtcp_thread),
(join_rtcp_thread), (gst_rtp_session_change_state):
Avoid a deadlock when joining the RTCP thread in PAUSED because it might
be blocked downstream. Also avoid spawning multiple rtcp threads.
Fixes#520894.
Original commit message from CVS:
Patch by: Stefan Kost <ensonic@users.sf.net>
* gst/rtpmanager/rtpjitterbuffer.c: (calculate_skew):
Don't try to reset the clock skew when we have no timestamps.
Fixes#519005.
Original commit message from CVS:
* gst/mpegtsparse/gstmpegdesc.h:
* gst/mpegtsparse/mpegtspacketizer.c:
Add parsing of cable delivery system descriptor.
Original commit message from CVS:
Patch by: Jens Granseuer <jensgr at gmx dot net>
* gst/mve/gstmvedemux.c: (gst_mve_audio_data),
(gst_mve_demux_get_type):
Fix audio discontinuity that happens when silent chunks are
followed by real data again. Fixes bug #519905.
Original commit message from CVS:
* gst/mpegtsparse/mpegtsparse.c:
Only send PMTs to program pads that the PMT is for even if
on same pid.
As a by-product, we now no longer hardcode any psi pid numbers.
Also remove pcr stream from old pmt when we apply a new pmt.
Original commit message from CVS:
* gst/selector/gstinputselector.c:
* gst/selector/gstinputselector.h:
Added "select-all" property to make it work like aggregator in 0.8.
* gst/selector/gstoutputselector.c:
Fix resend-latest behavoiur.
* tests/check/Makefile.am:
* tests/check/elements/.cvsignore:
* tests/check/elements/selector.c:
Add unit tests for selector.
Original commit message from CVS:
* configure.ac:
* ext/mpeg2enc/Makefile.am:
* ext/soundtouch/Makefile.am:
* gst/modplug/Makefile.am:
Check for and define ERROR_CXXFLAGS and GST_CXXFLAGS and use them
when building C++ code.
Original commit message from CVS:
Patch by: Olivier Crete <tester@tester.ca>
* gst/rtpmanager/gstrtpbin.c: (gst_rtp_bin_associate):
Ignore streams that did not receive an SR packet when doing
synchronisation. Fixes#516160.
Original commit message from CVS:
* gst/dvdspu/gstdvdspu.c: (gst_dvd_spu_handle_new_spu_buf):
Set n_line_ctrl_i to 0 whenever we free line_ctrl_i. Patch based
on an idea by Jan Schmidt, fixes bug #516436.
Original commit message from CVS:
* gst/mpegtsparse/gstmpegdesc.c:
* gst/mpegtsparse/gstmpegdesc.h:
* gst/mpegtsparse/mpegtsparse.c:
Make sure the gstmpegdesc debug lines do not critical
when GST_DEBUG is enabled and also actually output.
Thanks to Alessandro Decina for spotting.
Fixes#516448
Original commit message from CVS:
* ext/xvid/gstxvidenc.c:
* gst/vmnc/vmncdec.c:
* sys/glsink/glimagesink.c:
* sys/glsink/gstgldisplay.c:
Fix some finalize leaks by chaining up to the parent method.
Original commit message from CVS:
* gst/selector/Makefile.am:
Listing the marshal.h in the nodist_HEADERS breaks distcheck, so
let's not do that
* tests/check/Makefile.am:
Disable the crashing cdaudio plugin from the states test so I can make
pre-releases.
Original commit message from CVS:
* gst/multifile/gstmultifilesink.c:
* gst/multifile/gstmultifilesrc.c:
Use g_file_[sg]et_contents() instead of using stdio functions.
Should be less error prone.
* tests/check/elements/multifile.c:
Create a temporary directory using standard functions instead of
creating a directory in the current dir.
Original commit message from CVS:
* configure.ac:
* docs/plugins/gst-plugins-bad-plugins-docs.sgml:
* docs/plugins/gst-plugins-bad-plugins-sections.txt:
* docs/plugins/inspect/plugin-xingheader.xml:
* gst/xingheader/Makefile.am:
* gst/xingheader/gstxingmux.c:
* gst/xingheader/gstxingmux.h:
* tests/check/Makefile.am:
* tests/check/elements/.cvsignore:
* tests/check/elements/xingmux.c:
* tests/check/elements/xingmux_testdata.h:
Remove the xingmux plugin, as the element has moved into
mpegaudioparse in -ugly.
Original commit message from CVS:
* ext\neon\gstneonhttpsrc.c:
Include unistd.h only if _HAVE_UNISTD_H is defined
* gst\mpegvideoparse\mpegvideoparse.c:
Use G_GUINT64_CONSTANT GLIB macro for constant
* sys\dshowsrcwrapper\gstdshowaudiosrc.c:
* sys\dshowsrcwrapper\gstdshowvideosrc.c:
* sys\dshowdecwrapper\gstdshowaudiodec.c:
* sys\dshowdecwrapper\gstdshowaudiodec.h:
* sys\dshowdecwrapper\gstdshowdecwrapper.c:
* sys\dshowdecwrapper\gstdshowdecwrapper.h:
* sys\dshowdecwrapper\gstdshowvideodec.c
* sys\dshowdecwrapper\gstdshowvideodec.h:
Add a DirectShow decoder wrapper.
* win32\MANIFEST:
Add new win32 files to MANIFEST
* win32\vs6\gst_plugins_bad.dsw:
* win32\vs6\libgstdshow.dsp:
* win32\vs6\libgstdshowdecwrapper.dsp:
* win32\vs6\libgstflv.dsp:
Add new projects to bad workspace
Original commit message from CVS:
* gst/mpegtsparse/mpegtspacketizer.c:
Parse component descriptor.
* gst/mpegtsparse/mpegtsparse.c:
Add SI pids to every program (but hardcoded currently).
Original commit message from CVS:
* gst/multifile/gstmultifilesink.c:
Add a fixme comment.
* gst/selector/gstoutputselector.c:
Fix same leak as in input-selector.
* tests/icles/output-selector-test.c:
Improve the test.
Original commit message from CVS:
* gst/mpegtsparse/mpegtspacketizer.c:
Add flag to both sdt and nit structures to say
whether the table is for the actual network/ts
or not.
Original commit message from CVS:
* gst/selector/gstinputselector.c: (gst_selector_pad_event):
Don't leak event on pads that are not linked. Fixes#512826.
