The printf extension mechanism changed in glibc 2.10, and the older
register_printf_function is deprecated. Detect and use the new
mechanism where available.
When its disabled, we poison some symbols to force a build error if they are
used. Dunno how useful this acually is, but we need to disable the poisoning
when we include this ourself. Also don't define some of the dummies, as they
are getting replaced with defines and that creates code that does not compile.
gst_static_pad_template_get_caps () actually returns a reference to the
caps and it's cleaner to unref them after usage. The core will, however,
always hold a reference to the caps so this didn't result in a memory
leak.
And, for our release manager, the in-commit-message keywords
for the previous commit:
API: GST_PARAM_MUTABLE_READY
API: GST_PARAM_MUTABLE_PAUSED
API: GST_PARAM_MUTABLE_PLAYING
API: gst_param_spec_is_mutable
When we get an ASYNC_DONE message when a state change was busy, set the
pending_async_done flag so that after the state change completes, the bin can
check if all async elements are finished. Don't only do this for the bin itself
but for all elements.
This fixes some bins in bins that simulate async state changes by posting ASYNC
messages (such as sdpparse in uridecodebin/playbin2).
Two new log levels to dump FIXMEs into the log and to log data
in form of a hex dump (#578114).
API: GST_CAT_FIXME_OBJECT
API: GST_CAT_MEMDUMP_OBJECT
API: GST_CAT_FIXME
API: GST_CAT_MEMDUMP
API: GST_FIXME_OBJECT
API: GST_MEMDUMP_OBJECT
API: GST_FIXME
API: GST_MEMDUMP
If that block is entered, then start_time becomes GST_CLOCK_TIME_NONE.
Since start_time is invalid, the code will enter the block at line 434 and
new_base_time will be set there.
Move the checks for using an unscheduled entry from the unsafe GstClock to the
SystemClock object so that we can perform the correct locking.
fix a leak and potential deadlock then the async thread fails to start.
Sprinkle some G_LIKELY around because we can.
Keep a counter for the amount of outstanding wakeups that we produce and only
perform a write/read to the control socket when 1 or 0 respectively.
don't poll when waiting for the entries to be unblocked and clear their wakeup
counts, just act on the signal when the wakeup count is 0.
unscheduled entries will clear their wakeup count themselves.
Keep track of when we wakeup the async thread because the list of entries has
changed.
don't try to see if the list changed because we can't really know when one entry
is added multiple times.
Only wake up the async thread when we add an async entry to the head of the list
and the old entry was BUSY.
when an entry being waited on in the async thread is unscheduled, clear the
wakeup queue so we can continue waiting on other entries.
When an entry being waited on in the async thread is unlocked because an earlier
entry was added to the list, set the entry to OK again. This makes sure that
only the entries being waited on have the BUSY flag set and wake up the timer
poll when they are unscheduled.
Yes, 'codec' isn't exactly the best word, but let's be consistent with AUDIO_CODEC
and VIDEO_CODEC (which may be 'raw' formats as well after all). Prerequisite for
bug #576552.
Running configure with e.g. --disable-dst-debug was compiling out the debug
system (ABI break). Now stubs are added and only if one does e.g.
make CFLAGS="-DGST_REMOVE_DISABLED" the symbols are ommitted.
Move the gst_pad_can_link() implementation from gstutils to gstpad and use
gst_pad_link_prepare() to make it work correctly and also check the caps.
Make the broken implementation in gstutils static.
Small cleanups in the _get_fixed_caps() function.
Fixes#575682.
Fixes build with -Werror caused by '_FILE_OFFSET_BITS redefined' warning on
OpenSolaris where _FILE_OFFSET_BITS may be defined both in our config.h
and via stdio.h (#575695).
Bin collects segment-start messages and segent-done messages. it posts a
segment-done message to its parent, once it has received a segment-done for
each segment-start. Imho it should also send a segment-start if it receives the
first segment start and if parent is !=NULL. This is needed for bins in bins,
so that also higher order bins can group segment-starts and segment-dones.
Right now higher order bins will post a segment-done for each segment-done
received.
Long caps fields like enums are ellipsised. If caps are not negotiated, use
head- and taillabel to place them closer to the pads. Use smarter way to indent.
Keep track of when a new callback is installed in the callback and call the new
callback in that case.
Add unit test for checking pad blocking.
Fixes#573823.
When de/activating a pad in pull mode the pad needs to de/activate the
peer pad it is connected to, failure to be able to do this in activation mode
is an error.
However if there is no peerpad, we can still deactivate the pad correctly and
assume the application will deactivate the unlinked peer pad eventually.
Fixes#574163.
This will be mostly useful in all elements that have some kind of internal
seek/index table. Currently almost all of them (or even all of them)
are using a linear search although the used array is already sorted,
wasting some CPU time without good reason.
Fixes bug #573623.
unref and copy functions are sometimes used as function
pointers for example in the case of g_hash_table_new_full
as a GDestroyNotify function.
Currently GstBuffer, GstEvent, GstMessage and GstQuery
define their respective unref and copy functions as
macros, making use of gst_mini_object_unref/copy.
This approach works very well for most cases, except
for some automatically generated bindings (currently Vala),
where the memory management semantics are defined
declaratively.
The possible solutions would be to either convert all
the macros into static inline function, or change the
signature of gst_mini_object_unref to take a void*
instead of a GstMiniObject*.
Fixes bug #572480.
Add a GST_MESSAGE_REQUEST_STATE that can be posted by element when they would
like to have the application change the state of the pipeline. the primary use
case is to pause the pipeline when an audio mixer is mixing a higher priority
stream but it can also be used for other purposes.
Add some docs and a unit test.
Implement the REQUEST_STATE message in gst-launch.
API: gst_message_new_request_state()
API: gst_message_parse_request_state()
API: GST_MESSAGE_REQUEST_STATE
When the target pad disappears (because it was explicitly unlinked or the
element was removed/unreffed) make sure we receive a notify with the unlink
function on the proxy pad and clear the target. We use a simple flag to not do
this and cause deadlocks when the target was changed explicitly using the
ghostpad functions.
Update the unit test because we now unref the target sooner (and correctly).
This tag will list a homepage for the media,
i.e. the artist's or movie's homepage.
This is different to GST_TAG_LOCATION as the latter
lists the original location of the media.
Fixes bug #571227.
The previous commit was bogus, as was the check before. We just point m to the file data,
so neither it nor its members will be NULL. Better check if we have enough data.
Add a property to select the clock type, currently REALTIME and MONOTONIC when
posix timers are available.
Implement the systemclock with GstPoll instead of GCond. This allows us to
schedule timeouts with nanosecond precission on newer kernels and with ppoll
support. It's also resilient to changes to the systemclock because of NTP or
similar.
Add a special timer mode in GstPoll that makes it only use the control socket
with a timeout to schedule timeouts. Also add a pair of methods to wakeup the
timeout thread.
API: GstPoll::gst_poll_new_timer()
API: GstPoll::gst_poll_write_control()
API: GstPoll::gst_poll_read_control()
Remove class-to-interface-struct cast macros which don't work,
don't make sense, and in some cases wouldn't even compile if
used. Removal should be ok seeing that code which uses any of
these is broken and bound to crash. Fixes#565607.
API: remove GST_IMPLEMENTS_INTERFACE_CLASS
API: remove GST_IS_IMPLEMENTS_INTERFACE_CLASS
API: remove GST_URI_HANDLER_CLASS
Allocate every structure that is directly written to the binary
registry with g_malloc0(). Otherwise some parts of it will be
uninitialized (struct padding because of alignment, etc) and
valgrind will complain about it.
Add a new variable that keeps track of the status of the control socket. This
allows us to avoid doing a read() on the control socket when we did not write
anything to it.
Fixes#568438.
Link libgstreamer with $(LIBM) as it uses math functions.
Add a configure check for socket and nsl library and add
them to LIBS if they're found. This is needed on Solaris
for socket() and gethostbyname().
Original commit message from CVS:
* gst/gstbus.c: (gst_bus_set_main_context), (gst_bus_create_watch):
Fix order of members in GstBusSource structure - the first member
must be the parent structure ie. GSource. Should make bus sources
attached to non-default main contexts work in all cases now (ie.
primarily in cases where the callback has a non-NULL user data
argument). Fixes#562170.
* tests/check/gst/gstbus.c: (test_custom_main_context):
Add unit test for the above, based on code by
Justin Karneges <justin at affinix com>.
Original commit message from CVS:
* gst/gst.c:
Create a registry if there is none also when the option for
gst-disable-registry-update has been selected. Fixes#567002
Original commit message from CVS:
* gst/gstbus.c: (gst_bus_wakeup_main_context):
The lock order should be maincontext > OBJECT_LOCK so we need to release
the object lock when waking up the mainloop to avoid deadlocks.
Original commit message from CVS:
* gst/gstbin.c: (gst_bin_set_index_func), (gst_bin_set_clock_func),
(gst_bin_change_state_func):
Use an iterator to set the clock and the index so that we can release
the object lock appropriately. Fixes#566393.
Original commit message from CVS:
* gst/gstindex.c: (gst_index_get_type):
Add a debugging category for GstIndex, first little step in making
indexing top-notch.
Original commit message from CVS:
* gst/gstelement.c: (gst_element_message_full),
(gst_element_pads_activate):
* gst/gstobject.c: (gst_object_dispatch_properties_changed):
* gst/gstutils.c: (gst_pad_proxy_getcaps), (gst_pad_proxy_setcaps),
(gst_pad_add_data_probe_full), (gst_pad_add_event_probe_full),
(gst_pad_add_buffer_probe_full), (gst_pad_remove_data_probe),
(gst_pad_remove_event_probe), (gst_pad_remove_buffer_probe):
Assign debug statements to relevant categories instead of the 'default'
category so they don't get lost in debugging.
Original commit message from CVS:
* gst/gstdebugutils.c:
Add some ideas, how to make the graph smaller.
* gst/gstutils.c:
Add a comment from a debug session.
* libs/gst/base/gstbasetransform.c:
Log more context.
* libs/gst/controller/gstinterpolationcontrolsource.c:
Indet.
* plugins/elements/gstcapsfilter.c:
Fix typo in docs.
Original commit message from CVS:
* gst/gstbus.c: (gst_bus_dispose), (gst_bus_get_property),
(gst_bus_wakeup_main_context), (gst_bus_set_main_context),
(gst_bus_post), (gst_bus_source_prepare), (gst_bus_source_finalize),
(gst_bus_create_watch):
Make GstBusSource work with non-default main contexts (#562170).
* tests/check/gst/gstbus.c: (message_func_eos), (message_func_app),
(test_watch), (test_watch_with_custom_context), (gst_bus_suite):
Add test case for GstBusSource with a non-default main context.
* tests/check/libs/.cvsignore:
Ignore more.
Original commit message from CVS:
* gst/gstregistrybinary.c: (unpack_element), (unpack_const_string),
(unpack_string)::
Wrap multi-line macros in G_STMT_{START|END}.
Original commit message from CVS:
* docs/gst/gstreamer-sections.txt:
* gst/gstquark.c:
* gst/gstquark.h:
* gst/gstquery.c: (gst_query_new_uri), (gst_query_set_uri),
(gst_query_parse_uri):
* gst/gstquery.h:
API: Add URI query type. This is useful to query the URI
of a sink/source element and can be used by demuxers that
need to get data from other files.
This query should go upstream by default.
Fixes bug #562949.
* plugins/elements/gstfdsink.c: (gst_fd_sink_query):
* plugins/elements/gstfdsrc.c: (gst_fd_src_class_init),
(gst_fd_src_query):
* plugins/elements/gstfilesink.c: (gst_file_sink_query):
* plugins/elements/gstfilesrc.c: (gst_file_src_class_init),
(gst_file_src_query):
Implement URI query.
Original commit message from CVS:
* gst/gstghostpad.c:
* tests/check/gst/gstghostpad.c:
Don't forward gst_pad_set_caps() on a source ghostpad to its target.
That would cause the ghostpad to emit notify::caps two times (fist
from gst_pad_set_caps() and after from on_src_target_notify()).
Original commit message from CVS:
* gst/gstghostpad.c:
* tests/check/gst/gstghostpad.c:
In a source ghostpad, when caps are changed in the target pad, the
change needs to be reflected in the ghostpad.
Fixes#564863.
Original commit message from CVS:
* gst/gstutils.c: (gst_element_found_tags_for_pad):
Add FIXME for 0.11 to set the pad as message source and not
the element. Otherwise it's impossible to detect for which
pad the tags were found without adding an event probe
or something similar to the pad.
Original commit message from CVS:
* docs/gst/gstreamer-sections.txt:
* gst/gsttagsetter.c:
* gst/gsttagsetter.h:
Rename api added in previous commit and add since tag to docs.
API: gst_tag_setter_reset_tags()
Original commit message from CVS:
* docs/gst/gstreamer-sections.txt:
* gst/gsttagsetter.c:
* gst/gsttagsetter.h:
Add function to reset tagsetter for element reuse.
API: gst_tag_setter_flush()
Original commit message from CVS:
* gst/gsttaglist.c:
* tests/check/gst/gsttag.c:
More complete unit tests. Fix handling of empty taglists (they were
not merged before).
Original commit message from CVS:
* gst/gsttaglist.h:
* gst/gsttagsetter.c:
Update GstTagSetter and GstTagMergeMode documentation. Mention
that tags can come from events and from application. Fix example.
Original commit message from CVS:
* gst/gstutils.c: (gst_element_get_compatible_pad):
Check if the caps of the pads are compatible before returning
a pad and claiming it is compatible. This, among other things,
fixes a bug with gst-launch where an incompatible pad is chosen
and linking fails. Fixes bug #544003.
Original commit message from CVS:
Patch by: 이문형 <iwings at gmail dot com>
* gst/gstpoll.c: (gst_poll_fd_ctl_write), (gst_poll_fd_has_error):
Adds support for FD_CONNECT event (win32). See #562258.
Original commit message from CVS:
Patch by: Simon Holm Thøgersen <odie at cs dot aau dot dk>
* gst/gstregistry.c: (gst_registry_scan_path_level):
Reduce the number of stat() calls for every file from three times
to one time. Fixes bug #560360.
Original commit message from CVS:
* gst/gstregistrybinary.c:
In win32 codepath, if we fail to write the registry, create the
directory for it and try again, matching the behaviour in non-win32
codepaths.
Original commit message from CVS:
* gst/gstquery.c:
* gst/gstquery.h:
Add GstQueryType for custom queries instead of having to use the
not-so-very-convenient registration infrastructure to register new
types.
Original commit message from CVS:
Patch by: Andrew Feren <acferen at yahoo dot com>
* gst/gstobject.c: (gst_object_default_deep_notify):
Unref the GEnumClass after usage again. Fixes bug #561501.
Original commit message from CVS:
* gst/gstbin.c: (_gst_boolean_accumulator), (gst_bin_class_init),
(gst_bin_recalculate_latency), (gst_bin_do_latency_func),
(gst_bin_change_state_func):
* gst/gstbin.h:
Add do-latency signal with the old default fallback implementation. This
allows for custom latency calculations for when the default is not
sufficient.
API: GstBin::do-latency signal.
Original commit message from CVS:
* docs/gst/gstreamer-sections.txt:
* gst/gstbin.c: (gst_bin_recalculate_latency),
(gst_bin_change_state_func):
* gst/gstbin.h:
Add method to recalculate and redistribute the latency on a bin.
API: gst_bin_recalculate_latency().
Original commit message from CVS:
* gst/gstbuffer.c: (gst_buffer_finalize):
Avoid costly typechecking for trivially correct pointers.
* gst/gstpoll.c: (gst_poll_wait):
Add some G_LIKELY here and there.
* libs/gst/base/gstadapter.c: (gst_adapter_push):
Add some debug info.
Original commit message from CVS:
* gst/gstregistrybinary.c: (gst_registry_binary_write_chunk),
(gst_registry_binary_initialize_magic),
(gst_registry_binary_write_cache),
(gst_registry_binary_check_magic):
* gst/gstregistrybinary.h:
Don't write and check a CRC for the binary registry file. It's
guaranteed that the registry is completely written (it's first written
to a temporary file and then moved) and if the registry was corrupted
by some hardware failure we would have bigger problems.
Bump binary registry version to 0.10.21.1 for this as it's an
incompatible change and to ensure that the registry gets rebuild
after the update.
This saves some milliseconds for reading/writing the registry.
Fixes bug #560399.
Original commit message from CVS:
* gst/gstvalue.c: (gst_type_is_fixed), (gst_value_is_fixed):
Reorganize some more, be more conservative with the GST_TYPE_ARRAY not
being fixed and inline the trivial check.
Original commit message from CVS:
* gst/gstcaps.c: (gst_caps_copy), (_gst_caps_free),
(gst_caps_merge_structure), (gst_caps_get_structure),
(gst_caps_copy_nth), (gst_caps_set_simple),
(gst_caps_set_simple_valist), (gst_caps_is_fixed),
(gst_caps_is_equal_fixed), (gst_caps_intersect),
(gst_caps_subtract), (gst_caps_normalize), (gst_caps_do_simplify),
(gst_caps_to_string):
Callgrind micro optimisations.
Avoid array bounds checks and force inline of trivial function.
* gst/gstobject.c: (gst_object_set_name_default):
-1 is equivalent to letting glib to the strlen but then there is more
room for optimisations and it's not our fault.
* gst/gststructure.c: (gst_structure_id_empty_new_with_size):
no need to clear the array, we're cool.
* gst/gstvalue.c: (gst_type_is_fixed), (gst_value_is_fixed):
The most common _is_fixed() check is done on fundamental glib base
types so we check this first instead of doing a huge amount of
useless GST_TYPE_ARRAY calls.
Original commit message from CVS:
* gst/gststructure.c: (gst_structure_id_empty_new_with_size):
No need to memset, we can clear the value ourselves.
* gst/gstvalue.c: (gst_type_is_fixed),
(gst_value_get_compare_func):
Some optimisations from a few callgrind sessions:
When checking if a type is fixed, check for trivial fundamental types
first before checking types for which we need to get the type followed
by the heavy duty type checks, this reduces the amount of
g_type_fundamental() calls a lot.
When getting the compare function, first check for our registered types.
If that fails, do the heavy duty g_type_is_a() checks, reduces the
amount of g_type_is_a() considerably.
Original commit message from CVS:
* gst/gstbin.c: (bin_handle_async_start),
(gst_bin_handle_message_func), (gst_bin_query):
* libs/gst/base/gstbasesink.c: (gst_base_sink_render_object),
(gst_base_sink_event), (gst_base_sink_change_state):
* libs/gst/base/gstbasesrc.c: (gst_base_src_perform_seek),
(gst_base_src_loop), (gst_base_src_change_state):
Copy seqnums from events to messages so that they can all be related
back to eachother.
Original commit message from CVS:
2008-11-04 Andy Wingo <wingo@pobox.com>
* gst/gstutils.c (gst_util_seqnum_next): Refactor for clarity.
Also add API: to previous changelog entry.
Original commit message from CVS:
2008-11-04 Andy Wingo <wingo@pobox.com>
Add sequence numbers to events and messages. See #559250.
* gst/gstutils.c (gst_util_seqnum_next, gst_util_seqnum_compare):
New functions.
* gst/gstevent.h:
* gst/gstevent.c (_gst_event_copy, gst_event_new): Initialize new
events with a new sequence number, and copy it when copying.
(gst_event_get_seqnum, gst_event_set_seqnum): Accessors for an
event's sequence number.
* gst/gstmessage.h:
* gst/gstmessage.c (_gst_message_copy, gst_message_new_custom):
(gst_event_get_seqnum, gst_event_set_seqnum): As with events, so
with messages.
* docs/gst/gstreamer-sections.txt: Add new functions to the docs.
Original commit message from CVS:
* gst/gstsegment.c: (gst_segment_set_newsegment_full):
Set the last_stop to a more meaningful position when configuring the
segment. ie. the start/stop of the segment or clipped against the
updated segment boundaries.
* tests/check/gst/gstsegment.c: (GST_START_TEST):
Add some unit tests for the last_stop.
Original commit message from CVS:
* docs/gst/gstreamer-sections.txt:
* gst/gstutils.h:
API: Move float endianness conversion macros from libgstfloatcast
to core as it's useful in general, even in core. Fixes bug #555196.
This adds GDOUBLE_FROM_BE, GDOUBLE_FROM_LE, GDOUBLE_TO_BE,
GDOUBLE_TO_LE, GDOUBLE_SWAP_LE_BE, GFLOAT_FROM_BE, GFLOAT_FROM_LE,
GFLOAT_TO_BE, GFLOAT_TO_LE, GFLOAT_SWAP_LE_BE.
Also add GST_READ_ and GST_WRITE_ macros for floats and doubles:
GST_READ_FLOAT_LE, GST_READ_FLOAT_BE, GST_READ_DOUBLE_LE,
GST_READ_DOUBLE_BE, GST_WRITE_FLOAT_LE, GST_WRITE_FLOAT_BE,
GST_WRITE_DOUBLE_LE, GST_WRITE_DOUBLE_BE.
Original commit message from CVS:
* gst/gstpad.c: (gst_pad_configure_src):
When we use gst_pad_alloc_buffer() without wanting to set the caps we
also don't need to check if the caps are compatible because the caller
presumably is going to perform its own custom checks. Fixes some cases
where basetransform elements would error out when it was not needed.
Original commit message from CVS:
* docs/design/part-TODO.txt:
Remove item from the todo list because it was fixed with the latency
state change rewrites.
* docs/design/part-seeking.txt:
* docs/design/part-segments.txt:
Update some docs.
* gst/gstevent.c: (gst_event_new_new_segment_full),
(gst_event_parse_new_segment_full), (gst_event_new_buffer_size),
(gst_event_parse_buffer_size), (gst_event_new_qos),
(gst_event_parse_qos), (gst_event_new_seek),
(gst_event_parse_seek), (gst_event_new_latency),
(gst_event_parse_latency):
Use quarks to construct and parse events.
* gst/gstquark.c: (_priv_gst_quarks_initialize):
* gst/gstquark.h:
Add some more quarks to the table.
Emit a warning when the quark tables are not in sync.
* tests/check/gst/gstbus.c: (GST_START_TEST):
Add an assert.
Original commit message from CVS:
* gst/gstbus.c: (gst_bus_source_finalize),
(gst_bus_add_watch_full_unlocked), (gst_bus_add_watch_full),
(gst_bus_enable_sync_message_emission),
(gst_bus_disable_sync_message_emission),
(gst_bus_add_signal_watch_full), (gst_bus_remove_signal_watch):
Fix deadlock, g_source_get_id() cannot be called in finalize.
Keep track of the watch source by keeping a pointer to the source object
instead.
Use the bus lock to protect access to the pointer to the current
watch source.
Original commit message from CVS:
Base on Patch by: Olivier Crete <tester at tester dot ca>
* gst/gstbus.c: (gst_bus_source_finalize),
(gst_bus_add_watch_full), (gst_bus_add_signal_watch_full):
Only allow one bus watch to be set at a time. This is necessary
because the dispatcher pops the message from the bus and the second
watcher will then get NULL or the next message (and the first won't
get this next message then, etc). If more than one "watcher" is
required signal watches should be used. Fixes bug #526044.
Original commit message from CVS:
* gst/gstxml.c:
Clarify that the save_thyself() and restore_thyself() virtual
functions of GstObject need to be overriden, not
gst_object_(save|restore)_thyself() which is impossible.
Fixes bug #555700.
Original commit message from CVS:
* gst/gstpad.c: (gst_pad_get_range), (gst_pad_pull_range):
Revert a patch from 21 months ago that broke caps negotiation in pull
mode. Basically, having a buffer pass over a pad will trigger the
setcaps function when caps change, just like in push mode.
Original commit message from CVS:
* gst/gstbin.c: (gst_bin_remove_func), (update_degree),
(gst_bin_handle_message_func):
The message src can be NULL, don't try to print the object names in that
case.
* libs/gst/base/gstbasesink.c: (gst_base_sink_pad_activate):
Add some more debug info.
* tests/check/pipelines/simple-launch-lines.c: (run_pipeline),
(GST_START_TEST):
Add some debug.
Fix the test, pull based sinks go ASYNC to PAUSED, just like other
scheduling modes.
Original commit message from CVS:
2008-10-08 Andy Wingo <wingo@pobox.com>
* gst/gstghostpad.h (GST_GHOST_PAD_CAST):
* gst/gstghostpad.c (GST_GHOST_PAD_CAST): Fix unintended API
removal: re-add GST_GHOST_PAD_CAST to the header.
Original commit message from CVS:
2008-10-08 Andy Wingo <wingo@pobox.com>
* gst/gstghostpad.h (GstProxyPad, GstProxyPadClass, GstGhostPad)
(GstGhostPadClass): Publically expose these structures so as to
allow easy subclassing from C. Hide the member data behind a
private opaque data pointer.
* gst/gstghostpad.c: Adapt to store instance data in the type
instance's private data region, not in the public struct.
Original commit message from CVS:
* gst/gstregistrybinary.c: (gst_registry_binary_write_cache):
If we can't get a cache file don't try to save something to it.
Dereferencing NULL pointers usually isn't a good idea.
