meson: dist pot file in tarballs

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/618>

aggregator: Document that samples_selected() must only be called from the aggregate() function

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/607>

aggregator: Don't automatically adjust segment if subclass provided one

On the first buffer the base class would update the segment position
based on the start-time-selection. If the subclass provides its own
segment this will caused unexpected behaviour and override segment
information that was explicitly set by the subclass.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/600>

Release 1.17.90

aggregator: fix documentation for samples-selected and buffer-consumed

GI expects the instance parameter to be documented, omitting it
leads to a msismatched output in the gir.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/592>

aggregator: Add optional GstStructure info parameter to "samples-selected" signal

Subclasses can use this to provide more information, for example
audioaggregator could provide the offset into the output buffer where
the next data is going to be filled.

See https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/805

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/590>

aggregator: add segment, pts, dts and duration to samples-selected

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/588>

Meson: Override gst-tester-1.0 program to find it in other modules

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/586>

fix clang 10 warnings

the typesystem checks in g_atomic_pointer_compare_and_exchange
seem to trigger some false positives with clang 10

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/584>

aggregator: fix iteration direction in skip_buffers

Subclasses use the pad segment to determine whether a buffer
should be skipped, we thus don't want to check if a buffer
needs to be skipped before processing the segment it's part
of.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/585>

gsturi: Add new API for storing unmodified userinfo / fragment

New API: gst_uri_from_string_escaped()

Identical to gst_uri_from_string() except that the userinfo and
fragment components of the URI will not be unescaped while parsing.

This is needed for correctly parsing usernames or passwords with `:`
in them such as reported at:
https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/831

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/583>

tests: Add more tests for gsturi

Add tests that exercise unescaping of userinfo and fragments.

Also convert to a modular macro-based definition so that we can reuse
the list of tests in the next commit.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/583>

inspect: Print preset description when available

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/582>

aggregator: expose sample selection API

See https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/771
for context.

This exposes new API that subclasses must call from their
aggregate() implementation to signal that they have selected
the next samples they will aggregate: gst_aggregator_selected_samples()

GstAggregator will emit a new signal there, `samples-selected`,
handlers can then look up samples per pad with the newly-added
gst_aggregator_peek_next_sample.

In addition, a new FIXME is logged when subclasses haven't actually
called `selected_samples` from their aggregate() implementation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/549>

gsturi: unescape '=' in http query

Don't use percent-encoding for '=' in http queries.
'=' in the following kind of http query should be maintained.
example:
?token=exp=123~acl=/QualityLevels(*~hmac=0cb ...

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/580>

basetransform: handle invalid subclass implementation for fixate_caps

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/575>

baseparse: Don't push pointless new segment events

In 1.0, there is no concept of segment update, so don't push new
identical segments.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/578>

baseparse: Fix seqnum handling in pull mode

After a seek in pull mode, we should use the seek seqnum for all
following operations, not some random seqnums

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/577>

check: suppress g-ir-scanner warnings

Make g-ir-scanner skip all those check macros that are
not useful for or usable from bindings.

gstcheck.h:209: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_message_error'
gstcheck.h:212: Warning: GstCheck: Unknown namespace for symbol 'assert_message_error'
gstcheck.h:251: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_int'
gstcheck.h:267: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_int'
gstcheck.h:280: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_int_hex'
gstcheck.h:299: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_int_hex'
gstcheck.h:310: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_int64'
gstcheck.h:327: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_int64'
gstcheck.h:340: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_int64_hex'
gstcheck.h:358: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_int64_hex'
gstcheck.h:369: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_uint64'
gstcheck.h:386: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_uint64'
gstcheck.h:399: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_uint64_hex'
gstcheck.h:417: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_uint64_hex'
gstcheck.h:428: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_string'
gstcheck.h:444: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_string'
gstcheck.h:455: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_float'
gstcheck.h:474: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_float'
gstcheck.h:487: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_pointer'
gstcheck.h:506: Warning: GstCheck: Unknown namespace for symbol 'assert_equals_pointer'
gstcheck.h:517: Warning: GstCheck: Unknown namespace for symbol 'fail_unless_equals_clocktime'
gstcheck.h:534: Warning: GstCheck: Unknown namespace for symbol 'MAIN_START_THREADS'
gstcheck.h:547: Warning: GstCheck: Unknown namespace for symbol 'MAIN_START_THREAD_FUNCTIONS'
gstcheck.h:555: Warning: GstCheck: Unknown namespace for symbol 'MAIN_START_THREAD_FUNCTION'
gstcheck.h:626: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_CRITICAL'
gstcheck.h:628: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_CRITICAL'
gstcheck.h:640: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_WARNING'
gstcheck.h:652: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_OBJECT_REFCOUNT'
gstcheck.h:661: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_OBJECT_REFCOUNT_BETWEEN'
gstcheck.h:676: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_CAPS_REFCOUNT'
gstcheck.h:679: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_BUFFER_REFCOUNT'
gstcheck.h:682: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_MINI_OBJECT_REFCOUNT'
gstcheck.h:690: Warning: GstCheck: Unknown namespace for symbol 'ASSERT_SET_STATE'
gstcheck.h:729: Warning: GstCheck: Unknown namespace for symbol 'tcase_skip_broken_test'
gstcheck.h:740: Warning: GstCheck: Unknown namespace for symbol 'tcase_skip_broken_loop_test'

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/579>

utils: silence g-ir-scanner warnings about float conversion macros

663: Warning: Gst: symbol='GFLOAT_TO_LE': Unknown namespace for symbol 'GFLOAT_TO_LE'
664: Warning: Gst: symbol='GFLOAT_TO_BE': Unknown namespace for symbol 'GFLOAT_TO_BE'
665: Warning: Gst: symbol='GDOUBLE_TO_LE': Unknown namespace for symbol 'GDOUBLE_TO_LE'
666: Warning: Gst: symbol='GDOUBLE_TO_BE': Unknown namespace for symbol 'GDOUBLE_TO_BE'
669: Warning: Gst: symbol='GFLOAT_TO_LE': Unknown namespace for symbol 'GFLOAT_TO_LE'
670: Warning: Gst: symbol='GFLOAT_TO_BE': Unknown namespace for symbol 'GFLOAT_TO_BE'
671: Warning: Gst: symbol='GDOUBLE_TO_LE': Unknown namespace for symbol 'GDOUBLE_TO_LE'
672: Warning: Gst: symbol='GDOUBLE_TO_BE': Unknown namespace for symbol 'GDOUBLE_TO_BE'
678: Warning: Gst: symbol='GFLOAT_FROM_LE': Unknown namespace for symbol 'GFLOAT_FROM_LE'
679: Warning: Gst: symbol='GFLOAT_FROM_BE': Unknown namespace for symbol 'GFLOAT_FROM_BE'
680: Warning: Gst: symbol='GDOUBLE_FROM_LE': Unknown namespace for symbol 'GDOUBLE_FROM_LE'
681: Warning: Gst: symbol='GDOUBLE_FROM_BE': Unknown namespace for symbol 'GDOUBLE_FROM_BE'

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/579>

math-compat.h: silence g-ir-scanner warnings

Easier to just make g-ir-scanner skip this header via #ifndef __GI_SCANNER__
than maintain different sets of headers in the meson.build file.

Warning: Gst: symbol="rint": Unknown namespace for symbol "rint"
Warning: Gst: symbol="rintf": Unknown namespace for symbol "rintf"
Warning: Gst: symbol="isnan": Unknown namespace for symbol "isnan"
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/579>

event, query: fix g-ir-scanner warnings

gstevent.h:72: Warning: Gst: symbol='FLAG': Unknown namespace for symbol 'FLAG'
gstquery.h:76: Warning: Gst: symbol='FLAG': Unknown namespace for symbol 'FLAG'

Use _FLAG(xyz) instead of FLAG(xyz) to silence g-ir-scanner
warnings about this internal helper define.

It's also slightly more hygienic.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/579>

tracer: declare GstTracer API stable

It's been around for more than 4 years and people have built
lots of stuff on top of it, doesn't really make sense to keep
it marked as unstable. We're unlikely to change it now, and
we can always deprecate it and make a new one if needed.

This stabilises the following API:
- gst_tracer_register()
- gst_tracing_get_active_tracers()
- gst_tracing_register_hook()
- gst_tracer_record_new()
- gst_tracer_record_log()

Might also help a bit with #424

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/576>

harness: unref sink/src caps after deactivating pads

Otherwise, access to the harness' sink/src caps is racy between any caps
query performed by an element and gst_harness_teardown().

Fixes: https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/794
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/574>

check: Always mark _ck_assert_failed as noreturn

So that we can use `fail` like `g_assert_not_reached`.

The comment is apparently wrong or outdated, as GCC considers it legal
for noreturn-marked functions to return using longjmp.

See the thread at
https://gitlab.freedesktop.org/gstreamer/gst-plugins-ugly/-/merge_requests/59#note_576422

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/573>

buffer: improve seqnum fallback warning message

Print target CPU we're building for.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/565>

buffer: fix meta sequence number fallback on rpi

The global seqnum variable wasn't actually increased in
the fallback code path, leading to all buffers getting
a seqnum of 0. Which also made the unit test fail.

This affects platforms/toolchains that don't have
64-bit atomic ops such as when compiling for armv7 rpi.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/565>

meson: add a plugin summary

This summary displays a list of plugins which
have been enabled.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/560>

gst-inspect: fix memory leak

With meson configure option: -Db_sanitize=address, the following
issue is seen while running the test "tools_gstinspect":

Running suite(s): gst-inspect

=================================================================
==20880==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 51 byte(s) in 9 object(s) allocated from:
    #0 0x7ffb4dbb0b40 in __interceptor_malloc (/usr/lib/x86_64-linux-gnu/libasan.so.4+0xdeb40)
    #1 0x7ffb4cdf1ab8 in g_malloc (/usr/lib/x86_64-linux-gnu/libglib-2.0.so.0+0x51ab8)

SUMMARY: AddressSanitizer: 51 byte(s) leaked in 9 allocation(s).
0%: Checks: 1, Failures: 0, Errors: 1

GOptionEntry man page states that: "Please note that parsed arguments need to be freed separately (see GOptionEntry)."
Thus, free the 'min_version' string that has been allocated but never freed.

Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/572>

gst/bufferpool: only resize in reset when maxsize is larger

Only resize the buffer if the maxsize is larger then the configued pool
size.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/570>

build/coretracers: add dep on threads

Fixes the following build error and missing '-pthread' argument when
linking:

subprojects/gstreamer/plugins/tracers/libgstcoretracers.so.p/gstleaks.c.o: In function `gst_leaks_tracer_setup_signals':
/work/build32/../subprojects/gstreamer/plugins/tracers/gstleaks.c:919: undefined reference to `pthread_atfork'

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/571>

directcontrolbinding: Properly initialize default `last_value`

It was zero and in some condition it means that the control binding
values where ignored (as shown in the test). Setting it to MAXDOUBLE
so that the first time we sync the values from a a timestamp in the
right range the proper value is computed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/564>

info: Fix possible broken debug output on Windows

Depending on Windows codepage setting, some characters could
be broken when printing on terminal. Fortunatly g_print* family will
take care Windows codepage.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/566>

pad: More explicitly explain how to post errors on GST_FLOW_ERROR

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/568>

latency tracer: Fix leaks in the reported latency trace

The stack item was not freed as it was supposed, causing leaks.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/269>

latency tracer: Fix unsafe and NULL pointer accesses

Use thread-safe accesses to pad peers and parent objects. This
fixes some crashers and all the non-safe access patterns I could
spot. There's still some weirdness when using the latency
tracer on pipeline chains that aren't yet linked, but this
at least stops it segfaulting.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/269>

Revert "gstinfo: Check threshold for category from macro"

This reverts commit dcece2a878b88335fd1990dbeeb88bdeacba0f06.

This increased the code size and number of branches for all debug statements.

Fixes #564

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/563>

aggregator: expose gst_aggregator_finish_buffer_list API

See https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/1276

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/562>

tests/elements/leaks.c: check get_tracer_by_name return value

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/557>

basesrc: Deprecate gst_base_src_new_seamless_segment()

It can be replaced by gst_base_src_new_segment()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/490>

basesrc: Add new API for handling GstSegment update by subclass

Add API gst_base_src_new_segment() for subclass to be able to
signalling new GstSegment which should be applied to following
buffers.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/490>

taglist: Stop inlining gst_tag_list_copy

This way it gets exposed to bindings through GObject Introspection.

Same logic as with d1b2d3429c66d80b8d38f9afc6a8dfca49f3a71a

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/561>

check: Use g_thread_yield instead of g_usleep(1)

Since the commit
https://gitlab.gnome.org/GNOME/glib/-/commit/01c02ac08b682de622930b1278c9c14d0ffe6c49,
g_usleep(1) will be translated to Sleep(1) on Windows which means
sleep in 1 millisecond. But GLib provides g_thread_yield() API
which is exactly what we required here for thread context switching.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/559>

deviceprovider: Do static probe on start as fallback

For providers that don't support dynamic probing, just fall back to doing
a static one on start() to make the UI developers life easier.

This also means that the monitor doesn't need to call _can_monitor() before
calling start.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/353>

gstelementfactory: Fix missing features in case a feature moves to another filename

In case a plugin filename was renamed with the plugin being in the registry cache
the features were not loaded after the rename:

1) Cache of old/gone filename was loaded, features added
2) New filename was loaded, features where not added because
   they were already found in the registry.
3) In the end stale cache entries for files which are no longer there
   are removed, including the wanted features.
4) The cache gets updated without the features.

Fix this by also checking at (2) that the found feature is from the loaded plugin
and not from some stale cache entry.

This affected directsoundsink where libgstdirectsoundsink.dll was renamed
to libgstdirectsound.dll, losing the directsoundsink element in the process.

Fixes #290

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/102>

meson: set release date from .doap file for releases

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/555>

po: update POTFILES

xgettext: error while opening "gst/parse/grammar.y" for reading: No such file or directory

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/586

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/556>

ptp: Add GNU Hurd to the list of supported platforms and fix the Solaris name

https://mesonbuild.com/Reference-tables.html#operating-system-names has
the table of all supported names right now.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/583

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/554>

element: When removing a ghost pad also unset its target

Otherwise the proxy pad of the ghost pad still stays linked to some
element inside the bin, which is not allowed anymore according to the
topology.

In 2.0 this should be fixed more generically from inside GstGhostPad but
currently there is no way to get notified that the ghost pad is
unparented.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/553>

Back to development

Release 1.17.2

stream: Don't use GST_FIXME_OBJECT() when generating a random stream id if the caller didn't provide one

That would call into gst_info_describe_stream(), which takes the same
mutex a second time and then deadlocks.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/552>

gstutils: fix link in parse_bin_from_description_full doc

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/475>

gstbin: Fix docs typo

element-added-deep -> deep-element-added in the GstBin doc header

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/547>

docs: Specify UTF-8 encoding everywhere

Otherwise some non-English character might be broken depending on
OS and/or locale.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/545>

hooks: add opt-in cache update hook

CI now checks that plugin caches are up to date, for example
when adding a new property to an element.

This is something pretty easy to forget, and it can be checked
in a pre-commit hook provided the cache generator generates no
unnecessart diff on the developer's machine. This is now the
case for me in core, -base and -good, and eventually all the
repositories should behave appropriately, at least on my
development machine.

The new check in the pre-commit hook is only performed when
the user is in the development environment (it checks
$MESON_BUILD_ROOT to determine that), and when the developer
has opted-in by setting `GST_CACHE_HOOK=enabled`. That is
because the hook will actually rebuild the cache with ninja,
and modify it in the source directory, that's not a behaviour
we want to enable by default.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/546>

gst-inspect: Use gst_info_strdup_vprintf to print string

g_vprintf() will write a string binary to stdout directly using fwrite().
So, depending on character in the string, fwrite to stdout can
print broken one but printf family might not cause the issue.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/544>

bus: clear bus->priv->signal_watch immediately when the source is removed

There is a race-condition that can trigger the assertion in
gst_bus_add_signal_watch_full():

If gst_bus_add_signal_watch_full() is called immediately after
gst_bus_remove_signal_watch() then bus->priv->signal_watch may still be set
because gst_bus_source_dispose() or gst_bus_source_finalize() was not yet
called.
This happens if the corresponding GMainContext has the source queued for
dispatch. In this case, the following dispatch will only unref and delete
the signal_watch because it was already destroyed. Any pending messages
will remain until a new watch is installed.

So bus->priv->signal_watch can be cleared immediately when the watch is
removed. This avoid the race condition.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/543>

gst-hotdoc-plugins-scanner: serialize interfaces

gst-hotdoc-plugins-scanner: don't instantiate base classes

gst-hotdoc-plugins-scanner: serialize parents in hierarchy when needed

docs: Document signals and properties only for current type

doc: Stop documenting properties from parents

registry: Print the pointer when printing features

This is what we do everywhere else too, useful for debugging.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/542>

registry: Use a toolchain-specific registry file on Windows

If we load a plugin registry for MinGW plugins when running with MSVC,
we will have to write out the whole cache again, and vice-versa. Just
use separate cache files so that the cache is actually useful.

Closes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/427

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/542>

ghostpad: Deprecate gst_ghost_pad_construct()

Instead do everything it did as part of GObject::constructed() and
change the function to always return TRUE.

gst_ghost_pad_construct() was meant to be called by subclasses right
after construction of the object to finish construction as it can fail
in theory. In practice it's impossible for it to fail, even more so if
called directly from GObject::constructed(): The only failure condition
is if the newly created proxy pad already has a parent, which is
impossible at this point as nothing else can have a reference to it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/540>

tools: gst-inspect, add option '--color' with short option '-C'

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/498>

info: Add a printf extension for ClockTime/ClockTimeDiff pointers

Using both GST_TIME_FORMAT+GST_TIME_ARGS and friend is cumbersome, this
makes it sensibly more user friendly.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/535>

Back to development

Release 1.17.1

doc: Stop recording if building from source/release in plugin cache

Closes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/537

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/537>

Fix up and add various "Since" markers and other related docs fixes

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/536>

systemclock: Fix clock time conversion on Windows/xbox

The returned ratio can be bigger than GST_SECOND, in which case we would
forever return 0 for the system clock time. Even in other cases if it's
close to GST_SECOND it would result in accuracy loss.

Instead of doing the division by GST_CLOCK_TIME_NONE during
initialization once, do it every time the clock time is requested.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/575

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/534>

gst: Delay creation of threadpools

Since glib 2.64, gthreadpool will start waiting on a GCond immediately upon
creation. This can cause issues if we fork *before* actually using the
threadpool since we will then be signalling that GCond ... from another process
and that will never work.

Instead, delay creationg of thread pools until the very first time we need
them. This introduces a minor (un-noticeable) delay when needing a new thread
but fixes the issues for all users of GSTreamer that will call gst_init, then
fork and actually start pipelines.

See https://gitlab.gnome.org/GNOME/glib/-/issues/2131 for more context.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/531>

inputselector: Push event on selected pad only when one is selected

Calling `gst_input_selector_get_active_sinkpad` triggers sinkpad
selection but won't notify about it, leading  to breaking code that
relying on it. This new code added as part of
63ccf45395ce734a2044a58193939a3eda50eb0c was thought to be triggered only
when a pad was already selected and not change the behavior otherwise
so this commit makes sure it is actually the case.

Fixes https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/766

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/522>

check: Fix datetime unit test for builds without assert

If built with assertions disabled, we need to ensure the variable is properly
reset before testing

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/530>

message: Add unit test for GST_MESSAGE_RATE_CHANGE

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/528>

event: Add unit test for INSTANT_RATE_CHANGE and INSTANT_RATE_SYNC_TIME

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/528>

segment: Add unit test for GST_SEEK_FLAG_INSTANT_RATE_CHANGE

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/528>

docs: Update plugins cache

docs: Fix the way we mark properties mutability

When nothing is specified, we should default to NULL, not PLAYING

aggregator: Fix StartTimeSelection enum type registration

Make it thread-safe and use the actual C identifiers for the "name"
field, as otherwise gobject-introspection will fall apart.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/527>

docs: Update plugins cache

tracers/leaks: fix reentrancy issues with the custom signal handlers

The signal handlers were performing mutex operations in the signal handlers
which is bad idea that may lead to deadlocks.

1. Implement a separate signal thread to handle the signals.
2. Use the glib provided signal GSource to avoid performing operations in
   the signal handler.

Fix #186

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/487>

gst_private.h: increse padding in struct _GstClockEntryImpl

When compiling for 32bit ios arm, the static assert that the
GstClockEntryImpl smaller or equal to the struct _GstClockEntryImpl
triggered. (they were 12bytes off).

To fix this, the padding is increased by 12 bytes (on 32bit).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/525>

paramspecs: add 'Since: 1.18' markers for new param spec flags

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/524>

doc: Add GstObject specific GParamFlags

Document "controllable", "mutable-{ready, paused, playing}" and
"conditonally-available" GParamFlags

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/283>

paramspecs: Add a GParamSpecFlag to indicate the property might not always exists

Add new flag for users to notice that the property is not guaranteed
to exist depending on environment.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/283>

check: Avoid race with leaks test

The problem is that the taskpool might not have completely drained by the time
we check for leaks.

Instead, ensure all tasks have stopped before testing for valid results.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/523>

queue: protect against lost wakeups for iterm_del condition

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/513>

docs: Prevent potential NULL pointer dereference when serializing plugin object types

CID 1464007

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/516>

padtemplate: Directly unreference the documentation caps

The public-facing API has a (valid) protection against NULL caps. We can just
directly remove it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/515>

utils: fix markdown link to #GstPluginAPIFlags

padtemplate: expose getters and setters "documentation caps"

This can be used in elements where the caps of pad templates
are dynamically generated and dependent on the environment.

An example is x265enc.

utils: expose GstPluginAPIFlags

These can be passed to gst_type_mark_as_plugin_api, to inform
plugin cache generation.

For now a single flag is specified, "IGNORE_ENUM_MEMBERS", it
can be used for dynamically generated enums to avoid documenting
environment-specific enumeration members. An example is
GstX265EncTune.

identity, clocksync: implement provide_clock

Since those are using the clock for sync, they need to also
provide a clock for good measure. The reason is that even if
downstream elements provide a clock, we don't want to have
that clock selected because it might not be running yet.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/509>

structure: Quickly document serialization format

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/510>

structure: Reflow the SECTION comment

Removing trailing whitespaces and avoiding to exceed 80chars

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/510>

queue2: Defer downstream bitrate query to the streaming thread.

When we want to perform a downstream bitrate query, just
set the reconfigure flag on the srcpad and get the streaming
thread to do it. That avoids emitting a downstream query
when receiving the upstream RECONFIGURE event - which can
lead to deadlocks if downstream is sending the event from
within a lock - e.g. input-selector.

If querying the downstream bitrate changes the cached
value, then make sure to update our buffering state
and potentially post a BUFFERING message to the application.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/566
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/501>

inputselector: Avoid deadlock when requesting pads

The deadlock was the following:
* One thread requests a new pad, the internal lock is kept while adding the pad
* Another thread (or the same one) requests the internal links of a pad (could
be that pad)... which also requires that lock.

That internal lock is not required when adding the pad to the element (which is
the last action when requesting a new pad). The fact it will be actually used
will be *after* the request pad function is released.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/512>