Original commit message from CVS:
* gst/schedulers/gstoptimalscheduler.c: (gst_opt_scheduler_init),
(create_chain), (destroy_chain), (create_group), (destroy_group),
(add_to_group), (merge_groups), (group_elements), (group_inc_link),
(group_dec_link), (gst_opt_scheduler_pad_link),
(group_inc_links_for_element), (group_migrate_connected):
Call group_inc_link with the proper src->sink ordering --
break this, and we break sort_chain. patch from wingo for bug
147713.
Partially revert patch 1.89. When adding a loop based element to
the scheduler, the links to other groups are automatically followed
and incremented. This should not happen because the bin will call
pad_link explicitly for those connection, resulting in them counted
twice. Results in assertion failure on pipeline cleanup.
Original commit message from CVS:
* gst/gstvalue.c: (gst_greatest_common_divisor):
use ints and return ints, fractions only use ints, too, so this
avoids accidently casting multiplications to unsigned
(gst_value_lcopy_fraction): it's ints, not uint32
(gst_value_set_fraction): disallow minint, multiplying and negation
are broken with it
(gst_value_fraction_multiply): fix to make large numbers work and get
rid of the assumption that the multiplication of two ints fits an
int64 - dunno if that's true for all systems
* testsuite/caps/Makefile.am:
* testsuite/caps/fraction-multiply-and-zero.c:
(check_multiplication), (check_equal), (zero_test), (main):
add tests for all the stuff above
* testsuite/caps/value_compare.c: (test1):
fix comment
* tests/.cvsignore:
* testsuite/caps/.cvsignore:
* testsuite/debug/.cvsignore:
* testsuite/dlopen/.cvsignore:
* testsuite/states/.cvsignore:
get up to date
Original commit message from CVS:
* gst/gstelement.h:
Don't GST_ERROR_OBJECT smpty strings - Solaris doesn't like NULL
strings.
* gst/gstelement.c (gst_element_class_init):
GError's are boxed, not objects
* gst/gstmarshal.list:
update list for the fixed error signal
Original commit message from CVS:
2004-07-14 Andy Wingo <wingo@pobox.com>
* gst/gsttag.c: Add a tag merge func for pointers. The header was
there all along, but the function wasn't. (guile-gstreamer's build
system uses the address of the function -- I wasn't actually
trying to use this.)
2004-07-13 Andy Wingo <wingo@pobox.com>
* gst/gstpad.c (gst_pad_try_set_caps): Naive link functions (such
as gst_pad_proxy_pad_link) just link to every other pad when they
are called. In the case where the graph has cycles, this will mean
that a call to try_set_caps will recurse. Allow this recursion
and return OK, while we wait for the first try_set_caps to give a
proper return value.
(gst_pad_link_call_link_functions): Since this function is the
only one to set the NEGOTIATING flag on a pad, if the flag is set
it means that the link functions have indirectly recursed. If this
happens, error out to avoid infinite recursion and an eventual
SEGV.
(gst_real_pad_class_init): Remove a crufty GtkObject comment.
(gst_pad_proxy_getcaps): Intersect the result with the template
caps to ensure that the return value is valid.
2004-07-11 Andy Wingo <wingo@pobox.com>
* gst/gstdata.c (gst_data_is_writable): s/>=/>/. If there is only
one refcount, the calling function is the owner of the buffer.
Original commit message from CVS:
* gst/schedulers/gstoptimalscheduler.c: (group_dec_link),
(gst_opt_scheduler_pad_link), (group_migrate_connected):
Fix stupid warning when an element is to be migrated but
is already migrated.
Original commit message from CVS:
* gst/schedulers/gstoptimalscheduler.c: (group_dec_link),
(gst_opt_scheduler_pad_link), (group_migrate_connected):
Make sure that a single non-loop-based element does not
end up in a group. This fixes the testsuite again.
Original commit message from CVS:
* gst/schedulers/gstoptimalscheduler.c: (create_group),
(add_to_group), (merge_groups), (schedule_group),
(gst_opt_scheduler_get_wrapper), (group_elements),
(group_dec_link), (gst_opt_scheduler_pad_link),
(group_migrate_connected), (gst_opt_scheduler_pad_unlink),
(gst_opt_scheduler_iterate):
move isolated groups to a new chain.
Emit a warning instead of segfaulting in some error cases.
Fix a bug where the link count between groups was not calculated
correctly. Fixes#144510.
Original commit message from CVS:
compatibility fixes for Solaris 8/gcc 2.95
* configure.ac:
include libintl libs in LDFLAGS
* gstvalue.c (gst_value_deserialize_buffer):
cast isxdigit stuff to int to silence compiler warning
Original commit message from CVS:
* gst/gsttypes.h:
get rid of GST_O_READONLY, GST_FILE_MODE_READ and
GST_FILE_MODE_WRITE, I don't want them in the exported headers. It
just causes support madness
* gst/elements/gstfilesrc.c: (gst_filesrc_open_file):
make it work without this
* gst/indexers/gstfileindex.c: (_file_index_id_save_entries),
(gst_file_index_commit):
glib IO channels don't want binary mode
* testsuite/bytestream/filepadsink.c: (main):
* testsuite/bytestream/test1.c: (read_param_file):
use "rb" instead of GST_FILE_MODE_READ, it works on POSIX systems
Original commit message from CVS:
* gst/gstelement.c: (gst_element_class_init),
(gst_element_set_state), (gst_element_set_state_func):
virutalize gst_element_set_state, use set_state member in class
struct that was already added in 0.7 for this.
* gst/gstbin.c: (gst_bin_foreach), (set_kid_state_func),
(gst_bin_change_state):
make gst_bin_foreach works similar to other foreach functions, plug
memleaks in it. Make functions using it work with the new approach.
Document gst_bin_foreach, so it can be exported if we want to
* gst/gstbin.c: (gst_bin_class_init), (gst_bin_set_state):
use virtualized set_state to make set_state on bins set the state of
all its children.
Original commit message from CVS:
* configure.ac:
require valgrind >= 2.1 (fixes Gentoo bug 53967, see
http://bugs.gentoo.org/show_bug.cgi?id=53967)
* gst/gstpad.c: (gst_pad_alloc_buffer):
allow buffer_alloc functions to return NULL and allocate a normal
buffer in that case
Original commit message from CVS:
* gst/gstclock.h:
GST_SECOND shouldn't cause a conversion to unsigned.
* testsuite/clock/.cvsignore:
* testsuite/clock/Makefile.am:
* testsuite/clock/signedness.c: (main):
make sure it never will again
Original commit message from CVS:
2004-07-11 Andy Wingo <wingo@pobox.com>
* gst/gstbin.c (gst_bin_add_func): If we're adding an element
whose state is higher than the bin state, raise the bin state to
ensure that bin state := highest child state.
Original commit message from CVS:
2004-07-11 Andy Wingo <wingo@pobox.com>
* gst/gstbin.c (gst_bin_foreach): New static function. Calls a
procedure on the children of a bin. Assumes that the procedure can
change the set of children.
(set_kid_state_func): New static function.
(gst_bin_change_state): Use gst_bin_foreach to call
set_kid_state_func. Fixes a bug: if a child had a state-change
handler that removes it from the bin, there would be a segfault.
Hopefully it should also work in the case where the state-change
handler on one child adds or removes other children. In any case,
fixes should go to gst_bin_foreach.
Original commit message from CVS:
* gst/gstbin.c: (gst_bin_remove), (gst_bin_dispose):
* gst/gstthread.c: (gst_thread_dispose), (gst_thread_catch),
(gst_thread_main_loop):
Since remove is virtual in GstBin we must not assume the
elements GList to have anothing usefull.
Add some more logging to GstThread and be a bit more paranoid
when resetting the scheduler.
Set the state of the bin to NULL before removing the children.
Original commit message from CVS:
* gst/schedulers/gstoptimalscheduler.c:
(gst_opt_scheduler_pad_link), (group_elements_set_visited),
(element_get_reachables_func), (element_get_reachables),
(debug_element), (rechain_group), (group_migrate_connected),
(gst_opt_scheduler_pad_unlink):
Do not try to migrate decoupled elements to a new group since
they are not added to groups.
Original commit message from CVS:
* gst/schedulers/gstoptimalscheduler.c:
(chain_recursively_migrate_group), (create_group),
(schedule_group), (gst_opt_scheduler_pad_link),
(group_elements_set_visited), (element_get_reachables_func),
(element_get_reachables), (group_can_reach_group), (debug_element),
(rechain_group), (group_migrate_connected),
(gst_opt_scheduler_pad_unlink):
* testsuite/schedulers/Makefile.am:
Implemented group splitting and rechaining.
Fixes 143777 and 143777-2 in the testsuite.
Original commit message from CVS:
* gst/autoplug/gstspideridentity.c: (gst_spider_identity_chain):
extra debugging
* gst/gstevent.h:
* gst/gstinfo.c: (gst_debug_log_default):
print time nicely. add thread pointer until someone figures out
a completely portable way of getting at thread id's.
* gst/gstpad.c: (gst_pad_set_active), (gst_pad_link_try),
(_invent_event), (gst_pad_pull), (gst_pad_event_default_dispatch),
(gst_pad_call_chain_function):
extra debugging
* gst/schedulers/gstoptimalscheduler.c:
(get_group_schedule_function), (loop_group_schedule_function),
(gst_opt_scheduler_loop_wrapper), (gst_opt_scheduler_get_wrapper),
(pad_clear_queued), (gst_opt_scheduler_iterate):
rename BUFPEN and friends to DATAPEN since that's what they are.
Original commit message from CVS:
* configure.ac:
* gst/gstvalue.c: (gst_value_compare_enum),
(gst_value_serialize_enum), (gst_value_deserialize_enum),
(gst_value_can_compare), (gst_value_compare):
* testsuite/Makefile.am:
Fix enum serialization, deserialization, comparison in caps, add
a test to ensure that this continues working in the future.
Original commit message from CVS:
* gst/gstcaps.c: (gst_caps_save_thyself), (gst_caps_load_thyself):
Actually do something in these functions, like before the big
caps change. (bug #145137)
Original commit message from CVS:
* gst/gstinfo.h: wrap #pragmas in #ifdefs for the correct compiler
* gst/gstobject.h:
* gst/gstparse.h:
* gst/gsttrace.h:
* gst/gstxml.h:
Original commit message from CVS:
* gst/gstpad.c: (gst_pad_check_schedulers),
(gst_pad_can_link_filtered), (gst_pad_link_filtered),
(gst_pad_link_prepare):
check that pads that get linked belong to the same manager. The old
code allowed linking elements before putting them into bins, so it
worked to link them and then put them in different threads, which
lead to weird behaviour.
Since this effectively disallows linking elements before putting
them in a bin, some applications might not work after this and error
out. If these applications are too critical, we might need to revert
that patch. Please test this before the next release...
