When getting backtraces, we were always creating a new Dwfl context and then
discarding it. The problem with that is that it resulted in having to re-scan a
lot of information for every single backtrace.
In order to fix that issue, use a global on-demand Dwfl context and use it with
a lock.
Furthermore, we were scanning the mappings of the
process (dwfl_linux_proc_report) for *every single step* in the backtrace, and
that function is horrendously expensive (does sscanf on /proc/PID/maps
...). While there is a possibility that new mappings might be available (new
plugins being loaded for example), we can limit ourselves to just do it once per
backtrace.
These two modifications speed up the elements_leaks unit test (which traces all
pads with full backtraces) by a factor of 6.
Partially fixes#567
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/504>
... and update meson file so that enable it only using required headers.
"dependency(...)" is unlikely successful for Windows SDK libraries
since it doesn't ship pkg-config file. So it needs to be changed
to "find_library()" to link corresponding .lib file. That would
result to most MSVC build system will link dbghelp.dll. However,
one drawback of the change is that gstreamer-1.0.dll will mandate
dbghelp.dll although it should be optional. So g_module_open() way
can be the most safe way in this case.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/626>
If you're using a custom log handler, you had to reverse-engineer the
debug log format and create your own format function. Now, you can
call `gst_debug_log_get_line()` and it will return a string (without
ANSI escape color codes) representation instead.
This is useful in situations when you need to log the ordinary
gst_debug log to a resource that can't be opened as a `FILE` handle.
Also includes a test.
On Windows, concurrent colored gstreamr debug output and usual
stdout/stderr string will cause broken output on terminal.
Since it's OS specific behavior, that's hard to completely avoid it
but we can protect it at least among our printing interfaces side.
gst_ring_buffer_logger_log calls several functions while formatting
the message which may in turn log a message while we already hold
the mutex. Do all formatting first before acquiring the mutex to
avoid this and reduce the time we hold the mutex.
And change it to do nothing at all.
As debug categories don't use reference counting and they can be
retrieved from anywhere at any time by name, it is fundamentally unsafe
to free them at any point in time except for right before the end of the
process.
No code apart from a unit test seems to be currently using the function,
so deprecate it and also change it to do nothing at all.
Ensure that the code paths for HAVE_UNWIND and HAVE_DBGHELP are never
taken at the same time, even if the build file code changes.
Prefer DbgHelp over libunwind on Windows in case both are somehow
available because DbgHelp is only available when building with the
MSVC toolchain, and libunwind won't give us debug symbols from objects
built with the MSVC toolchain.
Also, print slightly more useful messages for the level of stack trace
support enabled, and document what each if conditional does.
The code implicitly uses this value when the stack trace is not FULL.
Mostly useful for documenting the behaviour when each flag is passed
and for translating to/from strings.
* Making sure that `static inline` function are in the GIR (by first
defining them, and make sure to mark as skiped)
* Do not try to link to unexisting symbols
* Also generate GIR information about gst_tracers
And add strduped function pointer names to the global quark
table, so that they don't get reported as lost by valgrind.
This allows us to use GST_DEBUG when running tests under
valgrind.
When outputting debug logs on Windows, some sections are protected
with a non-recursive lock. Turns out though that gst_debug_message_get()
might indirectly, via our printf format extensions, call code which
in turn would try to log something when it can't handle something. If
that happens we end up in gst_debug_log_default() again recursively and
try to again take the lock that's already taken, thus deadlocking.
Format the debug message string outside of the critical section
instead to avoid this.
https://bugzilla.gnome.org/show_bug.cgi?id=784382
Remove unneeded reapplication of patterns. Besides being
superfluous (gst_debug_reset_threshold already applies
patterns) it was also wrong and didn't stop checking patterns
after the first match (broken in 67e9d139).
Also fix up unit test which checked for the wrong order.
https://bugzilla.gnome.org/show_bug.cgi?id=794717
Replace the while+goto with a for+break and check walk to determine
whether we had a match. Move up the unlock to keep the locked section as
small as possible.
https://bugzilla.gnome.org/show_bug.cgi?id=794717
Move the match, logging and set_threshold to a new function.
The log levels are different, so choose the higher one (LOG). Having two
equivalent messages at two different levels seems like a bad idea
anyway.
https://bugzilla.gnome.org/show_bug.cgi?id=794717
When registering a new debug category after gst_init(), simply check
the existing patterns against that new category.
No need to iterate over all categories and recheck them all against
the existing patterns.
Also, no need to re-parse the existing pattern string set via GST_DEBUG
and add the same set of match patterns all over again to the existing
list of match patterns every time we register a new debug category.
Combined with iterating all debug categories on a change this would
make adding debug categories after gst_init() very very very slow.
This stores debug logs in memory per thread and uses up to a
configurable amount of bytes per thread for the logs. Inactive threads
are timed out after a configurable amount of time.
https://bugzilla.gnome.org/show_bug.cgi?id=785035
When registering a new debug category after _debug_init(), we need to
re check the GST_DEBUG filter settings again.
In addition when parsing the filter setting, we need to already bump up
the min-debug level to not suppress debug log statments that dynamically
register a category. This happens in libraries that use a function to
register a category on first use.
This is an API break but that API has not been released yet.
We are passing a flag rather than a simple boolean as we can imagine
to implement more features in the future for example to retrieve a
stack trace for all the threads, etc..
Retrieving source file and line numbers is pretty
expensive while getting a stack trace, this new argument
allows the user to decide to retrieve a backtrace
without those infos instead which is much faster.
For example running $ GST_LEAKS_TRACER_STACK_TRACE=1 GST_DEBUG=GST_TRACER:7 \
GST_TRACERS=leaks time gst-launch-1.0 videotestsrc num-buffers=1 ! fakesink:
* With simple stack traces:
0.04s user 0.02s system 99% cpu 0.060 total
* With full stack traces:
0.66s user 0.23s system 96% cpu 0.926 total
https://bugzilla.gnome.org/show_bug.cgi?id=775423
We were creating a new session to retrive each line of a stack trace
and we are supposed to start it once for a whole stack trace.
And pass the whole file to gst-indent.
https://bugzilla.gnome.org/show_bug.cgi?id=775365
It's useful to be able to set a name pattern for GST_DEBUG_FILE so that
the same environment variable can be used for multiple processes and
still write to different files. Especially useful if these processes
run simultaneously.
%p: Replaced with PID
%r: Replaced with random number
%p is obviously useful. %r is useful when for instance running two
processes with same PID but in different containers.
https://bugzilla.gnome.org/show_bug.cgi?id=773092
