Add support for some modules to use a module.$NAME.args config section
to override the arguments of the module. Update the docs.
Add more conditions to the module loading, use !false to also load the
module when the condition is not set and so that we don't have to
explicitly set each condition to true.
Improve module-profiler example config.
Add a new followerClock block in the profiler info. This is only set
when the follower could be a driver and it contains the clock info used
for following the driver, mostly the rate difference and delay.
Dump this info in pw-profiler -J
Make sure we always set the info in the clock, especially also when we
are following.
Allows us to get error counts and occasional samples, which can be handy
if we are CPU-bound. There is room to improve this by having
module-profiler create aggregates, which might be more useful that
periodic instaneous samples. However, this would need more logic to
store node data in between emissions, so we'll punt that to later.
Also handle the relation between a node and the driver with pw_node_peer,
like we do with the links.
Because these are refcounted, we only make one peer for a node that is
linked to another node that is also the driver (pw-play -> sink) and we
save some fds as well as some admin stuff and overhead for the refcounts.
This in return then results in less problems getting all the refcounts
right when adding/removing nodes.
Don't use a timer but use an event to trigger a flush of the collected
data. We are now using a per-driver buffer and we don't want to make
this too large so make sure we flush the data quickly.
Make a ringbuffer per driver because each driver might run in a
different threads and because a shared ringbuffer can not be written to
from multiple threads. Assemble all the driver stats into one buffer
before sending out the profile info.
Remove the context_driver events and replace them with realtime node
events. The problem is that the realtime node events are emitted from
the node data thread, which can be different for each node and
aggregating them into context_driver events is not a good idea.
It's also nice for the stream drained event, which no longer needs to go
through the context_driver events.
For all the modules that include the private header we require that the
library and compiler versions match.
Otherwise we might end up poking into new or old fields that got moved or
changed in the private struct and crash.
See #3243
Remove some includes of private.h
Add some methods to get the mempool of client and context so that we can
remove direct access.
Move some things around.
Use methods to get pw_loop variables.
See #3243
Make a copy of the node name into a statically allocated array. This is
for debugging purposes only but might crash if we do a name change while
the data thread is reading it.
Make it possible to do reposition on the client side by copying the id
to the target. The client side does not have a node in the target so we
can't deref it.
Even though all event sources are kept in a list in the current
implementation of the event loop in the SPA support plugin, relying
on it freeing all sources could lead to unbounded memory use
if the profiler module is loaded and unloaded over and over again.
Handle if the global is destroyed (e.g. `pw-cli destroy X`) to
avoid a use-after-free in `pw_global_destroy()` when it is called
with a dangling pointer from `module_destroy()`.
This also brings the advantage that all tools, examples, modules, components
can also be compiled standalone out-of-tree using libpipewire from the system
Write the profiler data to an allocated buffer instead of the stack
so that we can make it a little larger.
Don't try to process the data when the builder had to truncate it
because it didn't fit.
Just like the real free() we should just ignore a NULL pointer, makes the
caller code easier for those instances where properties are optional.
Patch generated with concinelle with a few manual fixes.
If the profiler manages to fill the complete buffer, we would have 8MB
of data that we need to allocate on the stack. Don't do that and instead
use a preallocated buffer large enough to copy things into.
SPA_MEMBER is misleading, all we're doing here is pointer+offset and a
type-casting the result. Rename to SPA_PTROFF which is more expressive (and
has the same number of characters so we don't need to re-indent).
Make a new complete event and use it instead of the start event
Use the start event at the start of the cycle
Make the profiler also log incomplete graph cycles
Add Profiler object and fields.
Add profiler extension API. It notifies Profiler objects with
real-time performance data.
Add module that implements the profiler extension.
Add pw-profiler tool that binds to the profiler API and dumps the
data into a log file, gnuplot files, a html page and a script to
generate svg graphs. This is almost the same as what JACK2
JackEngineProfiling does.
