This adds API to allow clients to schedule a callback in the mainloop
thread without the mainloop lock being held. This is meant for a case
where the client might be dealing with locking its own objects in
addition to the mainloop thread itself. In this case, it might need ton
control the locking order of the two, to match the order in other
threads, as it might not always be able to allow for its objects to be
locked after the mainloop thread lock.
We move over helper functions to get rate, channels, channel map and
sample format (if PCM) in the public API, so users of the extended API
are more easily able to pull out these values from pa_format_info.
The internal operation_set_state function already returns early if the
new state is the same as the existing state. The attached patch extends
this to return early if already in a finalised (done/cancelled) state,
i.e. blocks attempts to re-finalise into a different state.
This helps avoid unlinking more than once (or crashing on ref count
assertion).
I was not certain whether an assertion would be a better alternative -
with such a crash helping highlight usage problems...
The situation that lead to this was the thought of someone stupidly
trying to pa_operation_cancel() a callback within the callback
execution itself, while designing a solution for a memory leak related
to cancellation within my Rust binding. While no-one should do such a
thing, if they did, they'd either trip up a ref count assertion, or the
operation would be unlinked twice, which would be bad. It's a simple
thing to catch and mitigate, and could prove to be a useful
bulletproofing measure for this function in general.
This allows constifying public API functions that report their errors
via the context error but don't modify the context in any other way.
Philosophical arguments could be made why this is wrong, but I believe
in practice this is a net positive change.
Paves the way towards more of the API using const pointers.
Some pa_context_* functions return their errors by setting the context
error, even when there's no other change in the context state. This
prevented constifying the pa_context arguments of such functions. This
patch puts the error in its own struct behind a pointer, so that setting
the error doesn't any more count as modifying the pa_context object.
If the given proplist is NULL, the function creates a new (empty)
proplist. That caused a compiler warning after the constification, which
is why the new proplist is now created using a separate variable.
Existing documentation was unclear about which property list would be the
one changed (merged into), making it seem (along with the non-const
proplist pointer param, which needs changing seperately), that the proplist
object for which a pointer is given will be the one merged into, instead of
the internal cached entry's proplist.
This should make it easier for clients to elevate their audio threads to
real time priority without having to dig through much through specific
system internals.
