Bug 54270

I have fixed the TplLogWalker API so that the filter is always called from the same thread that invoked the walker.

Added some test cases that use a filter for get_events and rewind too.

See the attached branch.

I have simplified your branch a bit. I removed the callback indirection, by setting the real callback/user_data in the GSimpleAsyncResult. I also fixed events leak in an unlikely case that get_events_finish() is not called and used complete_in_idle() instead of g_idle_add().

http://cgit.collabora.com/git/user/nicolas/telepathy-logger.git/log/?h=event-filter

The remaining issue is the lock in get_events_async(). Calling this method a second time before the previous call has finish has undefined result according to mutex documentation.

The lock was used to serialize multiple calls get_events and rewind at a time when both the async operations were implemented in a thread. Now only rewind is done in a thread, but not get_events.

So you are correct that two successive calls to get_events will now cause problems because it is not implemented in a thread anymore.

I was looking at using a queue to do the serialization now. So if you make the following calls without returning to the main loop:
 - get_events
 - get_events
 - rewind
 - get_events
... they will be kept in the queue and executed in FIFO order one by one.

However I am wondering how we could hook into the completion of an operation (eg., rewind which is done using g_simple_async_result_run_in_thread). When an operation completes we want to check if there are pending operations in the queue and if so, execute them.

As discussed in #telepathy, I restored the wrapper callbacks for get_events and rewind (but not fill_cache) to hook into their completion stage.

The latest version of the code is in my branch:
http://cgit.freedesktop.org/~debarshir/telepathy-logger/log/?h=event-filter

Only problem I see atm is that GQueue is not threadsafe. I know there exist GAsyncQueue, which is thread safe queue specialised to thread communication. If that does not work, then I guess protecting the GQueue with a mutex would be sufficient.

Accesses to the queue are not made from run_in_thread, but only from the main thread, so it should be ok.

From #telepathy:

<rishi> stormer: Any other comments on the walker refactoring?
<stormer> rishi: nop, you got my ++