The criterias are now a mapping between field names and one or more values.
This aligns tracklist.{filter,remove} with the API of
library.{find_exact,search}, and allows for e.g. batch removals.
An exception is raised immediately if the API is used in the old way to ease
migration and debugging.
- Move imports to top as they should be.
- Report state based on publish return value.
- Remove overly broad except clauses.
- Unpublish before stopping servers.
- Prepare for handling missing dbus directly in module.
- Constants should always be all caps.
- Extracted helpers from class to convey intent via their naming.
- Moved imports out of class, imports should always be on the top.
- Made sure calling publish mutiple times does not re-convert text.
- Made sure calling unpublish without publish does not break.
This change just patches over the worst of the inconsistencies in how the
scanner mixes local and file uris to get us to a working state again. Ideally,
this still needs a real cleanup when we finish the plugable library providers
and/or json library work.
- Unlike the old python version we do not wait for the first audio handoff, we
only progress until the PAUSED state. This ensure we don't block on empty files.
- Instead of using the signal watch and running the main loop we simply poll
the messages from the bus directly allowing for a synchronous code flow.
- Between each file the pipeline is always returned to NULL, this is done as we
found that gst 0.10 will slow down as the uribin does not cleanup the
children it creates for handling each file. This issue is not present in 1.0.
- This also works around a segfault that was likely caused by a race condition
that seems to trigger in the 0.10 version of the pbutils discoverer.
- This version of the scanner also fixes the per track slow down, and works out
to be considerably faster than even the built in discoverer from 1.0.
- Finally this removes the WMA hack as I kan no longer find any evidence of it
being needed.
