Make _pending_overwrite atomic (butler + process thread).
This also addresses a potential seek before override race.
Seeking will fill the buffers and by the time overwrite_existing_buffers()
is called from there is no space to overwrite anymore.
When speed is -1, start_sample >= end_sample and _transport_sample
needs to be decremented.
Session::process_with_events() did this correctly, this change makes
Session::process_without_events() behave identically.
Keep track of safe reservation:
Data has been read (or was skipped) previously can be read again
up to the allocated "reservation" (which is never overwritten).
This is mainly a NO-OP, introducing a new PlaybackBuffer type
and preparing for its use.
At this point in time, the buffer is just a power-of-two sized
ringbuffer and the disk-reader's read-logic is still unchanged.
Eventually the read and write sample position that are currently
private to the disk-reader can be migrated to be owned by the buffer.
Also Diskreader::read() positions can be matched to read-position ..
+/- buffer reservation and de-click can read w/o committing the read.
Trust that ::reset() works for all transport masters, and call it when engine is stopped. This way
the transport masters are ready to be called again as soon as the engine restarts.
(bool) false == 0 == (const char*) NULL
error: ISO C++ says that these are ambiguous, even though the worst
conversion for the first is better than the worst conversion for the second:
actions.h:92: note: candidate 1: Glib::RefPtr<Gtk::Action> ActionManager::get_action(const char*, const char*, bool)
actions.h:91: note: candidate 2: Glib::RefPtr<Gtk::Action> ActionManager::get_action(const std::string&, bool)
Theoretically this could be lock-free by using a queue of device
ports to be added/remove in sync in the process-callback, but
realistically adding/removing devices doesn't have to be rt-safe.
