13
0

don't do a pending save if slave is syncing.

13:00 < rgareus> When a slave does initial sync, it sets speed=0, pos=XXX (required by session_process.cc state-machine to init)
13:01 < rgareus> This triggers a locate(roll=false) which in turn triggers a realtime_stop() which in turn triggers a non_realtime_stop().
[..]
13:06 < rgareus> las: the problem I have with non_realtime_stop() is that it does save a pending state IFF  get_record_enabled() is true.
13:06 < rgareus> The save can take ages (seconds), which will void the initial sync of the slave.
13:07 < rgareus> The slave enters a live-lock: sync, save, re-sync, save...
13:07 < las> rgareus: understood
13:07 < rgareus> las: I propose to workaround this: only save pending state if there is no slave or the slave is not locked.
13:07 < las> rgareus: another reason why recording + slave == bad idea :(
13:07 < las> rgareus: but yes, that sounds fine to me
13:07 < rgareus> AFICT this is not harmful. It only affects pre-record settings.
13:07 < rgareus> 'did_record' is used to save a full state after each recording.

git-svn-id: svn://localhost/ardour2/branches/3.0@13288 d708f5d6-7413-0410-9779-e7cbd77b26cf
This commit is contained in:
Robin Gareus 2012-10-16 20:53:08 +00:00
parent 4c9fcd407c
commit 625ee55df0

View File

@ -618,10 +618,20 @@ Session::non_realtime_stop (bool abort, int on_entry, bool& finished)
}
if ((ptw & PostTransportLocate) && get_record_enabled()) {
/* This is scheduled by realtime_stop(), which is also done
* when a slave requests /locate/ for an initial sync.
* We can't hold up the slave for long with a save() here,
* without breaking its initial sync cycle.
*
* save state only if there's no slave or if it's not yet locked.
*/
if (!_slave || !_slave->locked()) {
DEBUG_TRACE (DEBUG::Transport, X_("Butler PTW: pending save\n"));
/* capture start has been changed, so save pending state */
save_state ("", true);
saved = true;
}
}
/* always try to get rid of this */