time-control function object does not respect underlying function object
When a function-object uses timeStart, timeEnd to control its activation, these values are used to define if execution or writing occurs, but can also mean that the underlying end() function is never called.
This can be problematic if the end() method is being used to free up resources.
Activity
mentioned in commit Community/catalyst@24f17124
My concern was that it may be desirable to only run for a certain period and then end() the function object to free up cached meshes, disconnect from catalyst etc. One idea could be that the time control part of the function object remembers its state (not started, active, done) and triggers an end when it transitions from active to done. When it is done, I guess it could also delete itself from the list (don't know if this is useful or necessary). But this also raises the issue of what to do on startup if the simulation is already beyond the timeEnd. Maybe the function object needs to remove itself from the list before even creating the rest of the backend.
-
One more item that we may wish to consider would be the ability to tag certain function objects as being runtime-only (ie, suppressed when using the -post-processing option). The Catalyst and ADIOS function objects come to mind as things that will likely only work properly during a simulation but not as postpro.
-
Another point just raised by @Prashant - for a scalarTransport with a particular timeStart, it is initialized during construction. This means that it is being carried about during mesh motion etc, even before timeStart actually wants it.
