Satish Balay <[email protected]> writes:
Thanks for the patch. I can add it in.
Wrt TSAdaptCheckStage() - there is one more warning - its associated with lines Jed changed - so you didn't see this.
Well, it's caused by his change and it does appear to change the semantics. When I wrote the code, scale_solve_failed was associated with a solver failing. With the change, it is associated with any reason that the stage is rejected. Does that make sense? If so, the name needs to be changed to reflect what it actually does. If not, we need a possibly-different scaling. Also, when we exceed ts->max_snes_failures, the user will now see one line saying the number of failures is greater than the max allowed followed by a line saying the step is being retried, and then it won't be retried. It's clearly a bug with Pierre's patch.