Excessive time step choping for Norne when using flexible solvers.

[blattms]

Runing flow --linear-solver-configuration=ilu0 NORNE_ATW2013 results in excessive time step chopping for current master. Same for CPR. Using the default, i.e. without the flexible solver bx skipping the argument, works as before.

[akva2]

I can reproduce.

However, results are ballpark similar (and no apparent excessive time step cutting) if --matrix-add-well-contributions=true is given with the flexible solver (ilu0). Maybe this is an implicit requirement with the flexible solver?

[blattms]

the connection with the wells might hint at the changes in #2694.

[GitPaean]

I guess --matrix-add-well-contributions=true is required for cpr.

[akva2]

as stated, this was with ilu0, not cpr.

[blattms]

@GitPaean at least for ILU0 I thought this was resolved with #2694. For CPR you might be right, but currently no exception is thrown when not passing this option and I am pretty sure we did throw one some time ago.

[blattms]

Seem like when running with 1 process there is an if-else missing in ISTLSolverEbos.hpp#L771-L775.

[blattms]

Will try to fix

[akva2]

beat me by 1 min ;)

[blattms]

Well, I still have to compile and run norne before posting a pr. hence if yours is tested....

[atgeirr]

This is indeed a bug as described above, #2706 has also fixed this I think, but has become stale.

[blattms]

Maybe we should not hide bug fixes in feature PRs?

[atgeirr]

Maybe we should not hide bug fixes in feature PRs?

I was actually not aware it would have fixed a bug in the existing code, I just made sure it worked in the new...