Why did it disappear? Epetra is still quite useful, and Tpetra is still a work in progress. Why couldn't the Epetra code stay while y'all are doing the Tpetra interface? Now that it's been 5 months, it might be time to bring back the Epetra...
Epetra was removed because the burden of supporting it in the light of other changes in DOLFIN was too great. Contributions towards a Tpetra interface would be welcome.
Groovy. Thanks for the info!
I'm having trouble with FEniCS+MPI (that I'm somewhat disinclined to debug), so I don't know if the thing that I did is robust to parallel code, but I managed to get a Tpetra::CrsMatrix that I created by faking my way to creating reasonable Tpetra::Map objects for the row, column, domain, and range maps and then pulling apart a uBLAS matrix by its CRS pointers and constructing and expert-static-fill-completing a Tpetra::CrsMatrix that way. It looks right? I hope? I'm on somewhat of a tight timetable, and I only care about serial code for this particular application, so that seemed like the path of least resistance. Perhaps once that is all wrapped up, I will look at a lower-level implementation that skips over the uBLAS nonsense entirely. Based on my 3-work-days old knowledge of FEniCS, everything that Tpetra could need for maps, etc, is in the dolfin classes, but, like I said, shortened timetable, so I didn't want to do that thorough of an implementation. I'll be sure to follow up if I come up with something I think could be useful to you.
