#1549 Coords storage in Interface is too brittle
It was written at a time a \a Coords got a unique index that was unchanged, and the same data is actually stored twice (the idea was to prevent to allocate a vector at each call that specifically required indexes, but as those are very rare and don't occur during the main computation it is not an issue).
The problem I met (related to the changes in #1545 (closed)) is that after the \a Coords are renumbered the default ordering chosen might not be respected, and a test checking the consistency between original and reconstructed data from prepartitioned data might therefore not follow the same convention (with in fact weirdly enough the latter being the most consistent choice...)