[SciPy-Dev] Re: scipy.linalg.lu considerations

Without looking into the details, how about using some form of what scipy stats does (or was doing?) with make_tuple_bunch --- basically a class which behaves like a tuple of a fixed number of arguments, and then P or a permutation matrix becomes an extra attribute?

Yes, that's probably how we are going to do it, if, we decide to do it.

That said, why do we carry custom LU decomposition code at all, is this not a single LAPACK call (maybe plus a thin wrapper to preserve backcompat).

That code is going away regardless. Det is done, with LU we can remove _flinalg.py stuff.