TODO List for NLPy as of August 30th, 2004
==========================================

- Generate Fortran timings on the fly with the
  selected compiler

- Allow for re-entry into GLTR when candidate step
  is rejected (GLTR type object?)

- Allow for user-specified pivot sequence in MA27

- AmplModel class should really be a subclass of NLPModel

- Interface to L-BFGS-B + separate functions to create
  limited-memory matrices

- Pycfs could be speeded up if ICFS did not use the
  compressed sparse column format, but compressed sparse
  row (then, Pysparse's SSS format would fit perfectly).
  Instead, we could introduce a "SSSC" format in Pysparse,
  where the strict lower triangle is stored in CSC instead
  of in CSR.

11 Aug 05 (MPF): Pyma27_fetch_perm (_pyma27.c) build the permutation
vector by appending element by element.  Probably should allocate and
populate an integer Numeric array and return that to Python space.
