# TODO
- CollectionsM: complete the subset of MapM and uncomment/complete CollectionM.
- See if needGC hack can be improved. It seems its eating CPU time.

- First RBR test: 1/3 running time when using MapL (via Map2). Not sure
  if the results are correct (they seem to be). Check with Ketil about usage
  of Map2, if its enough.

- Names names names
  - HashIO is too similar to Hash. Stringable is strange and not totally true,
    maybe CStringable?

- Other missing malloc error checking
- Tests that stress MiniGC (w/ two maps at same time would be nice too)

- Check how GHC do interning
- Check how's sparse array implementation of Parrot.

# "CRAZY" BUGS (X means "solved")
X Dummy/Refeable Bug: need to ask in #haskell and reply Simon's email.
  It dissapeared recently. Dunno the reason.
X Bug on HSIter stuff when use a lot of memory.
X Finalizer may be not working right => test with sequencing a lot of operations
  causes error in GHC runtime. (seems to be fixed by having ONLY ONE gc
  for everyone)


# NEXT
- Data.HashTable
- http://shootout.alioth.debian.org/gp4/benchmark.php?test=knucleotide&lang=all
  Did a little naive test, but need to compare with other implementations. It's
  worst than hawiki Ketil entry, but better than one of three Haskell entries
  on the page (are they updated?).
- Implementations specific to unboxed values like Int and Double: Int seems to
  be ok now, but Double and other unboxable still need to check

- Discuss APIs
-- Unbounded arrays: see Bulat work on this.

- Refeable vs. HashIO => unify?
- Better testing / More test programs
- Document this work.
- Create a LICENSE file (I think BSD3) and put license info in the
  necessary files. Mention that John Meacham as original author (how?).
- Find something other than WordPtr (ghc 6.5) that works in GHC 6.4.x?
  need to have conversion from/to Ptr and from/to Enum to work right
  (not really sure about).


# THINGS TO CONSIDER

nice mail from Bulat:
http://www.haskell.org/pipermail/haskell-cafe/2006-May/015915.html

## API SUGGESTIONS

Bulat: "Judy implements _mutable_ array/map/set, so closest interfaces for your
lib will be MArray and HashTable. DiffArray is a general way to make IArray
(interface) on the base of MArray (interface), that is usable with any
datastructure implementing MArray interface"


## STABLEPTR POSSIBLE OVERHEAD

StablePtr is used to store references of Haskell objects in external
datastructures. Maybe GHC isn't very efficient dealing with a lot of
StablePtrs, if that's true three obvious ways are:
 - GHC 6.5/6 is better at this.
 - Turn GHC better at this (wow!).
 - Find another way to do it (without StablePtr).


## REEVALUATE BENEFITS?

> all Haskell software that uses common data structures like Maps and
> Arrays would benefit from a faster implementation

Bulat says "sorry, i don't think so. especially for Arrays ;)"

Judy is useful for applications for (very-)sparse Arrays, which is Pugs' case.