Most of these will probably be postponed...

(1) Fix the x(1)=y bug by adding a virtual instruction for assignment.

(2) Add a syntax for trace operations; actually it would be better

to go for an indexing syntax that allows you to specify that one

index equals another. The following would even allow more than

two indices to be set equal:

x(=,i,=,j,...)

This even works with the =:start:stop:step syntax, which is very

nice, and it should work on assignment as well, since all you are

doing is adding the strides of the participating dimensions.

Three minor problems:

(a) Where does the new index go in the index list?

Suggestion: at the position of the first marked index

(b) What happens if only one index is marked?

Suggestion: noop

(c) Must the index lengths agree?

(3) L'Ecuyer random numbers, compiled Bessel, gamma, and elliptic funcs

(4) Investigate sort algorithm -- can performance for big numerical

sorts be improved by actually moving the data during the sort?

(5) Work on new binary file model.

(a) Directories? (dictionaries?)

(b) Interpreted close hooks? Other hooks?

-- should there be a destroy hook for any interpreted variable?

(c) Support for opening family as array of filenames.

(d) Attributes? With dictionaries, maybe these can be interpreted.

(e) Better low level support for interpreted open routines.