wishlist

Editor:
	load/save a subtree
		as you can clip a subtree to a workspace, so should you be
		able to save.
	labels with leading zeros. 
		pascal says leading zeros in a label don't matter.  We may
		switch to full text labels, in which case we won't 'fix' this,
		or will we?
	window for immediate mode - need this at least
		and in general this is part of the multiple editing window
		bit.  Or we should implement the "statement on the command
		line bit"
	Editing of with, fix errors that are signaled (*)
		The whole business with withs and fields is poor and needs
		major rework.  this includes the ability to enter a field
		name without the record define etc. 
	windows (*)
		This means multiple editing windows.
	cursor move after null type can still go quite a distance (*)
		The whole question of TAB, SPACE and ESC has never been
		generally dealt with.
	fancier macros (2 or more char) (*)
		If we have a meta key we don't need this.
	lines longer than 255 chars - treeprint buffer is fixed size (*)
		This is the total refix of oledit, treeprint's output and
		the output routines.
	multi-d arrays - special representation *
		I doubt we need this, but it lets the user display the
		array as array[1..2,1..2] if entered that way.  Needs
		special bits
	search user interface (*)
		Which means putting search on a key, something like the
		slash style of vi for fast searching
		also make token search and regular search the same thing
	special local variable for function assign instead of actual symbol
		ie. stop recognizing the function name on the left of the
		assignment outside the function body.
	reading in templates from a file (*)
		Jan is keen on this, but our current output scheme isn't
		flexible enough to make it really useful.
Interp:
	Subrange and other checks on value parameters
		This can only be done by scanning over the parameters at
		input time.  This would also be a good time to do the
		conversion of string parameters which we currently do
		at push time, and general support for any other funny
		passing types.
	some non checked errors on goto
		We don't catch them all, I am not sure which.
	goto INTO an if
		we don't let this happen.  The block the goto is in must
		be active.
		actually, further checking shows this may be illegal anyway
	check for duplication of label - this is hard...
		since labels uses are actually references.  Need a bit
		or something.
	proper creation of frames in immediate execution (how do I duplicate
	  this bug)
	function & procedure params - checks, 
		We don't check the parameters on procedures that are passed
		as parameters, WHEN they are passed.  We allow the passing
		if the number of params and the return type, if any, match.
	printpoints - variable assign, routine call (*)
		This will be done by the general "debug block" feature I
		have designed.
	Assigning to for loop variable caught (*)
		This needs a bit set in the variable for the duration of
		the loop.  Time consuming - is it that important?
	variant records checking(*)
		Yes, we should check variant assignments.
	writeln of records, small arrays * - for debug mostly
		Would be handy
	new & dispose with variants. - check them at least (*)
		We currently don't check the extra arguments passed to
		new that say what variants are to be set.  I am not
		sure I care.
	new & dispose with table of pointers & double indirection
		We should switch new and dispose over to this scheme.
		it's doubly safe and then we can have a real dispose, only
		wasting one pointer for every item disposed.
	traceback statements (*)
		ie. traceback of the last N statements done, or the last N
		function calls. Kind of like reverse execution - "how did
		I get here?"
	start up again after error? (*)
		not important
	start up again from different point (manual goto) (*)
		Could be done but on the back burner
	duplicate cases detection (*)
		this is quadratic but should be done at entry time.  The
		point is that cases are probably small so it won't be bad.


Parser
	doesn't do withs properly yet.
	
General
	"C functions"
		support call to a general C function kept in a big array,
		but no memory wasted, let the user load in an "external"
		declaration for the routine.

		Include here window routines, message passing, operating
		system interface and have loadable libraries with these.
		this can be done as a whole new node, or as a variant on
		syscall/sysfunc.  On the ibm, perhaps even an absolute
		address?  investigate dynamic loading of a .com file although
		that can't be large model...
	Generic types
		Let the user write a routine (or declare a c routine) that
		can take a generic ordinal.  We allow it for our builtins
	Absolute variables
		variables located somewhere in memory, or in i/o space
		This might include a pre-defined "memory" array and "io"
		array.  On the large model, the memory array would have to
		be two dimensional.  64K by 16 bytes or 16 rows of 64K?
	writestr procedure (and readstr)
		line writeln but writes into a string, or perhaps returns
		one, with the problem that you get restricted in size and
		in the number that can be in one expression.
		best stick with the procedure unless we do something fancier
		with strings, like
	String operators
		use of string expressions and string building with '+'
		as in basic.
	byte operators, upper, lower, shift left, shift right

	pointer type
		that you can build, as as substitute for the absolue variable
		and perhaps an in/out procedure, although they exist in c
	A string val function like val() in basic (or readstr)

	TRACE block
		to trace assignment to a given memory loc, or breakpoints
		within the block, causing the code in the block to
		be executed when it is trapped.  This is a very handy
		and general debug feature.	Also perhaps trap
		certain runtime errors, particularly input errors

	use of typenames in error messages
		many of the typecheck errors could be more specific
		about the types involved, using a variant of the new
		treeprint.

	dynarray of <<type>> - user supplied upper and lower bound when
		created, dynarray of char as a string, compatible with other
		arrays - automatic release on function termination.