* generalize process_file() and put it in input.c (ie. part of the library)

* error handling and reporting
  x structure for error location (filename, line, offset, item_name, item_num)
  - suppress printing and store errors for application to query later
  - document this mechanism

* stack overflows:
  - really big entries can blow up either the attrib or AST stack
    (if same size, attrib stack will blow up much sooner)
  - with stacksize==400, 97 fields are enough to blow up the attrib stack
  - can't do much about this without seriously attacking the generated
    code (yuck), and I think I had enough of that in fixing the static
    lexical buffer; look into ANTLR 2.x ...?

* test suite:
  - need a "catch-all" test program for putting in tests for known
    (and suspected) bugs

* external interface to post-process strings

* "smart" (or maybe "smart minimal") post-processing option: in regular
  entries, collapse whitespace if only one simple value (ie. where
  collapsing can't hurt)

* convert AST to a structure, with number of fields, list of field names, 
  list of field values, and so on (probably also need a structure
  for value and simple value -- entry will include list of values,
  each of which is a