 * Compile environment:
    - Check for latex in order to (optionally) build the documentation.

 * General source modifications:
    - (Better) source comments.
    - The sources could use some name changes and indentation in my
      opinion.

 * Compiler technical changes:
    - Synchronise cdlc command line parameters with documentation (or
      vice versa) and POSIX-ify them a bit while we're at it
    - (Taken from previous README:) bug concerning arrays and globals:
      the compiler does NOT check if a non-array global variable is
      indexed (as if it were an array).
    - Polymorphism checks do sometimes report ambiguity while there is
      none (as far as I can see).
    - Check the tail-recursion optimization, I suspect it not to find
      all tail recursions.
    - Check the import/export checks, or (in case this is correctly 
      implemented) modify the error message "import of blah after
      export".
    - Check the Sparc code generation if it still works. The code
      generation could use some lines of comment on the jumptable used
      for failing rules etc.
    - Only after the checking the Sparc code generation: fix the i386
      code generation as it breaks with recent binutils' gas.

 * New additions:
    - It would be nice to use the CDL3 routines as an add-on library to
      other programs. The current problem is that the RTS and the code
      generation assume that it is the CDL3 program that is in control.
      Things that should be looked into is the main() in the CDL3 RTS,
      the name generation of the (generated) routines and the names of
      the constructors and destructors. Some documentation and an
      example would be very usefull.
    - The documentation could use a bit more explanation on the routines
      in the standard CDL3 library.
    - Describe in the documentation the behaviour of undefined values.
      (Taken from previous README:) Local variables will be initiated
      upon use. Global variables will not be initiated, and will contain
      Undefined as their value. Upon first write these globals will get
      initiated and get a value. An undefined variable can very well be
      read, but doing a split will fail for all (defined) alternatives.

 * This document needs a check - what has been done? - and a timestamp

-- From todo 
TEST head (>TEXT, TEXT2>).
TEST tail (>TEXT, TEXT2>).
guards on formal or actual parameter positions
solve transient parameter bug
