- disallow empty prefixes for infinitives in the enumeration. probable reason is
  a bug in dmask2ps.
- put desc_sizes.h into runtime-read file.
- give better names to functions.

Really important stuff to be done:
* Extend the list of known verbs, nouns, proper nouns, relation words, etc.
  In this one, the general public could really help.
* We no longer add prefixes without any judgment, and now reject non-words like
   (to guarded) or  (the + ladies, learn!). However, we should be
  even smarter and reject wrong prefixes before milot/extrawords as in +
  and refine our judgement in order to reject + but accpept +

Quite important things:
* Support the infinitive with different prefixes (not just Lamed) and without
  prefixes ("absolute infinitive"). For example, not only  (to pass) is
  correct, but so is  (when he passes) and  (their passing). 
  The "fat" version of release 0.5 supports most, but not all, of these.
* Add library API to the C-based spell-checker
* Have a better and more complete ispell-like interface to hspell. Version 0.3
  has a rudimentary one.
* Setup to work with KWrite, LyX and friends. Version 0.3 works under LyX, but
  in a less-than-favorable fashion. Version 0.6 works well also withc emacs,
  KDE, and geresh.

Good to have:
* Compression technique that preserves the stem information of words (for the -l
  option).
* Rewrite the whole thing using Uzzi Ornan's transcript internally. This would
  probably induce a simpler, more elegant code, and enable us to add Niqqud
  (Hebrew accentation) to the game sometime in the future.
* Output morphological information about the conjugation. This can help
  understanding the suggested corrections, and have many other implications.
  (see Far-fetched section). This is mostly done by release 0.5's -d option
* Add statistics to each conjugation.  (for your repatriation) is less 
  probable word than  (to sit), and it might help the user to know that.

Far fetched:
* Add support for Niqqud.
* Make use of the morphological information about the conjugations. Outputing
  morphological information can help the user understand a corrected word, and
  may facilitate word analysis, automatic translations, understanding,
  intelligent search, and more.
* Use the Niqqud support for text-to-speech application.
