=================
TODO: optcomplete
=================


- IMPORTANT: if the option is a store option and takes e.g. an int or a float,
  override the default completion to not list the files.

- bug: we need a better way to pass the COMP_WORDS array to the program, lest
  the space delimited arguments will be split (bleh). perhaps use a random
  string to do this.

- provide more completers.


- Add this contribution from Mark Eichin <eichin@gmail.com>:

    Was experimenting with optcomplete (1.2 from debian) and noticed that
    the default completer *only* worked on local files, unlike the way
    bash completion works without a hook.  This wasn't to hard to change
    with an explicit FileCompleter::


     # we don't need state, so make it a function...
     def FileCompleter(pwd, line, point, prefix, suffix ):
         """Completes by listing all possible files, here or matching 
         substrings"""
         startdir = os.path.dirname(prefix)
         lhs = os.path.basename(prefix)
         if not startdir:
             startdir = pwd
         return [os.path.join(startdir, f)
                 for f in os.listdir(startdir) if f.startswith(lhs)]




.. end

- how will it work with my multiple subcommand framework?

- see if we can support csh.

- we could add an option that would make the short options automatically expand
  to long options, this depends how the shell treats our answers.

- optcomplete: write and publish the insight document about optcomplete, with 3
  volets:

  1. documentation, in file
  2. options parsing and help
  3. graying/filter vs. programmable completion

  - Unix has had the great tradition of the command-line interface (note: just
    another interface);

  - this has nurtured an environment where genericity has been given
    importance, and incidentally, modularity. these conditions make for
    software that is more robust: there is more reuse since the functionality
    is more generic, and modularity forces the programmer to have to deal with
    interfaces, which in turn indices the need for better design and thinking
    of the code.





  
- Reported by Graham Davies::

    the module gives exactly the same symptoms as the regular bash
    completion code (not that surprising).

    These do NOT work as expected:

    :work[tab]
    p:work
    =work
    p=work  etc.

    However, these all work fine with or without the prefix 'p':

    @work
    p-work
    p+work
    p%work
    p_work
    p^work
    p[work
    p]work
    p{work
    p}work
    p?work

    Can you envision a fix for this - or are the : and = symbols fundamental
    to bash completion?