Re: Bug in getopt-long?

[Thien-Thi Nguyen]

   From: Daniel Skarda <address@hidden>
   Date: 02 Aug 2001 11:32:34 +0000

    I think that getopt-long arguments should be independent and one
    should not imply another. If I had needed to make '--test' argument
    mandatory, I would have wrote '(required #t).

see latest cvs (in either stable or unstable branch) for a mechanism
that allows you to express what you expect to see and also test guile's
actual behavior: test-suite/tests/getopt-long.test.  because the spec
definition is somewhat confusing (to me at least), i would prefer to
center discussion around the concrete tests in this file rather than
easy-to-misunderstand (by the sleep-deprived ;-) descriptions.  the
tests there are just a start -- they should be expanded greatly to cover
all the nooks and crannies of (ice-9 getopt-long).

    There are a lot of optional arguments, which you would like to check
    for correctness (using `file-exist?' - for example) and it does not
    make sense to check optional arguments in different way than
    mandatory arguments.

    Maybe we also can improve getopt-long error reporting by adding one
    option - predicate-failed - something like this...

      (getopt-long args
         `((input-file 
               (value #t) 
               (predicate ,file-exist?) 
               (preciade-failed ,(lambda (x) (%% "Can not find file ~S" x))))))

    ... because 'option predicate failed' is not too informative :)

    (I use @@ as a shortcut for (apply format #t args) and %% for (apply
    format #f args))

this allows the right amount of flexibility for the user.  if you're
feeling motivated, consider cleaning up getopt-long.scm, beefing up
getopt-long.test, and submitting a patch.  i think such a patch would be
well-received.  personal note: i'm incommunicado for almost a week and
won't be responsive to guile-devel discussion for the duration; please
don't construe my silence as lack of interest...

thi