Re: Regexps from shell wildcards

Warner Losh <imp@Boulder.ParcPlace.COM> writes:
      if you wanted to do /bin/csh shell expressions, then you'll find that
      things like "*.{c,C,H,h,cf}" cause problems and cause the output string
      length to grow wildly.


Worse than that, the csh {foo,bar} construct is not a file glob and
in general has semantics that cannot be duplicated with REs:
  - Order is preserved, so *.{h,c} is NOT the same as *.[hc]
  - Is expanded regardles of matches, so "echo {foo,bar}.c" will work
      whether or not foo.c or bar.c exist.


Of course, in any one application these may not be a problem, and
more-or-less mechanical conversion to (foo|bar) might be acceptable.


Just as a hint, here is some perl code I use to convert sh-type globs
to REs in a Perl package. The input glob pattern is known to contain
no '/' characters (the handling of which is "interesting" recursion).


I make no promises about this, but it hasn't failed me yet.


    # Convert shell-style glob pattern to regex
    $pat =~ s/[.=<>+_\\-]/\\$&/g;
    $pat =~ s/\?/./g;
    $pat =~ s/\*/.*/g;
    # Hide leading . from wildcards
    $pat =~ s/^\.\*/[^.].*/; # .* -> [^.].*
    $pat =~ s/^\.([^\*])/[^.]$1/; # .x -> [^.]x
    $pat =~ s/^\*/[^.]*/;
    # Anchor the pattern
    $pat = "^$pat\$";
    # could do some optimising here, but leave it to perl!
    # e.g. "^.*" => ""
    # ".*$" => ""
--
Gregory Bond <gnb@bby.com.au>
Burdett Buckeridge & Young Ltd Melbourne Australia
--

Return to the comp.compilers page.
Search the comp.compilers archives again.