================================================================
Perl-on-NT implementation notes

John Kerl
12/27/2000 
================================================================

----------------------------------------------------------------

The following NT bug appears to exist.  I make the following association:

  assoc .pl=PerlScript
  ftype PerlScript=perl.exe %1 %*

Now, if I type 

  script a b c

or 

  script.pl a b c

then script.pl's output goes to the screen. However, if I type 

  script a b c > x

or 

  script.pl a b c > x

then x is an empty file.  This is *not* OK.  But if I type 

  perl script.pl a b c > x

then x has the desired contents.

Similarly, "dir | script" produces no output, but "dir | perl script.pl"
produces the desired output.  It appears that NT associations simply do not
work with redirected I/O.

The workaround I employ (to avoid having to type "perl" in front of the name of
all my perl scripts) is this:

For each script.pl in my bin directory, I also have a script.cmd wrapper file
which looks like this:

  C:> type c:\bin\script.cmd
  @echo off
  c:\bin\perl c:\bin\script.pl %*

That is, script.cmd simply invokes perl on script.pl, passing the argument list
straight through.  In effect, I'm doing my own association, since NT's
associations are botched.  This way, "script a b c" and "script a b c > x" both
work as desired.

However, this depends on my PATHEXT having .CMD *before* .PL: For this scheme
to work when I type "script", the cmd shell must find script.cmd before it
finds script.pl.

The "toc" utility copies given Perl script(s) to my bin directories and creates
the wrapper .cmd files as well.  For example, I would type "toc script.pl".

----------------------------------------------------------------

Perl file globbing doesn't work on NT till version 5.6.  For version 5.6 and
above, Perl does the globbing itself.  But for versions before 5.6 (e.g. I'm
using 5.005), Perl simply calls the Unix (!) shell (e.g. csh or sh) to perform
globbing.  csh and sh of course don't exist on NT.

The workaround I use is the following:

I am using the nsperl distribution, which contains a program named
perlglob.exe, which globs its argument list and writes its result delimited by
null (!) characters.

Suppose a script is invoked as "script.pl *.c *.h", and the current directory
contains apple.c, buggy.c, cat.c, dog.h and elephant.h.

My script will contain two loops (not just one as you'd have on Unix, since the
Unix shell globs for you):  The outer loop loops over the arguments, e.g. the
array [*.c *.h].  Just inside the inner loop I call perlglob on the current
argument and split its output into an array, splitting on the null character.
The inner loop loops over this array, e.g. the array [apple.c buggy.c cat.c] on
the first iteration of the outer loop, or the array [dog.h elephant.h] on the
second iteration of the outer loop.

The code looks like this:

	foreach $argument (@ARGV) {
		foreach $filename (split /\0/, `perlglob $argument`) {
			... process $filename ...
		}
	}

In contrast to what you'd use if you had an intelligent shell:

	foreach $argument (@ARGV) {
		... process $filename ...
	}

Alternatively, the w.pl program globs its arguments, then invokes the specified
command.  E.g. "w dir *\*\*.c".  So, prefixing a shell command with "w "
provides the shell command with arguments already globbed for it.