Oh My Glob

My recent posts on the inner workings of perl have made me reflect on some of the more advanced, lesser-known parts of the Perl language. One of these features that I'd like to talk about today is the typeglob.

Typeglobs are involved in pretty much every Perl program, even if you're not aware of them: they're used for I/O, global variables, the export system, and much more.

What the Stuff is a Typeglob?

If you've been exposed to older Perl, you may be familiar with the following construct:

open FH, '<', '/dev/null';
# Modern Perl would prefer open my $fh, '<', '/dev/null';

Even if you've never seen that form of open, you are no doubt familiar with STDIN, STDOUT, etc. Now, if you have to pass a filehandle to a subroutine, how do you do it? If you didn't know, this is how:

my_sub(\*STDOUT);

What's with that wacky syntax? I know that backslash creates a reference, but what's with the * character? That, my friend, indicates a typeglob.

Every global variable (note, not locals declared with my; those are on the pad) is associated with a typeglob in what's called a stash. There is one stash per package, and each global variable with the same "root" name (ex. $name, @name, %name, and &name) are associated with the same typeglob in memory. Typeglobs also contain information on the associated filehandle (like STDIN) and format (which I will not discuss here).

Typeglobs. Huh. Yeah. What Are They Good For?

Absolutely nothing? I think not!

Aliasing

One thing you can do with typeglobs is aliasing:

*ff_name = *File::Find::name;

Now, if you want to refer to $File::Find::name, you can refer to it as $ff_name instead. However, this links every entry in the name typeglob in the File::Find package to ff_name. We can tell Perl to be more particular by putting a reference on the right-hand side instead:

*ff_name = \$File::Find::name;

This links $File::Find::name and $ff_name as before, but leaves @ff_name and the others alone. This is actually the underyling mechanism that Exporter uses to implement the importing of symbols.

Dynamic definition of subroutines

I don't write a lot of JavaScript, but it really irks me when I come across something like this:

eval("this." + key + " = value");

Just the same, this makes me angry in Perl:

eval <<"END_PERL";
sub $name {
  my ( \$self ) = @_;
  return \$self->{$name};
}
END_PERL

Because just like you can avoid an eval in JavaScript in the above example, you can avoid the eval in the Perl example too:

# Not so dynamic...
*name = sub {
  my ( $self ) = @_;

  return $self->{$name};
};

# Now we're talking!
do {
  my $title = 'name';
  no strict 'refs'; # lexically enable symbolic refs; don't do this lightly!

  *$title = sub {
    my ( $self ) = @_;

    return $self->{$name};
  };
};

This snippet of code shows how we can dynamically create a subroutine with a name determined at runtime (or compile time, if you wrap the whole thing in a BEGIN block). Note that code like this may seem cool and handy, but it can really fail The Grep Test. So use this technique sparingly!

You may also notice that I defined the subroutine using symbolic references, a faux pas in modern Perl. There is a way to do this without symbolic references; they're called symbol tables, but that's a story for another time.

Introspection

If you have a reference to a glob object, you can ask it for its various associated data:

my $glob = get_glob('STDIN');
*{$glob}{SCALAR} # \$STDIN
*{$glob}{ARRAY}  # \@STDIN
*{$glob}{HASH}   # \%STDIN
*{$glob}{CODE}   # \&STDIN
*{$glob}{IO}     # \*STDIN

You can also ask its name, and which stash it belongs to:

say *{$glob}{NAME};   # 'STDIN';
say *{$glob}{PACKAGE} # 'main'

If the corresponding datum doesn't exist in your Perl program, the reference will be undef, except in the case of SCALAR, which will be a reference to a scalar containing undef. It's impossible to tell if that's an existing scalar containing undef or a previously non-existing entry.

Of course, getting your hands on the glob is the hard part; this is another job for which you need those magical symbol tables if you want to avoid symbolic references.

Subroutine Localization

If you've been using Perl for a while, you're no doubt familiar with the local keyword. It allows you to do cool stuff like this:

do {
  local $Data::Dumper::Terse = 1; # enable terse output lexically

  print STDERR Dumper($my_big_data_structure);
};
# $Data::Dumper::Terse has had its value restored!

However, did you know you can do this with subroutines, too?

package Coworkers::Code {
    sub _some_private_routine {
        ...
    }

    sub some_public_routine {
        ...
        _some_private_routine(@some_arguments_you_want_to_know_about);
        ...
    }
}

my $private_routine = \&Coworkers::Code::_some_private_routine;
# intercept calls to Coworkers::Code::_some_private_routine
local *Coworkers::Code::_some_private_routine = sub {
    use Data::Printer;
    p \@_; # print arguments

    return $private_routine->(@_); # dispatch to actual routine
};

Coworkers::Code::some_public_routine();

This seems like it can cause some serious action at a distance (which it can), but it can be an invaluable tool when debugging, or when carefully used in things like testing code without spending hours mocking LWP::UserAgent. Use wisely!

More Information

I hope you learned something from today's post; if you'd like to know more about typeglobs, please consult perldata and perlref.

(EDIT: By popular demand, the obligatory gif of LSP:)