Why does Perl open() documentation use two different FILEHANDLE styles?

Question

The documentation for the open function shows the syntax of open() as:

• open FILEHANDLE,EXPR
• open FILEHANDLE,MODE,EXPR
• open FILEHANDLE,MODE,EXPR,LIST
• open FILEHANDLE,MODE,REFERENCE
• open FILEHANDLE

Down in the examples they have places where a normal $-prefixed variable is used for the file handle:

open(my $fh, "<", "input.txt")

and also examples where a bareword is used:

open(FOO, "|tr '[a-z]' '[A-Z]'");

One question is what is the name of each style as in "I am using __ for a filehandle" in each case? The other is, why did they start using bare words for open() in the documentation? It appears to be the later uses all don't involve normal filename open()s. Is the $-prefixed form not acceptable in those instances?

Answer 1

The bareword form is, essentially, just historical legacy for backward compatibility. Using a lexical variable is pretty much always The Right Thing To Do in new code.

→ incidentally, $x is a lexical scalar variable, where FOO is, as you said, called a bareword

Details/Digression

Just for completeness, as @Joe_Z pointed out in the comments, the lexical filehandle objects are “relatively new,” as a part of the rather major rewrite between Perl 5.005 and 5.6 (they even gained whole orders of magnitude in that version number…).

However, technically, the bareword FOO (or, e.g. STDIN) is interpreted in a separate namespace just for filehandles. Since there isn't a sigil (like $ @ % &) for the filehandle namespace, there are only two ways to reference a filehandle in that namespace:

• You can refer to it in the indirect-object slot of certain functions, like print, who will (behind the scenes) infer that a bareword must refer to a filehandle, for historical reasons;
• You can use a typeglob, like *FOO, which refers to “anything in any namespace which happens to be bound to the symbol FOO.

Note that in some languages, like C or Scheme, a single symbol has no type sigils, and so all symbols can be bound only in one way (e.g. one cannot have a variable named printf and a function named printf in C … generally), whereas in Perl or (e.g.) Common Lisp, the same symbol foo can be bound to many different things; the distinction is that Perl actually requires you to use sigils to disambiguate “which foo you mean” in most contexts. $foo, @foo = @foo[ $x .. $y], $foo[ $n ], %foo = @foo{ $k1, $k2 } = $foo{ $k }, &foo and the like.

By using barewords as filehandles, though, you lose some abilities:

Significantly, in order to bind them locally or lexically (instead of globally), you need to bind every symbol in every namespace, because there is no sigil available. Thus, my $foo and my @foo can live in two different scratchpads (scopes), where perhaps one outlives the other; but my *foo would include both of these, as well as the filehandle foo (and potentially other obscure corner-cases, like format specifiers, although I wouldn't swear to it).

It's also immensely difficult to pass a bareword-style filehandle into a function, and the like.

Basically, barewords inherit all of the downsides of global scope and have none of the advantages of lexical variables.

perldoc perldata has a nice section on Typeglobs and Filehandles which probably explains these things a bit more clearly, as well. I don't have my copy handy, but I believe the Camel goes into more detail on the subject, as well.