So, you want only those things that are either numeric scalars or which inherit from overload and can provide the numeric comparisons. That's a good definition of what you feel a number is.

To achieve that, you will probably want to do something along these lines:

use Scalar::Utils qw( blessed );
use DBI qw( looks_like_number );
use overload;

sub my_is_number
{
    my $x = shift;

    unless ( blessed( $x ) )
    {
        return looks_like_number( $x );
    }

    if ( overload::Method( $x, '<=>' ) )
    {
        return !!1;
    }

    return;
}
[download]

If it acts just like a number as far as Perl is concerned, then it's a number as far as I'm concerned.

You can let Perl tell you what it thinks then...

#!perl

use strict;
use warnings;

sub seems_like_number {
    my $thing = shift;
    use warnings qw/FATAL all/; # Promote warnings to fatal, so
                                # they can be trapped.  The effect is
                                # lexically scoped.
    eval {
        $thing += 0;
    };
    return $@ ? 0 : 1;
}

while ( <DATA> ){
    chomp;
    if( seems_like_number( $_ ) ) {
        printf "%10s seems like a number.\n", $_;
    } else {
        printf "%10s doesn't seem like a number.\n", $_;
    }
}

__DATA__

1234
1.234
-2.345
1-1+2
0.032
.321
ASDF
ASDF1234
1234ASDF
[download]

The first line after __DATA__ is intentionally blank, to test whether or not an empty string will qualify as a number... and of course it won't.

Ok, now if Perl thinks it's not a number, you'll know about it, and if Perl doesn't object to it being considered a number, you'll know that too.

Note that looks_like_number() tells you exactly the same thing, just without all of the overhead. (:

- tye        

For the record, you're absolutely right. Scalar::Util's looks_like_number() is 655% faster than my eval trap fatal warning approach. But for some reason I do get a sort of a kick out of the idea of letting perl (the interpreter) tell me if a scalar is a number, rather than Perl (the script). Consider my solution purely an academic enveavor, and use the module for production code.

The benchmark script:

Read more... (939 Bytes)

The results:

        Rate Seems Looks
Seems 38.8/s    --  -87%
Looks  297/s  665%    --
[download]




Dave