Re^4: Preferred technique for named subroutine parameters?

[akho]

Quite the opposite: moving func to a separate module makes Carp work.

And it gives an error that is more specific (for example, it includes the name of the function) than the generic message returned by the anonymous hash.

Besides, it actually dies, not prints something and then goes on as if nothing happened.

[tilly]

use strict;
Foo::bar();

package Foo;
use Carp qw(croak);

sub bar {
    baz(one => 'uno', two => 'dos', three => ); 
}

sub baz { 
    croak "wrong number of arguments for baz(); has to be even" if sca
+lar(@_) % 2;
    my %args = @_; 
    print "func: one translates to $args{one} \n"; 
}
__END__
wrong number of arguments for baz(); has to be even at - line 2

Seems like more typing for a worse result to me. Oh, and if you rename your function, you have to synchronize the function with the error message. A little laziness on that part could lead to all sorts of puzzlement.

[shmem]

Using Carp::confess solves that and gives an accurate picture:

wrong number of arguments for baz(); has to be even at - line 12
    Foo::baz('one', 'uno', 'two', 'dos', 'three') called at - line 8
    Foo::bar() called at - line 2

...except where it doesn't ;-) - there are edge cases where Carp fails to backtrace properly.

[Porculus]

More typing for you, less typing for the poor souls who are actually calling your function. Swings and roundabouts.

Frankly, all the arguments based on catching odd numbers of arguments ring a bit hollow. The most likely case for this kind of bug is where someone carelessly writes

some_function(foo => $foo, bar => @bar);

in which case using an anonymous hashref will only catch the problem if @bar has an even number of elements. That's a pretty marginal gain.

If you actually want your function to be robust, you still have to validate the parameters, or you're asking for all kinds of interesting fun when @bar is three elements long and $bar[1] eq 'foo' (or any other valid parameter name). So you still have to do extra typing, and you still have to throw runtime errors. In this case, having the parameters as a flat list is actually helpful: it means you can detect and warn about duplicate keys, instead of having them silently swallowed by the conversion to a hash.

Or you could be lazy and not validate ... but if you're not worried about even-lengthed garbage inputs, why do you care about odd-lengthed garbage inputs?

[bluto]

++. If you are going to write verification code, then it's worth verifying with something like Params::Validate. You can always disable it in production level code if performance becomes an issue.

[AnomalousMonk]

Besides, it actually dies, not prints something and then goes on as if nothing happened.

Ah, but if you  use warnings FATAL => 'all';, then everything comes to a screeching halt! Again, a very broad benefit for what seems a minimal investment.

(Actually, I'm starting to really like the practice of making all warnings fatal.)