NAME
    Sub::Nary - Try to count how many elements a subroutine can return in
    list context.

VERSION
    Version 0.03

SYNOPSIS
        use Sub::Nary;

        my $sn = Sub::Nary->new();
        my $r  = $sn->nary(\&hlagh);

DESCRIPTION
    This module uses the B framework to walk into subroutines and try to
    guess how many scalars are likely to be returned in list context. It's
    not always possible to give a definitive answer to this question at
    compile time, so the results are given in terms of "probability of
    return" (to be understood in a sense described below).

METHODS
  "new"
    The usual constructor. Currently takes no argument.

  "nary $coderef"
    Takes a code reference to a named or anonymous subroutine, and returns a
    hash reference whose keys are the possible numbers of returning scalars,
    and the corresponding values the "probability" to get them. The special
    key 'list' is used to denote a possibly infinite number of returned
    arguments. The return value hence would look at

        { 1 => 0.2, 2 => 0.4, 4 => 0.3, list => 0.1 }

    that is, we should get 1 scalar 1 time over 5 and so on. The sum of all
    values is 1. The returned result, and all the results obtained from
    intermediate subs, are cached into the object.

  "flush"
    Flushes the Sub::Nary object cache. Returns the object itself.

PROBABILITY OF RETURN
    The probability is computed as such :

    * When branching, each branch is considered equally possible.
        For example, the subroutine

            sub simple {
             if (rand < 0.1) {
              return 1;
             } else {
              return 2, 3;
             }
            }

        is seen returning one or two arguments each with probability "1/2".
        As for

            sub hlagh {
             my $x = rand;
             if ($x < 0.1) {
              return 1, 2, 3;
             } elsif ($x > 0.9) {
              return 4, 5;
             }
            }

        it is considered to return 3 scalars with probability "1/2", 2 with
        probability "1/2 * 1/2 = 1/4" and 1 (when the two tests fail, the
        last computed value is returned, which here is "$x > 0.9" evaluated
        in the scalar context of the test) with remaining probability "1/4".

    * The total probability law for a given returning point is the
    convolution product of the probabilities of its list elements.
        As such,

            sub notsosimple {
             return 1, simple(), 2
            }

        returns 3 or 4 arguments with probability "1/2" ; and

            sub double {
             return simple(), simple()
            }

        never returns 1 argument but returns 2 with probability "1/2 * 1/2 =
        1/4", 3 with probability "1/2 * 1/2 + 1/2 * 1/2 = 1/2" and 4 with
        probability "1/4" too.

    * If a core function may return different numbers of scalars, each kind
    is considered equally possible.
        For example, "stat" returns 13 elements on success and 0 on error.
        The according probability will then be "{ 0 => 0.5, 13 => 0.5 }".

    * The "list" state is absorbing in regard of all the other ones.
        This is just a pedantic way to say that "list + fixed length =
        list". That's why

            sub listy {
             return 1, simple(), @_
            }

        is considered as always returning an unbounded list.

        Also, the convolution law does not behave the same when "list"
        elements are involved : in the following example,

            sub oneorlist {
             if (rand < 0.1) {
              return 1
             } else {
              return @_
             }
            }

            sub composed {
             return oneorlist(), oneorlist()
            }

        "composed" returns 2 scalars with probability "1/2 * 1/2 = 1/4" and
        a "list" with probability "3/4".

EXPORT
    An object-oriented module shouldn't export any function, and so does
    this one.

CAVEATS
    The algorithm may be pessimistic (things seen as "list" while they are
    of fixed length) but not optimistic (the opposite, duh).

    "wantarray" isn't specialized when encountered in the optree.

DEPENDENCIES
    perl 5.8.1.

    Carp (standard since perl 5), B (since perl 5.005) and XSLoader (since
    perl 5.006).

AUTHOR
    Vincent Pit, "<perl at profvince.com>", <http://www.profvince.com>.

    You can contact me by mail or on #perl @ FreeNode (vincent or
    Prof_Vince).

BUGS
    Please report any bugs or feature requests to "bug-b-nary at
    rt.cpan.org", or through the web interface at
    <http://rt.cpan.org/NoAuth/ReportBug.html?Queue=Sub-Nary>. I will be
    notified, and then you'll automatically be notified of progress on your
    bug as I make changes.

SUPPORT
    You can find documentation for this module with the perldoc command.

        perldoc Sub::Nary

    Tests code coverage report is available at
    <http://www.profvince.com/perl/cover/Sub-Nary>.

ACKNOWLEDGEMENTS
    Thanks to Sebastien Aperghis-Tramoni for helping to name this module.

COPYRIGHT & LICENSE
    Copyright 2008 Vincent Pit, all rights reserved.

    This program is free software; you can redistribute it and/or modify it
    under the same terms as Perl itself.