10 use B qw/class ppname svref_2object OPf_KIDS/;
14 Sub::Nary - Try to count how many elements a subroutine can return in list context.
31 my $sn = Sub::Nary->new();
32 my $r = $sn->nary(\&hlagh);
36 This module uses the L<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).
42 The usual constructor. Currently takes no argument.
44 =head2 C<nary $coderef>
46 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. A few special keys are also used :
52 C<'list'> is used to denote a possibly infinite number of returned arguments ;
56 C<'exit'> gives the probability for C<exit> to be called somewhere in the code.
60 The return value hence would look at
62 { 1 => 0.2, 2 => 0.4, 4 => 0.25, list => 0.1, exit => 0.05 }
64 that is, we should get C<1> scalar C<1> time over C<5> and so on. The sum of all values is C<1>. The returned result, and all the results obtained from intermediate subs, are cached into the object.
68 Flushes the L<Sub::Nary> object cache. Returns the object itself.
70 =head1 PROBABILITY OF RETURN
72 The probability is computed as such :
78 When branching, each branch is considered equally possible.
80 For example, the subroutine
90 is seen returning one or two arguments each with probability C<1/2>.
102 it is considered to return C<3> scalars with probability C<1/2>, C<2> with probability C<1/2 * 1/2 = 1/4> and C<1> (when the two tests fail, the last computed value is returned, which here is C<< $x > 0.9 >> evaluated in the scalar context of the test) with remaining probability C<1/4>.
106 The total probability law for a given returning point is the convolution product of the probabilities of its list elements.
111 return 1, simple(), 2
114 returns C<3> or C<4> arguments with probability C<1/2> ; and
117 return simple(), simple()
120 never returns C<1> argument but returns C<2> with probability C<1/2 * 1/2 = 1/4>, C<3> with probability C<1/2 * 1/2 + 1/2 * 1/2 = 1/2> and C<4> with probability C<1/4> too.
124 If a core function may return different numbers of scalars, each kind is considered equally possible.
126 For example, C<stat> returns C<13> elements on success and C<0> on error. The according probability will then be C<< { 0 => 0.5, 13 => 0.5 } >>.
130 The C<list> and C<exit> states are absorbing in regard of all the other ones.
132 This is just a pedantic way to say that C<list + fixed length = list>, C<exit + fixed length = exit>, but note also that C<exit + list = exit>.
136 return 1, simple(), @_
139 is considered as always returning an unbounded list.
141 Also, the convolution law does not behave the same when C<list> or C<exit> elements are involved : in the following example,
152 return oneorlist(), oneorlist()
155 C<composed> returns C<2> scalars with probability C<1/2 * 1/2 = 1/4> and a C<list> with probability C<3/4>.
163 XSLoader::load(__PACKAGE__, $VERSION);
167 croak 'First argument isn\'t a valid ' . __PACKAGE__ . ' object'
168 unless ref $_[0] and $_[0]->isa(__PACKAGE__);
173 $class = ref($class) || $class || __PACKAGE__;
174 bless { cache => { } }, $class;
180 $self->{cache} = { };
189 return ($self->enter(svref_2object($sub)))[1];
193 local $SIG{__DIE__} = \&Carp::confess;
195 $n eq 'null' ? substr(ppname($_[0]->targ), 3) : $n
199 my ($p, $n, $c) = @_;
200 return unless defined $p;
201 return { 0 => $c } unless $n;
203 my $z = delete $p->{0};
204 return { 'list' => $c } unless $z;
205 return { 0 => $c } if $z == 1;
206 return { 0 => $c * $z, list => $c * (1 - $z) };
208 my $r = combine map { { %$p } } 1 .. $n;
209 $r->{$_} *= $c for keys %$r;
215 $ops{$_} = 1 for scalops;
216 $ops{$_} = 0 for qw/stub nextstate pushmark iter unstack/;
217 $ops{$_} = 1 for qw/padsv/;
218 $ops{$_} = 'list' for qw/padav/;
219 $ops{$_} = 'list' for qw/padhv rv2hv/;
220 $ops{$_} = 'list' for qw/padany/;
221 $ops{$_} = 'list' for qw/match entereval readline/;
223 $ops{each} = { 0 => 0.5, 2 => 0.5 };
224 $ops{stat} = { 0 => 0.5, 13 => 0.5 };
226 $ops{caller} = sub { my @a = caller 0; scalar @a }->();
227 $ops{localtime} = do { my @a = localtime; scalar @a };
228 $ops{gmtime} = do { my @a = gmtime; scalar @a };
230 $ops{$_} = { 0 => 0.5, 10 => 0.5 } for map "gpw$_", qw/nam uid ent/;
231 $ops{$_} = { 0 => 0.5, 4 => 0.5 } for map "ggr$_", qw/nam gid ent/;
232 $ops{$_} = 'list' for qw/ghbyname ghbyaddr ghostent/;
233 $ops{$_} = { 0 => 0.5, 4 => 0.5 } for qw/gnbyname gnbyaddr gnetent/;
234 $ops{$_} = { 0 => 0.5, 3 => 0.5 } for qw/gpbyname gpbynumber gprotoent/;
235 $ops{$_} = { 0 => 0.5, 4 => 0.5 } for qw/gsbyname gsbyport gservent/;
238 my ($self, $cv) = @_;
240 return undef, 'list' if class($cv) ne 'CV';
244 return undef, { %{$self->{cache}->{$tag}} } if exists $self->{cache}->{$tag};
246 # Anything can happen with recursion
247 for (@{$self->{cv}}) {
248 return undef, 'list' if $tag == tag($_->ROOT);
251 unshift @{$self->{cv}}, $cv;
252 my $r = add $self->inspect($op->first);
253 shift @{$self->{cv}};
255 $self->{cache}->{$tag} = { %$r };
260 my ($self, $op) = @_;
263 return add($self->inspect_kids($op)), undef if $n eq 'return';
265 my $meth = $self->can('pp_' . $n);
266 return $self->$meth($op) if $meth;
268 if (exists $ops{$n}) {
270 $l = { %$l } if ref $l;
274 if (class($op) eq 'LOGOP' and not null $op->first) {
278 my ($r1, $l1) = $self->inspect($op);
279 return $r1, $l1 if defined $r1 and zero $l1;
283 my ($r2, $l2) = $self->inspect($op);
288 # If the logop has no else branch, it can also return the *scalar* result of
292 ($r3, $l3) = $self->inspect($op);
295 my $r = add $r1, scale $c / 2, add $r2, $r3;
296 my $l = scale $c / 2, add $l2, $l3;
300 return $self->inspect_kids($op);
304 my ($self, $op) = @_;
306 return undef, 0 unless $op->flags & OPf_KIDS;
309 return undef, 0 if null $op;
310 if (name($op) eq 'pushmark') {
312 return undef, 0 if null $op;
317 for (; not null $op; $op = $op->sibling) {
319 if ($n eq 'nextstate') {
323 if ($n eq 'lineseq') {
328 my ($rc, $lc) = $self->inspect($op);
330 $r = add $r, scale $c, $rc if defined $rc;
331 if (not defined $lc) {
335 push @l, scale $c, $lc;
338 my $l = scale +(1 - count $r), normalize combine @l;
343 # Stolen from B::Deparse
345 sub padval { $_[0]->{cv}->[0]->PADLIST->ARRAYelt(1)->ARRAYelt($_[1]) }
348 my ($self, $op) = @_;
349 if (class($op) eq 'PADOP') {
350 return $self->padval($op->padix)
351 } else { # class($op) eq "SVOP"
357 my ($self, $op) = @_;
359 # the constant could be in the pad (under useithreads)
360 $sv = $self->padval($op->targ) unless $$sv;
365 my ($self, $op) = @_;
367 $op = $op->first while $op->flags & OPf_KIDS;
368 # First must be a pushmark
370 # Next must be non null - at worse it's the rv2cv
374 for (; not null $op->sibling; $op = $op->sibling) {
375 my ($rc, $lc) = $self->inspect($op);
376 return $rc, $lc if defined $rc and not defined $lc;
377 $r = add $r, scale $c, $rc;
381 if (name($op) eq 'rv2cv') {
385 my $next = $op->sibling;
386 while (not null $next) {
388 $next = $next->sibling;
391 } while ($op->flags & OPf_KIDS and { map { $_ => 1 } qw/null leave/ }->{$n});
392 return 'list', undef unless { map { $_ => 1 } qw/gv refgen/ }->{$n};
393 local $self->{sub} = 1;
394 my ($rc, $lc) = $self->inspect($op);
395 return $r, scale $c, $lc;
398 return $r, { 'list' => $c };
403 my ($self, $op) = @_;
405 return $self->{sub} ? $self->enter($self->gv_or_padgv($op)->CV) : (undef, 1)
409 my ($self, $op) = @_;
411 return $self->{sub} ? $self->enter($self->const_sv($op)) : (undef, 1)
415 my ($self, $op) = @_;
418 if ($op->flags & OPf_KIDS) {
419 ($r, my $l) = $self->inspect($op->first);
420 return $r, $l if defined $r and zero $l;
421 $r->{exit} = 1 - count $r;
423 $r = { 'exit' => 1 };
430 my ($self, $op) = @_;
432 my ($r, undef) = $self->inspect_kids($op);
434 my $c = 1 - count $r;
435 $r->{die} = $c if $c;
444 my ($self, $op) = @_;
447 while ($op->flags & OPf_KIDS) {
448 my $nop = $op->first;
450 if ($nn eq 'pushmark') {
451 $nop = $nop->sibling;
454 if ($n eq 'rv2cv' and $nn eq 'gv') {
455 return $self->enter($self->gv_or_padgv($nop)->CV);
461 return undef, 'list';
465 my ($self, $op) = @_;
467 return undef, 0 unless $op->isa('B::SVOP');
469 my $sv = $self->const_sv($op);
474 } elsif ($c eq 'HV') {
481 sub pp_aslice { $_[0]->inspect($_[1]->first->sibling) }
484 *pp_hslice = *pp_aslice{CODE};
486 sub pp_lslice { $_[0]->inspect($_[1]->first) }
489 my ($self, $op) = @_;
492 if (name($op) eq 'gv') {
493 return undef, { list => 1 };
500 my ($self, $op) = @_;
502 my $r = ($self->inspect($op->first))[0];
504 my $c = 1 - count $r;
505 return $r, $c ? { 1 => $c } : undef
509 my ($self, $op) = @_;
513 # Can't assign to return
514 my $l = ($self->inspect($op->sibling))[1];
515 return undef, $l if not exists $l->{list};
521 my ($self, $op) = @_;
523 my ($r, $l) = $self->inspect_kids($op);
525 my $d = delete $r->{die};
526 return $r, $l if not defined $d;
528 my $z = delete $l->{0};
529 $l = { %$l, 0 => $z };
540 my ($self, $op) = @_;
545 if (name($op) eq 'enteriter') { # for loop ?
547 ($r1, $l1) = $self->inspect($op);
548 return $r1, $l1 if defined $r1 and zero $l1;
553 if (name($op->first) eq 'and') {
554 ($r2, $l2) = $self->inspect($op->first->first);
555 return $r2, $l2 if defined $r2 and zero $l2;
557 return { list => 1 }, undef if !$for and defined $r2;
558 my ($r3, $l3) = $self->inspect($op->first->first->sibling);
559 return { list => 1 }, undef if defined $r3 and defined $l3;
560 $r2 = add $r2, scale $c, $r3;
562 ($r2, $l2) = $self->inspect($op);
563 return { list => 1 }, undef if defined $r2 and defined $l2;
566 my $r = (defined $r1) ? add $r1, scale +(1 - count $r1), $r2
568 my $c = 1 - count $r;
569 return $r, $c ? { 0 => $c } : undef;
573 my ($self, $op) = @_;
576 return $self->inspect($op) if name($op) ne 'range';
579 my $begin = $op->first;
580 if (name($begin) eq 'const') {
581 my $end = $begin->sibling;
582 if (name($end) eq 'const') {
583 $begin = $self->const_sv($begin);
584 $end = $self->const_sv($end);
586 no warnings 'numeric';
587 $begin = int ${$begin->object_2svref};
588 $end = int ${$end->object_2svref};
590 return undef, $end - $begin + 1;
592 ($r, $l) = $self->inspect($end);
595 ($r, $l) = $self->inspect($begin);
598 my $c = 1 - count $r;
599 return $r, $c ? { 'list' => $c } : undef
603 my ($self, $op) = @_;
606 return $self->inspect($op) if name($op) ne 'grepstart';
607 $op = $op->first->sibling;
609 my ($r2, $l2) = $self->inspect($op->sibling);
610 return $r2, $l2 if defined $r2 and zero $l2;
611 my $c2 = count $l2; # First one to happen
613 my ($r1, $l1) = $self->inspect($op);
614 return (add $r2, scale $c2, $r1), undef if defined $r1 and zero $l1
618 $l2 = { $l2 => 1 } unless ref $l2;
621 add map { scale $l2->{$_}, cumulate $r1, $_, $c1 } keys %$l2;
622 my $c = 1 - count $r;
623 return $r, $c ? { ((zero $l2) ? 0 : 'list') => $c } : undef;
627 my ($self, $op) = @_;
630 return $self->inspect($op) if name($op) ne 'mapstart';
631 $op = $op->first->sibling;
633 my ($r2, $l2) = $self->inspect($op->sibling);
634 return $r2, $l2 if defined $r2 and zero $l2;
635 my $c2 = count $l2; # First one to happen
637 my ($r1, $l1) = $self->inspect($op);
638 return (add $r2, scale $c2, $r1), undef if defined $r1 and zero $l1
642 $l2 = { $l2 => 1 } unless ref $l2;
645 add map { scale $l2->{$_}, cumulate $r1, $_, $c1 } keys %$l2;
646 my $c = 1 - count $r;
647 my $l = scale $c, normalize add map { power $l1, $_, $l2->{$_} } keys %$l2;
653 An object-oriented module shouldn't export any function, and so does this one.
657 The algorithm may be pessimistic (things seen as C<list> while they are of fixed length) but not optimistic (the opposite, duh).
659 C<wantarray> isn't specialized when encountered in the optree.
665 L<Carp> (standard since perl 5), L<B> (since perl 5.005) and L<XSLoader> (since perl 5.006).
669 Vincent Pit, C<< <perl at profvince.com> >>, L<http://www.profvince.com>.
671 You can contact me by mail or on C<irc.perl.org> (vincent).
675 Please report any bugs or feature requests to C<bug-sub-nary at rt.cpan.org>, or through the web interface at L<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.
679 You can find documentation for this module with the perldoc command.
683 Tests code coverage report is available at L<http://www.profvince.com/perl/cover/Sub-Nary>.
685 =head1 ACKNOWLEDGEMENTS
687 Thanks to Sebastien Aperghis-Tramoni for helping to name this module.
689 =head1 COPYRIGHT & LICENSE
691 Copyright 2008 Vincent Pit, all rights reserved.
693 This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
697 1; # End of Sub::Nary