9 use List::Util qw/reduce sum/;
11 use B qw/class ppname svref_2object OPf_KIDS/;
15 Sub::Nary - Try to count how many elements a subroutine can return in list context.
32 my $sn = Sub::Nary->new();
33 my $r = $sn->nary(\&hlagh);
37 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).
43 The usual constructor. Currently takes no argument.
45 =head2 C<nary $coderef>
47 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 C<'list'> is used to denote a possibly infinite number of returned arguments. The return value hence would look at
49 { 1 => 0.2, 2 => 0.4, 4 => 0.3, list => 0.1 }
51 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.
55 Flushes the L<Sub::Nary> object cache. Returns the object itself.
57 =head1 PROBABILITY OF RETURN
59 The probability is computed as such :
63 =item * All the returning points in the same subroutine (i.e. all the explicit C<return> and the last computed value) are considered equally possible.
65 For example, the subroutine
75 is seen returning one or two arguments each with probability C<1/2>.
87 it is considered to return 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), C<2> or C<3> arguments each with probability C<1/3>.
89 =item * The total probability law for a given returning point is the convolution product of the probabilities of its list elements.
97 returns C<3> or C<4> arguments with probability C<1/2> ; and
100 return simple(), simple()
103 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.
105 =item * If a core function may return different numbers of scalars, each kind is considered equally possible.
107 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 } >>.
109 =item * The C<list> state is absorbing in regard of all the other ones.
111 This is just a pedantic way to say that "list + fixed length = list".
115 return 1, simple(), @_
118 is considered as always returning an unbounded list.
120 Also, the convolution law does not behave the same when C<list> elements are involved : in the following example,
131 return oneorlist(), oneorlist()
134 C<composed> returns C<2> scalars with probability C<1/2 * 1/2 = 1/4> and a C<list> with probability C<3/4>.
142 XSLoader::load(__PACKAGE__, $VERSION);
146 croak 'First argument isn\'t a valid ' . __PACKAGE__ . ' object'
147 unless ref $_[0] and $_[0]->isa(__PACKAGE__);
152 $class = ref($class) || $class || __PACKAGE__;
153 bless { cache => { } }, $class;
159 $self->{cache} = { };
168 return $self->enter(svref_2object($sub));
173 $n eq 'null' ? substr(ppname($_[0]->targ), 3) : $n
179 my $la = delete $a->{list};
180 my $lb = delete $b->{list};
181 if (defined $la || defined $lb) {
184 $res{list} = $la + $lb - $la * $lb;
186 while (my ($ka, $va) = each %$a) {
188 while (my ($kb, $vb) = each %$b) {
189 my $key = $ka + int $kb;
190 $res{$key} += $va * $vb;
194 }} map { (ref) ? $_ : { $_ => 1 } } grep defined, @_;
198 my ($p, $n, $c) = @_;
199 my $s = sum values %$p;
200 return { 0 => $s * $c } unless $n;
202 my $z = delete $p->{0};
203 return { 'list' => $c } unless $z;
204 return { 0 => $c } if $z == 1;
205 return { 0 => $c * $z, list => $c * ($s - $z) };
207 my $r = combine map { { %$p } } 1 .. $n;
208 $r->{$_} *= $c for keys %$r;
214 $a->{$_} += $b->{$_} for keys %$b;
216 } map { (ref) ? $_ : { $_ => 1 } } grep defined, @_;
221 $ops{$_} = 1 for scalops;
222 $ops{$_} = 0 for qw/stub nextstate/;
223 $ops{$_} = 1 for qw/padsv/;
224 $ops{$_} = 'list' for qw/padav/;
225 $ops{$_} = 'list' for qw/padhv rv2hv/;
226 $ops{$_} = 'list' for qw/padany/;
227 $ops{$_} = 'list' for qw/match entereval readline/;
229 $ops{each} = { 0 => 0.5, 2 => 0.5 };
230 $ops{stat} = { 0 => 0.5, 13 => 0.5 };
232 $ops{caller} = sub { my @a = caller 0; scalar @a }->();
233 $ops{localtime} = do { my @a = localtime; scalar @a };
234 $ops{gmtime} = do { my @a = gmtime; scalar @a };
236 $ops{$_} = { 0 => 0.5, 10 => 0.5 } for map "gpw$_", qw/nam uid ent/;
237 $ops{$_} = { 0 => 0.5, 4 => 0.5 } for map "ggr$_", qw/nam gid ent/;
238 $ops{$_} = 'list' for qw/ghbyname ghbyaddr ghostent/;
239 $ops{$_} = { 0 => 0.5, 4 => 0.5 } for qw/gnbyname gnbyaddr gnetent/;
240 $ops{$_} = { 0 => 0.5, 3 => 0.5 } for qw/gpbyname gpbynumber gprotoent/;
241 $ops{$_} = { 0 => 0.5, 4 => 0.5 } for qw/gsbyname gsbyport gservent/;
244 my ($self, $cv) = @_;
246 return 'list' if class($cv) ne 'CV';
250 return { %{$self->{cache}->{$tag}} } if exists $self->{cache}->{$tag};
252 # Anything can happen with recursion
253 for (@{$self->{cv}}) {
254 return 'list' if $tag == tag($_->ROOT);
257 unshift @{$self->{cv}}, $cv;
258 (my $r, undef) = $self->expect_any($op->first);
259 shift @{$self->{cv}};
261 $r = { $r => 1} unless ref $r;
262 my $total = sum values %$r;
263 $r = { map { $_ => $r->{$_} / $total } keys %$r };
264 $self->{cache}->{$tag} = { %$r };
269 my ($self, $op) = @_;
271 return ($self->expect_kids($op))[0] => 1 if name($op) eq 'return';
273 if ($op->flags & OPf_KIDS) {
274 for ($op = $op->first; not null $op; $op = $op->sibling) {
275 my ($p, $r) = $self->expect_return($op);
276 return $p => 1 if $r;
284 my ($self, $op) = @_;
287 return ($self->expect_kids($op))[0] => 1 if $n eq 'return';
289 my $meth = $self->can('pp_' . $n);
290 return $self->$meth($op) if $meth;
292 if (exists $ops{$n}) {
294 $r = { %$r } if ref $r;
298 if (class($op) eq 'LOGOP' and not null $op->first) {
303 ($p, $r) = $self->expect_return($op);
304 return $p => 1 if $r;
307 push @res, ($self->expect_any($op))[0];
309 # If the logop has no else branch, it can also return the *scalar* result of
315 push @res, ($self->expect_any($op))[0];
318 return (add @res) => 0;
321 return $self->expect_kids($op);
325 my ($self, $op) = @_;
327 return 0 unless $op->flags & OPf_KIDS;
331 for ($op = $op->first; not null $op; $op = $op->sibling) {
333 next if $n eq 'pushmark';
334 if ($n eq 'nextstate'
335 and not null(($op = $op->sibling)->sibling)) {
336 ($p, $r) = $self->expect_return($op);
337 return $p => 1 if $r;
339 ($p, $r) = $self->expect_any($op);
340 return $p => 1 if $r;
345 return (combine @res) => 0;
348 # Stolen from B::Deparse
350 sub padval { $_[0]->{cv}->[0]->PADLIST->ARRAYelt(1)->ARRAYelt($_[1]) }
353 my ($self, $op) = @_;
354 if (class($op) eq 'PADOP') {
355 return $self->padval($op->padix)
356 } else { # class($op) eq "SVOP"
362 my ($self, $op) = @_;
364 # the constant could be in the pad (under useithreads)
365 $sv = $self->padval($op->targ) unless $$sv;
370 my ($self, $op, $exp) = @_;
373 while ($next->flags & OPf_KIDS) {
374 $next = $next->first;
376 while (not null $next) {
378 my ($p, $r) = $self->expect_return($op, $exp);
379 return $p => 1 if $r;
380 $next = $op->sibling;
383 if (name($op) eq 'rv2cv') {
387 my $next = $op->sibling;
388 while (not null $next) {
390 $next = $next->sibling;
393 } while ($op->flags & OPf_KIDS and { map { $_ => 1 } qw/null leave/ }->{$n});
394 return 'list' unless { map { $_ => 1 } qw/gv refgen/ }->{$n};
395 local $self->{sub} = 1;
396 return $self->expect_any($op, $exp);
404 my ($self, $op) = @_;
406 return $self->{sub} ? $self->enter($self->gv_or_padgv($op)->CV) : 1
410 my ($self, $op) = @_;
412 return $self->{sub} ? $self->enter($self->const_sv($op)) : 1
416 my ($self, $op) = @_;
419 while ($op->flags & OPf_KIDS) {
420 my $nop = $op->first;
422 if ($nn eq 'pushmark') {
423 $nop = $nop->sibling;
426 if ($n eq 'rv2cv' and $nn eq 'gv') {
427 return $self->enter($self->gv_or_padgv($nop)->CV);
437 my ($self, $op) = @_;
439 my $sv = $self->const_sv($op);
442 return $sv->FILL + 1;
443 } elsif ($c eq 'HV') {
444 return 2 * $sv->FILL;
450 sub pp_aslice { $_[0]->expect_any($_[1]->first->sibling) }
453 *pp_hslice = *pp_aslice{CODE};
455 sub pp_lslice { $_[0]->expect_any($_[1]->first) }
458 my ($self, $op) = @_;
461 return (name($op) eq 'const') ? $self->expect_any($op) : 'list';
465 my ($self, $op) = @_;
469 # Can't assign to return
470 my ($p, $r) = $self->expect_any($op->sibling);
471 return $p => 0 if not exists $p->{list};
473 $self->expect_any($op);
476 sub pp_leaveloop { $_[0]->expect_return($_[1]->first->sibling) }
479 my ($self, $op) = @_;
482 return 'list' if name($op) ne 'range';
484 my $begin = $op->first;
485 if (name($begin) eq 'const') {
486 my $end = $begin->sibling;
487 if (name($end) eq 'const') {
488 $begin = $self->const_sv($begin);
489 $end = $self->const_sv($end);
490 no warnings 'numeric';
491 return int(${$end->object_2svref}) - int(${$begin->object_2svref}) + 1;
493 my ($p, $r) = $self->expect_return($end);
494 return $p => 1 if $r;
497 my ($p, $r) = $self->expect_return($begin);
498 return $p => 1 if $r;
505 my ($self, $op) = @_;
508 return 'list' if name($op) ne 'grepstart';
510 $op = $op->first->sibling;
511 my ($p, $r) = $self->expect_any($op);
512 return $p => 1 if $r;
515 ($p, $r) = $self->expect_any($op);
516 return $p => 1 if $r;
522 my ($self, $op) = @_;
525 return 'list' if name($op) ne 'mapstart';
527 $op = $op->first->sibling;
528 my ($p1, $r) = $self->expect_any($op);
529 return $p1 => 1 if $r;
532 (my $p2, $r) = $self->expect_any($op);
533 return $p2 => 1 if $r;
534 $p2 = { $p2 => 1 } unless ref $p2;
536 return add map { power $p1, $_, $p2->{$_} } keys %$p2;
541 An object-oriented module shouldn't export any function, and so does this one.
545 The algorithm may be pessimistic (things seen as C<list> while they are of fixed length) but not optimistic (the opposite, duh).
547 C<wantarray> isn't specialized when encountered in the optree.
553 L<Carp> (standard since perl 5), L<B> (since perl 5.005), L<XSLoader> (since perl 5.006) and L<List::Util> (since perl 5.007003).
557 Vincent Pit, C<< <perl at profvince.com> >>, L<http://www.profvince.com>.
559 You can contact me by mail or on #perl @ FreeNode (vincent or Prof_Vince).
563 Please report any bugs or feature requests to C<bug-b-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.
567 You can find documentation for this module with the perldoc command.
571 Tests code coverage report is available at L<http://www.profvince.com/perl/cover/Sub-Nary>.
573 =head1 ACKNOWLEDGEMENTS
575 Thanks to Sebastien Aperghis-Tramoni for helping to name this module.
577 =head1 COPYRIGHT & LICENSE
579 Copyright 2008 Vincent Pit, all rights reserved.
581 This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
585 1; # End of Sub::Nary