TL;DR

An iterator-based implementation of Permutations with Heap’s Algorithm. You saw it coming.

As implicitly promised, we will leverage our iterative implementation to get an iterator-based implementation.

# Starting from the iterative implementation

This is where we left in Permutations with Heap’s Algorithm:

``````sub permutations {
my @indexes = 0 .. \$#_;
my @stack = (0) x @indexes;
output(@_[@indexes]);
my \$sp = 0;
while (\$sp < @indexes) {
if (\$stack[\$sp] < \$sp) {
my \$other = \$sp % 2 ? \$stack[\$sp] : 0;
@indexes[\$sp, \$other] = @indexes[\$other, \$sp];
output(@_[@indexes]);
\$stack[\$sp]++;
\$sp = 0;
}
else {
\$stack[\$sp++] = 0;
}
}
}
``````

We will turn this into an iterator-based implementation, which means that we will have a function that returns another function, that will provide us a new permutation each time it is called, until they have been all emitted.

One little thorn in the side is the fact that `output` is called in two different places, one over the very first arrangement, then during the loop. This forces us to treat the “initial state” as something special, i.e. skip all computations for the first call to the sub and then consider the stuff in the `while` loop from the second call on.

Another observation that comes to our help is that the placement of the call to `output` inside the loop can be moved a little ahead, like this:

``````sub permutations {
my @indexes = 0 .. \$#_;
my @stack = (0) x @indexes;
output(@_[@indexes]);
my \$sp = 0;
while (\$sp < @indexes) {
if (\$stack[\$sp] < \$sp) {
my \$other = \$sp % 2 ? \$stack[\$sp] : 0;
@indexes[\$sp, \$other] = @indexes[\$other, \$sp];
\$stack[\$sp]++;
\$sp = 0;
output(@_[@indexes]);
}
else {
\$stack[\$sp++] = 0;
}
}
}
``````

In this way, it becomes the last statement in its branch of the `if` condition, which will come handy later.

# Turning into an iterator

The generic structure of our iterator factory function is the following:

``````sub iterator_creation_factory_function {

# declaration and initialization of state-tracking variables

return sub { # this is the actual iterator

# move the state on to the next step

# return the proper value for the current state

};
}
``````

In our iterative implementation, variables `@indexes`, `@stack`, `\$sp`, and `@_` too are all part of the state, hence they need to be declared outside the iterator, so that they will be closed by the iterator function.

Additionally, we have to take care of the `@_` variable here, because we cannot “see” the `@_` of the factory inside the itertor (because the iterator is a function with its own `@_`). Hence, we’ll change the interface to take a reference to an array in parameters passed as either hash or hash reference:

``````sub permutations_iterator {
my %args = (@_ && ref(\$_)) ? %{\$_} : @_;
my \$items = \$args{items} || die "invalid or missing parameter 'items'";
my \$filter = \$args{filter} || sub { wantarray ? @_ : [@_] };
my @indexes = 0 .. \$#\$items;
my @stack = (0) x @indexes;
my \$sp = undef;
return sub { ... }
}
``````

While we’re at it, we also support a `filter` function that will be passed the permutation as input, and whose output will be returned by our iterator.

Note that we explicitly set `\$sp` to start as `undef`. This will let us distinguish the very first call from the other ones, as we saw that we need to do this. Hence, in our first call we will just set `\$sp` to the real initialization value, and skip everything the `while` would normally do:

``````sub permutations_iterator {
my %args = (@_ && ref(\$_)) ? %{\$_} : @_;
my \$items = \$args{items} || die "invalid or missing parameter 'items'";
my \$filter = \$args{filter} || sub { wantarray ? @_ : [@_] };
my @indexes = 0 .. \$#\$items;
my @stack = (0) x @indexes;
my \$sp = undef;
return sub {
if (! defined \$sp) { \$sp = 0 }
else { ... }
return \$filter->(@{\$items}[@indexes]) if \$sp < @indexes;
return;
}
}
``````

As anticipated, the iterator will return the output of calling the `\$filter` function over the current arrangement of the data, or nothing if we got past the last permutation.

We can now take a look at what’s executed from the second iterator call on, that is our adaptation of the `while` loop in the original iterative implementation. The key here is that we have to enter the loop, but only execute as much work as to get to the right state for generating the needed output and then exit the loop:

``````sub permutations_iterator {
my %args = (@_ && ref(\$_)) ? %{\$_} : @_;
my \$items = \$args{items} || die "invalid or missing parameter 'items'";
my \$filter = \$args{filter} || sub { wantarray ? @_ : [@_] };
my @indexes = 0 .. \$#\$items;
my @stack = (0) x @indexes;
my \$sp = undef;
return sub {
if (! defined \$sp) { \$sp = 0 }
else {
while (\$sp < @indexes) {
if (\$stack[\$sp] < \$sp) {
my \$other = \$sp % 2 ? \$stack[\$sp] : 0;
@indexes[\$sp, \$other] = @indexes[\$other, \$sp];
\$stack[\$sp]++;
\$sp = 0;
last;
}
else {
\$stack[\$sp++] = 0;
}
}
}
return \$filter->(@{\$items}[@indexes]) if \$sp < @indexes;
return;
}
}
``````

It’s now clear why it was so useful to move the call to `output` to be the last statement in the block… in this way we can substitute it with a literal `last` and exit the loop exactly when we need to provide an output.

Thanks to the closure mechanism, the next time the iterator is called, it will enter the `while` loop just as if it never exited it… allowing us to reach the following state, then the next one, etc. until the last one.

# Let’s put it to work

A little example of using our iterator-based implementation:

``````my \$it = permutations_iterator(
items => [ qw< howdy you all > ],
filter => sub { join ' ', @_ },
);
while (my \$message = \$it->()) { say \$message }
``````

Now… let’s run it:

``````howdy you all
you howdy all
all howdy you
howdy all you
you all howdy
all you howdy
``````

It works! The good thing is that we can stop when we want, e.g. consider the following arrangement that will cease to look for new permutations as soon as we reach a lexicographically ordered one:

``````my @items = qw< howdy you all >;
my \$sorted = join ',', sort { \$a cmp \$b } @items;
my \$it = permutations_iterator(
items => \@items,
filter => sub {
my \$candidate = join ',', @_;
say \$candidate;
return 1 unless \$candidate eq \$sorted;
say "---> found, stopping here";
return 0;
}
);
1 while \$it->();
``````

When we run it, we can appreciate the fact that we don’t have to go through all permutations:

``````howdy,you,all
you,howdy,all
all,howdy,you
---> found, stopping here
``````

# The inevitable ending…

… is that this function ended up in cglib, inside Permutations.pm 😄

Stay safe!