AoC 2016/11 - New successors


On with Advent of Code puzzle 11 from 2016: a new function for finding successors of a node.

The New representation forces us to update the successors_for function to read the new state layout and, more importantly, produce new states that adhere to the new convention.

While we’re at it, anyway, we’ll also chip in some enhancements. Hopefully.

Now buckle up, because it’s going to be a longish journey!

The function

The function implementing the new successors_for is the following. As we already introduced, it takes an input state and returns a list of other states that can be reached from the input one with one allowed move of the elevator.

 1 sub successors_for ($state) {
 2    my ($elevator, $generators, $microchips) =
 3      $state->@{qw<elevator generators microchips>};
 4    my $floor_start_mask = 0x01 << 8 * $elevator;
 5    my @retval;
 6    for my $ne ($elevator - 1, $elevator + 1) {
 7       next unless 0 <= $ne && $ne <= 3;
 9       # I can move (g), (m), (gg), (mm), (gm)*
10       # (gm)* means matching and only 1 move makes sense (prune others)
11       my $outer_mask = $floor_start_mask;
12       my $did_mixed  = 0;
13       for my $outer_element (1 .. $state->{n_elements}) {
14          my @masks_prefix = ();
15          for my $type (qw< generators microchips >)
16          {    # (g), (gg), (m), (mm)
17             if ($state->{$type} & $outer_mask) {
18                push @retval,
19                  new_candidate($state, $ne, @masks_prefix, $outer_mask)
20                  ;    # (x)
21                my $inner_mask = $outer_mask << 1;
22                for my $inner_element (
23                   $outer_element + 1 .. $state->{n_elements})
24                {
25                   if ($state->{$type} & $inner_mask) {
26                      push @retval,
27                        new_candidate($state, $ne, @masks_prefix,
28                         $outer_mask, $inner_mask);    # (xx)
29                   }
30                   $inner_mask <<= 1;
31                } ## end for my $inner_element (...)
32             } ## end if ($state->{$type} & ...)
33             push @masks_prefix, 0, 0;
34          } ## end for my $type (qw< generators microchips >)
35          if (  !$did_mixed
36             && ($generators & $outer_mask)
37             && ($microchips & $outer_mask))
38          {
39             $did_mixed = 1;
40             push @retval,
41               new_candidate($state, $ne, $outer_mask, 0, $outer_mask, 0);
42          } ## end if (!$did_mixed && ($generators...))
44          $outer_mask <<= 1;    # move to next position
45       } ## end for my $outer_element (...)
46    } ## end for my $ne ($elevator -...)
47    return @retval;
48 } ## end sub successors_for ($state)

Lines 2 and 3 extract the current relevant values from the input state to avoid having to do indirections on the input hash reference all the time.

Line 4 introduces a mask that marks the starting position for the current floor. As you might remember, floors are represented by individual octets in a 32-bits integer, with the fourth floor (indexed with 0) sitting at the least significant octet, and the first floor (indexed with 3) sitting at the most significant one. We keep this variable only because we will need to reuse this starting value in line 11, possibly 2 times. So it’s not much the efficiency gain that we get here, but more the readability (i.e. *let’s start from the beginning of the source floor).

An observation is due at this moment about what movements are possible and/or meaningful (this is by no means efficient but it’s at least complete and correct). All of the following possible moves are of course subject to additional checking for feasibility (i.e. not frying up any microchip in the process):

  • we might do a single-step move of each item in the source floor. This is what the comments refer to as (g) and (m);
  • we might move two same-typed elements, i.e. two generators ((gg)) or two microchips ((mm)).
  • we might move one pair of same-element items, i.e. one generator and one matching microchip ((gm)*). If the starting floor has more than one such pair, it only makes sense to consider one of them (all other alternatives would be isomorphic and a total waste of resources). This is why variable $did_mixed exists: it tracks whether we considered such a pair and skips any following ones for a possible elevator move.

These are the only three possibilities. We are explicitly cutting out the possibility to move two different-element items of different type, e.g. a plutonium generator together with a curium microchip, because it would always lead to frying the microchip.

Not convinced?

If the starting state is correct, and the microchip is still alive even in presence of the other-element generator in the same floor, it means that the microchip is protected by the corresponding generator in that starting floor. Hence, if we move the microchip away from a floor where it is protected by its corresponding generator, and we also make sure that we bring a generator of a different element together with it… we’re going to fry it. Hence, we know beforehand it’s a bad idea.

So, we’re left with the five alternatives described above.

The loop in line 13 (ending online 45, which is pretty long) goes through each slot in the floors. We leverage the count of number of elements here, but we might just as well use 8, possibly at the expense of some efficiency (but avoiding a bugCOUGH).

At each slot, we might either decide to focus on microchips or on generators (we’ll save the mixed case for later at line 35, don’t worry), so we iterate over generators and microchips (line 15) to consider them. At this point, anyway, it only makes sense to go on if the specific element type is present in the current floor, which is established by the $outer_mask. “Outer” here is in reference to the fact that we’re considering a outer loop (lines 15 to 34).

The first thing we do is to consider the single-movement alternantive, i.e. (g) or (m) (depending on $type). Function new_candidate here is supposed to return either a new neighbor of the current state, or an empty list if the candidate is not good; hence, the push in line 18 might actually not push anything in the @retval array used to collect all return values.

Now it’s time to consider the (gg) or (mm) pair (again, depending on $type), which is why we have to setup an inner loop (lines 22 through 31). To avoid duplications, we only iterate after the current slot, so we initialize the $inner_mask to the following slot with respect to $outer_mask (following means doing a left shift, line 21).

Then, again, for the pair to be valid we have to check that the slot in the inner loop is filled in (check at line 25) and in case repeat the call to new_candidate with the same contact we discussed above (i.e. it returns either a valid new neighbor, or the empty list).

A little word on the new_candidate function is due here; its calling interface is the following:

my @new_items = new_candidate(
    $starting_state,         # this has elevator, microchips, ...
    $target_elevator_index,  # where the elevator is heading to

The four *_mask variables are either false (in which case they are ignored) or contain a valid mask to detect which item to move for the specific type. At any given call, only one or two of these masks are different from a false value, accounting for one or two-item movements.

For this reason, the @masks_prefix is initially empty, making sure that $outer_mask in line 19 and $outer_mask/$inner_mask in line 28 refer to generators.

After the first loop in line 15 ends, it’s time for microchips, right? This is why at the very end of the first loop (line 33) we set @masks_prefix to (0, 0): in this way, the following round will put $outer_mask at line 19 and $outer_mask/$inner_mask at line 29 in the positional places for the microchip masks. Maybe it’s not that elegant… but it works.

Line 35 addresses the mixed case where we might move one generator and one microchip together. As we already saw, the must be of the same type (which is why lines 36 and 37 do checks against the same $outer_mask, i.e. the same slot) and we must not have done this attempt before for this elevator movement (so we check against $did_mixed, initially set to a false value in line 12 and then set to a true value in line 39).

Lines 40 and 41 are two old friends at this point… you will notice that we’re calling new_candidate again, this time with the mask for one generator and one microchip.

Note that we set $did_mixed to 1 in line 39 independently of whether the mixed move is allowed or not. Considering that we’re moving two same-element items, either a pair can be moved, or no pair can be moved.

Line 44, at last, sets the $outer_mask on the next item by shifting our *aiming bit” one position to the left.

After all these loops, array @retval collected all possible and admissible neighbors… so we’re only left to return them in line 47.

Whew, what a ride!

Crafing and checking a new candidate

The new_candidate function follows. As a matter of fact, it should have probably been called new_neighbor instead, because this funciton not only builds up a candidate, but it also tests it for being feasible. Anyway.

 1 sub new_candidate ($state, $ne, @masks) {
 2    my $target_shift = 8 * ($ne - $state->{elevator});    # shift: <<
 3    my %retval = (elevator => $ne, n_elements => $state->{n_elements});
 4    for my $type (qw< generators microchips >) {
 5       my $v = $state->{$type};
 6       for (1 .. 2) {
 7          my $mask = shift @masks or next;
 8          $v = ($v & ~$mask) | ($mask << $target_shift);
 9       }
10       $retval{$type} = $v;
11    } ## end for my $type (qw< generators microchips >)
13    # now check if the new candidate is viable
14    state $mf4 = 0xFF;
15    state $mf3 = $mf4 << 8;
16    state $mf2 = $mf3 << 8;
17    state $mf1 = $mf2 << 8;
18    my $generators       = $retval{generators};
19    my $naked_microchips = $retval{microchips} & ~$generators;
20    return
21      if ((($naked_microchips & $mf1) && ($generators & $mf1))
22       || (($naked_microchips & $mf2) && ($generators & $mf2))
23       || (($naked_microchips & $mf3) && ($generators & $mf3))
24       || (($naked_microchips & $mf4) && ($generators & $mf4)));
25    return \%retval;
26 } ## end sub new_candidate

As we already explained, we can get four masks in, but only two of them will be actually filled. This is anyway how we use this function, but this is in no way enforced here. Ah… the joy of very specialized functions!

The input masks allow us to pinpoint a bit in the starting floor, but we also need to know the right mask/bit for the landing floor. The shift will be by 8 bits (because each floor is an octet), while the direction will be given by the difference between the landing floor identifier and the source floor identifier. This should explain line 2; the note refers to the fact that the sign of $target_shift is such that it works properly when used with a left shift.

Did you know it? Doing a left bit shift by a negative amount actually yields a right shift by the corresponding positive amount! Amazing!

Varible %retval holds our candidate, and is initialized with the same n_elements as the source one (this is an invariant) and with the elevator in the target floor. This all happens at line 3.

The loop in lines 4 through 11 considers each input mask and does an action only for true ones (line 7, note the or next to skip false values). Line 8 is an obfuscated way to say that we set the bit in the starting floor to 0 ($v & ~$mask) and we set the corresponding bit on the landing floor to 1 (with | ($mask << $target_shift)). This effectively moves the item across the two floors.

When we arrive at line 12, our candidate %retval is built, but is it feasible? Or would it fry any microchip?

A correct, modular approach here would require us to encapsulate the check in its own function. Alas, my Perl-fu is a bit rusty, and I’m not sure that tail call optimization has been implemented at all, so we’ll spare the cost of calling another sub here and just put the check in the same function. At the end of the day… it’s the only place in the code where we need this check, so it’s not a big deal.

Well, maybe it’s a big aesthetic deal.

But we have a puzzle to solve here, not to do decorations!

The check in lines 14 to 24 makes sure that unfeasible states are pruned out, returning the empty list (because we’re calling new_candidate in list context) if applicable (line 20).

Lines 14 to 17 declare some handy masks to isolate each single floor. This is needed because the check for “lonely microchip in the same floor as a different-element generator) has to be done floor by floor. as you can see, these masks are “all bits high” (0xFF in line 14), properly shifted (from floor 4 down to floor 1).

We’re testing for generators in our candidate, so our $generators in line 18 is initialized to that value.

Also, we’re looking for naked microchips, i.e. microchips that have no corresponding generator in the same floor. This is calculated at line 19, by doing an & bitwse operation between the microchips themselves and the inverse of the generators. This is the right way to detect a naked microchip, because…

  • if the microchip bit is 0, the & operation will yield 0 (so, no naked microchip in that bit position);
  • otherwise, if the corresponding generator bit is 1, inverting it will yield 0 and the & operation will yield 0 as well. This is correct, because the microchip is protected by the generator and is not naked;
  • otherwise, we have a 1 in the microchip and a 0 in the generator, the output of the expression is 1 in that bit position… and it marks a naked microchip.

So… it seems that those years studying electronic engineering finally gave some result, yay!

Now that we located all naked microchips, it’s time to do the test floor by floor. In each floor, the frying condition is that we have naked microchips in that floor and (boolean and) we also have generators in that floor. Naked microchips in a floor without generators are fine!

So, in the first floor:

  • $naked_microchips & $mf1 tells us whether the floor has naked microchips or not, while
  • $generators & $mf1 tells us whether the floor has generators or not.

Doing a boolean and between these two conditions does the trick for this floor. Applying the same approach to the other floors (using their respective mask $mf2, $mf3, and $mf4) does the trick overall.

If we manage to get past the dreaded test in lines 21 to 24… congratulations, we have a feasible new state, and we can happily return it in line 24, yay!

This was an intense ride…

… and we’re stopping it here.

No, I’ll save running this for the next post, MBWAHAHAHAAHAH!


I’m not this bad.

If you’re curious, you can try this local version here and see it by yourself if it works or not.

Until then… stay safe!

Comments? Octodon, , GitHub, Reddit, or drop me a line!