Dave.

to make this clearer for beginners:
• in Perl a sub returns the result of the last statement¹
• to avoid unwanted results a last statement can be an "empty" return (recommended by PBP)
• an empty list () as last statement is faster because two opcodes for return are avoided
• a LABEL: without following code is even faster because a new statement is enforced without any opcode
• "_" is a legal identifier but has no magic meaning as label, IOW it's a convention
I wouldn't call the effects undocumented, just an idiomatic approach (and micro optimization)

FWIW using a label can have unwanted global side-effects, because a nested sub could have a goto _;

some experiments:

$ perl -MO=Concise,ret,empty,label -e'sub ret {return}; sub empty {()}
+; sub label {LABLE:};' 
main::ret:
4  <1> leavesub[1 ref] K/REFC,1 ->(end)
-     <@> lineseq KP ->4
1        <;> nextstate(main 2 -e:1) v ->2
3        <@> return K ->4
2           <0> pushmark s ->3
main::empty:
7  <1> leavesub[1 ref] K/REFC,1 ->(end)
-     <@> lineseq KP ->7
5        <;> nextstate(main 4 -e:1) v ->6
6        <0> stub P ->7
main::label:
9  <1> leavesub[1 ref] K/REFC,1 ->(end)
8     <;> nextstate(LABLE: main 6 -e:1) P ->9
-e syntax OK
[download]

Cheers Rolf
_{(addicted to the Perl Programming Language :)
see Wikisyntax for the Monastery}

UPDATE

¹) jdporter pointed out: you say "last statement", but you mean "last expression evaluated". Not the same thing!

well yes not every statement is an expression leading to a value, and it's the last in the current code flow which counts (think if/else), not in the sub.

That's what the docs say:

If no return is found and if the last statement is an expression, its value is returned. If the last statement is a loop control structure like a foreach or a while, the returned value is unspecified. The empty sub returns the empty list

I would have expected the empty return to have been optimized away during compilation two decades ago at least.

Jenda
1984 was supposed to be a warning,
not a manual!

Thanks, everyone. FWIW (I usually distrust multi-millions-per-sec benchmarks):

cmpthese 5e7, {
    1 => sub { my $x = 1 + 1; return },
    2 => sub { my $x = 1 + 1; () },
    3 => sub { my $x = 1 + 1; _: },
};

1 27367269/s   -- -13% -22%
2 31367629/s  15%   -- -11%
3 35161744/s  28%  12%   --
[download]

"Perl subs are expensive"? The cheapest adjustment is just above. Otoh, now I have a justification (not that I need any or anyone cares) not to follow PBP's "Place the label on the line preceding the loop keyword" -- just in case, easier and safer while debugging and commenting-out this or that, to keep label and code on the same line.

Sometimes, I prefer to invert the numbers to get the time saved to put things into perspective.

27367269 calls/s = 37 ns/call
31367629 calls/s = 32 ns/call
35161744 calls/s = 28 ns/call
[download]

So _: is 9 ns faster than return (assuming the numbers are meaningful).

After 100 million calls, you still wouldn't have saved a second by switching from return to _:.

Update: I had moved a decimal point in the wrong direction. Fixed.

> Otoh, now I have a justification (not that I need any or anyone cares) not to follow PBP's "Place the label on the line preceding the loop keyword" --

That's a weird logical construct

There was no indication or benchmark in this thread so far that placing a label at a previous line comes with a notable performance overhead.

Cheers Rolf
_{(addicted to the Perl Programming Language :)
see Wikisyntax for the Monastery}