It isn't completely consistent.

#!/usr/bin/perl
use strict;
use warnings;

sub x { local$,=" "; print "$_[ 0 ]:", map( "[$_]", @_[ 1 .. $#_ ] ), 
+"\n" }

x "for",          sub { for( 5 .. 7 ) { 1 } }->();
x "while(5)",     sub { while( 5 ) { 1; last } }->();
x "while(0)",     sub { while( 0 ) { 1 } }->();
x "if(5)",        sub { if( 5 ) { 1 } }->();
x "if(0)",        sub { if( 0 ) { 1 } }->();
x "do{}while(0)", sub { do { 1 } while 0 }->();
x "do{}until(5)", sub { do { 0 } until 5 }->();

__END__
for: [] 
while(5): 
while(0): 
if(5): [1] 
if(0): [0] 
do{}while(0): [1] 
do{}until(5): [0]
[download]

I admit the value returned from the function for surprised me. It returns an honest-to-goodness false (ie. an empty string) -- not even undef. (I checked.) Interesting.

Also somewhat surprising, but not nearly as much, is that the functions with while return nothing at all.

Because I knew about the behaviour of if – when the condition is the last thing evaluated, it is returned, while when the body runs, its last expression's value is returned.

Which is exactly as expected by the model I described. Which is exactly what perlsub describes:

The return value of a subroutine is the value of the last expression evaluated by that sub, or the empty list in the case of an empty sub.

I guess Perl is simply inconsistent here.

Update: per Re: "last expression" quiz, I added do{} pseudo-loop tests. They return a code reference. It’s probably safe to say that Perl simply has no rule, and that what you get back from a sub depends on implementation details of perl.

Update: gack, I forgot to invoke the do subs using ->. Fixed, thanks to robin in Re^4: "last expression" quiz. No wonder I got back coderefs. The actual behaviour does not seem very consistent either, though.

Makeshifts last the longest.