Is this a bug?

The regexp debugger shows that in the /g case, the last match in the "implicit loop" of /g is a failed one (see also Repeated Patterns Matching a Zero length Substring), probably leading to that result in @-; note that $+ and $^N are also undef in this case. In addition to the below, try perl -MRegexp::Debugger -e '$_="a";s/(?=(a))/b/g'.

However, using those variables after a /g match feels a little sketchy to me, so this may be an XY problem, so if you could explain the bigger picture, perhaps we can help solve that.

$ perl -Mre=debug -e '$_="a";s/(?=(a))/b/g'
Compiling REx "(?=(a))"
Final program:
   1: IFMATCH[0] (11)
   3:   OPEN1 (5)
   5:     EXACT <a> (7)
   7:   CLOSE1 (9)
   9:   SUCCEED (0)
  10: TAIL (11)
  11: END (0)
minlen 0 
Matching REx "(?=(a))" against "a"
   0 <> <a>                  |   0| 1:IFMATCH[0](11)
   0 <> <a>                  |   1|  3:OPEN1(5)
   0 <> <a>                  |   1|  5:EXACT <a>(7)
   1 <a> <>                  |   1|  7:CLOSE1(9)
   1 <a> <>                  |   1|  9:SUCCEED(0)
                             |   1|  SUCCEED: subpattern success...
   0 <> <a>                  |   0| 11:END(0)
Match successful!
Matching REx "(?=(a))" against "a"
   0 <> <a>                  |   0| 1:IFMATCH[0](11)
   0 <> <a>                  |   1|  3:OPEN1(5)
   0 <> <a>                  |   1|  5:EXACT <a>(7)
   1 <a> <>                  |   1|  7:CLOSE1(9)
   1 <a> <>                  |   1|  9:SUCCEED(0)
                             |   1|  SUCCEED: subpattern success...
   0 <> <a>                  |   0| 11:END(0)
END: Match possible, but length=0 is smaller than requested=1, failing
+!
   1 <a> <>                  |   0| 1:IFMATCH[0](11)
   1 <a> <>                  |   1|  3:OPEN1(5)
   1 <a> <>                  |   1|  5:EXACT <a>(7)
                             |   1|  failed...
                             |   0| failed...
Match failed
Freeing REx: "(?=(a))"
[download]

> END: Match possible, but length=0 is smaller than requested=1, failing!

Dunno which length is meant, but I have a hunch, this is a mechanism to avoid endless loops restarting from the same pos.

edit

seems so: Repeated Patterns Matching a Zero-length Substring

> WARNING: Difficult material (and prose) ahead. ...

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

It was golfing exercise, where I tried everything possible presumably correct from language spec POV as I see it, be it sketchy or not. In the end I didn't need @-, so no problem at all, thank you, just curiosity. Perhaps I know less about regular expressions than I thought.

Consistent behaviour would be $#- and $#+ both the same and either 0 or 2 here. And all three $1, $+, $^N ending up the same here, also. It's not what I observe.

Can you explain debug output, why the

Matching REx "(?=(a))" against "a"

and then

SUCCEED: subpattern success...

are mentioned twice?

Same with perl -MRegexp::Debugger -e '$_="a";s/(?=(a))/b/g', I hit Enter too many times, why are steps 0..7 repeated twice?

$#- and $#+ both the same and either 0 or 2

please read "either -1 or 1"

$-[0] is the offset of the start of the last successful match. $-[n] is the offset of the start of the substring matched by n-th subpattern, or undef if the subpattern did not match.

edit

It's hard to tell which result you expect, but the way pos is changed after replacing a zero length assertion is tricky. It's meant to avoid endless loops.

use v5.12;
use warnings;
use Data::Dump 'dd';


$_ = 'aa';
s/(?=(a))(?{say "\$1='$1' \@-=(@-) pos=",pos})/b/g;
dd [$1];
dd \@-;
dd $_;

__END__
$1='a' @-=(0 0) pos=0
$1='a' @-=(0 0) pos=0
$1='a' @-=(1 1) pos=1
$1='a' @-=(1 1) pos=1
["a"]
[1]
"baba"
[download]

update

this may shed some light

use v5.12;
use warnings;
use Data::Dump 'dd';


$_ = 'ab';
s/(?=([ab]))(?{say "\$1='$1' \@-=(@-) pos=",pos})/X/g;
dd [$1];
dd \@-;
dd $_;
__END__
$1='a' @-=(0 0) pos=0
$1='a' @-=(0 0) pos=0
$1='b' @-=(1 1) pos=1
$1='b' @-=(1 1) pos=1
["b"]
[1, 1]
"XaXb"
[download]