Regex of a non-greedy match different behavior

Question

I found that non-greedy regex match only become non-greedy when anchoring to the front, not to the end:

$ echo abcabcabc | perl -ne 'print $1 if /^(a.*c)/'
abcabcabc
# OK, greedy match

$ echo abcabcabc | perl -ne 'print $1 if /^(a.*?c)/'
abc
# YES! non-greedy match

Now look at this, when anchoring to the end:

$ echo abcabcabc | perl -ne 'print $1 if /(a.*c)$/'
abcabcabc
# OK, greedy match

$ echo abcabcabc | perl -ne 'print $1 if /(a.*?c)$/'
abcabcabc
# what, non-greedy become greedy?

why is that? how come it doesn't print abc as before?

(The problem was found in my Go code, but illustrated in Perl for simplicity).

Answer 1

$ echo abcabcabc | perl -ne 'print $1 if /(a.*?c)$/'
abcabcabc
# what, non-greedy become greedy?

Non-greedy means it'll match the fewest characters possible at the current location such that the entire pattern matches.

After matching a at position 0, bcabcab is the least .*? can match at position 1 while still satisfying the rest of the pattern.

"abcabcabc" = /a.*?c$/ in detail:

1. At pos 0, a matches 1 char (a).
   1. At pos 1, .*? matches 0 chars (empty string).
      1. At pos 1, c fails to match. Backtrack!
   2. At pos 1, .*? matches 1 char (b).
      1. At pos 2, c matches 1 char (c).
         1. At pos 3, $ fails to match. Backtrack!
   3. At pos 1, .*? matches 2 chars (bc).
      1. At pos 1, c fails to match. Backtrack!
   4. ...
   5. At pos 1, .*? matches 7 chars (bcabcab).
      1. At pos 8, c matches 1 char (c).
         1. At pos 9, $ matches 0 chars (empty string). Match successful!

"abcabcabc" = /a.*c$/ in detail (for contrast):

1. At pos 0, a matches 1 char (a).
   1. At pos 1, .* matches 8 chars (abcabcabc).
      1. At pos 9, c fails to match. Backtrack!
   2. At pos 1, .* matches 7 chars (abcabcab).
      1. At pos 8, c matches 1 char (c).
         1. At pos 9, $ matches 0 chars (empty string). Match successful!

Tip: Avoid patterns with two instances of a non-greediness modifier. Unless you are using them as an optimization, there's a good chance they can match something you don't want them to match. This is relevant here because patterns implicitly start with \G(?s:.*?)\K (unless cancelled out by a leading ^, \A or \G).

What you want is one of the following:

/a[^a]*c$/
/a[^c]*c$/
/a[^ac]*c$/

You could also use one of the following:

/a(?:(?!a).)c$/s
/a(?:(?!c).)c$/s
/a(?:(?!a|c).)c$/s

It would be inefficient and unreadable to use these latter three in this situation, but they will work with boundaries that are longer than one character.