Surprising, yes, but I guess I'd call it a pitfall of both * and a zero-width assertion. s/o*$/O/gm first matches "hello\n" and the $ assertion is satisfied, and since it's zero-width the matching position is left just before the \n. Then it matches "hello \n" because o* and $ are satisfied, and then the regex engine matches the same thing again but the rules in Repeated Patterns Matching a Zero length Substring kick in, advancing the position past the \n instead. Update: You can see this in action with: perl -MRegexp::Debugger -e '"hello\nfoo"=~s/o*$/O/gmr' Update 2: Note the same thing happens again with foo - it first replaces oo with O, and then adds the second O because it matched the $ again. You can see this with "o\no".

Simplification

I tried to simplify the case to avoid misunderstandings

  DB<32> p "hello" =~ s/o*$/O/gr;
hellOO
  DB<33> $_="hello"; s/o*$/O/g; print       # for older Perls
hellOO
  DB<34>
[download]

Surprise: the o is replaced twice.

Explanation so far

You and Hauke already explained that

• pos isn't changing after the first match b/c of the zero-width of $
• the empty o* is matching again °

(And I agree that the referenced perlre#Repeated-Patterns-Matching-a-Zero-length-Substring needs a rewrite)

  DB<41> $_="hello"; say pos,"($1)" while m/(o*$)/g;   # pos doesn't c
+hange
5(o)
5()

  DB<42> p "hello" =~ s/x*$/O/gr;                      # empty match (
+no x)
 helloO
[download]

Disappointments

Now, why is it surprising?

I think your case is that $ in combination with the /m modifier should act differently. Correct?

• Would this be consistent?
• Are there already examples of zero-width assertions who does it that way?
• Are there work-arounds to achieve what you want? (i.e. skipping zero-length matches)

Workarounds

Here a guess for the last question

  DB<44> p "hello\nfoo" =~ s/o*\n/O/gmr;
hellOfoo
  DB<45> p "hello\nfoo\n" =~ s/o*\n/O/gmr;    # added \n at the end of
+ input
hellOfO
  DB<46>
[download]

Meta

Question @all: Is the problem better understood now? :)

edit

added more code

update

added headlines for structuring

°) because empty patterns are always matching

compare

  DB<59> p "12345" =~ s/x*/ /gmr;
 1 2 3 4 5
  DB<60>
[download]

Are there work-arounds to achieve what you want?

I sometimes use (?:\n|\z) to be explicit that I want the line endings to be consumed by the engine.

Thanks! :)

But please note the second fOO

  DB<55> p "hello\nfoo" =~ s/o*(?:\n|\z)/O/gmr;
hellOfOO
  DB<56>
[download]

I'm busy right now, but I seem to remember that one could use features for atomic matches...

I'll try later...°

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

update

°) nah doesn't help, since it's not a backtracking problem.

  DB<48> p "hello\nfoo" =~ s/o*$/O/gmr;
hellOO
fOO
  DB<49> p "hello\nfoo" =~ s/o+$/O/gmr;
hellO
fO
  DB<50>
[download]

This is perl 5, version 30, subversion 3 (v5.30.3) built for x86_64-li
+nux-gnu-thread-multi
[download]