Regular expressions with backreferences to a group including .* and \n are matching where they shouldn’t

[chris-morgan]

Describe the bug
Backreferences are supposed to match the text that the nominated group matched. But once .* and \n are involved in the group, the regular expression engine can match different text in the backreference, allowing the backreference to kind of expand the .* a second time.

To Reproduce
Detailed steps to reproduce the behavior:

1. Run vim --clean (or gvim --clean, etc.)
2. Insert the following:

   foo
   bar
   barnaby
   baz
   

3. Search for any duplicated lines: /\(^.*\n\)\1<Enter>
4. Observe that this matches bar\nbarnaby\n. (It’s like it searched for \(^.*\)\n\1.*\n instead.)

Expected behavior
There should be no matches: in the case that did match, \1 is bar\n, which is different from “barn\n”.

(\(^.*$\)\n\1\n does not exhibit this bug.)

Environment (please complete the following information):

• Vim 8.2.814
• OS: Arch Linux

[chrisbra]

can confirm and the 'regexpengine' setting does not seem to make a difference

[jlittlenz]

As well, (^.*\n)\1 does not match duplicated lines at the end of the file.

[mvduin]

It gets weirder, the pattern ^\(.*\n\)\1xyz matches neither

foo
foo
xyz

nor

foo
foobar
xyz

yet the pattern ^\(.*\n\)\1fyz matches both of these.

[mvduin]

It looks like the bug has nothing to do with .*, just with backreferences to groups ending in a newline: @chris-morgan's example still works if you search for ^\(bar\n\)\1 instead of ^\(.*\n\)\1. Similarly, replacing .* by foo in my previous comment does not change the results.