homepage

In certain cases a zero-width /Z match that should be replaced isn't.
An example might help:
 re.compile('(?m)(?P<trailing_ws>[ \t]+\r*$)|(?P<no_final_newline>(?<=[^\n])\Z)').subn(lambda m:next('<'+k+'>' for k,v in m.groupdict().items() if v is not None), 'foobar ')
this gives
 ('foobar<trailing_ws>', 1)
I would have expected
('foobar<trailing_ws><no_final_newline>', 2)
Contrast this with the following behavior:
 [m.span() for m in re.compile('(?P<trailing_ws>[ \t]+\r*$)|(?P<no_final_newline>(?<=[^\n])\Z)', re.M).finditer('foobar ')]
gives
 
 [(6, 7), (7, 7)]
The matches are clearly not overlapping and the re module docs for sub say "Return the string obtained by replacing the leftmost non-overlapping occurrences of pattern in string by the replacement repl.", so I would have expected two replacements.
This seems to be what perl is doing:
 echo -n 'foobar ' | perl -pe 's/(?m)(?P<trailing_ws>[ \t]+\r*$)|(?P<no_final_newline>(?<=[^\n])\Z)/<$&>/g' 
gives
 foobar< ><>%

It's a bug caused by trying to avoid getting stuck when a zero-width match is found. Basically the fix is to advance one character after a zero-width match, but that doesn't always give the correct result.
There are a number of related issues like issue #1647489 ("zero-length match confuses re.finditer()").

@Serhiy can you take a look at this as I recall you've been doing some regex work?

This bug was fixed in Python 3.7, see issue32308.
Python 3.6.8 (tags/v3.6.8:3c6b436a57, Dec 23 2018, 23:31:17) [MSC v.1916 32 bit (Intel)] on win32
>>> re.compile('(?m)(?P<trailing_ws>[ \t]+\r*$)|(?P<no_final_newline>(?<=[^\n])\Z)').subn(lambda m:next('<'+k+'>' for k,v in m.groupdict().items() if v is not None), 'foobar ')
('foobar<trailing_ws>', 1)
Python 3.7.3rc1 (tags/v3.7.3rc1:69785b2127, Mar 12 2019, 22:37:55) [MSC v.1916 64 bit (AMD64)] on win32
>>> re.compile('(?m)(?P<trailing_ws>[ \t]+\r*$)|(?P<no_final_newline>(?<=[^\n])\Z)').subn(lambda m:next('<'+k+'>' for k,v in m.groupdict().items() if v is not None), 'foobar ')
('foobar<trailing_ws><no_final_newline>', 2)