[Moved to https://github.com/brannerchinese/notes/blob/master/CONTENT/Python/Zero-or-one_operator_within_group_in_perl_vs_python_regex.md. Any future edits will be done there. 20140215.]
One feature of Perl-style regular expressions that is not implemented in
Python is the use of the zero-or-one operator (metacharacter “question
mark”, ?
) with a parenthesis-delimited “group” or marked
sub-expression. In Python, if such a group is found (below, (ghi)?
),
it can be referred back to with a back-reference (below, \2
; code
below is Python 3.3 in Ipython 0.13.2):
1 2 | In [1]: re.sub('(abc)\t(ghi)', r'\2', 'abc\tghi')
Out[1]: 'ghi'
|
but if it is not found an error is raised instead of an empty string being returned (which is the norm):
1 2 3 | In [2]: re.sub('(abc)\t(ghi)?', r'\2', 'abc\t')
...
error: unmatched group
|
I’m aware of two work-arounds for this curious situation.
First, rewrite the optional sub-expression as an optional look-ahead
marked sub-expression: ((?:ghi)?)
. There are two pairs of parentheses:
one for the look-ahead syntax (?:...)
and one for the optional
matching group (...?)
:
1 2 3 4 5 | In [3]: re.sub('(abc)\t((?:ghi)?)', r'\2', 'abc\tghi')
Out[3]: 'ghi'
In [4]: re.sub('(abc)\t((?:ghi)?)', r'\2', 'abc\t')
Out[4]: ''
|
Second, discard the zero-or-one operator and put a set-union operator
(metacharacter “pipe”, |
) inside the parentheses so that the
sub-expression now means “either nothing or else the target in question”
(below, (|ghi)
):
1 2 3 4 5 | In [5]: re.sub('(abc)\t(|ghi)', r'\2', 'abc\tghi')
Out[5]: 'ghi'
In [6]: re.sub('(abc)\t(|ghi)', r'\2', 'abc\t')
Out[6]: ''
|
The first work-around is marginally faster than the second:
1 2 3 4 5 6 7 8 | $ python -m timeit -n 1000000 -s "from re import sub"
"sub('(abc)\t((?:ghi)?)', r'\2', 'abc\tghi')"
1000000 loops, best of 3: 6.48 usec per loop
$
$ python -m timeit -n 1000000 -s "from re import sub"
"sub('(abc)\t(|ghi)', r'\2', 'abc\tghi')"
1000000 loops, best of 3: 6.97 usec per loop
$
|
though it’s longer to type and more complex to understand and remember.
I vote for the second work-around.
I've just noticed that the replacement version of regex now at PyPi addresses this issue (“regex 2013-06-26”):
Unmatched group in replacement: An unmatched group is treated as an empty string in a replacement template.
[end]
Comments are enabled.