Try fixing regex incompatibility

[goodspark]

Fixes #1045

Based on #1045 (comment)

[Gallaecio]

Why is DIGIT_GROUP_PATTERN replaced by pattern? Isn’t that a change in behavior?

[DJRHails]

Can confirm this has no change in behaviour (and should be merged). The aim of this code is to wrap digits with named groups. The desired behaviour is preserved as:
(\\d+[.,]?\\d*) days ago should be translated into (?P<n>\\d+[.,]?\\d*) days ago

The current breakage is due to some extra escaping being desired by re.sub.

Currently the code uses re.sub to do a replacement of \\d+ -> ?P<n>\\d+ (or equivalently in raw strings r'\d+' -> r'?P<n>\d+'. It uses the regex DIGIT_GROUP_PATTERN = r'\\d\+' (a literal \\d+ eventually)

For some reason, that still escapes me, re.sub now requires an additional set of escapes:

# Fix 1: sub with triple escape
DIGIT_GROUP_PATTERN.sub('?P<n>\\\\d+', pattern) # => (?P<n>\\d+[.,]?\\d*) days ago

# Fix 2: sub with raw double escape
DIGIT_GROUP_PATTERN.sub(r'?P<n>\\d+', pattern) # => (?P<n>\\d+[.,]?\\d*) days ago

# Fix 3 (as in this PR): ditch sub and do it using builtins
pattern.replace('\\d+', '?P<n>\\d+') # => (?P<n>\\d+[.,]?\\d*) days ago

It might also be wise to expand the pattern slightly to:

pattern.replace('(\\d+', '(?P<n>\\d+')

as if anyone adds a \d+ in the text instead of a \d* everything will break

[Gallaecio]

Makes sense.

However:

• DIGIT_GROUP_PATTERN does not seem used anymore, so we should remove it.
• pattern.replace('(\\d+', '(?P<n>\\d+') indeed sounds better. However, seeing how the old code did it without it, I am OK with not improving on this.

[DJRHails]

Made those changes #1095 here 👍