named capture

[tdhock]

Hi, I saw that you recently added perl functionality to stringr. I just hacked a little change to str_match() that switches to regexpr(,perl=TRUE) instead of regexec() for perl regexps. That way we can take advantage of named capture regular expressions in stringr, and so I added 1 test to make sure the capture group names get copied to the colnames() of the resulting match matrix.

[hadley]

Do named captures only work with perl regular expressions?

[tdhock]

Yes, named captures only work with perl regexp because it uses a
special regexp named group syntax (?pattern) that is only valid
using the PCRE library. Those named groups give syntax errors under
regular regexps.

[hadley]

And regexec doesn't automatically name the results? Annoying. Maybe it would be a good idea to send that as a request to r-devel?

[tdhock]

Well, regexec doesn't support perl regexps, and perl is required for
named capture, so it is normal that regexec doesn't name the results.

The g?regexpr(,perl=TRUE) functions DO return the name specified in
the (?pattern), but the match+group start and end locations are
in a different format.

[rubenarslan]

Maybe isn't completely clear from this focus on named captures, but currently the reliance on regexec means "perl()" is broken for str_match (and maybe other functions, didn't test).
That's not just named captures but also lookahead etc. Had me stumped and maybe it isn't intended, because it's not documented? Tdhock's commit fixes this for me, i.e. regexp with lookahead works as expected now.

Perl regexps not supported by regexec
Error in regexec("(([a-zA-Z0-9_]+)(?=q[1-4])|([a-zA-Z0-9]+))(_q([1-4]))?y([0-9][0-9])?(m([0-1][0-9]))?", :
regcomp error: 'Invalid regexp'

[gagolews]

Confirming, this is a good fix - could you please merge that?

[hadley]

Closing since ICU doesn't appear to support this syntax. @gagolews can you please confirm?

[gagolews]

@hadley Unfortunately: YES. No capture group names support in ICU regex engine yet.