On 14.09.2011 1:36, Jonathan M Davis wrote:
On Tuesday, September 13, 2011 14:04 Dmitry Olshansky wrote:
Speaking of source there are few artifacts that should end up in std.uni
and not in std.regex:
CodepointSet and CodepointTrie, and unicode property tables.
I'd rather have them all accessible to user, but their interface is
admittedly clunky, I'm open to ideas on better API.

I suspect that there's value in reviewing the unicode stuff separately, and if
it affects std.regex' API at all (as opposed to being simply an implementation
detail), I'd suggest that we review and sort out the unicode stuff before we
review then new std.regex, since that would then have definite impact on
std.regex, and it affects a lot more than just std.regex.

It doesn't affect API at all, but an important implementation detail.
Particularly one can't get reasonable performance w/o unicode incorporated as an implementation detail yet I feel there should be a more general interface to it. Otherwise everyone willing to tackle unicode would have to, for instance, duplicate all of unicode property tables, and that's some 100Kb we are looking at.


If the unicode stuff is entirely an implementation detail of std.regex and is
not in the API, then we can review the unicode stuff separately and make
std.regex use the stuff in std.uni once it's in there, but if you're doing
major unicode stuff, I think that that's significant enough to merit its own
review.

It could be done in two steps, if I was to choose which way to go I'd first folded in std.regex then moved all its hidden unicode stuff to std.uni. It's mostly about character classification and about doing it fast.


--
Dmitry Olshansky

Reply via email to