Re: master ea93326: Add `union' and `intersection' to rx (bug#37849)

["Mattias Engdegård" <mattiase@acm.org>]

13 dec. 2019 kl. 18.13 skrev Stefan Monnier <monnier@iro.umontreal.ca>:

>    (rx-define-charset ident-chars "a-z0-9")
>    (rx-define-charset op-chars (union "+*/" ?-))
> 
> and then
> 
>     ... (any (union ident-chars op-chars)) ...
> 
> Is it worth the trouble?

Not if a special set of definition constructs are required, no.

> We could try and automatically determine inside `or` when regexp-opt can
> be used.  That doesn't sound like much fun, tho.

That is done now, to some extent. Some history, for those not keeping up:

(or STRING1 STRING2 ...) did use regexp-opt for some time, but it had the unfortunate effect that the match order wasn't preserved, which made the outcome unpredictable. For example, if the target string is "abc", then (or "a" "ab") would match "ab", but (or "a" "ab" digit) would match "a".

The code was then changed to only use regexp-opt when it doesn't affect observed behaviour, which is what we have today.

However, typically the longest match is desired, not the first. The user can order his strings carefully before putting them inside an 'or', but that (1) doesn't compose well, (2) is error-prone, (3) useless manual work that is done by regexp-opt anyway.

Something like (longest-or STRING...) has been proposed, but was not well received. A solution is still sought (preferably one that doesn't involve rewriting the regexp engine).