Re: BUG #13538: REGEX non-greedy is working incorrectly (and also greedy matches fail if non-greedy is present)

I wrote:
> As David says, these examples appear to be following what's stated in
> http://www.postgresql.org/docs/9.4/static/functions-matching.html#POSIX-MATCHING-RULES
> The Spencer regex engine we use has a notion of greediness or
> non-greediness of the entire regex, and further that that takes precedence
> for determining the overall match length over greediness of individual
> subexpressions. That behavior might be inconvenient for this particular
> use-case, but that doesn't make it a bug.

BTW, perhaps it would be worth adding an example to that section that
shows how to control this behavior. The trick is obvious once you've seen
it, but not so much otherwise: you add something to the start of the regex
that establishes the overall greediness you want, but can never actually
match any characters. "\0*" or "\0*?" will work fine in Postgres
use-cases since there can never be a NUL character in the data.

regression=# select regexp_matches('abc01234xyz', '(.*)(\d+)(.*)');
regexp_matches
-----------------
{abc0123,4,xyz}
(1 row)

regression=# select regexp_matches('abc01234xyz', '(.*?)(\d+)(.*)');
regexp_matches
----------------
{abc,0,""}
(1 row)

regression=# select regexp_matches('abc01234xyz', '\0*(.*?)(\d+)(.*)');
regexp_matches
-----------------
{abc,01234,xyz}
(1 row)

regards, tom lane