Robert Haas <robertmhaas@gmail.com> writes:
> Actually, though, varstr_levenshtein_less_equal() never gets called
> from parse_relation.c with a distance bound of more than four, so it
> can't actually go quadratic. There's another call in that file to
> varstr_levenshtein(), which in retrospect looks silly, but I'm pretty
> sure that could also be changed to use a distance bound of 4 (well,
> MAX_FUZZY_DISTNACE + 1) without changing the behavior at all. Given a
> constant distance bound, the algorithm is, I believe, only O(max(m,
> n)) not O(mn). Knowing that, we could let the internal code just
> bypass the length check and only enforce the length restriction when
> the code is called from SQL via fuzzystrmatch.
Done that way; thanks for the advice.
regards, tom lane