----- Цитат от Oleg Bartunov ([email protected]), на 28.01.2012 в 21:04 -----
> I suggest you work on more general approach, see > http://www.sai.msu.su/~megera/wiki/2009-08-12 for example. > > btw, I don't like you changed ts_rank_cd arguments. Hello Oleg, Thanks for the feedback. Is it OK to begin with adding an exta argument and check in calc_rank_cd? I could change the function names in order not to overload ts_rank_cd arguments. My proposition is : at sql level: ts_rank_lcd([weights], tsvector, tsquery, limit, [method]) at C level: ts_ranklcd_wttlf ts_ranklcd_wttl ts_ranklcd_ttlf ts_ranklcd_ttl Adding the functions could be done as an extension but they are just trampolines into calc_rank_cd(). I agree that what you describe in the wiki page is more general approach. So this : SELECT ts_rank_lcd(to_tsvector('a b c'), to_tsquery('a&c'),2 )>0; could be replaced with SELECT to_tsvector('a b c') @@ to_tsquery('(a ?2 c)|(c ?2 a) '); but if we need to look for 3 or more nearby terms without order the tsquery with '?' operator will became quite complicated. For example SELECT tsvec @@ '(a ? b ? c) | (a ? c ? b) | (b ? a ? c) | (b ? c ? a) | (c ? a ? b) | (c ? b ? a)'::tsquery; is the same as SELECT ts_rank_lcd(tsvec, 'a&b&c'::tsquery,2)>0; So this is the reason to think that the general approach does not exclude the the usefulness of the approach that I am proposing. Best regards -- Luben Karavelov
