I meant that the expansion of 'hu%' is done before and outside of tsearch, so the question is how efficient will tsearch be for searching for hudreds or thousands of words in one expression.
Ok, I see. The answer - bad. Index structure is signature tree with constant signature length, by default 2016 bits. Siganture makes by hashing word and sets bits number HASHVAL % 2016 to 1. So, if query has many terms and all terms are ored then there is a lot of signatures that matched by query. This means a lot of pages in index will be readed.
No, one heap tuple - one index tuple.How hard (or sensible ;) would be creating such an index using GiST ? As proved by tsearch GiST can cope well with many-to-many indexes.
Sorry, I don't understand. Do you mean that GiST supports one heap tuple in several index tuple? If yes then no :). GiST doesn't support this feature. I don't think that GiST may help in this situation.
but tsearch seems to support this, and tsearch uses GiST. Is this functionality added entirely by tsearch ?
I'll try to explain index structure used by tsearch (three levels just for example): Root page internal tuple 1 -> second level page 1 internal tuple 1.1 -> internal tuple 1.2 -> internal tuple 2 -> second level page 2 internal tuple 2.1 -> internal tuple 2.2 -> third level (leaf) page 2.2 leaf tuple 2.2.1 -> heap tuple leaf tuple 2.2.2 -> heap tuple
leaf tuple contains one of two types of predicats: 1 just lexemes (without psition information) 2 if store size of first type is too big then tuple stores signature as described above.
internal tuple contains ored (super-imposed) signatures of childs.
-- Teodor Sigaev E-mail: [EMAIL PROTECTED]
---------------------------(end of broadcast)--------------------------- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])