[HACKERS] B-Tree emulation for GIN

[Teodor Sigaev]

We'd like to improve usefulness of a new multicolumn feature of GIN index ( already in CVS HEAD ), which is currently a bit limited because of lack of GIN support for most data types. For example, it'd be very useful to combine native GIN-supported data types, like text search, hstore, arrays with other ones, for example, date, integers, etc. Right now, in the best case, we'll have BitmapAnd scan over *two* indexes - GIN and btree.

New interface (only changed functions is shown):
Datum *extractQuery(Datum query, int32 *nkeys,
        StrategyNumber n, bool *pmatch[],
        void ** extra_data[])
bool consistent(bool check[],
        StrategyNumber n, Datum query, bool *recheck,
        void *extra_data[])
int comparePartial(Datum partial_key, Datum key, void *extra_data)

extractQuery() might allocate an array (with nkeys elements) of any data,
GIN provides that array as an 5-th argument for consistent() and corresponding
element for comparePartial().



TODO/issues
- add support of other data types to btree_gin


--
Teodor Sigaev                                   E-mail: [EMAIL PROTECTED]
                                                   WWW: http://www.sigaev.ru/


--
Teodor Sigaev                                   E-mail: [EMAIL PROTECTED]
                                                   WWW: http://www.sigaev.ru/

We suggest to emulate btree index by GIN to make possible to use only one index scan. For that reason we developed btree_gin contrib module, as we do btree_gist for GIN index. Current implementation provides support for int4 and text data types, but it's straightforward and easy to add support for other data types. However, GIN interface should be extended for effective realization of such emulation. We've already changed GIN interface, but it was done during 8.4 development cycle, so the new changes can be done safely. Patch also includes changes in pg_trgm to cache the number of query's trigrams in extra_data instead of fcinfo->flinfo->fn_extra usage in consistent(), to improve situation when the same index can be used for several conditions, like ( foo.bar % 'book' and foo.bar 'look' ) - current bulk insert algorithm (used in ginbuild and fast_insert_gin patch) doesn't optimized for ordered or near ordered table, so time of btree_gin's index creation is much worse than that for btree.

gin_extra-0.1.gz
Description: Unix tar archive

btree_gin-0.3.tgz
Description: GNU Unix tar archive

-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers