Sam Mason <sam@samason.me.uk> writes:
> On Tue, Aug 18, 2009 at 03:50:47PM -0600, Bob Gobeille wrote:
>>> CREATE INDEX "ufile_name_search" ON "public"."uploadtree" USING GIN
>>> ("ufile_name");
>>> ERROR: data type text has no default operator class for access method
>>> "gin"
> Not sure if understand very well myself, but GIN indexes can only speed
> up specific access patterns and these are exposed through various
> different operators.
What GIN indexes are good for is indexing equality queries on the
components of something the database otherwise thinks of as a single
object. For instance you can GIN-index searches for arrays containing
a particular value as a member.
Now type text doesn't have any built-in notion of a component, other
than individual characters, which aren't normally that interesting
to search for. What I suppose the OP has in mind is full-text
searching, which is looking for component *words*. But "word" is a
very language- and context-dependent concept. And defining which words
are to be considered equal for searching purposes is even more so.
If we'd hard-wired one notion of "word" into datatype text, it wouldn't
be very flexible. The point of the tsvector layer is to have a
configurable way to extract searchable words from a chunk of text.
There are also some implementation advantages like not having to repeat
that processing constantly during a search --- but the main point is
having a place to define what a word is and what search equality means.
regards, tom lane
