|| at Aug 27, 2012 at 7:38 pm
Fujii Masao writes:
After pg_trgm extracts the trigrams as GIN index keys, generate_trgm()
removes duplicate index keys, to avoid generating redundant index entries.
Also ginExtractEntries() which is the caller of pg_trgm does the same thing.
Why do we need to remove GIN index entries twice? I think that we can
get rid of the removal-of-duplicate code block from generate_trgm()
because it's useless. Comments?
I see eight different callers of generate_trgm(). It might be that
gin_extract_value_trgm() doesn't really need this behavior, but that
doesn't mean the other seven don't want it.
Also, seeing that generate_trgm() is able to use relatively cheap
trigram-specific comparison operators for this, it's not impossible
that getting rid of duplicates internal to it is a net savings even
for the gin_extract_value case, because it'd reduce the number of
much-more-heavyweight comparisons done by ginExtractEntries...
regards, tom lane