Itagaki Takahiro <itagaki.takahiro@gmail.com> writes:
> contrib/pg_trgm in 9.1 becomes more attractive feature by index supports
> for LIKE operators, but only alphabet and numeric characters are indexed
> by default. But, we can modify KEEPONLYALNUM in the source code to
> keep all characters in n-gram words.
> However, the limitation and KEEPONLYALNUM are not documented in the page:
> http://developer.postgresql.org/pgdocs/postgres/pgtrgm.html
> An additonal documentation patches acceptable? The issues would be a FAQ for
> non-English users. I heard that pg_trgm will be one of the *killer features*
> of 9.1 in Japan, where N-gram based text search is preferred.
I'm not sure it's really a great idea to encourage people to use custom
builds with modified versions of that symbol. And those not using
custom builds will just be frustrated. If we think this is an important
feature then we ought to work out a better way to expose the
functionality.
(Personally I wonder how useful pg_trgm is at all in multibyte
encodings. Its idea of a trigram is 3 bytes, not 3 characters...)
regards, tom lane