On Wed, Aug 17, 2022 at 2:58 PM Bruce Momjian <br...@momjian.us> wrote:
> On Wed, Aug 17, 2022 at 12:41:03PM -0400, Jeff Janes wrote: > > This was recently added to CREATE INDEX reference page: > > > > + The optional <literal>USING</literal> clause specifies an index > > + type as described in <xref linkend="indexes-types"/>. If not > > + specified, a default index type will be used based on the > > + data types of the columns. > > > > But I think this is wrong, the default type is BTREE, it does not depend > on the > > data type. Or at least, I've never witnessed the claimed behavior. The > claim > > also conflicts with what is said at https://www.postgresql.org/docs/10/ > > indexes-types.html > > You are correct --- parser/gram.y has: > > access_method_clause: > USING name { $$ = $2; } > | /*EMPTY*/ { $$ = > DEFAULT_INDEX_TYPE; } > > and from include/catalog/index.h: > > #define DEFAULT_INDEX_TYPE "btree" > > Patch attached. > > > That looks good to me. But now looking over the linked pages more, it seems like https://www.postgresql.org/docs/current/indexes-types.html was never changed to admit the possibility of custom index access methods (like bloom) and neither was the 'replaceable class="parameter">method' section of https://www.postgresql.org/docs/current/sql-createindex.html. Also, is it odd that we say essentially the same thing for literal USING as we say for the replaceable /method/? Cheers, Jeff
