Convert CREATE INDEX SQL into pgroll operation

[ryanslade]

Convert the majority of CREATE INDEX statements into OpCreateIndex

The following cases are covered:

CREATE INDEX idx_name ON foo (bar)
CREATE INDEX idx_name ON foo (bar ASC)
CREATE INDEX idx_name ON foo USING btree (bar)
CREATE INDEX idx_name ON foo USING brin (bar)
CREATE INDEX idx_name ON foo USING gin (bar)
CREATE INDEX idx_name ON foo USING gist (bar)
CREATE INDEX idx_name ON foo USING hash (bar)
CREATE INDEX idx_name ON foo USING spgist (bar)
CREATE INDEX CONCURRENTLY idx_name ON foo (bar)
CREATE INDEX idx_name ON schema.foo (bar)
CREATE INDEX idx_name ON foo (bar, baz)
CREATE UNIQUE INDEX idx_name ON foo (bar)
CREATE INDEX idx_name ON foo (bar) WHERE (foo > 0)
CREATE INDEX idx_name ON foo (bar) WHERE foo > 0
CREATE INDEX idx_name ON foo (bar) WITH (fillfactor = 70)
CREATE INDEX idx_name ON foo (bar) WITH (deduplicate_items = true)
CREATE INDEX idx_name ON foo (bar) WITH (buffering = ON)
CREATE INDEX idx_name ON foo (bar) WITH (buffering = OFF)
CREATE INDEX idx_name ON foo (bar) WITH (buffering = AUTO)
CREATE INDEX idx_name ON foo (bar) WITH (fastupdate = true)
CREATE INDEX idx_name ON foo (bar) WITH (pages_per_range = 100)
CREATE INDEX idx_name ON foo (bar) WITH (autosummarize = true)
CREATE INDEX idx_name ON foo (bar) WITH (fillfactor = 70, deduplicate_items = true)

And the following unsupported cases fall back to RAW SQL:

CREATE INDEX idx_name ON foo (bar) TABLESPACE baz
CREATE INDEX idx_name ON foo (bar COLLATE en_US)
CREATE INDEX idx_name ON foo (bar DESC)
CREATE INDEX idx_name ON foo (bar NULLS FIRST)
CREATE INDEX idx_name ON foo (bar NULLS LAST)
CREATE INDEX idx_name ON foo (bar) INCLUDE (baz)
CREATE INDEX idx_name ON foo (bar opclass (test = test))
CREATE INDEX idx_name ON foo (bar opclass)
CREATE INDEX idx_name ON ONLY foo (bar)
CREATE INDEX idx_name ON foo(a) NULLS NOT DISTINCT
CREATE INDEX IF NOT EXISTS idx_name ON foo(a)
CREATE INDEX idx_name ON foo(LOWER(a))
CREATE INDEX idx_name ON foo(a, LOWER(b))

[andrew-farries]

I think this is good to go now @ryanslade. Changes I made:

• Use raw SQL for indexes created with ONLY
• Use raw SQL for NULLS NOT DISTINCT indexes
• Use raw SQL for IF NOT EXISTS index creation
• Use raw SQL for indexes on expressions
• I made a change to our fork of pg_query_go (Add support for deparsing AST reloptions nodes pg_query_go#6) to allow deparsing of storage_options clauses. This allows us to remove the manual deparsing of those nodes from this PR.
• Removed the restriction on having > 1 storage parameter
• Added comments
• Updated the PR description

I also made #565 to simplify the use of the OpCreateIndex struct and remove the needless pointer fields.

This PR is now stacked on:

• Stop using pointers for OpCreateIndex fields #565
• Update pg query go dependency #564

[ryanslade]

Nice! The two PRs lower in the stack look good so we can merge those into main and then prepare this for merge too 👍