Add support for explicitly selecting from an index.

[petermattis]

Added support for "SELECT * FROM table@index". This is an extension
which allows us to more explicitly test our index scanning. The index
must exist within the table and the only columns accessible are those
the index is defined on.

See #2046.

[tbg]

comments for these? Especially whether they can both be true.

[tamird]

+1. this should probably be an enum

[tamird]

alternatively, maybe it's time to split this into separate types.

[petermattis]

Splitting into separate types is problematic due to the grammar changes. (I actually started out trying that approach and quickly gave up). Changing to an enum. It started as a single bool and then evolved.

[tamird]

Mostly LGTM. Some concern about the runtime assertions on QualifiedName

[petermattis]

@tamird The panics in QualifiedName should never happen. Seems reasonable to panic instead of returning an error up the stack.

[tamird]

can you add SELECT x FROM test.xyz@foo to exercise the error case? also nit from should be capitalized

[petermattis]

Done and done.

[tamird]

@tamird The panics in QualifiedName should never happen. Seems reasonable to panic instead of returning an error up the stack.

Yeah, that seems fine. My comment was intended to suggest that Normalize{Table,Column}Name could have return values of type {Table,Column}Name, which would be slightly cleaner.

[tamird]

LGTM

[tamird]

did you mean to add the select x case i described?

[petermattis]

Yes, though really I just misunderstood your comment and missed that error case. Good thing to add it because it found a bug.