Column resolution from aliases

[ajnavarro]

From this query:

select hash,name,language(name,content) as lang from blobs inner join tree_entries on blobs.hash = tree_entries.entry_hash where lang = 'Go' limit 1000

I have the following error:
ERROR 1105 (HY000): unknown error: column "lang" could not be found in any table in scope

[erizocosmico]

This is a bit more complex than I initially thought, so I'm gonna make a proposal before after thoroughly thinking about it.

The problem

We have two problems (the obvious one pointed out by the error and another one masked by it):

• Can't resolve the name of the column to the proper one of an alias in the projection.
• Can't actually access the column in the projection because projection nodes are always the topmost nodes.

So, if we did something like:

SELECT foo AS bar FROM baz WHERE bar = 1

The generated tree would be:

Project(foo AS bar)
 |- Filter(bar = 1)
     |- Table(baz)

In this case an optimizer rule could resolve that foo and bar are actually the same thing and move on, no need for any other change.

But if we did this:

SELECT MONTH(date) AS bar FROM baz WHERE bar = 1

The generated tree would be:

Project(MONTH(date) AS bar)
 |- Filter(bar = 1)
     |- Table(baz)

We cannot resolve bar in the Filter node to any field in the row because that part has not been processed yet.

The solution

So, for this bug to be effectively solved we might need to implement a few things:

• A rule that resolves simple field aliases that are already available.
• A rule that reorders the projection and moves the projection before the filter.

We could only do the second but not the first and not the second one.

### Caveats

Note that even if *plan.GroupBy is technically a projection we do not need to do that, because we don't support HAVING yet and the group by columns can't be used in the WHERE clause.