Virtual column proof of concept

[zachmu]

This needs a lot more tests, but this demonstrates the approach within can work well enough

[max-hoffman]

The main two things that stand out are 1) the indexing changes are kind of confusing, like the whole entrypoint function is centralized around virtual columns now, when the stuff that matters is now relegated to a helper function, and 2) i'm trying to remove the "resolved" qualifiers because plans are by default resolved now; the qualifiers are kind of redundant. Like I'm pretty sure sql.Node.Resolved() is mostly deprecated by now. For (1) I'd index upfront, in planbuilder, hide the special logic in the place it matters to avoid general indexing being even more confusing. Or just reorganize in a way that doesn't hide the hard part behind default expression boilerplate. And then (2) I'd prefer we use tablescan instead of resolvedTable as a conceptual progression of removing resolved stuff, but not really going to die on that hill.

[zachmu]

The main two things that stand out are 1) the indexing changes are kind of confusing, like the whole entrypoint function is centralized around virtual columns now, when the stuff that matters is now relegated to a helper function, and 2) i'm trying to remove the "resolved" qualifiers because plans are by default resolved now; the qualifiers are kind of redundant. Like I'm pretty sure sql.Node.Resolved() is mostly deprecated by now. For (1) I'd index upfront, in planbuilder, hide the special logic in the place it matters to avoid general indexing being even more confusing. Or just reorganize in a way that doesn't hide the hard part behind default expression boilerplate. And then (2) I'd prefer we use tablescan instead of resolvedTable as a conceptual progression of removing resolved stuff, but not really going to die on that hill.

I'll take a pass at 1). 2) seems out of scope provided I get rid of *ResolvedTable in the indexing logic.

[max-hoffman]

Thanks for fixing indexing. Are there any tests? Presumably need plan + correctness tests and run on the Dolt side, though I don't see any reason why Dolt side would't work.

[zachmu]

Thanks for fixing indexing. Are there any tests? Presumably need plan + correctness tests and run on the Dolt side, though I don't see any reason why Dolt side would't work.

Just the new engine tests I wrote. This code path should only impact tables with virtual columns, so no immediate impact to dolt