Improve support for SQL functions that return tables

[jrperritt]

This change adds an IsArray field for function parameters and creates structs for PosgreSQL functions that return tables.

There are a couple of examples in internal/endtoend/testdata/func_return_table/, in addition to the existing examples in func_return that didn't regress.

closes #1322

[jrperritt]

@kyleconroy Is this change out of scope for this project? If so, I can go ahead and close this to clear up the queue. If not, I'll try to find some time to rebase and resolve conflicts.

[andrewmbenton]

I don't think it's out of scope necessarily. When @kyleconroy and I last discussed I think we had settled on support for queries using the RETURNS TABLE(columns) syntax. He may correct me though.

I haven't had a chance to look closely through your PR to determine whether it meets that requirement though.

[jrperritt]

I don't think it's out of scope necessarily. When @kyleconroy and I last discussed I think we had settled on support for queries using the RETURNS TABLE(columns) syntax. He may correct me though.

I haven't had a chance to look closely through your PR to determine whether it meets that requirement though.

Great. Yes, that's exactly what it's supposed to do. You can see a couple of examples here.

[jrperritt]

The 2 force-pushes are for

1. Rebasing on main and,
2. Changing the enum names that were flagged during linting.

[kyleconroy]

A few pieces of feedback on these changes. Instead of generating structs inside models.go for these functions, I'd like to use per-query row structs. The Go codegen package already handles this for queries that don't cleanly map to an existing table.

This change simplifies the code as well, as we no longer need to pass function information to the codegen plugin.

const booksByTagsViaFunc = `-- name: BooksByTagsViaFunc :many
SELECT book_id, title, name, isbn, tags FROM sqlc_book_schema.books_by_tags_func($1::VARCHAR[])
`

type BooksByTagsViaFuncRow struct {
	BookID sql.NullInt32
	Title  sql.NullString
	Name   sql.NullString
	Isbn   sql.NullString
	Tags   sql.NullString
}

func (q *Queries) BooksByTagsViaFunc(ctx context.Context, dollar_1 []string) ([]BooksByTagsViaFuncRow, error) {
	...
}

As for the tests you've added, can you move all the changes for the examples directory to individual testcases into the end-to-end directory? The examples lean towards simplicity and most users of sqlc don't use stored procedures.

[jrperritt]

A few pieces of feedback on these changes. Instead of generating structs inside models.go for these functions, I'd like to use per-query row structs. The Go codegen package already handles this for queries that don't cleanly map to an existing table.

This change simplifies the code as well, as we no longer need to pass function information to the codegen plugin.

Indeed, that turned out to be a much better approach.

As for the tests you've added, can you move all the changes for the examples directory to individual testcases into the end-to-end directory? The examples lean towards simplicity and most users of sqlc don't use stored procedures.

Sure, I've added an internal/endtoend/testdata/func_return_table/ directory.

[jrperritt]

@kyleconroy This is ready for another look.

[jrperritt]

I rebased this on main and added support for multidimensional array function parameters, which was implemented some time in the last year without me realizing it.