[thomasp85]

Fix #516

This PR moves the training of free positional scales in facets to core so that writers can read the required information directly rather than reimplement scale training logic

[teunbrand]

LGTM, but there are two things we should pay attention to; either move to new issues or address here.

1. Currently either all panels or no panels have free scales. There is no partial sharing of free scales that ggplot2::facet_grid(free) has (within column for x, within row for y). I think it'd be nice if users can choose all/none/some for this, but may have to overhaul settings a bit for sugary syntax.
2. It is currently valid to set MAPPING 'Adelie' AS panels but not valid to set SETTING panels => 'Adelie', whereas I think they should both be valid (likewise row and column aesthetics). In particular for PLACE clauses it'd be sweet to do panel-specific annotations in this way.

[teunbrand]

Also related, but can be deferred to separate PR:

According to the grammar, facet_vars is defined as 1 or multiple column identifiers.

ggsql/tree-sitter-ggsql/grammar.js

Lines 1013 to 1028 in 92bf584

	facet_clause: $ => seq(
	caseInsensitive('FACET'),
	$.facet_vars,
	optional(seq(
	alias(caseInsensitive('BY'), $.facet_by),
	$.facet_vars
	)),
	optional($.setting_clause) // Reuse from DRAW/SCALE
	),
	facet_by: $ => 'BY',
	facet_vars: $ => seq(
	$.identifier,
	repeat(seq(',', $.identifier))
	),

However, we are only handling one aesthetic per dimension. We should either grammatically accept a single identifier, or warn/error when multiple are given. For table stuff there isn't a reason why we couldn't handle multiple identifiers in the wrap situation.

[teunbrand]

I think we have is_facet_aesthetic() and is_user_facet_aesthetic() to capture these patterns

[thomasp85]

oh yeah... I guess that is in a later batch of knowledge I have to hydrate