Rethinking the . AST

[c42f]

I was working a bit on macro expansion - particularly quote (quasiquote) expansion with $ interpolations - and I've found that it's weird and inconvenient that we parse a.b into (. a (quote b)).

Specifically, the part that's weird here is that we emit (quote b) for the field name even though this is "not quote syntax": this should not yield a syntax literal during lowering, and is thus a semantic mismatch with actual quote syntax of the form :(a + b) or quote a+b end.

Q & A:

• Why is this a problem? It means we need special rules to distinguish actual syntax literals from field names.
• But can we really change this? Surely this AST form had a purpose? Yes! A long time ago Julia supported a.(b) syntax to mean getfield(a, b), which would naturally have been parsed as (. a b). However this was deprecated as part of adding broadcast syntax in WIP: implement f.(args...) as a synonym for broadcast(f, args...) julia#15032

I propose we just parse a.b as (. a b) with the second argument implied to be a field name.

[c42f]

CC @BenChung - I imagine you might have encountered this AST inconsistency before :-)

[BenChung]

I wholeheartedly support the idea of parsing a.b as (. a b) - the old (. a (quote b)) syntax doesn't add anything in my opinion and needs to be special cased out downstream.

[c42f]

Thanks for the second opinion!! PR almost done :-D