Inconsistent results of Crystal::ASTNode::Call#id

[HertzDevil]

After #14490, I checked if any other #to_macro_id implementation is also problematic. This is the one for Call:

crystal/src/compiler/crystal/macros/methods.cr

Lines 2158 to 2164 in 562fb7b

	def to_macro_id
	if !obj && !block && args.empty?
	@name
	else
	to_s
	end
	end

It looks like the @name branch is for Call nodes that have no arguments at all, but this misses a few cases:

macro foo(x)
  {{ x.id }}
end

foo(bar(&a))   # => bar
foo(bar(a: 1)) # => bar

It also most certainly suffers from the same issue as #14489 (comment) regarding global Calls:

foo(::bar)       # => bar
foo(::bar(1))    # => ::bar(1)
foo(::bar(a: 1)) # => bar

[straight-shoota]

The implementation should probably just delegate to #to_s.

I'm even wondering why we even have a separate #to_macro_id. Shouldn't #to_s always be correct?

[HertzDevil]

#to_s would preserve e.g. the single or double quotation marks of a CharLiteral or StringLiteral, which we do not want.

[straight-shoota]

Yeah right. Besides these exceptions?

[HertzDevil]

But those exceptions are precisely why #to_s isn't "always" correct?

[straight-shoota]

I mean for every other type than CharLiteral and StringLiteral, #to_macro_id could be identical to #to_s.

[HertzDevil]

There is also SymbolLiteral#to_macro_id which drops the leading :. But I think the point is that the defined overloads avoid having to create a fresh Crystal::ToSVisitor every time ASTNode#id is called.