[glsl-in] add support for .length()

[SpaceCat-Chan]

this fixes #2016

i see two reasons you may not want to merge this

1. this assumes that .length() is only valid on arrays and never has any other meaning
2. this assumes that .length() has the same semantics as Expression::ArrayLength, which it may not have

other than that there should be no problems, other than that this was done in a hasty way

[JCapucho]

this fixes #2016

thank you for working on this

i see two reasons you may not want to merge this

1. this assumes that .length() is only valid on arrays and never has any other meaning

According to the glsl specification this should be the case

2. this assumes that .length() has the same semantics as `Expression::ArrayLength`, which it may not have

There's a slight difference, Expression::ArrayLength follows wgsl semantics as almost all of naga's IR follows, in wgsl arrayLength returns a u32 which is glsl's uint, meanwhile .length() should return a int so you need to add a cast.

other than that there should be no problems, other than that this was done in a hasty way

I do have another issue with the current implementation, it specializes the parsing on .length() which is the only method on the base glsl specification, but I don't know if there any extensions that add other methods or might appear new extensions that do so, so I think we should parse a HirExprKind::Method and then handle this in lowering, this also has the added benefit that error messages can be better instead of no such field

[SpaceCat-Chan]

latest commit adds a cast to Sint, and makes the parsing parse any method and then only specializes on .length in lowering instead

[JCapucho]

Amazing work, I wasn't expecting you to include argument handling but nice that you added it.

There are some issues that need to be addressed and I also added some nits.

Also please add a test (you can add it to expressions.frag)

[JCapucho]

Amazing work, but I noticed while testing your changes that ArrayLength only works on pointer to dynamic arrays, while glsl expects .length to work on all arrays.

To fix this we would need to check the type of the expression and if it's an array of fixed size or a pointer to it pass it as a constant.

[JCapucho]

Almost perfect just left some nits and ask that you add a test for constant size arrays.

[JCapucho]

Prefer de referencing the match expression

Suggested change

	match array_type {
	&TypeInner::Array {
	match *array_type {
	TypeInner::Array {

[JCapucho]

Prefer using ellipsis ..

Suggested change

	base: _,
	size: crate::ArraySize::Constant(size),
	stride: _,
	size: crate::ArraySize::Constant(size),
	..

[JCapucho]

Use forced_conversion in this case because constant array sizes might already be Sint

Suggested change

	self.conversion(&mut array_length, meta, ScalarKind::Sint, 4)?;
	self.forced_conversion(&mut array_length, meta, ScalarKind::Sint, 4)?;

[JCapucho]

Sorry I might have misled you but since we are already lowering with a non Lhs context we don't need to check if the array is wrapped in a pointer.

[JCapucho]

I only meant the previous branch, in here we can be sure it will be an Uint and a cast will be neeeded

[JCapucho]

Perfect, thanks for your contribution.