I've done some initial work porting WGSL's grammar to Bison so we can determine whether its ambiguous or not. Here's the work-in-progress, as of 3c633ea:

wgsl.y.zip

A few observations:

1. The grammar is ambiguous today because of type_decl argument_expression_list casts. See the discussion about this in Rename 'as' to bitcast; remove 'cast' #1047 and https://docs.google.com/document/d/1LN-L9UEtLPFKfou7jrpiSLH42xGp5ybvKaBHFTc4VKg/edit#heading=h.hp3f2zbslxr9
2. The grammar isn't LALR(1) today because of func_call_statement. For example, in foo(...) = 5; vs foo(...);, the differentiator (= vs ;) is many tokens after the point where we have to pick a path. I think this is fixable by rejiggering the grammar because foo(...) = 5; is illegal in WGSL (for now) anyway.
3. Bison doesn't support the * and + and ? operators that are in the spec's grammar, so I had to replace them with additional helper rules. (Which is fine.)

In https://docs.google.com/document/d/1LN-L9UEtLPFKfou7jrpiSLH42xGp5ybvKaBHFTc4VKg/edit#heading=h.hp3f2zbslxr9 we decided to try to rejigger the grammar to be LALR(1) and see if it's possible without too much trouble, so we'll investigate doing that.

I don't think there was a resolution about what to do about casts. I expect we'll discuss this in the coming meeting.