cleanup AST builder functionality

[bungcip]

• remove unused function & impl
  • impl<'a> Make<Unsafety> for &'a str
  • impl<'a> Make<Constness> for &'a str
  • impl Make<TokenStream> for Vec<String>
  • impl Make<PathArguments> for ParenthesizedGenericArguments
  • fn parenthesized_args
• refactor mk().vis() usage so we can remove impl<'a> Make<Visibility> for &'a str
• refactor mk().set_mutbl() usage so we can remove impl<'a> Make<Mutability> for &'a str
• refactor mk().unary_expr() usage so we can remove impl<'a> Make<UnOp> for &'a str
• unify impl Make<TokenStream> for Vec<&str>, impl Make<TokenStream> for Vec<u64>, and impl Make<TokenStream> for Vec<Meta> functionality to use fn comma_separated

[kkysen]

If this can be hoisted to the top-level, that'd be preferable.

[bungcip]

maybe we can create a new module, lets call it transform, which contain all helper function that transform a rust AST to another rust AST. so we can put helper function like this not() and unparen() in there. However i prefer doing that in another PR rather than this

[kkysen]

With an iterator-based fn comma_separated, you can simplify this to:

impl Make<TokenStream> for Vec<&str> {
    fn make(self, _mk: &Builder) -> TokenStream {
        comma_separated(self.iter().map(|&s| Ident::new(s, Span::call_site())))
    }
}

impl Make<TokenStream> for Vec<u64> {
    fn make(self, _mk: &Builder) -> TokenStream {
        comma_separated(self.iter().map(|&s| Literal::u64_unsuffixed(s)))
    }
}

impl Make<TokenStream> for Vec<Meta> {
    fn make(self, _mk: &Builder) -> TokenStream {
        comma_separated(self.iter())
    }
}

[kkysen]

I'm not sure getting rid of all of these is helpful. "!" is pretty easy to read and understand (although things like "not" and the other non-standard spellings for things seem bad). It has the downside of not being as easily searchable since it's not tied to a symbol like UnOp::Not, but it's also easier to read and write in a bunch of ways.

[bungcip]

I think is more readable for all unary_expr() usage in transpile to just use mk().deref_expr(), mk().neg_expr(), and mk().not_expr(). I can create follow up PR for the refactor

[fw-immunant]

I believe the original motivation for this was to isolate the actual types of the syntax library from the code using the builder API as much as possible. This was nice when we changed from using the rustc syntax tree to syn. But I don't think we anticipate any such switch in the future, and I don't think we desperately need to keep this coupling loose anymore. I'm slightly in favor of the not_expr()/etc. refactor you suggest, but I don't think using UnOp enum values in core translator code is verboten at this point.

[ahomescu]

#1328 restores c2rust-refactor which also uses the AST builder. I need to check whether it's using any of the methods this PR is removing.

[bungcip]

@ahomescu, correct me if i am wrong, but i think c2rust-refactor dont use c2rust-ast-builder. it seems to use rustc_ast

[ahomescu]

@ahomescu, correct me if i am wrong, but i think c2rust-refactor dont use c2rust-ast-builder. it seems to use rustc_ast

Right, the new PR completely forks the AST builder used by the transpiler (based on syn) from the one used by c2rust-refactor.

[kkysen]

LGTM! Could you just fix up a couple of the newline separators that got messed up?