David Tolnay

If the caller may want to trigger errors pointing to the brace, they need to be able to access the span of the brace token. ```rust return Err(Error::new(m.brace_token.span, "this error...

Will do!

Quote is the easiest way. https://github.com/dtolnay/quote ```rust let wrapped = quote! { mod my_mod { #stream } };

I believe the `syn::parse` doc covers this. [If `MyObject::parse` is a Parse impl:](https://docs.rs/syn/0.15.4/syn/parse/index.html#the-synparse-functions) ```rust let tokens = quote! { HelloWorld }; let output: MyObject = syn::parse2(tokens).unwrap(); ``` [Otherwise](https://docs.rs/syn/0.15.4/syn/parse/index.html#the-parser-trait), effectively the...

I am on board with a PartialOrd impl, but I don't think this implementation works. With a `None` delimited group, comparing the following two cursors would give a random answer...

I am not pleased that this implementation adds bookkeeping overhead to every cursor operation (which underlies all parsing) even in programs that never use Speculative. That seems like the wrong...

> cursors can be arbitrarily many `None`-delimited groups deep. In practice they are not. You should be able to efficiently compute the depth of a cursor without using `skip`, by...

As for tests: someday we should organize those but for now you can create a new top-level integration test file in the tests dir. Thanks!

I'd be on board with pursuing a flat representation. I think your assessment is correct that it has odds in favor of performance positive, and obviously it makes PartialOrd easy...

Oh for the flat representation work: make sure to fetch from master since I was poking around in TokenBuffer construction recently, since it had to be paged in for reviewing...