Experience report

[tailhook]

Hello, I've just published another implementation of KDL (impl / PR). Generally I like the language, so wanted to make it best in class configuration language for Rust (and maybe good for non-config usage too). Major focus was for convenience and great error reporting.

Here I want to share some minor issues during the implementation, and also some proposals (in no particular order):

1. Some types are defined a bit differently in grammar and verbal specification. Would be nice to clarify. Namely:
   • In Number section it states "may be separted by _", but in grammar it's [0-9] [0-9_]* which is a bit different
   • In Identifier section, definition might actually match to grammar but I'm struggling to find out if it does. And also it would be nice to show some weird identifiers like - and --1.
2. Should there be any provisions to fix this issue: https://blog.rust-lang.org/2021/11/01/cve-2021-42574.html Should we do the same as Rust?
3. Was it intentional that node { other_node } doesn't work without semicolon (inside braces)? It doesn't look very straightforward since omission is allowed at the end of file.
4. Why there are no plain str and bool type names?
5. I think it should be safe to add i128 and u128 types?
6. My observation is that base64 isn't a type name in the same sense that all other type names that are in the list. Type name here is kinda "byte buffer" or similar. And "base64" is kinda encoding. It's the only such thing in builtin type name list. I don't think I'm proposing to add another category "encoding", but it bothers me a bit. Or am I misunderstanding KDL types?
7. There is a weird example in the spec:

   (published)date "1970-01-01"
   (contributor)person name="Foo McBar"
   

   Here "date" should be type name and "published" should be node name, no? The second line is very much the same (although, more controversial). Or is it only for me?