Visibility of Result<T>.Backtrack about superpower

[nblumhardt]

👍 both angles are definitely valid. The current shape of these APIs has been stable for quite a long time now, so I'm leaning more towards favouring completeness and consistency in this case, and exposing all of the interface.

[ewoutkramer]

To be more precise, this is the problematic part:

 failure = new Result<TKind>()(remainder, augmentedMessage, attempt.Expectations, attempt.Backtrack);

It's not unlikely that custom tokenizer would need to augment the error message and position, just like the above statement.

[ewoutkramer]

..and this is what I needed in the end: ewoutkramer@9c2674f

[nblumhardt]

Thanks @ewoutkramer - those look like reasonable modifications, I'll take a look at this.

[AndrewSav]

I have a different use case that needs Backtrack.

I want to write another combinator, similar to Many. This combinator would accept a lower and an upper bound, similar to regex {n,m}. I want my combinator to conform Many style but I cannot, since Many has access to Backtrack and I don't. @ewoutkramer solution unfortunately will not help that use case.

[nblumhardt]

Looking at this again, I wonder if we should consider just making all the internal properties and constructors on Result<T> (and TokenListParserResult<T,U>) public 🤔

I originally steered away from this because I think they're pretty tightly bound to the internal implementation, but in retrospect, making experimentation and add-on development easier seems like a reasonable trade-off.

[AndrewSav]

@nblumhardt As a user I support this, because it would allow me to do what I want. I'm not sure if I would be comfortable with it as a designer, but fortunately, I'm not the designer, you are. ;) So I'll leave it up to you.

[ewoutkramer]

As a fellow API designer we all know that you'll be stuck with your public interface for a long time. So, I'd do it just "on demand".

[AndrewSav]

It probably does not make sense to open just a single property, if for nothing else, then for consistency reasons. They were designed as a part of an internal "interface" (I do not mean C# interface here, I mean it in more general sense) that we are considering making public here. If we are comfortable with this, it certainly makes sense. I cannot think of better options right now. I also agree, that it helps with experimentation and the longer term library enhancements.