Add more general primitive combinator.

[scott-fleischman]

I came across a need for a more general primitive combinator than tokenPrimEx when parsing more complicated structures than characters. In that case, the test function argument may return more information in the error case. The new tokenPrimEx' (any ideas for a better name?) takes a function that returns Either instead of Maybe to allow for any type (instead of Nothing) as error information. The errMsg function argument (formerly showToken) is also more general so it takes the token, position and error type and returns a ParseError.

It's easy to define tokenPrimEx in terms of tokenPrimEx' where the error type is ().

[mrkkrp]

@scott-fleischman, Can you give an example of such a “more complicated structure”? As I understand you're returning error info in Left so it can be used in error message. I like this idea. Could you open an issue describing this consideration on Megaparsec's page? If you don't have time or desire, I can do it as well.

[scott-fleischman]

An example is parsing XML Events. One could have an EventParser that is ParsecT [Event] () Identity, so that the token is the Event. A simple combinator for the EventBeginElement case would want to check the name and attributes (the attributes themselves could be checked with an AttributeParser). Checking the name and especially attributes could return more error information than Nothing, such as which attributes were expected but not found.

Feel free to add the idea/code to the Megaparsec project. I don't intend to use that package.

[mrkkrp]

@scott-fleischman, Thank you. Good luck with this pull request.

[scott-fleischman]

Closing old pull requests. I still think this approach is useful though.