Is it even possible to make `tagTree` lazy? about tagsoup

[ndmitchell]

Yeah, the current API is probably not lazy by definition, and I think 2 is probably the right solution - but I'm not super sure. I've never used this API in anger, so it's really something best directed by real use cases. Having it not be lazy at all is probably a bad solution - it effectively caps the size of documents that can be processed, and it doesn't look like it should be that hard to make it lazy.

[ChristopherKing42]

Probably the cleanest solution to number 2 would look sort of like this:

data Trees = Branch {tag :: String, afterOpen :: Trees, afterClose :: Maybe Trees} | NoTrees

(that type is simplified of course). afterClose is Nothing if there is no closing tag. tag and afterOpen can be determined lazily. Evaluating afterClose to Nothing or Just _ would requires tracing the document until you reach the closing tag or the end of the enclosing tag or the end of the document.

We could implement option 3 in addition to option 1 right now without changing the API. It would be a function like:

tagTreeLazy :: Eq str => (str -> Bool) -> [Tag str] -> [TagTree str]

Where the str -> Bool is whether or not the tag should have a closing tag. The only thing left to consider for that solution is what happens when it's wrong.

[ndmitchell]

I'm not adverse to an API change if that's what we think is needed. I'm not sure I have enough a good model for an API. Perhaps the existing data type with:

tagTree :: Eq str => (str -> Bool) -> [Tag str] -> ([TagTree str], [str])

That gives you the pipe of TagTree values, plus the second component being the list of things that were expected to get shut but didn't. You would never emit a TagClose unless it was unmatched. Then the "sanity" condition would be that the first list doesn't contain TagClose and the second list is empty.