Does `write` have the right type?

[hdgarrood]

Currently write is:

write :: forall r eff a. Writable r eff String -> a -> Eff eff Unit -> Eff eff Boolean

Should it be:

write :: forall r eff a. Writable r eff a -> a -> Eff eff Unit -> Eff eff Boolean

?

[hdgarrood]

Also what is the third argument for? Would be nice to explain this in the docs.

[hdgarrood]

Oh of course, it's a callback.

[garyb]

Your suggested change seems sensible to me, but then I'm surprised having a in the types here at all is safe, I'd have thought we'd need some class constraint or foreign type to represent things that streams can deal with.

[paf31]

I think the change is correct. The tricky thing is that Node stream types can change by calling things like setEncoding, so the current API is a bit of an approximation. Maybe indexed monads are the way to go, but I'm really not sure.

[hdgarrood]

The mutating nature of setEncoding is not very PureScript-y, so perhaps trying to provide a principled, typed interface to it in PureScript is the wrong approach.

As far as I can tell, write can only accept Strings or Buffers, but a given stream can accept either at any point; it's fine to write a String, then a Buffer, then another String. What about:

• Remove the last type parameter of Stream
• Have separate writeBuffer and writeString functions, where both work on any writable stream?

Reading is a little harder. I think we could deal with the fact that we don't know what type we might get back by inspecting the object in the FFI, and converting if necessary.

[hdgarrood]

My proposition is essentially: the fact that stream APIs can take or return either of Buffer or String seems to only be there for convenience, and I think we may be better off trying not to replicate that here.

[hdgarrood]

I tried to implement this but unfortunately there doesn't appear to be a way to ignore the encoding once one has been set. :( Indexed monads might work (I don't really know), but I think that's a bit fancy for a low-level API. I'd rather do one of the following:

• Throw an error inside onData if we get anything other than a Buffer, and remove setEncoding
• Change the callback type in onData to Either String Buffer -> Eff eff Unit

It seems that some Node.js libraries have opted for the first option in this situation, for example: senchalabs/connect#966

[hdgarrood]

Fixed by #4