and_then doesn't re-wrap

[jariji]

julia> Try.and_then(Ok(1)) do x
           Ok(x + 1)
       end
Try.Ok: 2

This example requires the user to wrap the value in Ok manually. An alternative API would automatically re-wrap the value in Ok, making it easier to retain the disjoint sum-type Ok{T} | Err{E} style that I often want.

Undisciplined use of and_then allows creating T | Err{E}

julia> Try.and_then(Ok(1)) do x
           x + 1
       end
2

which might be better avoided.

Is this and_then worth the risk of misuse? There are other functions that could be provided, like these in Rust.

The only situation I can think of where this would be needed is in an exotic sum type like

True | False | Unknown

where there is more than one success branch. But I'd usually represent this concept with the isomorphic

Ok(true) | Ok(false) | Err(Unknown)

which doesn't require the re-wrapping. And even with the multi-success type, users could always do manual pattern matching themselves.

Am I wrong about this? If not, what do you think about either removing and_then or documenting a requirement that the function argument return a wrapper type?