Purescript: Prohibit referring to a variable when a variable with the same name is bound in a preceding statement in ado-notation

[rightfold]

Currently, ado-notation allows the following:

ado x <- f
    y <- g x
    in x + y

In this example, _x_ is a free variable of the ado expression.

While not problematic in and of itself, this makes a future change to allow monadic bind in an ado expression a backwards incompatible change.

In GHC, applicative do allows binds, and will add a monad constraint. But it will only use bind instead of ap when necessary.

Allowing monadic binds would be desirable for Haxl-like computations.

We still want to keep the ado–in syntax, rather than transforming all do expressions, because of the strictness of ap.

[LiamGoodacre]

That sounds reasonable to me.

[paf31]

to allow monadic bind in an ado expression

Isn't that what do is for?

I would really like to keep do and ado separate.

[hdgarrood]

I think there’s a strong argument to be made in favour of this even if we are keeping do and ado separate*, which is that it’s confusing. For a type which has lawful Applicative and Monad instances, I feel that changing a do to an ado (or vice versa) should have no observable effect.

In a similar vein I’d quite like to improve the error when code such as the above results in an unknown name error, perhaps with a hint reminding you that applicative do notation cannot refer to earlier bindings, and that you may want to consider using regular do instead.

* I am also very much in favour of keeping do and ado separate.