Hacker News Reader

Comment by ColinWright | Hacker News Reader

An excellent article, well worth the read. I would make one comment:

    In Haskell, when you see a single equals sign it means that
    the left-hand and right-hand side are substitutable for each
    other both ways. For example, if you see:

        x = 5

    That means that wherever you see an x in your program, you
    can substitute it with a 5, and, vice versa, anywhere you
    see a 5 in your program you can substitute it with an x.

This is missing a key term: "in scope."

Apart from that, I'm going to re-read this several times. Thank you.

View on HN

> This is missing a key term: "in scope."

True, of course.

Sometimes I wish Haskell would have made shadowing variables into a compile error. Shadowing is the number one reason that substitution doesn't always work like you expect. (everyone please use -Wall)

However, substitution also doesn't work sometimes because the types aren't polymorphic enough. Let bindings are monomorphic by default nowadays which can give some surprising results.

I guess shadowing produces warnings instead of errors is because sometimes it can be a good thing: you can use it to create the illusion of mutability. In GHC's source I've seen a couple of longer functions (say ~50 lines) that re-use names in let/bind "statements" to get around going from x to x'''; the "x" simply kept shadowing its previous value.

Not just prettier, but it is more correct to shadow if your intent is to make the original x unavailable.

It would be cleaner/safer if the language had a keyword like 'shadowlet'.

I've also done stuff like

    x <- m1;
    f x;
    x <- m 2;
    f x;

in a do-block, and it's not clear to me how to write 'shadowfrom' (or however you pronounce '<-' in that context).

Maybe this would be cool:

    x! <- m 2

Sort of like a strictness annotation. In think it would be a compatible language change, since only identifiers can currently appear left of a <-

It is isn't just shadowing.

    f1 = let x = 1 in x
    f2 = let x = 2 in x

It would be insane to ban that in the name of "equational reasoning".

Reply on news.ycombinator.com