On Mon, Nov 23, 2020 at 09:45:55PM +0100, Francesco Ariis wrote:
Il 23 novembre 2020 alle 21:18 Sven Panne ha scritto:
Yes, you can think of (X, Y) as "Y with context X" or "a one-element container with Y in it", but is this really what comes to your mind first? I still highly doubt that. The Foldable-Traversable-in-Prelude change was largely a good thing, but very surprising (and not really needed) changes coming stealthily with it were only communicated when it was already too late... :-/
I recall a long thread, with everyone top posting. This feature could have helped OP:
https://gitlab.haskell.org/ghc/ghc/-/issues/11796
but unfortunately is still in design stage
Fwiw there is at least some strong user interest in this feature if anyone is thinking of working on it. I find myself checking on its progress pretty often, and would use it in our production codebase in a heartbeat. The strongest motivation is bugs that I've caught during a refactor, where a type somewhere changes from e.g. [x] to (y, [x]), and deep in some function a call to length/maximum/whatever just starts returning useless values without a type error. Consider, for example: do x <- getValuesForId "foo" when (length x > 1) $ fail "Unlikely condition: more than one value returned" ... It may be a long time before you discover the 'when' clause will never fire because the type of x has changed to a 2-tuple. In the meantime, you've been doing something wrong with your conflicting data. Tom