We discussed this back when I implemented the implicit call-stacks, and decided to be conservative in their use in `base`. There are two reasons behind the decision: 1. There's a runtime cost to using implicit call-stacks, namely an extra parameter to each function that uses them (just like regular implicit parameters). 2. They can clutter otherwise simple type signatures. Compare, e.g., head :: [a] -> a vs head :: (?callStack :: CallStack) => [a] -> a Adding the implicit call-stack more than doubles the length of the type signature! This might seem like a silly complaint, but I think there's a real cognitive overhead there, especially when you consider that Prelude will be one of the first modules that novices might read the haddocks for. GHC HEAD now uses implicit call-stacks for functions that always diverge (ie error and undefined), but we decided to draw the line there. Of course, there are plenty of other functions where a call-stack would make sense, especially partial functions. So I've collected some of the more common ones in a supplementary package: https://hackage.haskell.org/package/located-base I envision this as a package that you pull in for development, but perhaps leave out of "production" builds, using some CPP magic. And I'll happily accept PRs for other functions! Eric On Tue, Oct 20, 2015, at 10:29, aditya siram wrote:
Since the latest base now has a way of getting generating stack traces via the `CallStack` implicit param, is there any reason not to add that param to the error messages of unsafe functions in the Prelude ( and even commonly used modules like Data.Maybe/Data.List)?
The type signature will change but for most users it should be a drop-in replacement and a great aid to debugging.
Thanks! -deech _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe