blacktip

the tin. Nobody uses it in practice in ordinary code. It's not something you have to incorporate when reasoning about code. Giving up reasoning about your Haskell code as if there were semantics-violating unsafePerformIOs lurking in the depths is similar to giving up reasoning about any code because a cosmic ray might flip a bit. Lets consider the rather remarkable backflips Standard ML and OCaml had to perform in order to make FP work with strictness and impurity. You can ignore unsafePerformIO because very very few people ever need it and you'll almost never have a reason to use it. Just don't use it. Similarly, writing total functions is the cultural default in Haskell. The compiler warns on incomplete pattern matches which you can -Werror as well. You can't define your own Typeable instances as of GHC 7.8 - only derive them, so that soundness issue is gone. Typeable itself is not popularly used, particularly as SYB has fallen out of favor. One critical aspect of why I like Haskell and the community around it is the willingness to fix mistakes. Meanwhile, ML implementations still use the value restriction. >the example once again calls into question the dogma Oh please. Haskell is an ensemble, multi-paradigm language with the right defaults (purity, types, immutability, expressions-only) and escape hatches for when you need them. Haskellers appreciate the value of modules a la ML, but typeclasses and polymorphism by default have proven more compelling in the majority of use-cases. There are still some situations where modules (fibration) would be preferred without resorting to explicit records of functions. For those cases, Backpack [3] is in the works. It won't likely satisfy serious ML users, but should hopefully clean up gaps Haskell users have identified on their own. Harper's case is over-stated, outdated, and not graced with an understanding of how Haskell code is written. Informal reasoning works if the foundations (purity, types, FP) are solid. [1]: https://github.com/bitemyapp/blacktip/blob/master/src/Databa... [2]: http://caml.inria.fr/pub/papers/garrigue-value_restriction-f... [3]: http://plv.mpi-sws.org/backpack/