Re: [Haskell-cafe] A use case for *real* existential types

A value has an indefinite extent if it's lifetime is independent of any block of code or related program structure, think malloc/free or new/gc. A value has a dynamic extent if is lifetime is statically determined relative to the dynamic execution of the program (e.g. a stack variable): in this case the type system ensures that no references to the inotify descriptor can exist after the callback returns. Best, Leon On Fri, May 10, 2013 at 6:52 PM, Alexander Solla wrote:

On Fri, May 10, 2013 at 3:31 PM, Leon Smith wrote:

...

On Fri, May 10, 2013 at 5:49 PM, Alexander Solla wrote:

...

I'm not sure if it would work for your case, but have you considered using DataKinds instead of phantom types? At least, it seems like it would be cheap to try out.

http://www.haskell.org/ghc/docs/7.4.2/html/users_guide/kind-polymorphism-and...

I do like DataKinds a lot, and I did think about them a little bit with respect to this problem, but a solution isn't obvious to me, and perhaps more importantly I'd like to be able to support older versions of GHC, probably back to 7.0 at least.

The issue is that every call to init needs to return a slightly different type, and whether this is achieved via phantom types or datakinds, it seems to me some form of existential typing is required. As both Andres and MigMit pointed out, you can sort of achieve this by using a continuation-like construction and higher-ranked types (is there a name for this transform? I've seen it a number of times and it is pretty well known...), but this enforces a dynamic extent on the descriptor whereas the original interface I proposed allows an indefinite extent.

I know what extensions (of predicates and the like) are, but what exactly does "dynamic" and "indefinite" mean in this context?