On 14/10/2016 15:08, Anil Madhavapeddy wrote:
>> On 14 Oct 2016, at 15:05, David Scott <scott...@gmail.com> wrote:
>> On Fri, Oct 14, 2016 at 2:51 PM, Anil Madhavapeddy <a...@recoil.org> wrote:
>>> We have quite a few base libraries that use the pattern of
>>> val foo_exn : ... -> 'a
>>> val foo: ... -> 'a option
>>> Gobbles the exception and returns Some/None
>>> Should we take the Mirage3 opportunity to port libraries like Ipaddr to
>>> using the Result type instead, so it would be
>>> val foo : ... -> ('a, [`Msg of string]) result
>>> instead, using the Result type? That would let libraries use combinators
>>> such as Rresult, and not gobble errors from parsing silently. It would be
>>> an incompatible API bump so we would need to bump all consumers of, e.g.
>>> Ipaddr http://docs.mirage.io/ipaddr/Ipaddr_unix.V4.html simultaneously.
>> Personally I like the sound of this-- it's nice to use combinators to parse
>> a more complicated structure and be able to emit something more useful than
>> "None" as an error. Should we write a "recommended conventions" doc or wiki
>> page to include things like `foo_exn` and `foo` examples above?
> Yeah, once we agree on the conventions :-) Once we have everything using
> Result.t, we also need to find the right set of combinators.
> - There is Rresult for basic Result.t handling:
> - Lwt_result has a slightly different set of combinators
> So I guess we need to decide if we publish an Rresult_lwt.t which lifts up
> "('a,'b) result" into an Lwt.t with the same API as Rresult otherwise.
Based on earlier discussion from January 2015, I put some combinators in
mirage-types.lwt (maybe they should live elsewhere)
MirageOS-devel mailing list