ptset/ptmap are broken under 4.05 (beta)

[gasche]

opam-builder reports:
http://opam.ocamlpro.com/builder/html/ptset/ptset.1.0.0/2df016054683ba0af52a1ddae1f33d8a
http://opam.ocamlpro.com/builder/html/ptmap/ptmap.2.0.1/37f00ce2a1b7c7371cfab67b79c734aa

Error: The implementation ptset.ml
does not match the interface dist/build/lib-ptset/opt-d/ptset.cmi:

   The value `find_last_opt' is required but not provided
   The value `find_last' is required but not provided
   The value `find_first_opt' is required but not provided
   The value `find_first' is required but not provided
   The value `find_opt' is required but not provided
   The value `choose_opt' is required but not provided
   The value `max_elt_opt' is required but not provided
   The value `min_elt_opt' is required but not provided

File "ptmap.ml", line 1:
Error: The implementation ptmap.ml
does not match the interface dist/build/lib-ptmap/ptmap.cmi:

   The value `find_last_opt' is required but not provided
   The value `find_last' is required but not provided
   The value `find_first_opt' is required but not provided
   The value `find_first' is required but not provided
   The value `choose_opt' is required but not provided
   The value `max_binding_opt' is required but not provided
   The value `min_binding_opt' is required but not provided

I suppose that you can either stop using the standard Set.S interface (by duplicating it) to not break on new releases, or implement the new functions. They come from ocaml/ocaml#869 .

[UnixJunkie]

Duplicating anything violates a software engineering principle.

[UnixJunkie]

I will use cppo to handle ocaml versions >= 4.05.

[gasche]

I don't think you need to use a preprocessor: why not implement these functions for all OCaml versions? At worst, they will be hidden by a more restrictive signature on older versions.

(In Batteries we do need a preprocessor because some of the features we expose depend on new runtime features that cannot be re-implemented externally. This is not the case here.)

[UnixJunkie]

Good idea, I will try.

[UnixJunkie]

related to #2

[backtracking]

I prefer the idea of abandoning the compatibility with Set.S, in particular because functions such as
min_elt, max_elt, or elements make no sense with Patricia trees.
I dislike the idea of duplicating a fragment of Set.S, but I prefer this solution anyway.