|
|
Same global ideas (in particular the use of modules/functors), but:
- frequent use of Type Classes inside interfaces/implementation.
For instance, no more eq_refl/eq_sym/eq_trans, but Equivalence.
A class StrictOrder for lt in OrderedType. Extensive use of Proper
and rewrite.
- now that rewrite is mature, we write specifications of set operators
via iff instead of many separate requirements based on ->. For instance
add_spec : In y (add x s) <-> E.eq y x \/ In x s.
Old-style specs are available in the functor Facts.
- compare is now a pure function (t -> t -> comparison) instead of
returning a dependent type Compare.
- The "Raw" functors (the ones dealing with e.g. list with no
sortedness proofs yet, but morally sorted when operating on them)
are given proper interfaces and a generic functor allows to obtain
a regular set implementation out of a "raw" one.
The last two points allow to manipulate set objects that are completely free
of proof-parts if one wants to. Later proofs will rely on type-classes
instance search mechanism.
No need to emphasis the fact that this new version is severely incompatible
with the earlier one. I've no precise ideas yet on how allowing an easy
transition (functors ?). For the moment, these new Sets are placed alongside
the old ones, in directory MSets (M for Modular, to constrast with forthcoming
CSets, see below). A few files exist currently in version foo.v and foo2.v,
I'll try to merge them without breaking things. Old FSets will probably move
to a contrib later.
Still to be done:
- adapt FMap in the same way
- integrate misc stuff like multisets or the map function
- CSets, i.e. Sets based on Type Classes : Integration of code contributed by
S. Lescuyer is on the way.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@12384 85f007b7-540e-0410-9357-904b9bb8a0f7
|
letouzey