I ended up re-implementing most of the functionality of this library
while trying to implement a categorical model of a programming
language I was working on, so I went ahead and copied most of it
over here.
The new version is still missing some features, such as
linear functions, monadic bind, the Unc typeclass, and haddock.
It also makes a few different design decisions, which come
with their own trade-offs.
Instead, expect `morph a b` to evidence that `a` and `b` are objects,
and use `morph a a` equivalently to how `Obj morph a` was used before.
This results in some redundant parameters in categories in which every
type of a kind is an object ('nice categories', `NiceCat`),
but those can often be avoided with synonyms, and it's far better
than the constraint hell and `Dict`-passing I had to deal with before.
I had to move back to explicit arguments rather than associated types
as my representation of categories because a functor may be a functor
in multiple categories (e.g. ((,) a) is a functor in both the category
of linear and intuitionistic implication). This also allowed me to
get rid of some bad boilerplate involving `(:~:)` and some related issues.
I also finally managed to get Monads as monoids in the category of endofunctors again.
I don't recall what prevented me from doing it in the last iteration,
but I think it's probably working now. It could just be that I
understand the theory better and it's easier for me to get right now.
There's still some unimplemented stuff that I'd like,
pretty bad usability issues that I'd like to work around,
and a dire lack of documentation. I intend to work on all that soon.
The stack stuff has been removed because stackage doesn't have GHC 9 yet,
and the template-haskell stuff has been removed until I can rewrite it,
seeing as it was essentially useless as it was.
The LinearTypes support is very incomplete, but it's supported in some places
and I don't think I'd have much trouble expanding to support it in general;
I just haven't had much time to experiment with it yet.
It doesn't quite work yet, but it's a start. It'll need a rewrite to work, hence the commit.
Current known bugs:
* References to `r` do not take into account the skip pos.
* Generation of `forall r` even when r is not used.
* The `forall r` isn't placed where I want it to be.
* The code is horrifically bad.
* There are certainly other bugs I don't know about yet.
James T. Martin