view paste/paste.12027 @ 9994:d8734db325b9

<hppavilion[1]> le/rn Rogue One//Any regular who gives the slightest Rogue One spoiler shall be hunted down in real life and have their intestines removed through their eye sockets. Members would not be exempt if they existed, which they don\'t.
author HackBot
date Sat, 17 Dec 2016 23:40:13 +0000
parents 8f79561a34c2
children
line wrap: on
line source

2013-01-27.txt:09:43:46: <shachaf> monqy: imo the Await part is like (r,CoYoneda k r)
2013-01-27.txt:09:44:24: <shachaf> monqy: btw are you a CoYoneda expert
2013-01-27.txt:09:45:12: <shachaf> monqy: well you know how CoYoneda IORef = a read-only IORef
2013-01-27.txt:09:47:24: <shachaf> CoYoneda f a = forall r. CoYoneda (r -> a) (f r)
2013-01-27.txt:09:49:50: <shachaf> monqy: anyway when f is a functor CoYoneda f is isomorphic to f??
2013-01-27.txt:09:50:51: <shachaf> maybe THIS CoYoneda will be more to your liking!!!!!!!!!! http://ncatlab.org/nlab/show/co-Yoneda+lemma
2013-01-28.txt:05:37:44: <shachaf> kmc: By the way, you know ho CoYoneda IORef gives you a "read-only IORef" that can't do any other effects?
2013-02-07.txt:08:55:26: <shachaf> monqy: should i prefer yoneda or coyoneda
2013-02-07.txt:09:06:02: <monqy> shachaf: idk i havent studied them much at all?? my only real "experience" with them is that one time you asked me about how yoneda looks like partially applied >>= and coyoneda like =>>, or something like that
2013-02-07.txt:09:09:06: <shachaf> but coyoneda seems "more obvious to me"
2013-02-07.txt:09:09:33: <shachaf> also coyoneda can be meaningful for things that arne't functors
2013-02-07.txt:09:13:21: <shachaf> ion: do you know what coyoneda is
2013-02-07.txt:09:16:06: <shachaf> "CoYoneda Tree" keeps a tree, and a function to be mapped over it.
2013-02-07.txt:09:17:52: <shachaf> So CoYoneda Tree a = (Tree x, x -> a), for some x.
2013-04-02.txt:23:20:43: <shachaf> CoYoneda is easy.
2013-04-02.txt:23:21:41: <shachaf> CoYoneda is related to Functor.
2013-04-02.txt:23:35:10: <shachaf> Bike: As you probably guessed, Foo is CoYoneda.
2013-04-02.txt:23:37:57: <shachaf> You can treat (CoYoneda IORef) as a read-only IORef, sort of.
2013-04-02.txt:23:38:23: <shachaf> Bike: However: When f is a Functor, CoYoneda f is isomorphic to f
2013-04-02.txt:23:40:54: <shachaf> Bike: btw Yoneda is like CoYoneda except backwards
2013-04-02.txt:23:41:17: <shachaf> I think CoYoneda is more intuitive than Yoneda
2013-04-02.txt:23:57:19: <shachaf> oerjan: What do you think: Yoneda or CoYoneda?
2013-04-02.txt:23:59:13: <shachaf> oerjan: data CoYoneda f a = forall x. CoYoneda (f x) (x -> a)
2013-04-02.txt:23:59:38: <shachaf> when f is a functor it's isomorphic to Yoneda f and to CoYoneda f
2013-04-03.txt:00:00:01: <shachaf> btw Bike did you write lift and lower for CoYoneda
2013-04-03.txt:00:07:28: <shachaf> Where CoYoneda has the *arguments* to fmap, Yoneda has the *result* of fmap.
2013-04-03.txt:00:09:51: <shachaf> I guess you could say: fmap :: CoYoneda f a -> f a
2013-04-03.txt:09:08:43: <shachaf> ais523: Do you prefer one of Yoneda/CoYoneda?
2013-04-03.txt:09:09:23: <shachaf> data CoYoneda f a = forall x. CoYoneda (f x) (x -> a)