Comments on: Introducing JsonGrammar

By: g

Mon, 11 Jun 2012 02:23:36 +0000

Hi, Quoridor doesn’t work on Lion.

Thx

By: Martijn

Martijn — Fri, 13 May 2011 10:37:12 +0000

Yes, this might be a problem. In this sense ‘inverse’ is a poor name. I wonder how hard Sjoerd’s suggestion would be to implement.

For JSON grammars it might actually be a feature: you can map two different JSON constructs to the same Haskell value, and on the way back consistenly map the Haskell value to the preferred of the two JSON constructs.

Btw, I fixed the missing <>s in your comment.

By: Sjoerd Visscher

Sjoerd Visscher — Fri, 13 May 2011 10:18:45 +0000

There’s a kind of underlying assumption that <> is used on disjoint pattern matches. It would be nice to have a way to check that this is always the case, and also that there are no cases missing when pattern matching.

By: Tillmann Rendel

Tillmann Rendel — Thu, 12 May 2011 23:50:36 +0000

I looked at your Semigroup instance, and I am not sure I like it. Consider the following two partial isomorphisms:

iso1, iso2 :: Iso Char ()
iso1 = unstack (unit . inverse (lit ‘x’))
iso2 = unstack (unit . inverse (lit ‘y’))

In isolation, these partial isomorphisms are fine. But (iso1 <> iso2) has a problem: It is not really inverse to (inverse (iso1 <> iso2)).

> convert (iso1 <> iso2) ‘y’
Just ()

> convert (inverse (iso1 <> iso2)) ()
Just ‘x’.

Oups. We successfully converted from Char to () and back to Char, but we arrived at a different character. I am not sure how serios that problem is in your setting.

From a theoretical point of view, I believe this problem is fundamental with partial isomorphisms: They do not really support choice. That is why in our paper, we used partial isomorphisms only for semantic actions, and built the additional layer of syntax descriptions on top of it.

By: Martijn

Martijn — Tue, 10 May 2011 21:10:11 +0000

I actually have already, Yitz! I hope his package gains some more popularity. Thank you for your comment!

By: Yitz

Yitz — Tue, 10 May 2011 13:20:47 +0000

Very nice ideas, Martijn!

> The <> is just another name for mappend from Data.Monoid

Can you make it be from Edward Kmett’s semigroups package instead? Otherwise, your module will be less convenient to use with any program that uses a datatype which is a semigroup but not a monoid. Now that I noticed the semigroups package, I am finding more and more of those…

By: Chris Done

Chris Done — Tue, 10 May 2011 06:56:29 +0000

solrize: Text.JSON.Generic generates based on Data and Typeable, but you don’t get any control over the mapping (which you really need most of the time).

By: Martijn

Martijn — Mon, 09 May 2011 19:19:30 +0000

@solrize Yup, it's called json.

By: solrize

solrize — Mon, 09 May 2011 16:33:39 +0000

I thought there was a json package from Galois that used SYB to generate the conversion functions. I actually remember using it but have forgotten what package it was or exactly what it did.

By: Martijn

Martijn — Mon, 09 May 2011 13:17:31 +0000

Good question, Chris. To be honest I haven’t given fclabels much thought (although I really should). Quick answer: I don’t think fclabels supports failure.