Comments on: Data.MemoCombinators and You http://www.amateurtopologist.com/blog/2009/11/22/data-memocombinators-and-you/ Everything but topology. Tue, 13 Dec 2011 18:17:01 +0000 hourly 1 http://wordpress.org/?v=3.4-alpha-19719 By: uberVU - social comments http://www.amateurtopologist.com/blog/2009/11/22/data-memocombinators-and-you/comment-page-1/#comment-19 uberVU - social comments Mon, 23 Nov 2009 17:27:46 +0000 http://www.amateurtopologist.com/?p=446#comment-19 <strong>Social comments and analytics for this post...</strong> This post was mentioned on Twitter by ranha: Data.MemoCombinators http://bit.ly/7jm2oE... Social comments and analytics for this post…

This post was mentioned on Twitter by ranha: Data.MemoCombinators http://bit.ly/7jm2oE...

]]> By: Luke Palmer http://www.amateurtopologist.com/blog/2009/11/22/data-memocombinators-and-you/comment-page-1/#comment-18 Luke Palmer Mon, 23 Nov 2009 17:26:29 +0000 http://www.amateurtopologist.com/?p=446#comment-18 The problem with MemoTrie is the standard typeclass problem of one instance per type. So, for example, MemoTrie's Integer memoizer is the same as integral from MemoCombinators, which is not the most efficient implementation (it just wraps it to a list of bits). A better implementation would, say, be a 4-bit patricia tree using fast bitshifts. You could happily write this as a new MemoCombinator of type Memo Integer, but MemoTrie already has a memoizer for Integer so you have to do some newtype madness to convince the type system that it is a different type when really it's not. So not only is it hard to memoize types it doesn't give you (well, you declare new instances, but that risks orphans), it is hard to build new memo tables for types it *does* give you. Note that I have author's bias. :-) The problem with MemoTrie is the standard typeclass problem of one instance per type. So, for example, MemoTrie’s Integer memoizer is the same as integral from MemoCombinators, which is not the most efficient implementation (it just wraps it to a list of bits). A better implementation would, say, be a 4-bit patricia tree using fast bitshifts. You could happily write this as a new MemoCombinator of type Memo Integer, but MemoTrie already has a memoizer for Integer so you have to do some newtype madness to convince the type system that it is a different type when really it’s not.

So not only is it hard to memoize types it doesn’t give you (well, you declare new instances, but that risks orphans), it is hard to build new memo tables for types it *does* give you.

Note that I have author’s bias. :-)

]]>