Implementing fundeps in the new solver has proved more complicated than we thought at first. I’m not aware of any obvious “oh, this is going to perform really
badly” parts. My gut feel is that the inference engine will work better using type functions, but I can’t put any real evidence behind that claim. If you have a low-cost way to experiment we’d love to hear what you find. But don’t burn a month converting
in case it makes no difference!
Simon
From: glasgow-haskell-users-bounces@haskell.org [mailto:glasgow-haskell-users-bounces@haskell.org]
On Behalf Of Corey O'Connor
Sent: 22 October 2010 19:50
To: Simon Peyton-Jones
Cc: glasgow-haskell-users@haskell.org
Subject: Re: context-stack GHC 7.0.1 Release Candidate 1
I was running into a similar issue and haven't noticed a dramatic improvement with the latest changes. The number of ticks taken to compile are approximately the same before and after the latest patch. However the system still compiles
just fine with a context stack of 200. Which is OK by me.
The system I'm working with uses functional dependencies and type families w/ type equalities. Could I improve the performance of my system by replacing the functional dependencies with type families & type equalities?
The part of the system that uses functional dependencies is monad-param:
The other part of the system I can't release the source to yet. :-\ I know, not very useful.
-Corey O'Connor
coreyoconnor@gmail.com
http://www.coreyoconnor.com
On Tue, Oct 19, 2010 at 8:57 AM, Simon Peyton-Jones <simonpj@microsoft.com> wrote:
Christian
Dimitrios and I (mainly D) have fixed this. Your system compiles nicely now. Can you try again with the HEAD?
Simon