On Wed, Feb 11, 2009 at 2:53 PM, Sebastian Sylvan
I think that would be difficult. You could probably store the continuation in a server-side cache if you aren't doing CGI but have a persistent server process, but eventually you'll need to discard unused continuations to avoid running out of memory. You may be able to use a WASH style continuation model in conjunction with this. So you store the session logs on disk, and if the continuation does not exist in memory you'd fetch the session log from disk, replay the whole session from that, and reproduce the continuation that way. That way most sessions would just work directly off of the cache and never touch disk, but if someone waits too long (or, say, bookmarks a page in the middle of the session!) there's still a fallback stored on disk.
I thought about something like that. In this regards, I narrowed the problem to a computation's internal binding to symbol names. With this respect, a point-free approach, like the one found in the factor language, would be simpler to handle as you'd have to serialize the state of stacks and those of any other monads used. Cristiano