[ENH][Modules] Delta Modules [was: Another version]
Les Tyrrell
tyrrell at iserve11.lis.uiuc.edu
Mon Oct 29 04:58:25 UTC 2001
----- Original Message -----
From: Mark van Gulik <ghoul6 at home.com>
To: <squeak-dev at lists.squeakfoundation.org>
Sent: Saturday, October 27, 2001 2:27 AM
Subject: Re: [ENH][Modules] Delta Modules [was: Another version]
> Here's an idea: I think the idea of being able to manipulate
> and analyze modules *without* loading them is sufficiently
> important that the initial release of the module system should
> *not* actual support module loading.
That's how I started out with Oasis. Initially, just a buffer in which I could load code and then do some quick and dirty analysis on it. It is quite a bit more than that now. ( http://sabine.canis.uiuc.edu:8080/Oasis ).
> Such a system will still
> be useful for *describing*, say, how to partition an image.
Yes- there are a number of things that I've built for Oasis before or am working on now to support just this sort of thing.
> Come to think of it, the actual loading should be a completely
> separate, er, module. That way you can use it for directly
> assembling production images from makefiles (without ever
> running a single bytecode of that image).
Well, technically there is some truth to that. Smalltalk/X does something like this, and I think that AOS also does this. But there is more to the story than drawing lines in the image- the code that is there has to change as well. The deepest non-modularities in Sqeuak are basically human programming practices which are at odds with a nice clean decomposition. Tools can't change that- but they can make it more apparent, and perhaps provide a bit more firepower for certain operations.
> You'll also have a
> nice *clean* structure (unfettered by the implementations of
> method dictionaries and symbols etc) with which to perform
> analysis. Actual loading of modules into an existing image
> should almost be an afterthought.
LIke Allen says, by the time the stuff has come back into the image, the hard work has been done.
> Sure, people will *want* it,
> but I think these other properties might be more important.
Initially, yes- I'm very hesitant to commit to a particular approach without having had the time to propose, explore, and throw away several of them. I've been at it off and on for 5 years, so I'm very wary of quick-fix solutions, and in particular any proposals that start with the word "just". ( or "simplest thing that could possibly work"... lets define "simplest", "thing", "could" and "work" why don't we? ).
- les
More information about the Squeak-dev
mailing list
|