Re: [pclasses-devel] an ideal deps tree

[Marc D. <bdu...@gm...>]

We could also move Prefs (and the old P::Config for that matter) out of Util 
into an App module. Once we have the file loading/saving stuff for Prefs we 
will need FileStreams anyway.

Marc

On Wednesday 29 Dec 2004 01:01, stephan beal wrote:
> Okay, this point is really bugging me, and i couldn't sleep...
>
> Please understand: i'm not complaining to anyone here, i'm just trying
> to express how important a solid dependencies model is on flexible
> software. At the end of this mail i detail the short list of things we
> need to do to reach a "well-dependencied" tree.
>
> i've been playing around on paper, and i've found a deps tree which:
>
> a) i really like.
>
> b) is extremely close to what we have now.
>
> Unfortunately, i can't copy my drawing into the mail. i will try to
> express it in "flat form" here.
>
> Assumption #0: Core depends on no other P Modules. Period.
>
> Assumption #1: all modules depend on the Core. This is a fair
> assumption, even if it isn't 100% technically true.
>
> Assumption #2: all modules except Core depend on Util. This is almost
> 100% true, but not quite.
>
> Assumption #3: all modules depend on themselves. Sounds silly, but it is
> significant.
>
> When i write:
> 	A ==> B
>
> That means module A has a direct dep on module B. A uses B. Thus:
>
> 	A <==> B
>
> is circular, and therefore Evil.
>
> Now, here are the deps as i BELIEVE them to be currently:
>
> (Reminder: Unicode and s11n's core are handled by Assumptions #1 and
> #2.)
>
> System ==> Unicode
> System ==> IO
> Net ==> IO
> Net <==> System (only timeout.h)
> SIO ==> (s11n and s11n::io)
> s11n::io ==> s11n (s11n
> Plugins ===> may freely dep on any P module
>
> That's all fairly good - i'm fairly happy with that :). Now comes the
> problem:
>
> Util ==> IO (only because of the Prefs' throw IOError)
> Util ==> System (only classes ThreadPool and WorkerQueue)
>
> That means that ALL other modules have a dep on IO and System. Nothing
> is standalone :(. Nothing!
>
> Actually, we're much closer than i thought! i thought Unicode had a dep
> on Util, and that IO had a dep on System, but this does not appear to
> be the case (after some grepping) :).
>
> So, here is what i propose we do:
>
> a) Change Util::Prefs to throw Error instead of IOError. IMO this is a
> much better solution for Prefs' API, anyway.
>
> b) Move the thread-related Util stuff into Thread (==> System) or into
> System.
>
> c) Is Net's dep on timeout.h really significant? This seems like an
> extremely low-level detail which we could overlook?
>
> If nobody objects, i will happily do (a) and (b).
>
> "i await your command..."
>
> (And now i'm really going to bed...)