[ccan] Module dependency graph

[Rusty Russell]

On Mon, 17 Jan 2011 06:24:10 pm Joseph Adams wrote:
> http://tinyurl.com/64s9r9d

Hey, can I steal this link for my CCAN talk next week?

> The second picture is much easier to understand.
> 
> Looking at the graph begs the question: should CCAN be a collection of
> snippets that are easy to use in isolation?  Or should it be a
> framework, as the dependency graph suggests it currently is?

I hate frameworks :)

But if you write a good low-level module, it's almost inevitable that
we'll get dependencies.

(Note that, almost everything depends on ccan/tap, too, for testing, though
we consider that implicit.  More thing will depend on failtest may soon
be the same way.)

> Although
> the former would be convenient for users, it would lead to maintenance
> difficulties (imagine having to fix a bug in multiple copies of ilog).

I've been pondering this for a few days now.  I think we want the standalone
tarballs to have their dependencies nested into subdirs (like my proposed
nested modules, which I ran into issues implementing).

We'd still have one ilog module, but the standalone tarballs would have
copies in subdirs (with include paths fixed appropriately).

Thoughts?
Rusty.