On Mon, Nov 29, 2004 at 02:23:11AM +0100, Goswin von Brederlow wrote:
> A bit OT but has anyone done an analysis of the Depends and
> Build-Depends graph? How many depends does a package have on average?
> Whats the longer Build-Depends chain? What is the biggest cycle? and
> so on ....
not a real analysis with hard numbers, but i have some facts:
- most packages are fine with no or very small cycles in the dependency graph
- there are a couple of real cycles but small, mostly in library packages
- and very few *very* large cycles that need to be broken up. this goes
up to cycles with 120 packages, not counting "suggests"
- some of these problems are probably because people tend to confuse "suggest"
and "recommends", so check you packages and remember that
"RECOMMENDS" IS STRONGER THAN "SUGGESTS"
- britney, the testing script currently can't handle cyclic dependencies very
well, i am trying to fix this [0]
- britney also makes "ands" out of "ors" in the dependency tree. this sounds
stupid but is really hard to avoid. this makes the dependency tree much
denser than it actually is.
cu robert
[0] http://www.semistable.com/files/britney-autohinter.diff
--
Robert Lemmen http://www.semistable.com
Attachment:
signature.asc
Description: Digital signature