On Mon, 1 Sep 2003, Adam Heath wrote: > As many of you might know, I've been working on a dpkg rewrite. So far, I've > been able to limit myself to low-level helper functions. However, last week, > I started working on the mid-level support routines. This included the > dependency parser.
Well, APT can't support this without huge amounts of effort. There are many places where it relies on being able to quicky look at a single dependency element to see if it is making progress. There are a whole bunch of optimizations that rely on this, and the 3sat approximation is based on this as well. Further, the dependency chains use the most memory out of everything, and they have been seriously space optimized. Adding all sorts of additional dynamic information is going to have a nasty overhead both for space and execution time. The dependency calculations are a major performance consideration in APT. I don't see that you are solving any actual problems that are worth this kind of cost. Ie: > An astute observer will note that, strictly speaking, arch predicates > are superflous. One could just generate a unique dependency line for > the target arch when the deb is generated. So why should the baggage be carried into APT/DPKG? Jason
