Visualizing Debian package clusters

The set of Debian packages and their inter-package relationships is a very interesting graph. Unfortunately, it is too big to be visualized using naively graph visualization techniques: it would be about 25'000 package nodes with more than 400'000 inter-package relation edges. Nevertheless, that graph contains interesting hidden structures which grasp semantic inter-package relations. One of such relation is "strong dependency", i.e., the inability of a package to be installed without another companion package (note that such relationship is not the usual dependency relation, due to alternative dependencies and virtual packages).

In the context of the Mancoosi project, which has just finished is 1st year of existence, we have studied such relationship which has turned out to have interesting properties. On one hand it can help in establishing Quality Assurance metrics that pinpoint important packages on which a lot of other packages rely. On another hand it can be used to graph concisely inherent sub-structures of our package archive.

The talk will present the strong dependency notion and showcase fancy package graphs representing relevant cluster of packages which exist, sometime without us knowing that, in the Debian archive.