Lets say the user wants to apt-get package "foo". Package "bar" is marked for install due to a long chain of dependencies for foo, e.g.
package: foo --> d1 --> d2 --> d3 --> d4 --> d5 --> bar
Today apt-get assumes that if d1 is missing, then foo will not work. This means the probability to break foo by not installing d1 is assumed to be 1.0. There is a similar relationship between all other packages and their next dependent package in this chain, i.e. the probability to break foo by not installing bar is assumed to be 1.0.
If we extend the Depends: field in the package header by a numeric dependency value for each listed package, describing the probability to break the requested package if the dependent package is not installed, then the dependency chain of the example would look like this:
package: foo --> d1 --> d2 --> d3 --> d4 --> d5 --> bar depvalue 1.0 1.0 1.0 1.0 1.0 1.0
But if each package would list a more realistic depvalue for it dependencies, then the chain might look like this
package: foo --> d1 --> d2 --> d3 --> d4 --> d5 --> bar depvalue 0.8 0.8 0.5 0.2 0.3 0.8
If you multiply the numbers, then the probability to break foo if bar is missing is calculated to be 0.015. Lets assume the user has set a threshold for installing dependent packages to 0.02, so bar is not installed. d[1-5] are installed, because they are above threshold.
Of course a missing package bar would break package (lets say) d4 with a probability of 0.24, which is much higher than the threshold. But please remember that the user has not requested d4 to be installed.
Regards
Harri
