Certainty and severity levels

[Jordà Polo <jorda@ettin.org>]

I have been reviewing lintian checks to find a good way to classify
them. I still don't have a final proposal, but I would like to get some
feedback about the possible certainty and severity levels.

For severity I'm thinking of using a classification similar to that of
bugs. The idea is basically the same and developers are already familiar
with bug severities. Also, it isn't that easy to define severities
without using those terms, and mixing different classifications could be
really confusing.

One problem, though, is that lintian doesn't cover exactly the same
range of severities: "wishlist" seems a bit confusing for most checks,
and there is probably nothing above "serious". (Note that I have not
read every single lintian check yet, so I may be wrong.) Following this
convention, most of the current E|W|I tags would be classified like
this:

  error: serious, important, normal
  warning: normal, minor
  info: minor, wishlist (?)

For certainty I prefer a less fine-grained classification, at least for
now. The main goal is to identify checks that are fully "verifiable", so
3 levels are probably enough: high (100% sure the problem is present),
medium (mostly sure), and low (false positives likely). I'm not
satisfied at all with the names, but I'm having a hard time trying to
come up with something better. Suggestions are welcome.