[code-quality] Advice regarding caching values as a side-effect of visiting classes

[Anne Mulhern]

Hi!

I've implemented a very simple analysis of the kind described in this enhancement
request: https://bitbucket.org/logilab/pylint/issue/201/pointless-attribute-override-checker.
The analysis is neither sound nor complete, but has proven useful for removing of bits of
dead code.

It can be made more sound by altering the algorithm so that it accumulates data by visiting classes,
and subsequently processes that accumulated data, preferably after all classes have been visited,
in order to determine messages to report.

Is there some infrastructure in pylint that can accomodate an analysis that works that way?
If there is, is there a working example that uses it?

Thanks for any help,

- mulhern