[code-quality] Whole program analysis

[Claudiu Popa]

On Mon, Dec 15, 2014 at 6:27 PM, Anne Mulhern <amulhern at redhat.com> wrote:
> Hi!
>
> Pretend I have classes:
>
> class A(object):
>     pass
>
> class B(object):
>     pass
>
> class C(A,B):
>     pass
>
> class D(object):
>     pass
>
> The MRO linearizations are:
>
> L(object) = [object]
> L(A) = [A, object]
> L(B) = [B, object]
> L(C) = [C, A, B, object]
> L(D) = [D, object]
>
> I'll say that L(X) > L(Y) if every
> c in L(Y) occurs in L(X). By that rule
> L(C) and L(D) are maximal linearizations.
>
> I want to write an analysis which would
> work best if I could gather up all the
> linearization of all classes in a possibly
> large package distributed over possibly
> many files and find only
> the maximal linearizations.
>
> This is really a kind of whole-program analysis.
>
> I don't believe that this fits the pylint model
> very well, since it is reasonable to expect that
> pylint would be run separately on each individual
> file in a library.
>
> Are there other analyses around that make use of
> astroid and are more whole-program like?
>
> Thanks!


Hi, Anne.

I'm currently working on adding mro support in astroid (need it to
solve a false positive in Pylint). After this, it will be easy to use
astroid for your task, by walking each file in your package, build an
astroid.module with AstroidBuilder.file_build, get all the classes
using Module.nodes_of_class and finally calling .mro for each one.
I'll push the change to astroid tonight.