On Tue, 08 Jan 2013 17:05:50 +0100, Stefan Krah <stefan@bytereef.org> wrote:
> Eli Bendersky <eliben@gmail.com> wrote:
> > Everything works fine. However, if I add import_fresh_module for the C module:
> >
> > from test.support import import_fresh_module
> > import pickle
> > C = import_fresh_module('xml.etree.ElementTree', fresh=['_elementtree'])
> > P = import_fresh_module('xml.etree.ElementTree', blocked=['_elementtree'])
>
> sys.modules still contains the C version at this point, so:
>
> sys.modules['xml.etree.ElementTree'] = P
>
>
> > tb = P.TreeBuilder()
> > print(pickle.dumps(tb))
>
> > This interaction only seems to happen with pickle. What's going on here? Can we
> > somehow improve import_fresh_module to avoid this? Perhaps actually deleting
> > previously imported modules with some special keyword flag?
>
> pickle always looks up sys.modules['xml.etree.ElementTree']. Perhaps we
> could improve something, but this requirement is rather special; personally
> I'm okay with switching sys.modules explicitly in the tests, because that
> reminds me of what pickle does.

Handling this case is why having a context-manager form of
import_fresh_module was suggested earlier in this meta-thread.  At
least, I think that would solve it, I haven't tried it :)