[MERGE] Move RemoteRepository caching logic to CachedParentsProvider

[Andrew Bennetts]

Aaron Bentley wrote:
> Andrew Bennetts wrote:
> > Aaron Bentley wrote:
> > We can (and arguably should) change the failureException on
> > bzrlib.tests.TestCase to be a subclass of AssertionError.  But in practice I
> > don't think anyone really cares about the difference between FAIL and ERROR;
> > either way the test failed, and the distinction only matters as a hint for
> > debugging why.
> 
> Some test runners draw distinctions that make the difference painful.

Really?  They sound buggy.  It's difficult to run bzr's tests with anything
other than the builtin runner at the moment anyway.

> Do you want me to apply the patch I attached?

I'd prefer just a plain AssertionError to further bloating errors.py.  I
believe Martin has previously stated that we should try to avoid adding
errors that we never intend to catch.  Given that errors.py is starting to
take a surprising amount of time to import (even excluding modules it
imports in turn) I think that's a good policy.

I'd be ok with a compromise of raising a InternalBzrError if you don't want
to use AssertionError.

-Andrew.