Comment 15 for bug 740759

Andrew, I believe you are correct about who is calling cleanFailure (Deferred). I tried a simple hack:

    def cleanFailure(self):
        """Remove references to other objects, replacing them with strings.
        """
        return
        self.__dict__ = self.__getstate__()

With that code in place, I got:
***** WITH cleanFailure ******
0.037 hpss request: [139767171196672] BzrDirFormat.initialize_ex_1.16
0.516 end: [139767171196672] 0.482s

0.037 hpss request: [139767171196672] BzrDirFormat.initialize_ex_1.16
0.768 end: [139767171196672] 0.736s

0.036 hpss request: [139767171196672] BzrDirFormat.initialize_ex_1.16
0.523 end: [139767171196672] 0.490s

***** WITHOUT cleanFailure ******
0.035 hpss request: [140723835176704] BzrDirFormat.initialize_ex_1.16
0.424 end: [140723835176704] 0.393s

0.044 hpss request: [140723835176704] BzrDirFormat.initialize_ex_1.16
0.275 end: [140723835176704] 0.236s

0.040 hpss request: [140723835176704] BzrDirFormat.initialize_ex_1.16
0.321 end: [140723835176704] 0.285s

0.036 hpss request: [140723835176704] BzrDirFormat.initialize_ex_1.16
0.264 end: [140723835176704] 0.231s

Best times, 231ms vs 482ms. So __getstate__ is accounting for ~251ms in my testing. Which is pretty frickin huge. My immediate thought is that we should just avoid exceptions-as-return-codes when going through a Deferred.