Comment 14 for bug 495000

The associated branch has a one-line fix for this, but I have to figure out how we can get it tested properly.

I think the issue is that testing this requires some amount of concurrency. The existing restart tests monkeypatch functions to make sure that the files are removed at appropriate times.

Also, the 'gc_repo.texts.get_record_stream()' code *does* do the right thing, because it gets a reload function from

self.revision_index = AggregateIndex(self.reload_pack_names, flush)

The failure was that GCCHKPacker creates temp versioned_file instances that only reference the pack files we are going to be repacking. And when doing so, it wasn't setting the reload function.

Still trying to sort out a tasteful way for how to test it.