Comment 3 for bug 172970

In the associated branch, I'm adding a function to tell the index that we are about to do something which will require most of the data (CombinedIndex.buffer_all()).

The associated branch is mostly a workaround, rather than fixing the root of the problem (we shouldn't be using get_revision_graph() here.)

Without buffering, it takes 36.521s to do the 'bzr pull'. With buffering it drops to 17.760s. Which is still way too high, but is a very, noticeable improvement. (Note it should be about 1s which is all startup time, because it really only has to set branch.last_revision_info(). Arguably it should check that the original revision is in the ancestry of the target revision, but that is it, and if we use --overwrite it shouldn't even have to check that.)