Comment 5 for bug 455696

On 2009-10-20 10:54 , John A Meinel wrote:
>> I'm seeing the following traceback from dirstate while driving bzrlib to
>> manipulate a tree. The application has a stochastic component, so this
>> condition is reproducible, but not every time.
>>
>
> Since this is based on a 'rename_one', my guess is that it is a bug that
> Robert Collins recently fixed. Which has to do with:
>
> add foo/b
> rename foo/b foo/a
>
> (renaming a directory to a name that sorts earlier, inbetween adding
> that directory and before you commit.)
>
> I don't have a bug offhand, but I recall it was fixed recently.

Unfortunately, I still see this with the latest bzr.dev. To reproduce:

1. pull my current project, MCREPOGEN: bzr pull lp:~nmb/+junk/mcrepogen
2. Make sure you have Numpy and Ian's bzr-fastimport plugin installed.
3. Do ./mcrepogen.py -r 1500 -o test.fi

This will generate 1500 random revisions and output them in fast-export
format to test.fi. I see the bug usually within the first 200 revisions.