fastimport KeyError bug

[Ian Clatworthy]

Guido Ostkamp wrote:
> On Fri, 9 May 2008, James Westby wrote:
>> On Fri, 2008-05-09 at 13:56 -0500, John Arbash Meinel wrote:
>>> That seems to be a typo in generic_processor.py, it should be
>>> "InventoryLink".
>>>
>>
>> I've just pushed a fix for this to
>>
>> http://bazaar.launchpad.net/~bzr/bzr-fastimport/fastimport.dev/
>>
>> Thanks for catching it,
> 
> James & Ian, I update fastimport plugin to get that fix and retried. I
> seem to have a great talent to run into bazaar traps, because it just
> bombed out on me again (see below) <sigh> ;-)
> 
> As always, any hints from your side are most welcome.
> 
> 10:21:48 Found 30000 commits already loaded - skipping over these ...
> 10:28:00 Generated the file-ids cache - 9599 entries
> 10:44:26 31000/41713 commits processed [0:07:51] (:61314)
> 11:00:59 32000/41713 commits processed [0:11:55] (:63289)
> 11:18:02 33000/41713 commits processed [0:14:52] (:65278)
> 11:35:19 34000/41713 commits processed [0:16:42] (:67201)
> 11:53:15 35000/41713 commits processed [0:17:33] (:69182)
> 12:11:22 36000/41713 commits processed [0:17:24] (:71181)
> 12:29:38 37000/41713 commits processed [0:16:17] (:73177)
> 12:48:04 38000/41713 commits processed [0:14:18] (:75149)
> bzr: ERROR: exceptions.KeyError:
> 'tsptdevconfigrtpcfgu-20080510082814-5ykt3wj1u9ojhqgc-909'

Guido,

This one will take a bit of debugging. There's a fair chance it's a
symptom of a known implementation bug to do with referencing a file
deleted in another branch but I'd like to check it's not something else.

I want to find the id of the revision in the input file causing the
breakage and ask you about its header fields (e.g. ref, from).
Firstly, let's save you the 2-3 hour wait each time to trigger the bug!
If you run fastimport with the checkpoint option set to 1000, it will
begin creating smaller packs of size 1000. When it next falls over, it
will have then checkpointed to 38000 instead of 30000. When it falls
over, start it again with a checkpoint of 10 and then again with a
checkpoint of 1. That will get us to the exact revision causing the break.

In the interest of noise reduction, ping me on IRC or email me directly
when you've done this please. In the meantime, I'll take a look at that
existing bug.

Ian C.