[MERGE] Knit Index should ignore empty files

[Dmitry Vasiliev]

John Arbash Meinel wrote:
> The attached patch just changes the knit index code, so that if it sees
> an empty .kndx file, it just treats it as though it doesn't exist. Which
> lets bzr continue, rather than failing and making people feel like their
> data has been corrupted.
> 
> If we decide not to accept the whole patch, I would at least like to ask
> that we accept the changes to the error class, so that when this error
> occurs, it is possible to investigate, and figure out what file needs to
> be changed rather than giving a semi-useless error.

+1 for the whole patch

> Also, should KnitError be considered an 'InternalError' so it gives a
> traceback? It is sort of an environmental error, so I'm okay leaving it
> as a short-error. But it is serious enough that a traceback might be
> warranted.

I think KnitError should results to a traceback.

-- 
Dmitry Vasiliev (dima at hlabs.spb.ru)
     http://hlabs.spb.ru