nested and partial tree

[Robey Pointer]

On 6 Sep 2006, at 2:25, David Allouche wrote:

> Aaron Bentley wrote:
>> David Allouche wrote:
>>>> That's pretty scary. But I would like to advise anyone
>>>> from going forward with that "good idea", there are too many
>>>> internally-inconsistent Arch repositories out there.
>>
>> The idea of turning changesets directly into knits is just too pretty
>> for me to let go of, but I'm doing TDD and I won't merge anything  
>> that
>> doesn't produce the same results as baz-import.
>
> The issue I have is that I think that many baz imports did not even  
> have
> a well defined result, because of things like:
> * incomplete ancestry information
> * corrupt repositories that tla could be tricked into building by  
> adding
> a cachedrev at the right place
> * broken revlibs (that affects checkout when restarting an import  
> IIRC)
>
> Since baz-import was a one-off transitional feature (if only  
> because it
> was so slow to run), and since I did not really see a way of doing it
> much better, I eventually shut up. But multiplying the importance of
> those problems by providing a "transparent" plugin seems like a bad  
> idea.
>
> The problem is not in the plugin code, but in the fact that a lot of
> (most?) significant Arch branches are hopelessly internally
> inconsistent. I am sure you are the most competent person on Earth to
> implement such a plugin, but it's still going to be garbage-in- 
> garbage-out.

It's good to hear that I wasn't alone in having a lot of trouble  
doing the upgrade from tla to bazaar 2.  My upgrade involved lots of  
hand twiddling of files because tla had subtly corrupted itself in  
tiny places that it never noticed during normal operation.  Just  
another data point or "me too".

robey