[MERGE][0.15] Speed up 'bzr status' in xml trees

[John Arbash Meinel]

Aaron Bentley wrote:
> John Arbash Meinel wrote:
> 
>> True. Though I don't expect a huge speedup. I guess for 'commit' it
>> would be a bit faster, because we would only have 1 serialize call,
>> rather than 2.
> 
>> I guess I would rather use format 6 for 0.15, since WT3 is otherwise
>> unaltered from earlier formats.
> 
> I don't see any point in using format 6.  The only difference between it
> and 7 is that 7 supports tree-references.  5 on the other hand is a
> match for the knit and weave inventory formats, so you do save some time
> whevenever you need to update the basis
> 
> You realize that the cached-basis code is designed to be robust, right?
>    You can put garbage there, or anything you like, and it won't fall over.
> 
> Aaron

Oh, I understand you can put garbage there. But when you are trying to
benchmark the performance of 'bzr status' for the last 8 revisions of
bzr, it is nice when you don't have to do a full add+commit just to
regenerate the basis-inventory-cache. (ATM, I don't believe we have any
way of forcing an update to the file outside of bzrlib.)

Anyway, format 5 would be great for compatibility, I'm not sure how we
would jump to format 7 when we actually need it. Since WT4 actually does
support things like nested trees, we probably should stick with format 7.

John
=:->