[merge][#288751] avoid creating text deltas spanning repository stacks

[Aaron Bentley]

John Arbash Meinel wrote:
> Aaron Bentley wrote:
>> Martin Pool wrote:
>>> On Sat, Nov 15, 2008 at 2:30 AM, Aaron Bentley <aaron at aaronbentley.com> wrote:
> 
> 
> ...
> 
> As I understand stacked branches, the idea is that we aren't pushing up
> a lot of ancestry anyway. So we'll probably get it wrong in a few
> places, and store more than is necessary, but not critically so.

It does seem like the problem scope is small.

>> No, I meant what I wrote.  You're doing: "Does this have a compression
>> parent?  Is it present?".  I'm doing: "Are all compression parents
>> present?"  For current formats, they are equivalent, but mine is more
>> amenable to multi-parent formats.
> 
> Actually, you brought up an interesting point, and both of you are
> slightly wrong. The issue is that we have a separate graph with
> "compression_parent", which is not identical to the left-hand file-graph
> ancestor in all cases.

I think that the repositories where the compression parent is not the
lefthand parent are wrong.  But yes, considering my code assumed we were
talking about compression parents and we weren't, it was wrong.

> Anyway, I was hoping to say "use compression_parent" and for
> multi-parent formats they will just have a list instead of a single
> reference. (Arguably we should upgrade it to a list from now, and just
> allow it to have a single value.)
> 
> It does seem that our get/insert record stream needs to be fixed anyway.

They don't seem to have compression parents, so I agree.

Aaron