Please check my thinking on bug 646979

[James Westby]

On Tue, 05 Oct 2010 09:16:36 -0500, John Arbash Meinel <john at arbash-meinel.com> wrote:
> I've been thinking about it, and I'm pretty confident that what you are
> trying to do is inherently "criss-cross". Specifically consider a
> semi-ideal case:
> 
>  upstream
> 
>   A	release 2.0
>   |
>   B	release 2.1
>   |
>   C	release 3.0
> 
> with debian doing:
> 
>   A
>   |\
>   B d   deb 2.0-1
>   | |
>   C |
>    \|
>     e   deb 3.0-1
> 
> And ubuntu doing:
> 
>   A
>   |\
>   | d	deb 2.0-1
>   | |
>   B u   ubuntu 2.0-1ubuntu1
>    \|
>     v   ubuntu 2.1-1ubuntu1
> 
> 
> These graphs are, inherently, criss-cross, because you don't have a
> single point of convergence. Both 'ubuntu' and 'debian' make new changes
> vs the upstream, and never does one fully supersede the other. ubuntu is
> constantly pulling from debian, but debian *never* pulls back from ubuntu.

This will be the case for the vast majority of branches for at least the
medium term I think.

> Let's draw it combined:
>  A
>  |\
>  | d
>  | |\
>  B | u
>  |\: |
>  | \ |
>  | :\|
>  C | v
>   \|
>    e
> 
> (and that is with the 'ideal' case that C is actually a descendant of B
> and not A, though if it isn't you can usually fake it:
>   A
>   |\
>   B C
>   |/
>   C'
> )

This "faking" is exactly what we are doing.

> Anyway, in the above graph, B and d are competing as common ancestors.
> If you were to strictly use d as the base, then you would get the
> changes from d => e, but that would include the A=>B changes. (d does
> not include B, so those changes would try to be 'replayed')
> 
> If you used B as the base, then you would be trying to merge in the 'd'
> changes a second time.
> 
> Now, what matters is what is explicitly changed by each revision.
> Because if "d - A" is strictly adding debian/, and not touching the
> actual content, then there are likely no conflicts in a per-file sense.
> 
> That is what '--weave' should be able to handle, since it does per-file
> merging.

Would '--weave' perform at least as well if "d - A" doesn't have that
restriction?

> However, if we can't introduce a new revision in the 'debian' line, then
> it is really hard to find a single common base between the import lines.

Thanks for looking at this. I realise my characterisation of "better" is
fuzzy, which doesn't help.