Avoiding massive merges

[John Levon]

On Fri, Sep 07, 2007 at 03:41:28PM -0700, Stuart W. Marks wrote:

> >Actually not quite correct - we had no changes to upstream at all. As
> >the ASCII art showed, the 3.0.4 tree had branched off, and we needed to
> >move onwards to 3.1 - since the 3.1 tree doesn't have that 3.0.4 branch,
> >a merge was needed (and always will be).
> 
> Oh. I guess I completely misinterpreted the ASCII diagram to represent the 
> common "merge local changes with upstream changes" problem. If you're not 
> doing this, then I guess I have the same question as Francesco -- I don't 
> understand what needs to be merged here. But since others have apparently 
> solved your problem, it might not be worth your effort to continue trying 
> to explain it. Though I am curious....

As Thomas said it's not so much a merge as chopping off a head. Imagine
we have:

   0
   |
   1\
     \
      a
       \
        b
    
This is what the 3.0.4 repository looks like. Now we hg pull from the
3.1 repository:

   0
   |
   1\
   | \
   2  a
   |   \
   3    b
   |
   4

And we're at 'b'. We want to get to '4'. Now if we did hg update -C 4,
then we'd have the dirstate represent the contents of '4' (xen 3.1), but
the repository would still have two heads, 'b' and '4'. So we need a
"merge" like so:

   0
   |
   1\
   | \
   2  a
   |   \
   3    b
   |    |
   4    /
   \   /
    \ /
     5

where the "merge" contains 1,2,3,4 but none of the effects of a and b.

See?

regards
john