url lookups should be able to use the same heuristic switch does

[Robert Collins]

bzr init left && bzr commit left -m foo --unchanged
bzr branch left right
bzr commit right --unchanged -m bar
bzr checkout --lightweight right working
cd working
bzr switch left # finds ../left
bzr merge right # finds '.' and errors ('nothing to do')

 affects bzr
--
GPG key available at: <http://www.robertcollins.net/keys.txt>.

[Aaron Bentley (abentley)]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Robert Collins wrote:
> Public bug reported:
>
> bzr init left && bzr commit left -m foo --unchanged
> bzr branch left right
> bzr commit right --unchanged -m bar
> bzr checkout --lightweight right working
> cd working
> bzr switch left # finds ../left
> bzr merge right # finds '.' and errors ('nothing to do')

So if you have a branch named "down" in a subdirectory, "bzr merge down"
wouldn't work. Would we do "bzr merge ./down" for that?

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFIZPpD0F+nu1YWqI0RAopIAJ0eV6lahgdd3D8BzfRoVSsEkq0z3QCfQuRs
hprH6ImAysXl94k4cgKie94=
=tsZM
-----END PGP SIGNATURE-----

[John A Meinel (jameinel)]

Well, 'bzr switch' is a bit different than "merge" in that it doesn't allow you to give the path to a subdirectory of a branch.
So it first tries the path as a relative path, and if it doesn't find anything, it falls back to relative to the current "other" branch.

So, for example:

 bzr switch foo

Is './foo' a branch, if so, switch to that
elif cur_target + '../foo' is a branch, switch to that
else fail

'bzr merge' is going to have a harder time, because it has to say:

if './foo' is a branch, use that
elif './foo' is an existing file inside a containing tree, use the containing tree
elif cur_target + '../foo' is a branch, switch to that
elif cur_target + '../foo' is a file inside of a containing tree # we could probably ignore this one
else fail

However, this seems fairly easy to extract into a helper function. The next problem is to know exactly when it would be active. Only in a checkout, all the time, etc.

[Robert Collins (lifeless)]

On Fri, 2008-06-27 at 14:33 +0000, Aaron Bentley wrote:

> So if you have a branch named "down" in a subdirectory, "bzr merge down"
> wouldn't work. Would we do "bzr merge ./down" for that?

So given three trees:
.
./down
../down

what should 'bzr merge down' do - should it grab ./down or ../down? I
think that ./down is going to be a nested branch - either that or its a
very rare layout. So grabbing ../down is actually more appropriate than
the current behaviour of grabbing down as we do today. But we could
think of this as a search path:
find_branch_heuristic(branch, branchname)
  if is_abs(branchname):
    # probe directly
  else:
    for prefix in lookup_locations(branch):
      # probe for prefix + '/' + branchname
      # and then ...
      if branchname != '.' and found_branch.base == branch.base:
          continue

If we chose to look in ['.', branch.base + '..'] we'd fine ./down
before ../down, still find ../down if there was no ./down that is a
branch.

-Rob

--
GPG key available at: <http://www.robertcollins.net/keys.txt>.