[merge][#261315][1.7] RemoteBranch sets up stacking (more) correctly

[Martin Pool]

On Sat, Sep 6, 2008 at 11:16 AM, John Arbash Meinel
<john at arbash-meinel.com> wrote:
> John Arbash Meinel has voted approve.
> Status is now: Approved
> Comment:
> It would be interesting to put this into the per_repository (per_branch?)
> tests for all formats that support stacking. You would probably need to
> special case Remote*, to set it up behind the scenes, and then access it
> only over Remote*.
>
> It does cause a permutation, though. As you now need to test that
> "Repo.foo()" works when the object is stacked, versus when it isn't stacked.

Right, and you possibly want to do some of the setup creating history
in the stacked-on branch.

This also reminds me of the idea I had a while ago of doing the Remote
calls in-process within a single thread.  (Just synchronously invoking
the server-side method on the request once it's packed up.)  That may
get more hairy later but it seems like it'll give better failure
messages.

> Maybe as an alternative, we just create a StackedBranch format that is
> tested?

I don't think we need to add a new format; hopefully we should just be
able to add a new scenario.  One interesting way to start might be to
use the get_readonly_url to say that after creating the setup, we look
at it through a remote connection to a branch stacked on the original.
 (Maybe this should be renamed to get_examination_url or something.)

> Anyway, the patch is fine, it just feels like there should be more
> systematic tests for how things are handled when we have a Stacked object.

I definitely agree with that.  I wanted to at least get this up for
1.7rc1 and then prod it some more.

-- 
Martin <http://launchpad.net/~mbp/>