[MERGE] Fix for bug 183831

[Geoff Bache]

>> In other words, is the above case the real reason you're opposed to
>> this change or are you still going to be opposed to it even if there
>> is no concrete case to address? I'm happy to try and fix this but I
>> don't want to spend the time if you're going to reject it anyway.
>
> The real reason is that I think you're doing it at the wrong layer, and
> that this will have undesirable side-effects.  So if you submitted a
> patch that addressed all the concerns I've listed, I'd be inclined to
> look for other undesirable side-effects.

I think we've hit the crux now.

Robert said something interesting earlier on this thread: "I'm pro
this change because it fixes the bug and doesn't break other tests".
Isn't that what Test-driven Development is about? We don't try and get
every pathological case right at once: we proceed in small steps and
fix one issue at a time.

There are 17000 tests: if they all work then surely that's a pretty
good indication we can proceed. How do we know when we're done
otherwise? We could sit in our rooms and dream up undesirable side
effects until next year, but we still won't think of everything that
will happen, and we'll probably think of plenty of things that won't.

I don't disagree that I may be doing this at the "wrong layer" from a
theoretical correctness point of view : I don't know the code
structure well enough to judge, really. In my world, the interesting
question is not "Is this correct in every conceivable respect?" but
"Is it better than what we currently have?" From a behaviour point of
view at least the answer would seem to me to be yes, in that it fixes
an issue with normal sensible usage and may lead to complications in
situations where it isn't particularly hard to trigger bugs with
today's bzr anyway (i.e. complex symbolic link usage).