On 04/21/2014 02:51 PM, Stefan Monnier wrote:
>>>> +        (condition-case err
>>>> +            (vc-call-backend backend 'root default-directory)
>>>> +          (vc-not-supported
>>>> +           (unless (eq (cadr err) 'root)
>>>> +             (signal (car err) (cdr err)))
>>>> +           nil)))))
>>> Why do we need this gymnastics?
>> So that we don't accidentally suppress errors we *don't* expect to get.
> 
> No: I see the hypothetical reason, but I was wondering about
> a practical reason.  IOW, I think this is overkill.  Note that the cost
> is not only in code complexity but catching&re-raising signals also
> defeats the backtraces in debug-on-error, which can be a pain.

We're not catching all errors --- just vc-not-supported --- and it
amounts to an assertion that nothing else went horribly wrong, because
we re-raise only in cases where we'd otherwise swallow the error, and an
error with a truncated stack is at least better than nothing at all.

> I even wonder if we should allow `root' to fail.  E.g. RCS/CVS could
> just walk up the directories until there's no RCS/CVS subdir.

Works for me. Backends could also just have their root functions
explicitly return nil if they can't figure it out.