[MERGE][0.8] break-lock command

[Aaron Bentley]

Robert Collins wrote:
>>q: would it be better to say 'skipped' if it's unimplemented, rather
>>than just return?
> 
> 
> No: Being unimplemented is valid for this interface : it means that the
> branch does not suffer from broken locks. . There is nothing 'Wrong' -
> and our definition for 'skipped' is 'There is something you could do to
> fix this.' Constrast with format 4 branches, where we can fix it by
> restoring the format 4 write logic.

I haven't been meaning it that way.  I've been using it for cases that
can be fixed (by installing things) and for cases that can't (e.g.
adding symlink support to Windows 2000, making the prevalent windows
diff3 binary-safe).

Perhaps there are finer distinctions to be made.  It would be nice if
the test runner explained why tests had been skipped, but we'd want
symbolic constants for the reasons, I think, to prevent the test runner
from writing the Great American Novel.

> (Can I call again for the nuking with extrem prejuidice of the SFTP
> connection cache ?)

Sounds like it might force us to do the same work at a higher level.  Is
that a valid concern?

Aaron