[MERGE] Make accessing a branch.tags.get_tag_dict use a smart[er] method rather than VFS calls and real objects.

[Andrew Bennetts]

Robert Collins wrote:
> This eliminates the last VFS call during 'bzr branch bzr+ssh://foo bar'
> where the source is not stacked.

bb:tweak

Overall this was a really nice patch, btw.  I feel it leaves things cleaner than
it found them, which is always nice :)

[...]
> +    def _set_tags_bytes(self, bytes):
> +        """Mirror method for _get_tags_bytes.
> +
> +        :seealso: Branch._get_tags_bytes.
> +        """
> +        self.lock_write()
> +        try:
> +            self._transport.put_bytes('tags', bytes)
> +        finally:
> +            self.unlock()

It's a touch ugly, but please use _run_with_write_locked_target (also defined in
branch.py) to avoid self.unlock() potentially obscuring an error during
put_bytes.

> === modified file 'bzrlib/smart/bzrdir.py'
[...]
> +class SmartServerRequestOpenBranchV2(SmartServerRequestBzrDir):
> +
> +    def do_bzrdir_request(self):
> +        """open a branch at path and return the reference or format."""
> +        try:
> +            reference_url = self._bzrdir.get_branch_reference()
> +            if reference_url is None:
> +                format = self._bzrdir.open_branch()._format.network_name()
> +                return SuccessfulSmartServerResponse(('branch', format))
> +            else:
> +                return SuccessfulSmartServerResponse(('ref', reference_url))
> +        except errors.NotBranchError:
> +            return FailedSmartServerResponse(('nobranch', ))

I changed the cloning_metadir RPC similarly, but used "direct" and "reference"
rather than "branch" and "ref".  We probably should be consistent... would you
mind updating my change to cloning_metadir to use "branch" and "ref" like yours
does?  IIRC the test and code fallout should be pretty modest.