Re: bug in copy-directory

[Michael Albinus <michael.albinus@gmx.de>]

Thierry Volpiatto <thierry.volpiatto@gmail.com> writes:

> Chong Yidong <cyd@stupidchicken.com> writes:

>> Given that we are well into pretest for 23.3, here are two options:
>>
>> 1. Keep the code that's in the emacs-23 branch.  From now on,
>>    copy-directory needs a non-nil COPY-AS-SUBDIR arg to get cp-like
>>    copying, when called from Lisp.  In the meantime, users of
>>    M-x copy-directory will be surprised, but that's life.
> Not anymore needed.
>
>> 2. Revert the code in the emacs-23 branch to what it was in 23.2.
>>    Change copy-directory on the trunk (probably by flipping the meaning
>>    of the optional arg, and changing Dired and Tramp accordingly.)
>
> I think you must fix that for 23.3, copying is important.
>
>> Any arguments for either choice?
>
> Finally, COPY-AS-SUBDIR is not needed, to remove it we used a second
> function called copy-directory-1 and call copy-directory-1 from dired.
> Now with these change the state is:
>
> So it fail only in one place, copying from dired with ssh method.
> Because it seem ssh is the only method that call copy-directory, 
> so dired call copy-directory-1 but the handler reuse copy-directory.
> Otherwise other methods seem to work fine (here default method scpc
> tested).

It is not only one Tramp method which is broken. Tramp has an own
implementation for scp- and rsync-based methods. Everything else fails,
because Tramp falls back to the default copy-directory, including such
methods like ftp, ssh, su(do), smb. I believe, this is not acceptable.

I would vote for continuing with the last patch from Thierry, not
committed yet. Maybe it is possible to throw copy-directory-1 away, and
to adapt the call of copy-directory in Dired instead of.

If this is not possible, we shall fall back to the Emacs 23.2
implementation, which is less broken compared with the current status.

Best regards, Michael.