Re: Tramp 2.0 -> 2.1 migration woes

[Michael Albinus]

"Trent W. Buck" <address@hidden> writes:

Hi Trent,

> Multi-hop now hides initial hops
> ================================
>
> The initial hops are no longer shown in the path, nor on the first
> line of dired buffers.  Already I have been VERY confused because
> a bad tramp-default-proxies-alist resulted in no proxy being used for
>
>     /sudo:address@hidden:/etc
>
> ...so that instead of editing leek's /etc as root, I was editing my
> laptop's /etc as root!  Since the path claims that I'm on address@hidden, I
> would have made disastrous changes to my local system if I hadn't
> noticed that there was no /etc/ssh/sshd_config (because sshd is not
> installed on my laptop).
>
> This accident makes me *extremely* afraid to trust TRAMP 2.1-style
> multi-hops for anything at all, since I'm bound to make alist mistakes
> that I won't catch so fortuitously.

That isn't related to multi-hops. It is simply an error in Tramp, which
shall inform you about the used host.

In case of su(do)?, the host name is ignored (except there is a setting
in tramp-default-proxies-alist). The Tramp documentation says for the
su(do)?  methods: "... With other words, a specified host name in the
file name is silently ignored."

Consequently, if you open a file "/sudo:address@hidden:/etc", it must be
expanded to "/sudo:address@hidden:/etc", and this shall be visible. A
warning (or an error?) shall be raised. I'll fix it.

> Shell-command is no longer a shell
> ==================================
>
> It seems that shell-command (M-!) no longer uses a shell in TRAMP
> buffers.  For example, in /sudo::/ if I do M-! whoami && pwd RET or
> M-! sh -c 'whoami && pwd' I get the following errors:
>
>     whoami: extra operand `&&'
>     Try `whoami --help' for more information.
>
>     &&: 1: Syntax error: Unterminated quoted string
>
> Another example was running M-! find -name \*~ -delete RET in
> /ssh:leek:, which should have deleted backup files in my home
> directory.  In Emacs 22, it behaves correctly.  In CVS Emacs, it
> silently fails.
>
> This means I can no longer trust basic sh commands to work correctly,
> and thus cannot use this feature in TRAMP buffers.

I'll fix it. Obviously, there was too much optimization in Tramp 2.1 -
the COMMAND shall still be called via "sh -c COMMAND".

> sudo -s (without -H) now aborts connection
> ==========================================
>
> My .bashrc (et al) is mode 0600 (readable only by me).  If I run "sudo
> -u zippy -s" in a shell, I get the (ignorable) error
>
>     bash: /home/twb/.bashrc: Permission denied
>
> Because $HOME is still /home/twb.  For this reason I use sudo -i or
> (on legacy systems) sudo -H -s in interactive shells.  In Emacs 22,
> TRAMP ignores this error and /sudo:zippy@: works fine.  In CVS Emacs
> it seems this error aborts the connection entirely:
>
>     $ emacs -batch -Q -nw /sudo:zippy@: -f kill-emacs
>     Tramp: Opening connection for address@hidden using sudo...
>     Tramp: Waiting 60s for local shell to come up...
>     Tramp: Sending command `sudo -u zippy -s -p Password: && exit || exit'
>     Tramp: Waiting for prompts from remote shell
>     File error: Permission denied
>     [hangs]
>
> I have worked around this problem successfully with the following code
> in my .emacs.  Note that deleting "-H" is merely for potential
> forwards compatibility.
>
>     (eval-after-load "tramp"
>       '(progn
>          (add-to-list 'tramp-methods
>            (cons "sudoHs" (copy-tree (cdr (assoc "sudo" tramp-methods)))))
>          (let ((x (cdr (cadr (assoc 'tramp-login-args (cdr (assoc "sudo" 
> tramp-methods)))))))
>            (setcar x (cons "-i" (delete "-H" (delete "-s" (car x))))))
>          (let ((x (cdr (cadr (assoc 'tramp-login-args (cdr (assoc "sudoHs" 
> tramp-methods)))))))
>            (setcar x (cons "-H" (delete "-H" (car x)))))))

I'll check whether this could be the default.

> Multi-hop now requires tedious alist editing
> ============================================
>
> It is often the case that I need to access a customer site that I have
> never accessed before, or whose network layout has changed since I
> last accessed it.  On Emacs 22 I would therefore by trial and error
> modify a multi: path until it was correct, do the job, and then forget
> the path.  In the short term, I can switch back to an open dired for
> that site, or use M-p at the C-x C-f prompt.
>
> In the long term, by the time I need to access that customer's site
> again, their network will probably have changed.  Writing it down
> would be at best bloat, and at worst misleading, since the correct
> information is stored elsewhere in a text file that is maintained by
> other staff who are in charge of that customer's network.
>
> Now, the trial and error step must be done by editing a setq in .emacs
> and repeatedly doing C-M-x, then trying C-x C-f again (possibly
> killing *tramp foo* buffers after each attempt to make sure no stale
> connections cause problems.)  To remove this bloaty, misleading
> information I must remember to edit .emacs again after successfully
> connecting.  This is quite tedious compared to editing a single multi:
> string at the C-x C-f prompt.
>
> (I could call add-to-list in M-x ielm or M-: instead of editing the
> SETQ in .emacs, but then I'd need to remove-hook (there's no
> remote-from-list function) each bad attempt until I got it right, so
> it's still pretty tedious.)

I admit, that in your use case it is a real regression. There have been
several reasons to change the multi-hop syntax. I remember, that for
people with *static* multi-hop paths the new syntax is shorter and more
convenient to type therefore; I also remember some problems in
implementation of all these Tramp 2.1 optimizations for the Tramp 2.0
multi-hop paths.

OTOH, setting up a working tramp-default-proxies-alist is kind of
complex, indeed.

On the Tramp todo list, there is the item 

;; * WIBNI there was an interactive command prompting for Tramp
;;   method, hostname, username and filename and translates the user
;;   input into the correct filename syntax (depending on the Emacs
;;   flavor) (Reiner Steib)

This could be expanded to multi-hops, where it seems to be even more
useful. I'll try to find the time to work on it (Threat: I will exploit
*you* as tester for this functionality!)

Thanks for the report, and best regards, Michael.