Re: Strange error when doing M-g on a group in Gnus

[Michael Albinus]

address@hidden (Kai =?iso-8859-15?q?Gro=DFjohann?=) writes:

> Michael Albinus <address@hidden> writes:
> 
> > That's the point (your other alternatives are handled without
> > last-input-char). I get something like "/home-local". If I always
> > complete, I would need to check whether this is is user or host name
> > possible to be completed with "ssh" (your default method, I
> > guess). And that is something I'ld like to avoid.
> 
> Ah, right.  So why do you want to avoid it?

2 reasons:

- If there is the basic I/O completion called for local files like
  "/home-local" it shall be expanded for tramp only if requsted. On my
  machine at work, a genious has created the file
  "/etc/ssh_known_hosts" with everything reachable around here. If I
  type "/" and hit <TAB> afterwards, it takes about 1 minute, and then
  I can choose between ~11.000 host names. ("ssh" as default
  method). Great.

  Remember the start of this thread: you've typed M-g, and you've
  never expected any file name completion; it happened behind the
  scenes. Delay (even for less than 11.000 hosts) wouldn't be
  acceptable. Response time matters.

- If you type "/ftp:" and <TAB> (your default method is still "ssh")
  you would expect the "ftp" should be a method, and I should invoke
  ange-ftp for host name completion. No connection to host "ftp"
  desired.

  If you would type <RET> instead, you really mean "Connect to host
  ftp by method". No host name completion from ange-ftp expected. How
  to distinguish?

And that's why I check last-input-char. Not the best idea I fear, but
I haven't seen a more natural way.

Btw, not necessary for XEmacs, because the 2 cases are distinct by
syntax already.

> kai

Best regards, Michael.

PS: Did my patch help? I'ld like to commit it to CVS tonight, if yes.