Re: tramp (2.0.51); tramp and file-name-all-completions

[Michael Albinus]

"Richard M. Stallman" <address@hidden> writes:

> I would say these problems are all bugs in tramp.

Yes.

> file-name-all-completions has a simple spec,
> and in principle it ought to work "right" for all meaningful args.

The problem is the syntactical ambiguity of Tramp file
names. (file-name-all-completions "" "/telnet:") can mean completion
of user and host names for method "telnet", or completion of file
names for method "ssh" on host "telnet".

When I wrote completion code for Tramp 3 years ago, I've introduced a
somehow obscure heuristic determining what is meant. This works more or
less the usual cases, but it breaks ido.

I've dropped a comment in the TODO "must be changed when we know it
better". It seems to be the time now.

> If there are some cases where it is just hopeless to ask tramp
> to do completion correctly, we could decide on some return value
> that means "unable to do the job right".

I don't think so. It is not a tragedy when file-name-all-completions
does not return the most complete list. Given the example above,
something like ("telnet:127.0.0.1:" "telnet:localhost.local:") would
be sufficient, even if there is a host called "telnet". Only when
completion of the method returns nil, one might try file name
completion.

I'm sitting already the whole day on this problem.

Best regards, Michael.