Re: A future for the SSH File Transfer Protocol?

[Simon Tatham <anakin%pobox.com@localhost>]

Jeffrey Altman <jaltman%columbia.edu@localhost> wrote:

> Why should a client ever have to understand a system dependent alias
> such as ~fred?
> 
> The only entity that should need to know how to map ~fred to what it
> points to is the sender of the file.

Ah; you think it would be better for servers to understand `~fred'
at the start of a pathname?

Fair enough, I suppose. But the rest of the ethos of SFTP seems to
be that whatever alias or synonym you use to refer to a directory,
you can always call FXP_REALPATH and get the full canonical name. So
in my command-line SFTP client, I don't want to type `cd ~fred' and
then have `pwd' just tell me `~fred'; I'd rather get the same
behaviour I get at the real shell prompt, where I type `cd ~fred'
and `pwd' tells me `/home/fred'.

I suppose that could be done just by making the server understand
`~fred' at the start of a pathname, and then having the client call
FXP_REALPATH("~fred") if it wanted the canonical form; but that
starts to eat into the available characters in filenames, which is
something SFTP has so far been scrupulous in not doing. So instead,
my proposal was to introduce a protocol extension which allowed the
client side to request the home directory of a particular user.

While I was at it, I also added name<->number mappings for users and
groups, on the grounds that it seems silly that a command-line
client can list the textual usernames of file owners by printing out
the long name field, but a GUI client wanting to split all the
details into separate columns has to make do with the numbers.

Cheers,
Simon
-- 
Simon Tatham         "A cynic is a person who smells flowers and
<anakin%pobox.com@localhost>    immediately looks around for a coffin."