Re: (still) Need some help with translators

[Richard Braun]

On Wed, Nov 07, 2012 at 08:46:36AM +0100, Svante Signell wrote:
> Well the code is a pipe, see the attachment in
> http://lists.gnu.org/archive/html/bug-hurd/2012-11/msg00006.html
> and roland refers to pflocal in his reply. Can it happen that the old
> pflocal is still used? I have added print statements to io.c in pflocal
> and replaced it. Still no output.

That's probably what happens. The pflocal server is a bit sensitive, as
Unix domain sockets are relied upon by numerous system applications. A
reliable way to change it is to install the new executable in
/hurd/pflocal, then reboot.

> I did the following:
> mv /hurd/pflocal.old
> cp -p /usr/pflocal.new /hurd/pflocal
> rebooted
> Looks like both /hurd/pflocal and /hurd/pflocal.new are running:
> #> ps -feM|grep pflocal
> root     9     3  0:02.69 /hurd/pflocal
> root   881     1  0:00.01 /hurd/pflocal.new
> But in this case this should point to the same file.
> How to get rid of one of them?

Servers don't "point" to files, it's actually the contrary: file nodes
refer to servers. But you're not looking at the right location. The node
that actually refers to pflocal is /servers/socket/1:
$ showtrans /servers/socket/1
/hurd/pflocal

This means that pflocal is a passive translator for /servers/socket/1.
It is automatically started on the first access to this node, and then
serves the following requests until stopped (usually by a system
shutdown). If you simply replace the executable file, it doesn't change
the fact that the server is still running with the previous code, and
the parent of /servers/socket/1 (normally the root file system server)
still redirects lookups for this node to the already existing pflocal
instance.

-- 
Richard Braun