Re: talk/talkd on solaris 2.8

[Alain Magloire]

Bonjour

  Sorry for the delay, almost miss this.

> 
> I needed some ntalk client/server for Solaris 2.8 since Sun only supports
> talk by default. After some searching the talk/talkd from inetutils seemed
> best suited for me and I installed 'em. They compiled okay but I needed
> the attached patch to make client and server finallay work correctly:

Was this patch against the latest cvs tree? or snapshot?

snapshots:
ftp://alpha.gnu.org/gnu/cvs/inetutils.tgz

anonymous CVS.  Set your CVSROOT string to:

:pserver:address@hidden:/gd/gnu/anoncvsroot

Hit enter when prompted for a password

issue:

cvs co inetutils.

> 
> The client doesn't by default accept carriage returns and instead outputs
> ^M to the console. I checked the Linux netkit talk and found that it
> checks for \r in addition to \n as line feed and added the check
> respectively.

Ok.

> 
> Also the client outputs '[Connection established]^G^G^G' since the 007's
> don't get interpreted by the terminal to ring the bell. I removed them and
> added a curses bell() call instead.

Ok.

> 
> The hardest part was to make talkd work since Solaris 2.8 doesn't have a
> /var/adm/utmp any more and instead only supports utmpx. I decided to use
> the system functions for accessing it instead of reading the entries
> directly. This can of course be changed again. I tried to implement my
> changes so that they shouldn't interfer with standard utmp behaviour but I
> didn't test it.

Yes, this is very annoying, utmp vs utmpx can become such a pain grrr...!!
I thing, I will go for utmpx solely and provide a converter utmp_to_utmpx
for platforms not having utmpx

> An accompanying problem arises through /dev/console handling on solaris
> since the graphical login screen adds a utmpx entry for it but redirects
> the output to /dev/null unless the user starts a xconsole. I therefore
> changed the console searching to remember /dev/console as last resort
> (e.g. if the user is logged in on text console only) but to prefer any
> other device.
> 
> This again messed up libinetutils/ttymsg.h since pseudo ttys are
> /dev/pts/<number> on Solaris which makes ttymsg refuse to print something
> on it since it includes a /. I changed that to accept slashed but check
> the final path to be a character device which seems equally secure to me.

I'll will take a look later.

If it possible can you do the patch against the latest snapshot.
I'm not sure against what the patch is, and the code was change
recently.  Let me know.

--
alain