Re: Much oddness

[George <jirka 5z com>]

On Thu, Mar 25, 2004 at 11:05:00AM -0800, Jerry Wall wrote:
> To second, George redhat really wants you to configure
> gdm to go ontop of its installation if you replace it.
> You have to do the full configuration line as he said
> and follow that by a make and make install otherwise
> the gdm subcomponents will refuse to talk with one another.
> 
> I have notice several items while working with GDM
> that affect its stability.
> 
> 1. there is a file descriptor leak whenever an X server/gdm slave exits
>    and gdm relaunches a new X server/gdm slave. This is not a big
>    deal for a single display system that is not left up for 
>    extended periods of time but for my work this easily shows up.
>    
>    my solution is to close whatever of d->socket_conn,
>    d->slave_notify_fd, and d->master_notify_fd are open
>    in whack_old_slave(...)

Good catch, socket_conn should really never be on at that point, but
it's good I suppose to clean it, just in case it is.  slave_notify_fd
will never be used from the main daemon (it should always be -1 or
some other bogus value there) so that 'close' should not be there.

> 2. The other issue that has caused trouble in testing is the
>    main gdm allows SIGPIPE and of course then exits when
>    a gdm slave dies while the main gdm is writing on the
>    pipe to that slave.
>   
>    my solution is to have the master gdm ignore SIGPIPE:
>    
> gdm.c:
> 2151a2086
> >     gdm_signal_ignore (SIGPIPE);

Sounds like a good idea I suppose.  I don't suppose we need SIGPIPE
anywhere.

Comitting both of these to CVS HEAD, thanks.

George

-- 
George <jirka 5z com>
   If I had my life to live over again, I'd be a plumber.
                       -- Albert Einstein