Re: Starting openoffice with URP port open fails
Hi Holger,
On Wed, Feb 26, 2003 at 08:56:02 +0100, Holger Krug wrote:
> $ killall soffice.bin
> $ ~/.openoffice/1.0.2/soffice \
> "-accept=socket,host=localhost,port=8100;urp;"
> X IO Error
>
> I have tested different upstream versions:
>
> - upstream 1.0.2 produces the same error
> - upstream 643 and 644m1 don't have this error
>
> So it seems to be fixed in the current upstream code.
No, it isn't. But at OOoCon we identified the cause.
The good news: there is a workaround, set the environment variable
STAR_OVERRIDE_DOMAINNAME to let it contain the domain name part only (!)
without the hostname, of the machine running the executable. This is
usually what /bin/dnsdomainname on a Debian system outputs.
And that's also the reason why it fails on Debian: if no FQDN (Fully
Qualified Domain Name) could be obtained from a call to gethostbyname(),
the heuristics look for /bin/domainname which isn't found. Now that
wouldn't all be that bad if the routine wouldn't terminate the child
process with an exit() call in this case, which calls all functions
registered with atexit() or on_exit() and closes all open handles, not
only of the child process but also those of the parent process,
including the socket for X communication.
If you didn't encounter the error with 643 and 644m1 you may have
changed something in your resolv.conf, for example, you included the
FQDN and not just a localhost entry.
Kudos to Philipp Lohmann and Chris Halls who tracked it down!
Eike
--
GnuPG key 0x293C05FD: 997A 4C60 CE41 0149 0DB3 9E96 2F1A D073 293C 05FD
Reply to: