On 11/29/2012 06:23 PM, Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>> ... Or maybe the
>> postmaster needs to check the length somehow before it calls
>> StreamServerPort() so we can give a saner error message.
> Hm. That's ugly, but a lot less invasive than trying to get the
> official API to pass the information through. Sounds like a plan to me.
>
> However, that's only addressing the reporting end of it; I think we
> also need to change the pg_upgrade test script as suggested by Noah.
>
>
The test script doesn't do anything. It's pg_upgrade itself that sets
the socket dir. See option.c:get_sock_dir()
cheers
andrew