unix_socket_directory not fully recognized by pg_ctl?

Dear Gurus,

I have a similar problem as discussed in
http://archives.postgresql.org/pgsql-novice/2003-06/msg00321.php

This probably should go to pgsql-novice too, but I'm not a member of that
list so please forgive my lazyness.

I use Debian Woody Linux and installed postgres (v7.2.1), but I needed a
newer version, so decided to try 7.4RC2 (which is not in the news yet, as I
checked today). It compiled and installed, and I could run the backend.
However, PHP could not run, since it searched for the socket in
/var/run/postgresql.

I found the unix_socket_directory option in postgresql.conf and set it to
/var/run/postgresql and restarted the server with

  bin/pg_ctl start -w -D data -l data/postgres.log

The socket file and socket lockfile was created in /var/run/postgresql
-- it's ok, but pg_ctl hung, and after quite some time it exited with
"pg_ctl: postmaster does not start"

However; I can connect the server with PHP's pg_connect(), or psql -- but
psql only works through TCP/IP (i.e. -h localhost). If run locally, it
searches for socket in /tmp. Using

  psql -h /var/run/postgresql

did work. My questions are:
What causes pg_ctl's strange behaviour?
Is this the only way to tell this to psql?
Is there a compile-time configuration parameter? (could not find it)
Is there a psqlrc setting for this?
Is there a complete description for psqlrc? (could not find it)

TIA,

--------------------- cut here ---------------------

G.
--
while (!asleep) sheep++;