unix_socket_directory vs pg_ctl?

Dear Gurus,


I *think* I found something that is at least strange, if not a bug.

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

Abstract: setting unix_socket_directory in postgresql.conf seems to confuse
pg_ctl so that it starts postmaster but can't see it running so reports as
"failed"

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

Details:

1. Debian "Woody" Linux, appropriate postgresql (7.2.1) and php4 packages
installed.

2. I did my best to put 7.2.1 out of the way without uninstalling it, since
the php4-pgsql package depends on it.

3. Downloaded and unzipped postgresql-7.4 (also tried 7.4RC2 last weekend).
Configured to default port 5440 ("40" as in "(7.)4.0"). Compiled,
regress-tested, installed, initdb.

4. Configured postgresql.conf to use
 unix_socket_directory = '/var/run/postgresql'
 since Debian's php package requires socket files to be there.

5. Starting with
 bin/pg_ctl -w -D data -l data/logfile
 causes postmaster to start, but pg_ctl keeps sleeping and re-checking,
then gives up. However, postmaster runs, with socket files in the given
directory.

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

I know workarounds (the ugliest being "recompile php", or the simpler
"symlink the socket files from /tmp"). What I'm concerned is if I'm doing
something wrong, or is it an existent bug in Postgresql.

If needed, I can give detailed information (please specify) about my system,
or even try compiling and installing different versions with different
settings. I just don't have any ideas what to try and what not to.

TIA,

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