Return-Path: pg_adm@postgres.berkeley.edu
Received: by postgres.Berkeley.EDU (5.61/1.29)
	id AA11506; Mon, 26 Apr 93 19:12:12 -0700
Date: Mon, 26 Apr 93 19:12:12 -0700
Message-Id: <9304270212.AA11506@postgres.Berkeley.EDU>
From: wolf@molout.tutkie.tut.ac.jp (Wolf-D. Ihlenfeldt)
Subject: oids
To: postgres@postgres.berkeley.edu
Sender: pg_adm@postgres.berkeley.edu




In some recent posts it has been hinted that oid's of tuples
may change on replace operations. Is this true or have I misunderstood
something ? Could somebody elaborate on the exact circumstances when
an oid begins to melt ? I tried with some trivial sample tuples, these
do NOT seem to change the oid on replace.

I am currently attempting to port postgresv4.1 to an old sun3/80 with
SunOs4.0.3 and to a DataGeneral AViiON running DG/UX4.32. I have run
into problems in both cases.
The Sun3 version compiles, but crashes with some strange errors in
the shared memory allocation (postmaster). 
The DG version does not compile due to braindamaged limitations
in system programs. After installing Gnu awk, Gnu sed, Gnu grep
I can compile quite a lot, but not all: Now it is sh which bombs
out. Currently I am trying to talk the sysadmin into substituting
sh by bash, but if there is anybody anywhere in netland with a working
DG/UX version willing to share his binary I would be glad to hear.
If have a working bmake for both sun and dg, if anybody cares (this
is not trivial for dg).

There have been complaints about libpq exiting if an error is
encountered. I second this. I have a program which needs to be
really careful when exiting. It is definitely better to return
an error code. I also miss a function like PQsethost()/PQsetport().
My program needs to contact multiple databases on different hosts.
These elementary functions should be part of libpq, even if they
are easy to simulate. Another item on the wish list is more
complicated: I would like to have a libpq which keeps more than
one communication channel (to different db servers) open. The overhead
of switching the server is quite large. 

The definition 
typedef uint32 Time;

in include/tmp/postgres.h collides with the Time definition in X11.
Couldn't it be changed to something like PgTime ?

In case somebody is interested: I have C source and Postquel definitions
for 16, 32 and 64 bit uint's. Tested on SparcII under SunOS4.1.2,
no guarantees, but free. Mail me.

WDI