HP-UX porting strategy (moved from PATCHES) - Mailing list pgsql-hackers

From Tom Lane
Subject HP-UX porting strategy (moved from PATCHES)
Date
Msg-id 22357.893347119@sss.pgh.pa.us
Whole thread Raw
List pgsql-hackers
John Fieber <jfieber@indiana.edu> wrote:
> Since 9.* and 10.* are apparently so different, maybe there would
> be less breakage if they were treated as separate ports?

I've thought about this and concluded that it'd probably just result
in duplication of effort.  There are differences, which have to be
taken care of by conditional tests in the Makefiles and/or #ifdefs.
But I think most of the discrepancies we are hearing about have to do
with other differences across installations.  Specifically,
  (a) whether people are using gcc or the vendor cc;
  (b) what patch level people are at for libc, libm, vendor cc, ...
HP issues separate patch streams for all these system components,
which is great for getting fast turnaround on bug fixes, but it's
a nightmare when it comes to guessing what someone else's "HP-UX 9.05"
installation really is.

Bruce suggested that I look at Stan Brown's back messages to the
hackers list, which I did.  As far as I can tell, Stan's major problem
was that he didn't know how to tell configure to use cc rather than
gcc when both are installed.  The environment-variable override trick
("CC=cc configure ..." or "setenv CC cc; configure ...") probably ought
to be documented in the INSTALL instructions.  The other problems he
mentioned all seem to be solved in the 6.3.2 release.  Most of the
problems I ran into were really a question of porting 6.3.2 to HPUX 9,
not 10 which is what Stan used.

I would like to recommend that y'all go ahead and apply the HPUX patches
I sent to the patches list on Tuesday.  I have confirmed that they work on
my local installations of HPUX 9 *and* 10.  I cannot guarantee that they
will work on every installation of HPUX, but I will be willing to take
responsibility for coordinating any tweaks needed to handle problems
that pop up elsewhere.

BTW, is anyone planning to fix src/backend/port/getrusage.c so that
it doesn't have to be hand-edited before use?  I'm nervous about messing
with it without knowing what systems it is needed on.  But ISTM that we
ought to be able to auto-configure it.

            regards, tom lane

pgsql-hackers by date:

Previous
From: "Thomas G. Lockhart"
Date:
Subject: Re: [HACKERS] Re: [PORTS] BSDI-3.1
Next
From: darrenk@insightdist.com (Darren King)
Date:
Subject: Removing PORTNAME from libpq/Makefile