postgres/doc/FAQ_HPUX

=======================================================
Frequently Asked Questions (FAQ) for PostgreSQL  V6.4
HP-UX Specific
TO BE READ IN CONJUNCTION WITH THE NORMAL FAQ
=======================================================
last updated:           Sat Nov 28 16:21:25 EST 1998

current maintainer:     Tom Lane (tgl@sss.pgh.pa.us)
original author:        Tom Lane (tgl@sss.pgh.pa.us)


Questions covered here:
1.1)	What do I need to install PostgreSQL on HP-UX?
1.2)	Anything special about the build/install procedure?
1.3)	yacc dies trying to process src/backend/parser/gram.y.
1.4)	Linking the main postgres executable fails, complaining that
	there's no "alloca" function. 
1.5)	OK, it seemed to build and install, but the regression test fails.


----------------------------------------------------------------------
Section 1:      Installing PostgreSQL
----------------------------------------------------------------------

1.1)	What do I need to install PostgreSQL on HP-UX?

PostgreSQL 6.4 is known to build and pass regression test on HPUX 9.03,
9.05, and 10.20, given appropriate system patch levels and build tools.
It should work on other HPUX 9.* and 10.* releases for Series 700/800
machines, too.  (No one has reported trying it with HPUX 11 yet.)
Since this is a new FAQ, I don't yet have a lot of information about the
exact prerequisites, but I'd appreciate hearing from anyone who fails to
build a working copy, so that we can add more info about exactly what is
needed.

Aside from PostgreSQL 6.4 or later sources, you will need GNU make
(HP's make will not do), and either GNU gcc or HP's full ANSI C compiler.
You must also get flex (GNU lex) 2.5.4 or later --- all versions of
HP's lex fail on the Postgres lexer files.

I'd also recommend making sure you are fairly up-to-date on HP patches,
particularly if you are using HPUX 9.  At a minimum, if you are on HPUX 9,
you *must* have PHSS_4630 (libm update) or a successor patch; otherwise
Postgres' date/time functions will misbehave.  On general principles you
should be current on libc and ld/dld patches, as well as compiler
patches if you are using HP's C compiler (but I don't currently know of
any specific failures due to not having recent patches for these files).
See HP's support websites, such as http://us-support.external.hp.com/,
for free copies of their latest patches.

PostgreSQL 6.3.2 and earlier required quite a few small tweaks to
install on HPUX, so I recommend you not bother with anything older
than 6.4.


1.2)	Anything special about the build/install procedure?

When you run configure, you will want to explicitly select either the
hpux_cc or hpux_gcc template depending on which compiler you plan to
use:
	./configure --with-template=hpux_cc
for HP's C compiler, or
	./configure --with-template=hpux_gcc
for GNU gcc.  (If you omit --with-template, configure may either
default to hpux_cc or give up entirely, depending on which HPUX and
PostgreSQL releases you have.)

You may want to tweak the CFLAGS setting in template/hpux_[g]cc before
you configure; the distributed files contain neither -O nor -g switches,
which is hardly optimal for any situation.  I've seen no problems using
-O with gcc 2.7.2.*.

The default install target location is /usr/local/pgsql, which
(particularly on HPUX 10) you might want to change to something under
/opt.  If so, use the --prefix switch to configure.

If you have both HP and GNU C++ compilers in your PATH, keep an eye on
whether configure picks the right one --- you want the HP c++ if you are
using HP C, or g++ if you are using gcc.  Mixing HP and GNU compilers
won't work.  You may need to provide a --with-CXX=compiler switch to
force configure to pick the matching C++ compiler, or even say
--without-CXX if you have a C++ compiler but it doesn't match the C
compiler you want to use.

Otherwise the standard build/install procedure described in the
PostgreSQL documentation works fine.


1.3)	yacc dies trying to process src/backend/parser/gram.y.

HP's yacc doesn't create its tables large enough to handle the Postgres
grammar (a lot of other vendors' yaccs have this problem too).  There
are three possible workarounds:

1. The quickest answer is just to "touch" src/backend/parser/gram.c
and src/backend/parser/parse.h and repeat the build.  Any PostgreSQL
distribution file should have up-to-date copies of those files included,
so you shouldn't need to run yacc on gram.y at all ... but sometimes
gram.y mistakenly has a newer timestamp in the distribution than the
derived files do.

2. Install "bison" (GNU yacc) and reconfigure.  Bison doesn't have a
problem with large grammars.  Note this is not the right choice if you
are using HP's cc on HPUX 9 --- see next item.

3. Increase yacc's table sizes enough to cope.  With a pre-6.4
PostgreSQL grammar, I was able to get HPUX 9's yacc to work by
setting YFLAGS to
	-d -Np2000 -Ns3000 -Nm100000 -Nl2000 -Na30000 -Nc10000
(You can edit YFLAGS either in the template file before running
configure, or in src/Makefile.global afterwards.)  Future PostgreSQL
releases might require even larger tables, but this should do for
a starting point.


1.4)	Linking the main postgres executable fails, complaining that
	there's no "alloca" function. 

If you're using HP's cc on HPUX 9, it's right: there's no alloca
function.  The only place in PostgreSQL that uses alloca is the parser
(gram.c), and that does so only if it was generated with GNU bison.
Unfortunately the distribution copy of gram.c is made with bison.
There are several possible answers:

  1. Remake gram.c with HP's yacc (see above item for switch settings).
     You might also need to remake src/backend/bootstrap/bootparse.c.

  2. Build with gcc, which treats alloca as a compiled-in-line function.

  3. Install HPUX 10, which has alloca.  You're gonna have to do that
     before Y2K anyway...


1.5)	OK, it seemed to build and install, but the regression test fails.

There are several "expected failures" due to differences between HPUX
and the regression test reference platform used by the PostgreSQL group.
A look at the textual differences between the expected and actual
outputs will usually reveal that the differences are minor.  You should
expect these differences:

TEST(S)			COMMENTS

int2, int4:		pg_atoi generates a differently worded error
			message for integer overflow.

float8:			In 6.4, float8 shows some differences due to
			different handling of overflow/underflow errors in
			exp() and pow().  This should be fixed in 6.4.1
			and later.

float8, geometry:	Lots of differences in the last digit or two
			because of different roundoff errors in floating
			arithmetic.  Also, HPUX does not distinguish
			-0 from 0 during printout, but the reference
			platform does.

horology:		HPUX time library does not know about daylight
			savings time before 1970, so there are some
			places in horology where a time will be shown
			in PST instead of PDT.

In addition, the int8 regression test will fail massively on HPUX 9,
because int8 doesn't actually work on this platform (sprintf/sscanf
don't cope with long long int).  Either upgrade to HPUX 10, or don't
use int8 data.

Any other error is cause for suspicion.  In particular, if you see
failures in the datetime test on HPUX 9, you probably forgot to
install the libm patch PHSS_4630 --- see item 1.1 above.