Solaris SPARC - Sun compiler 5.5 - 64 bit

Which errors?

Hi Peter,

This is the error I receive:

gmake[5]: Entering directory
`/usr1/clarka/src/postgresql-8.0.2/src/backend/storage/lmgr'
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o lmgr.o lmgr.c
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o lock.o lock.c
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o proc.o proc.c
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o deadlock.o deadlock.c
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o lwlock.o lwlock.c
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o spin.o spin.c
cc -Xa -xtarget=ultra2 -xarch=v9a -xchip=ultra2
-I../../../../src/include -c -o s_lock.o s_lock.c
/devtools/WS8.0/SUNWspro/prod/bin/fbe: "/tmp/yabeAAAKgainV", line 206:
error: can't compute difference between symbols in different segments
Failure in /devtools/WS8.0/SUNWspro/prod/bin/fbe, status = 0x100
Fatal Error exec'ing /devtools/WS8.0/SUNWspro/prod/bin/fbe
cc: acomp failed for s_lock.c
gmake[5]: *** [s_lock.o] Error 2
gmake[5]: Leaving directory
`/usr1/clarka/src/postgresql-8.0.2/src/backend/storage/lmgr'


Thanks,
Andrew


-----Original Message-----
From: Peter Eisentraut
Sent: July 29, 2005 5:51 AM
To: Clark, Andrew
Cc: pgsql-ports@postgresql.org
Subject: Re: [PORTS] Solaris SPARC - Sun compiler 5.5 - 64 bit

Am Dienstag, 26. Juli 2005 16:06 schrieb Clark, Andrew:
Which errors?

You really should mention which Postgres version you're using
... but if it's pre-8.0, possibly the problem is the lack of this patch:

2004-12-29 18:47 tgl

* src/: backend/storage/lmgr/s_lock.c, template/sunos4: Support
Sun's compiler on SunOS4 (a/k/a Solaris 9). Per ayan@ayan.net

regards, tom lane

Hi Tom,

I'm using PostgreSQL 8.0.2 on Solaris 8.

Thanks,
Andrew

-----Original Message-----
From: Tom Lane
Sent: July 29, 2005 10:18 AM
To: Clark, Andrew
Cc: Peter Eisentraut; pgsql-ports@postgresql.org
Subject: Re: [PORTS] Solaris SPARC - Sun compiler 5.5 - 64 bit

"Clark, Andrew" <Andrew.Clark@fnf.com> writes:
You really should mention which Postgres version you're using ... but if
it's pre-8.0, possibly the problem is the lack of this patch:

2004-12-29 18:47 tgl

* src/: backend/storage/lmgr/s_lock.c, template/sunos4: Support
Sun's compiler on SunOS4 (a/k/a Solaris 9). Per ayan@ayan.net

regards, tom lane

Is configure using the solaris template, or the sunos4 one?
If the former, does changing
CFLAGS="-v" # -v is like gcc -Wall
to
CFLAGS="-v -DSUNOS4_CC" # -v is like gcc -Wall
fix it? (Note you need to re-configure to make template changes
have any effect.)

regards, tom lane

Tom,

Well that definitely did something, but now I get a different error:

----------------

/usr1/clarka/bin/gmake -C backend/utils/mb/conversion_procs all
gmake[3]: Entering directory
`/usr1/clarka/src/postgresql-8.0.2/src/backend/utils/mb/conversion_procs
'
gmake[4]: Entering directory
`/usr1/clarka/src/postgresql-8.0.2/src/backend/utils/mb/conversion_procs
/ascii_and_mic'
cc -Xa -v -xtarget=ultra2 -xarch=v9a -xchip=ultra2 -DSUNOS4_CC -KPIC
-I../../../../../../src/include -c -o ascii_and_mic.o ascii_and_mic.c
"ascii_and_mic.c", line 38: warning: assignment type mismatch:
pointer to unsigned char "=" pointer to char
"ascii_and_mic.c", line 39: warning: assignment type mismatch:
pointer to unsigned char "=" pointer to char
"ascii_and_mic.c", line 54: warning: assignment type mismatch:
pointer to unsigned char "=" pointer to char
"ascii_and_mic.c", line 55: warning: assignment type mismatch:
pointer to unsigned char "=" pointer to char
cc -Xa -G -h libascii_and_mic.so.0 ascii_and_mic.o
-L../../../../../../src/port -o libascii_and_mic.so.0.0
ld: fatal: file ascii_and_mic.o: wrong ELF class: ELFCLASS64
ld: fatal: File processing errors. No output written to
libascii_and_mic.so.0.0
gmake[4]: *** [libascii_and_mic.so.0.0] Error 1
gmake[4]: Leaving directory
`/usr1/clarka/src/postgresql-8.0.2/src/backend/utils/mb/conversion_procs
/ascii_and_mic'

----------------

The command that's failing doesn't have the -x flags, so I tried the
following command manually:

cc -Xa -v -xtarget=ultra2 -xarch=v9a -xchip=ultra2 -G -h
libascii_and_mic.so.0 ascii_and_mic.o -L../../../../../../src/port -o
libascii_and_mic.so.0.0

And it worked...

So, I specified the -x flags in LDFLAG in my configure command. But I
still get the above error and I the LDFLAG isn't being used.

Here is my configure command:

./configure \
CFLAGS="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2 -DSUNOS4_CC" \
LDFLAGS="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2" \
--without-readline \
--without-zlib \
--prefix=$HOME/UAT/postgres \
--enable-thread-safety \
--enable-integer-datetimes


Thanks,
Andrew


-----Original Message-----
From: Tom Lane
Sent: July 29, 2005 10:37 AM
To: Clark, Andrew
Cc: Peter Eisentraut; pgsql-ports@postgresql.org
Subject: Re: [PORTS] Solaris SPARC - Sun compiler 5.5 - 64 bit

"Clark, Andrew" <Andrew.Clark@fnf.com> writes:
Is configure using the solaris template, or the sunos4 one?
If the former, does changing
CFLAGS="-v" # -v is like gcc -Wall
to
CFLAGS="-v -DSUNOS4_CC" # -v is like gcc -Wall
fix it? (Note you need to re-configure to make template changes have
any effect.)

regards, tom lane

OK, thanks for the confirmation. There was some discussion recently
of whether we shouldn't add -DSUNOS4_CC to the solaris template,
but we didn't have proof.
AFAIK you have to specify both LDFLAGS and LDFLAGS_SL if you want
manual configuration of link options. LDFLAGS is used for building
regular executables, LDFLAGS_SL for shared libraries.

regards, tom lane

Actually, this doesn't prove anything one way or the other. Because you
were manually specifying CFLAGS, and thereby overriding the template,
the lack of -DSUNOS4_CC in the manual CFLAGS caused the s_lock.c
failure.

So: which template did configure pick on your machine, anyway?
(This is reported in the first few lines of configure output)

regards, tom lane

Tom,

I added LDFLAGS_SL and everything compiled and all regression tests
passed.

Thanks!
Andrew

-----Original Message-----
From: Tom Lane
Sent: July 29, 2005 11:26 AM
To: Clark, Andrew
Cc: Peter Eisentraut; pgsql-ports@postgresql.org
Subject: Re: [PORTS] Solaris SPARC - Sun compiler 5.5 - 64 bit

"Clark, Andrew" <Andrew.Clark@fnf.com> writes:
OK, thanks for the confirmation. There was some discussion recently of
whether we shouldn't add -DSUNOS4_CC to the solaris template, but we
didn't have proof.
AFAIK you have to specify both LDFLAGS and LDFLAGS_SL if you want manual
configuration of link options. LDFLAGS is used for building regular
executables, LDFLAGS_SL for shared libraries.

regards, tom lane

Quick Summary:

1.

With the follow config:

./configure \
--without-readline \
--without-zlib \
--prefix=$HOME/UAT/postgres \
--enable-thread-safety \
--enable-integer-datetimes \

I have the following config:

configure:1298: checking build system type
configure:1316: result: sparc-sun-solaris2.8
configure:1324: checking host system type
configure:1338: result: sparc-sun-solaris2.8
configure:1348: checking which template to use
configure:1449: result: solaris

2.

With the following config:

./configure \
CFLAGS="-v -DSUNOS4_CC" \
--without-readline \
--without-zlib \
--prefix=$HOME/UAT/postgres \
--enable-thread-safety \
--enable-integer-datetimes

I have the following config:

configure:1298: checking build system type
configure:1316: result: sparc-sun-solaris2.8
configure:1324: checking host system type
configure:1338: result: sparc-sun-solaris2.8
configure:1348: checking which template to use
configure:1449: result: solaris

3.

With the following config:

./configure \
CFLAGS="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2 -DSUNOS4_CC" \
LDFLAGS="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2" \
LDFLAGS_SL="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2" \
--without-readline \
--without-zlib \
--prefix=$HOME/UAT/postgres \
--enable-thread-safety \
--enable-integer-datetimes

I have the following config:

configure:1298: checking build system type
configure:1316: result: sparc-sun-solaris2.8
configure:1324: checking host system type
configure:1338: result: sparc-sun-solaris2.8
configure:1348: checking which template to use
configure:1449: result: solaris

4.

./configure \
CFLAGS="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2" \
LDFLAGS="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2" \
LDFLAGS_SL="-v -xtarget=ultra2 -xarch=v9a -xchip=ultra2" \
--without-readline \
--without-zlib \
--prefix=$HOME/UAT/postgres \
--enable-thread-safety \
--enable-integer-datetimes

I have the following config:

configure:1298: checking build system type
configure:1316: result: sparc-sun-solaris2.8
configure:1324: checking host system type
configure:1338: result: sparc-sun-solaris2.8
configure:1348: checking which template to use
configure:1449: result: solaris


NOTES:

1 and 2 can compile and they both produce 32bit bins and libs.
3 can compile and it produces 64bit bins and libs.
4 can't compile (fbe fails)


-----Original Message-----
From: Tom Lane
Sent: July 29, 2005 12:18 PM
To: Clark, Andrew
Cc: Peter Eisentraut; pgsql-ports@postgresql.org
Subject: Re: [PORTS] Solaris SPARC - Sun compiler 5.5 - 64 bit

I wrote:
Actually, this doesn't prove anything one way or the other. Because you
were manually specifying CFLAGS, and thereby overriding the template,
the lack of -DSUNOS4_CC in the manual CFLAGS caused the s_lock.c
failure.

So: which template did configure pick on your machine, anyway?
(This is reported in the first few lines of configure output)

regards, tom lane

OK, so we do need to fix the solaris template too.

I have a feeling that there may need to be some sort of test on the
compiler version specifically, but without some more data it's hard
to say what. For the moment I'll just put it into template/solaris
unconditionally.

regards, tom lane

Let me know if you want to test something...


-----Original Message-----
From: Tom Lane
Sent: July 29, 2005 12:58 PM
To: Clark, Andrew
Cc: Peter Eisentraut; pgsql-ports@postgresql.org
Subject: Re: [PORTS] Solaris SPARC - Sun compiler 5.5 - 64 bit

"Clark, Andrew" <Andrew.Clark@fnf.com> writes:
OK, so we do need to fix the solaris template too.

I have a feeling that there may need to be some sort of test on the
compiler version specifically, but without some more data it's hard to
say what. For the moment I'll just put it into template/solaris
unconditionally.

regards, tom lane