I haven't tried building postgres with the Watcom compiler for 7.1 because

it does not support 64bit integers. AFAIK that's the default data type for
OIDs, so I am not surprised that this does not work. Use gcc instead.
BTW., 7.1 does not compile as is with gcc either, I believed the
required patches made it into the 7.1.1 release but obviously I missed
the deadline.
Since the ports mailing list does not seem to be archived I have attached
a copy of the patch (for 7.1 and 7.1.1).

I've just performed a build of a Watcom compiled version and found a couple
of bugs in the watcom specific part of that patch. Please use the attached
version instead.

Tegge, Bernd

Author: bmomjian

doc/FAQ_QNX4

@@ -1,53 +1,76 @@
 PostgresSQL on QNX 4
 --------------------
+last updated:           $Date: 2001/05/24 15:53:31 $
+
+current maintainer:     Bernd Tegge (tegge@repas-aeg.de)
+original author:        Andreas Kardos (kardos@repas-aeg.de)
 
 This port is an important step because PostgreSQL is now the only free
 relational database with full SQL and ODBC support available for QNX 4.
 The only commercial databases available are Empress RDBMS and Velocis
 Database Server (not supported for Digital Unix).
 
 The most effort required the emulation of System V semaphore sets,
-shared memory and IPC and of some IEEE floating-point functionality. For
-the Watcom compiler spinlocks have been implemented using POSIX semaphores.
+shared memory and IPC and of some IEEE floating-point functionality.
 
-It is recomended to use the GNU C compiler instead of the Watcom compiler
+It is recommended to use the GNU C compiler instead of the Watcom compiler
 because the Watcom compiler doesn't support a int8 datatype (long or
-long long int) and it does not have a C++ frontend.
-It could be managed to build postgres using the Watcom compiler but it crashes
-quite early. These problems probably could be solved. The only advantage would
-be support of Tk and pgaccess. For the Watcom compiler some more modifications
-have to be made which are not described here.
+long long int) and it does not have a C++ frontend. The only advantage
+using Watcom C would be support of Tk and pgaccess.
 
 QNX 4 does not offer native support of shared libraries. Therefore the related
 functionality cannot be used. Shared library support could probably be
 implemented in future.
 
-QNX 4 does not support UNIX domain sockets. Connections can only be made
-with TCP/IP sockets. Therefore postmaster must always be started with the
--i option. Furthermore it would be useful to set the PGHOST variable.
+QNX 4 does not support UNIX domain sockets. Clients must use TCP/IP sockets.
+Therefore postmaster must always be started with the -i option. Furthermore
+it would be useful to set the PGHOST variable.
 
 Prerequisites:
 --------------
 
 The following prerequisites have been used:
 
-QNX 4.25
-Watcom C 10.6
-GNU make
+QNX standard:
+QNX 4.25, Watcom C 10.6, GNU make
+
+Available from http://www.teaser.fr/~jcmichot/
 flex-2.5.4a.tar.gz (flex)
-egcs111-qnx4-r20.tar[.gz] (GNU C)
+egcs-112-qnx4-r20.tar[.gz] (GNU C)
+
+Available from ftp://ftp.visi.com/users/hawkeyd/qnx/
+bison-1.2.5.tar.gz
+
+Available from http://quics.qnx.com/cgi-bin/dir_find.cgi?/usr/free/
 perl 5.004_04
+
+from ftp://ftp.freesoftware.com/pub/infozip/zlib/
+zlib 1.1.3
+
 tcl8.0.3.tar.Z (Tcl)
 
+In order to achieve meaningful results for the regression tests, you need
+a different shell. The standard QNX shell is quite broken with regard to
+scripting (even QSSL suggested to use pdksh or bash). Both are available
+in source and binaries at
+http://quics.qnx.com/cgi-bin/dir_find.cgi?/usr/free/
+
 flex can be built without any problems using the Watcom or GNU C compilers.
 
+The current version of bison (1.2.8 at the time of this writing) does not
+build cleanly on QNX4 with gcc installed, but you can download src and
+binary of a previous version from the above mentioned ftp site.
+
+download source
 GNU C can be installed by
-/etc/install -u egcs111-qnx4-r20.tar[.gz]
+/etc/install -u egcs112-qnx4-r20.tar[.gz]
 
 Although Tcl can be built after some slight code changes using the Watcom
 compiler, the GNU C compiler should be used instead. Otherwise it would be
 impossible to link postgres built with GNU C together with libtcl8.0.a built
-with the Watcom compiler.
+with the Watcom compiler. However, if you are going to build a Watcom version
+you must build the tcl and tk libraries with Watcom too.
+
 To make unix/tclUnixTest.c compilable uncomment
 #include <sys/resource.h>
 in it or
@@ -79,17 +102,14 @@ ln -s /usr/src/pgsql/src/backend/port/qnx4/sem.h /usr/include/sys/sem.h
 ln -s /usr/src/pgsql/src/backend/port/qnx4/shm.h /usr/include/sys/shm.h
 
 For ecgs-2.91.60 the file
-/usr/local/lib/gcc-lib/i386-pc-qnx4/egcs-2.91.60/include/g++/stl_alloc.h
+/usr/local/lib/gcc-lib/i386-pc-qnx4/egcs-2.91.66/include/g++/stl_alloc.h
 had to be patched (extern "C++" { ... } // extern "C++").
 See attached patch.
 
-If you use another ecpg version the include/g++ path in Makefile.qnx4
-(CXXFLAGS) has to be adopted.
-
 If all prerequisites are available postgres can be built and installed by
 cd pgsql/src
-./configure --with-tcl --with-perl --with-x
-gmake all > make.log 2>&1
+./configure --with-tcl --without-tk --with-perl
+gmake > make.log 2>&1
 gmake install > make.install.log 2>&1
 su
 cd interfaces/perl5
@@ -113,11 +133,15 @@ Currently yacc fails on backend/parser/gram.y and
 interfaces/ecpg/preproc/preproc.y due to exceeded maximum table size. You can
 generate the gram.h, parse.h, preproc.h and preproc.c files on another platform
 and use them. This is only a problem when you use the current source tree since
-parse.h, gram.c, preproc.h and preporc.c are included in official
-distributions.
+parse.h, gram.c, preproc.h and preproc.c are included in official
+distributions. In order to avoid this, install the bison binary mentioned
+above.
 
 Regression tests:
 -----------------
+please don't use the QNX shell for the parallel regression tests. Download
+bash from quics.qnx.com and run 'make SHELL=/usr/local/bin/bash check'
+instead.
 
 The majority of regression tests succeeded. The following tests failed:
 
@@ -136,7 +160,8 @@ if a value of type money is inserted in string representation e.g. as in
 "insert into rtest_emp values ('wiech', '5000.00');"
 Subject of further investigation.
 
-create_function_2, triggers, misc, plpgsql:
+create_function_1, create_function_2, create_type, create_operator,
+create_view, select_views, triggers, misc, plpgsql:
 Error messages due to the lack of shared library support.
 
 numeric, numeric_big, sanity_check:
@@ -147,14 +172,19 @@ Subject of further investigation. Probably because of the missing indices
 these numeric tests take a long time.
 The diffence in sanity_check.out is a consequence of this problem only.
 
-The reached state of this port should be sufficient for lot of applications.
+The current state of this port should be sufficient for lot of applications.
+
+The Watcom version failed additional tests because of the lack of int8
+support.
 
 Have fun!
 
 Andreas Kardos
-kardos@repas-aeg.de
 2000-04-04
 
+Bernd Tegge
+tegge@repas-aeg.de
+2001-03-01
 
 ---------------------------------------------------------------------------
 

src/backend/Makefile

@@ -4,7 +4,7 @@
 #
 # Copyright (c) 1994, Regents of the University of California
 #
-# $Header: /cvsroot/pgsql/src/backend/Makefile,v 1.71 2001/04/23 20:27:55 petere Exp $
+# $Header: /cvsroot/pgsql/src/backend/Makefile,v 1.72 2001/05/24 15:53:32 momjian Exp $
 #
 #-------------------------------------------------------------------------
 
@@ -25,7 +25,8 @@ OBJS := $(DIRS:%=%/SUBSYS.o)
 ifeq ($(PORTNAME), qnx4)
 # This file crashes qnx4's wlink and is therefore not in
 # bootstrap/SUBSYS.o on that platform. (Wotta hack ... is it still
-# necessary?)
+# necessary?) [ Yes, until the Watcom compiler goes open source it's
+# effectively unsupported ]
 OBJS+= bootstrap/bootstrap.o
 endif
 

src/backend/port/qnx4/sem.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/port/qnx4/Attic/sem.c,v 1.4 2001/02/02 18:21:58 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/port/qnx4/Attic/sem.c,v 1.5 2001/05/24 15:53:33 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -23,14 +23,16 @@
 #include "storage/ipc.h"
 #include "storage/proc.h"
 #include <sys/sem.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
 
 
-#define SETMAX	((MAXBACKENDS + PROC_NSEMS_PER_SET - 1) / PROC_NSEMS_PER_SET)
+#define SETMAX	((MAXBACKENDS + PROC_NSEMS_PER_SET + 1) / PROC_NSEMS_PER_SET)
 #define SEMMAX	(PROC_NSEMS_PER_SET+1)
 #define OPMAX	8
 
 #define MODE	0700
-#define SHM_INFO_NAME	"SysV_Sem_Info"
+#define SHM_INFO_NAME	"PgSysV_Sem_Info"
 
 
 struct pending_ops
@@ -56,6 +58,17 @@ struct sem_info
 
 static struct sem_info *SemInfo = (struct sem_info *) - 1;
 
+/* ----------------------------------------------------------------
+ * semclean - remove the shared memory file on exit
+ *            only called by the process which created the shm file
+ * ----------------------------------------------------------------
+ */
+
+static void
+semclean( void )
+{
+  remove( "/dev/shmem/" SHM_INFO_NAME );
+}
 
 int
 semctl(int semid, int semnum, int cmd, /* ... */ union semun arg)
@@ -132,6 +145,7 @@ semget(key_t key, int nsems, int semflg)
 				semid,
 				semnum /* , semnum1 */ ;
 	int			exist = 0;
+  struct stat statbuf;
 
 	if (nsems < 0 || nsems > SEMMAX)
 	{
@@ -153,6 +167,26 @@ semget(key_t key, int nsems, int semflg)
 			return fd;
 		/* The size may only be set once. Ignore errors. */
 		ltrunc(fd, sizeof(struct sem_info), SEEK_SET);
+    if ( fstat( fd, &statbuf ) ) /* would be strange : the only doc'ed */
+    {                            /* error is EBADF */
+      close( fd );
+      return -1;
+    }
+    /*
+     * size is rounded by proc to the next __PAGESIZE
+     */
+    if ( statbuf.st_size != 
+         ((( sizeof(struct sem_info) /__PAGESIZE)+1) * __PAGESIZE) )
+    {
+       fprintf( stderr,
+         "Found a pre-existing shared memory block for the semaphore memory\n"
+         "of a different size (%ld instead %ld). Make sure that all executables\n"
+         "are from the same release or remove the file \"/dev/shmem/%s\"\n"
+         "left by a previous version.\n", statbuf.st_size,
+         sizeof(struct sem_info), SHM_INFO_NAME);
+         errno = EACCES;
+       return -1;
+    }
 		SemInfo = mmap(NULL, sizeof(struct sem_info),
 					   PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 		if (SemInfo == MAP_FAILED)
@@ -167,6 +201,7 @@ semget(key_t key, int nsems, int semflg)
 			for (semid = 0; semid < SETMAX; semid++)
 				SemInfo->set[semid].key = -1;
 			sem_post(&SemInfo->sem);
+      on_proc_exit( semclean, NULL );
 		}
 	}
 

src/backend/port/qnx4/shm.c

@@ -7,17 +7,19 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/port/qnx4/Attic/shm.c,v 1.4 2001/03/22 03:59:43 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/port/qnx4/Attic/shm.c,v 1.5 2001/05/24 15:53:33 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
 
 #include <errno.h>
 #include <fcntl.h>
 #include <stdlib.h>
+#include <string.h>
 #include <unistd.h>
 #include <sys/mman.h>
 #include <sys/shm.h>
+#include <sys/stat.h>
 
 
 #define MODE	0777
@@ -41,6 +43,12 @@ static int	shm_updinfo(int i, struct shm_info * info);
 static int	shm_getinfo(int shmid, struct shm_info * info);
 static int	shm_getinfobyaddr(const void *addr, struct shm_info * info);
 
+static char *
+keytoname(key_t key, char *name)
+{
+  sprintf( name,"PgShm%x", key );
+  return name;
+}
 
 static int
 shm_putinfo(struct shm_info * info)
@@ -172,25 +180,51 @@ shmctl(int shmid, int cmd, struct shmid_ds * buf)
 {
 	struct shm_info info;
 	char		name[NAME_MAX + 1];
+  int     result;
+  int     fd;
+  struct stat statbuf;
 
-	if (cmd == IPC_RMID)
+  
+	switch( cmd )
 	{
+    case IPC_RMID :
 		if (shm_getinfo(shmid, &info) == -1)
 		{
 			errno = EACCES;
 			return -1;
 		}
-		return shm_unlink(itoa(info.key, name, 16));
+      close( info.shmid );
+      keytoname(info.key, name);
+      return shm_unlink( name );
+      
+    case IPC_STAT :
+      /*
+       * we have to open it first. stat() does no prefix tracking
+       * -> the call would go to fsys instead of proc
+       */
+      keytoname(shmid, name);
+      fd = shm_open( name, 0, MODE );
+      if ( fd >= 0 )
+      {
+        result = fstat( fd, &statbuf );
+        /*
+         * if the file exists, subtract 2 from linkcount :
+         *  one for our own open and one for the dir entry
+         */
+        if ( ! result )
+          buf->shm_nattch = statbuf.st_nlink-2;
+        close( fd );
+        return result;
 	}
-	if (cmd == IPC_STAT)
+      else
 	{
-
 		/*
-		 * Can we support IPC_STAT?  We only need shm_nattch ... For now,
-		 * punt and assume the shm seg does not exist.
+         * if there's no entry for this key it doesn't matter
+         * the next shmget() would get a different shm anyway
 		 */
-		errno = EINVAL;
-		return -1;
+        buf->shm_nattch = 0;
+        return 0;
+      }  
 	}
 	errno = EINVAL;
 	return -1;
@@ -214,16 +248,21 @@ shmget(key_t key, size_t size, int flags)
 		else
 			oflag |= O_RDONLY;
 	}
-	info.shmid = shm_open(itoa(key, name, 16), oflag, MODE);
+	info.shmid = shm_open(keytoname(key, name), oflag, MODE);
 
 	/* store shared memory information */
 	if (info.shmid != -1)
 	{
 		info.key = key;
 		info.size = size;
 		info.addr = NULL;
-		if (shm_putinfo(&info) == -1)
+		if (shm_putinfo(&info) == -1) {
+      close( info.shmid );
+      if ( (oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL) ) {
+        shm_unlink( name );
+      }
 			return -1;
+    }
 	}
 
 	/* The size may only be set once. Ignore errors. */