Created by Bob McElrath (mcelrath@isp.nwu.edu) on Saturday, August 8, 1998 8:26:05 AM PDT Additional Details : include/mcom_db.h needs system include endian.h, which is in /usr/sys/include/arch/alpha. Can be fixed by adding the line: OS_INCLUDES = -I/usr/sys/include/arch/alpha to config/OSF1.mk also: running gmake in the dbm/src directory (after fixing the above problem): cc -ieee_with_inexact -std1 -o OSF1V4.0_OPT.OBJ/h_page.o -c -DXP_UNIX -O -Olimit 4000 -taso -D_ALPHA_ -DIS_64 -DOSF1 -DOSF1V4 -D_REENTRANT -DHAVE_LCHOWN -DNEED_CDEFS_H -DNTOHL_ENDIAN_H -DNEED_IOCTL_H -DMACHINE_ENDIAN_H -DHAVE_VA_LIST_STRUCT -DNEED_BYTE_ALIGNMENT -DMITSHM -DNEED_REALPATH -DHAVE_WAITID -DNEED_H_ERRNO -DNEED_SYS_TIME_H -DHAVE_SYSTEMINFO_H -DNEED_SYS_PARAM_H -DHAVE_INT32_T -DODD_VA_START -DHAVE_REMAINDER -DSW_THREADS -UDEBUG -DNDEBUG -DTRIMMED -DNSPR20 -DNETSCAPE -DOSTYPE=\"OSF1V4\" -DMOZILLA_CLIENT -DEDITOR -DMOZ_COMMUNICATOR_IIDS -DNO_SECURITY -DEDITOR -DEDITOR_UI -DLAYERS -DUNIX_EMBED -DX_PLUGINS -DUNIX_LDAP -DNSPR -DMOCHA -DUNIX_ASYNC_DNS -DDEVELOPER_DEBUG -DMEMMOVE -D__DBINTERFACE_PRIVATE -DNSPR20=1 -I../../include -I/usr/sys/include/arch/alpha -I../../dist/OSF1V4.0_OPT.OBJ/include -I../include h_page.c cc: Error: h_page.c, line 1160: In this statement, "SIG_BLOCK" is not declared. (void)sigprocmask(SIG_BLOCK, &set, &oset); --------------------------^ cc: Error: h_page.c, line 1171: In this statement, "SIG_SETMASK" is not declared. (void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL); --------------------------^ gmake: *** [OSF1V4.0_OPT.OBJ/h_page.o] Error 1 I can clearly see that signal.h is included, and that SIG_BLOCK and SIG_SETMASK are defined. I suspect something odd is happening in the 1000-odd lines between #include <signal.h> and the lines on which the error occurs. running: cpp h_page.c -I../../include -I/usr/sys/include/arch/alpha -I../../dist/OSF1V4.0_OPT.OBJ/include -I../include | grep SIG gives: (void)sigprocmask(SIG_BLOCK, &set, &oset); (void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)0L); So clearly the pre-processor is failing. (Fails for both DEC's cpp and gcc -E version 2.7.2.3)

I've taken care of the mozilla/config/OSF1.mk change. Our DEC machines all have /usr/include/alpha linked to /usr/sys/include/arch/alpha, so I hope that will work for you. The rest of this bug (I think) belongs to Lou.

I've taken care of the mozilla/config/OSF1.mk change. Our DEC machines all have /usr/include/alpha linked to /usr/sys/include/arch/alpha, so I hope that will work for you. The rest of this bug (I think) belongs to Lou.

Was lou's bug, assigning to dp for lack of anyone better.

Setting all current Open Critical and Major to M3

brian is all over the build situation. also, since this bug is pretty old, it might be INVALID ...

I think this is invalid now. I still can't build successfully on OSF/1 4.0D, but this problem doesn't seem to have anything to do with the current incarnation of the Mozilla tree. BTW: There is a (test) Tinderbox build for this platform running at: http://cvs-mirror.mozilla.org/webtools/tinderbox/showbuilds.cgi?tree=MozillaTest (if anyone cares....)

Marking Verified/Invalid.

I'm sorely tempted to reopen this bug. I experience it myself right now. When I attempt to build dbm on OSF/1, I get the same error reported in this bug, namely that SIG_BLOCK and SIG_SETMASK are not defined. If someone can build mozilla's DBM on OSF1V4.0, I'd like to see what their cc command looks like that builds h_page.c The problem is that in DEC's <signal.h> file, these symbols are only defined if _POSIX_C_SOURCE >= 1. That symbols is not defined in our builds. Defining it causes many other things to break. The "uint" type ceases to be defined, for example. BTW, Netscape/iPlanet's copy of the dbm sources contain a hack which works around the problem when OSF1V4D is defined by simply defining the two missing symbols. That hack is not in the mozilla tree. I can build iPlanet's copy of the dbm sources on OSF1, but not mozilla's copy.

Proposed patch to dbm/src/h_page.c follows. This is essentially the same hack now in iPlanet's dbm sources. dbm builds for me on OSF1 with this patch. =================================================================== RCS file: /cvsroot/mozilla/dbm/src/h_page.c,v retrieving revision 3.13 diff -r3.13 h_page.c 109a110,119 > #if defined(OSF1V4D) && !defined(SIG_BLOCK) > /* OSF1 V4.0D's signal.h only defines these symbols when _POSIX_C_SOURCE >= 1 > * Defining _POSIX_C_SOURCE breaks lots of other stuff. > * Including ncompat.h causes multiply defined symbols. > * So, we define them here for OSF1, even though we'd rather not. > */ > #define SIG_BLOCK 1 > #define SIG_SETMASK 3 > #endif >

Nelson, That's the wrong fix. The problem is the -I/usr/sys/include/arch/alpha compiler option. That was an attempt to make #include <endian.h> work on OSF1. The correct way is to say #include <machine/endian.h>, which has been checked in mozilla/dbm/include/mcom_db.h. Use this patch. Index: mozilla/security/dbm/src/Makefile =================================================================== RCS file: /cvsroot/mozilla/security/dbm/src/Makefile,v retrieving revision 1.2 diff -u -r1.2 Makefile --- Makefile 2001/05/08 23:31:07 1.2 +++ Makefile 2001/06/06 05:55:29 @@ -66,10 +66,6 @@ # (6) Execute "component" rules. (OPTIONAL) # ####################################################################### -ifeq ($(OS_CONFIG),OSF1V4.0D) -INCLUDES += -I/usr/sys/include/arch/alpha -endif - ####################################################################### # (7) Execute "local" rules. (OPTIONAL). # #######################################################################

Thanks Wan-Teh, I checked your change in on the trunk.