Date: Wed, 22 Dec 2004 00:04:18 EST-10EDT,10,-1,0,7200,3,-1,0,7200,3600 Subject: [UnixOS2_Archive] No. 480 ************************************************** Tuesday 21 December 2004 Number 480 ************************************************** Subjects for today 1 Re: postgreSQL v8.0 rc1 beta is released : lsunley at mb.sympatico.ca 2 Re: Openjade - GCC 3.3.5 - and a dumb question about linking : Arnd Hanses" 3 Re: postgreSQL v8.0 rc1 beta is released : John Poltorak 4 Re: postgreSQL v8.0 rc1 beta is released : lsunley at mb.sympatico.ca 5 Re: postgreSQL v8.0 rc1 beta is released : John Poltorak 6 Re: postgreSQL v8.0 rc1 beta is released : Dave Yeo" 7 Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released : lsunley at mb.sympatico.ca 8 Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released : Dave Yeo" 9 Debugging DLLs : Dave Yeo" 10 Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released : lsunley at mb.sympatico.ca 11 Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released : Dave Yeo" 12 Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released : John Poltorak 13 Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released : Knut St. Osmundsen" **= Email 1 ==========================** Date: Mon, 20 Dec 2004 07:50:45 -0500 From: lsunley at mb.sympatico.ca Subject: Re: postgreSQL v8.0 rc1 beta is released Hi Answers below In <20041220111623.G48 at warpix.org>, on 12/20/04 at 11:16 AM, John Poltorak said: >On Sun, Dec 19, 2004 at 07:41:59PM -0500, lsunley at mb.sympatico.ca wrote: >> Hi All >> >> For anyone interested, I put a beta of postgreSQL v8.0 rc1 on Hobbes. >Just gave it a quick try... >initdb worked OK, but when running postmaster, I get:- >[G:\pgsql\bin]postmaster >LOG: select() failed in statistics collector: Invalid argument LOG: >disabling statistics collector for lack of working socket The above always happens - some kind of bug.. >Process terminated by SIGILL - pid=0x0b07 - tid=0x01 - >G:\PGSQL\BIN\POSTMASTER.EXE That's ugly - I don't know... >What have I overlooked? -- ----------------------------------------------------------- lsunley at mb.sympatico.ca ----------------------------------------------------------- **= Email 2 ==========================** Date: Thu, 16 Dec 04 13:38:53 +0100 From: "Arnd Hanses" Subject: Re: Openjade - GCC 3.3.5 - and a dumb question about linking Hi, On Sun, 12 Dec 2004 11:05:49 -0500, lsunley at mb.sympatico.ca wrote: >The configure for opensp fails under gcc 3.3.5 when it compliles a simple >test for size_t using g++. (code is attached two files conftest.c and >conftest.cc). The .c and .cc files are identical other than the file name. > >When the program (.cc) is compiled with g++ the libc06a3 dll is linked in. >When the program (.c) is compiled with gcc the program works. > >Lorne I suspect the failure is only too correct, if we look at conftest.cc (a test for broken stack/register return value passing, catching bad values in case of integer overflow, impicit conversion, or whatever ??): (...) unsigned long ulongval () { return (long) (sizeof (size_t)); } ^^^^^^^^^^^^ ^^^^^^ (...) Any decent C++-Compiler will not tolerate incompatible return types that might lead to data loss. g++ is such a decent C++-Compiler ! On the other hand, ANSI C guarantees for the implicit conversion of function return values, even if this might lead to data loss. Greets, Arnd **= Email 3 ==========================** Date: Mon, 20 Dec 2004 14:26:01 +0000 From: John Poltorak Subject: Re: postgreSQL v8.0 rc1 beta is released On Mon, Dec 20, 2004 at 07:50:45AM -0500, lsunley at mb.sympatico.ca wrote: > > >Process terminated by SIGILL - pid=0x0b07 - tid=0x01 - > >G:\PGSQL\BIN\POSTMASTER.EXE > > That's ugly - I don't know... > > >What have I overlooked? Any suggestions about what I can/should try? Are there any debugging options or logs to look at? > > > > -- > ----------------------------------------------------------- > lsunley at mb.sympatico.ca > ----------------------------------------------------------- > -- John **= Email 4 ==========================** Date: Mon, 20 Dec 2004 19:11:47 -0500 From: lsunley at mb.sympatico.ca Subject: Re: postgreSQL v8.0 rc1 beta is released Hi Sorry for the too short note this morning, I was called away and pressed the "send" instead of saving it for later. If you run postmaster with a -d2 or -d3 switch it might show what is happening prior to the SIGKILL What version of Warp are you running on? I compiled it with OBJ_ANY to allocate memory in the high (above 512 MB) memory area. That might die on some systems. Is there a popupos2.log entry? Lorne In <20041220142601.A47 at warpix.org>, on 12/20/04 at 02:26 PM, John Poltorak said: >On Mon, Dec 20, 2004 at 07:50:45AM -0500, lsunley at mb.sympatico.ca wrote: >> >> >Process terminated by SIGILL - pid=0x0b07 - tid=0x01 - >> >G:\PGSQL\BIN\POSTMASTER.EXE >> >> That's ugly - I don't know... >> >> >What have I overlooked? >Any suggestions about what I can/should try? >Are there any debugging options or logs to look at? >> >> >> >> -- >> ----------------------------------------------------------- >> lsunley at mb.sympatico.ca >> ----------------------------------------------------------- >> -- ----------------------------------------------------------- lsunley at mb.sympatico.ca ----------------------------------------------------------- **= Email 5 ==========================** Date: Tue, 21 Dec 2004 02:16:58 +0000 From: John Poltorak Subject: Re: postgreSQL v8.0 rc1 beta is released On Mon, Dec 20, 2004 at 07:50:45AM -0500, lsunley at mb.sympatico.ca wrote: > > The above always happens - some kind of bug.. > > > >Process terminated by SIGILL - pid=0x0b07 - tid=0x01 - > >G:\PGSQL\BIN\POSTMASTER.EXE > > That's ugly - I don't know... Hmmm... Similar result when running the new NTPD:- http://hobbes.nmsu.edu/pub/new/ntp-4.2.0-b2-os2-emx.zip G:\eval\ntp>ntpd.exe Process terminated by SIGILL - pid=0x01bb - tid=0x01 - G:\EVAL\NTP\NTPD.EXE This looks gcc/libc related. Any hints as to what to look for Knut? > > -- > ----------------------------------------------------------- > lsunley at mb.sympatico.ca > ----------------------------------------------------------- > -- John **= Email 6 ==========================** Date: Mon, 20 Dec 2004 18:46:42 -0800 From: "Dave Yeo" Subject: Re: postgreSQL v8.0 rc1 beta is released On Tue, 21 Dec 2004 02:16:58 +0000, John Poltorak wrote: > >Hmmm... > >Similar result when running the new NTPD:- > >http://hobbes.nmsu.edu/pub/new/ntp-4.2.0-b2-os2-emx.zip > > >G:\eval\ntp>ntpd.exe > >Process terminated by SIGILL - pid=0x01bb - tid=0x01 - >G:\EVAL\NTP\NTPD.EXE > > >This looks gcc/libc related. I was getting similar results with a few things that I traced down to DLL conflicts. Some other things are also doing this for I don't know why. Dave **= Email 7 ==========================** Date: Mon, 20 Dec 2004 21:42:34 -0500 From: lsunley at mb.sympatico.ca Subject: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released In <20041221024632.11083BDB6A at generation.lgisp.net>, on 12/20/04 at 06:46 PM, "Dave Yeo" said: >On Tue, 21 Dec 2004 02:16:58 +0000, John Poltorak wrote: >> >>Hmmm... >> >>Similar result when running the new NTPD:- >> >>http://hobbes.nmsu.edu/pub/new/ntp-4.2.0-b2-os2-emx.zip >> >> >>G:\eval\ntp>ntpd.exe >> >>Process terminated by SIGILL - pid=0x01bb - tid=0x01 - >>G:\EVAL\NTP\NTPD.EXE >> >> >>This looks gcc/libc related. >I was getting similar results with a few things that I traced down to DLL >conflicts. Some other things are also doing this for I don't know why. >Dave Thanks... When I run the thing I have the GCC 3.3.5 stuff at the front of path and libpath. When I test on a eCS system I don't set anything in beginlibpath or endlibpath and have .\. in LIBPATH. There also appears to be a bug in GCC3.3.t setlocale(LC_MESSAGES, NULL) always returns a null string rather than the messages language setting. I don't know if that is a GCC bug or postgreSQL does some bad things when it finds that and does not recognize the set LANG value. When set lang=de_DE initdb will work fine. When set LAN=de_DE_EURO initdb fails... Is that trailing _EURO part of the standard for language settings? Lorne -- ----------------------------------------------------------- lsunley at mb.sympatico.ca ----------------------------------------------------------- **= Email 8 ==========================** Date: Mon, 20 Dec 2004 21:33:07 -0800 From: "Dave Yeo" Subject: Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released On Mon, 20 Dec 2004 21:42:34 -0500, lsunley at mb.sympatico.ca wrote: >>I was getting similar results with a few things that I traced down to DLL >>conflicts. Some other things are also doing this for I don't know why. >>Dave > >Thanks... > >When I run the thing I have the GCC 3.3.5 stuff at the front of path and >libpath. > >When I test on a eCS system I don't set anything in beginlibpath or >endlibpath and have .\. in LIBPATH. Yes so do I usually. Lately I had the experience that I needed a lib in libpath and forgot. Anyways I got lots of sigills due to a simillary named dll. I doubt that it is your enviroment, I was actually thinking of John who seemed to be getting sigils that no one else was getting. Interestingly I'd never seen a sigil before GCC 3.3.5. I guess better error handling. > >There also appears to be a bug in GCC3.3.t setlocale(LC_MESSAGES, NULL) >always returns a null string rather than the messages language setting. I >don't know if that is a GCC bug or postgreSQL does some bad things when it >finds that and does not recognize the set LANG value. > >When set lang=de_DE initdb will work fine. When set LAN=de_DE_EURO initdb >fails... >Is that trailing _EURO part of the standard for language settings? Don't remember ever seeing that one. Dave **= Email 9 ==========================** Date: Mon, 20 Dec 2004 21:38:12 -0800 From: "Dave Yeo" Subject: Debugging DLLs Lately been playing a bit with SD386. Couple of questions How do you link to the debug version of libc? How to get sdd to follow a function call into a DLL (where the program is in src and the DLL is in lib)? Dave **= Email 10 ==========================** Date: Mon, 20 Dec 2004 23:57:49 -0500 From: lsunley at mb.sympatico.ca Subject: Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released Hi I can believe that the problem is something in John's environment. That is the only one I have a report on... Other problems have surfaced with set LANG values and one person has reported a really strange error where a table create fails because the table is already present, but nothing else on the SIGKIL front. I think the table create is a problem with a symlink. The failure is in postgreSQL code that is only enabled for symlinks. I don't know if it is kernel or file system related. The person is running a 14.062 level kernel. Lorne In <20041221053302.DF056BD0FA at generation.lgisp.net>, on 12/20/04 at 09:33 PM, "Dave Yeo" said: >On Mon, 20 Dec 2004 21:42:34 -0500, lsunley at mb.sympatico.ca wrote: >>>I was getting similar results with a few things that I traced down to DLL >>>conflicts. Some other things are also doing this for I don't know why. >>>Dave >> >>Thanks... >> >>When I run the thing I have the GCC 3.3.5 stuff at the front of path and >>libpath. >> >>When I test on a eCS system I don't set anything in beginlibpath or >>endlibpath and have .\. in LIBPATH. >Yes so do I usually. Lately I had the experience that I needed a lib in >libpath and forgot. Anyways I got lots of sigills due to a simillary >named dll. I doubt that it is your enviroment, I was actually thinking >of John who seemed to be getting sigils that no one else was getting. >Interestingly I'd never seen a sigil before GCC 3.3.5. I guess better >error handling. >> >>There also appears to be a bug in GCC3.3.t setlocale(LC_MESSAGES, NULL) >>always returns a null string rather than the messages language setting. I >>don't know if that is a GCC bug or postgreSQL does some bad things when it >>finds that and does not recognize the set LANG value. >> >>When set lang=de_DE initdb will work fine. When set LAN=de_DE_EURO initdb >>fails... >>Is that trailing _EURO part of the standard for language settings? >Don't remember ever seeing that one. >Dave I think it is supposed to set the currency symbol but I have to do more research... Lorne -- ----------------------------------------------------------- lsunley at mb.sympatico.ca ----------------------------------------------------------- **= Email 11 ==========================** Date: Mon, 20 Dec 2004 23:23:52 -0800 From: "Dave Yeo" Subject: Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released On Mon, 20 Dec 2004 23:57:49 -0500, lsunley at mb.sympatico.ca wrote: > >I think the table create is a problem with a symlink. The failure is in >postgreSQL code that is only enabled for symlinks. I don't know if it is >kernel or file system related. The person is running a 14.062 level >kernel. Well one thing I've noticed is there seems to be a problem with creating symlinks to symlinks. eg [H:\usr\local\bin]ls -l l* -rwxrwxrwx 1 0 0 29184 Dec 20 22:35 ln.exe -rwxrwxrwx 1 0 0 58913 Dec 20 22:37 ls.exe [H:\usr\local\bin]ln -s ln.exe link.exe [H:\usr\local\bin]ls -l l* -rwxrwxrwx 1 0 0 29184 Dec 20 22:35 link.exe -rwxrwxrwx 1 0 0 29184 Dec 20 22:35 ln.exe -rwxrwxrwx 1 0 0 58913 Dec 20 22:37 ls.exe [H:\usr\local\bin]ln -s link.exe link2.exe [H:\usr\local\bin]ls -l l* ls: link2.exe: Not a directory -rwxrwxrwx 1 0 0 29184 Dec 20 22:35 link.exe -rwxrwxrwx 1 0 0 29184 Dec 20 22:35 ln.exe -rwxrwxrwx 1 0 0 58913 Dec 20 22:37 ls.exe [H:\usr\local\bin]cmd /k dir l* Volume in drive H has no label. The Volume Serial Number is 2962:BC14. Directory of H:\usr\local\bin 12-20-04 10:41p 14 26 link.exe 12-20-04 10:42p 16 28 link2.exe 12-20-04 10:35p 29184 61 ln.exe 12-20-04 10:37p 58913 61 ls.exe 4 file(s) 88127 bytes used 3893438 K bytes free [H:\usr\local\bin]rm link2.exe rm: link2.exe: Not a directory This could well be a problem with the port of ls (I have another one that shows the symlink. This one shows it almost as a link) but I have seen other programs complaining when attempting to create a symlink to a symlink. Looking at the EAs link2.exe does link to link.exe Dave **= Email 12 ==========================** Date: Tue, 21 Dec 2004 11:15:41 +0000 From: John Poltorak Subject: Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released On Mon, Dec 20, 2004 at 11:57:49PM -0500, lsunley at mb.sympatico.ca wrote: > Hi > > I can believe that the problem is something in John's environment. That is > the only one I have a report on... You could well be right... I've just copied NTPD to another machine and the previous error disappears. I suspect the same will happen when I do the same with pgsql. The problem is knowing what to look for. I guess it must be due to some DLL conflict... > Lorne > -- > ----------------------------------------------------------- > lsunley at mb.sympatico.ca > ----------------------------------------------------------- > -- John **= Email 13 ==========================** Date: Tue, 21 Dec 2004 13:01:27 +0100 From: "Knut St. Osmundsen" Subject: Re: Possible GCC errors was Re: postgreSQL v8.0 rc1 beta is released lsunley at mb.sympatico.ca wrote: > In <20041221024632.11083BDB6A at generation.lgisp.net>, on 12/20/04 > at 06:46 PM, "Dave Yeo" said: > > >>On Tue, 21 Dec 2004 02:16:58 +0000, John Poltorak wrote: > > >>>Hmmm... >>> >>>Similar result when running the new NTPD:- >>> >>>http://hobbes.nmsu.edu/pub/new/ntp-4.2.0-b2-os2-emx.zip >>> >>> >>>G:\eval\ntp>ntpd.exe >>> >>>Process terminated by SIGILL - pid=0x01bb - tid=0x01 - >>>G:\EVAL\NTP\NTPD.EXE >>> >>> >>>This looks gcc/libc related. > > >>I was getting similar results with a few things that I traced down to DLL >>conflicts. Some other things are also doing this for I don't know why. >>Dave > > > Thanks... > > When I run the thing I have the GCC 3.3.5 stuff at the front of path and > libpath. > > When I test on a eCS system I don't set anything in beginlibpath or > endlibpath and have .\. in LIBPATH. > > There also appears to be a bug in GCC3.3.5 setlocale(LC_MESSAGES, NULL) > always returns a null string rather than the messages language setting. As far as I can tell it's returning the current locale for the LC_MESSAGES category, and that should work. The code updating it looks fine too... I will have to test this. setlocale() shouldn't return NULL unless the category is invalid. > I don't know if that is a GCC bug or postgreSQL does some bad things > when it finds that and does not recognize the set LANG value. > > When set lang=de_DE initdb will work fine. When set LANG=de_DE_EURO initdb > fails... > > Is that trailing _EURO part of the standard for language settings? No, there is no de_DE_EURO locale for libc afaik. LIBC is using the the OS/2 unicode library, which takes it's data from :\language. IIRC there are only a handful locales for the OS/2 unicode library, and they are all encoded in a single dll. I'm not sure if there might be a euro modifier or the de_DE locale, you'll have to experiment with that. Btw. keep in mind that LANG affects all C applications, and there are watcom, emx, vac365, vac308, various older ibm C version, and libc, which all may have their own set of locales. Kind Regards, knut PS. Do me a favor and don't call LIBC GCC. I mean there are no setlocale in GCC. setlocale is in the CRT, which is LIBC. So, refere to LIBC+versions, just the LIBC DLL.