Обсуждение: SunOS patches
Included are patches for SunOS port based on the Feb 24 snapshot. Remaining problems are some minor compiling errors of ecpg and strange NOTICE messages from initdb (known bug?). I have tested the patches on FreeBSD and Solairs2.6 to make sure they do not break other ports. o added SunOS support to template/.similar o changed UNIXSOCK_PATH macro in include/libpq/pqcomm.h. The macro supposes sprintf() return the length of the result string. Unfortunately on some platforms such as SunOS sprintf() never returns value (I guess that behavior was inherited from BSD4.2). So I changed in more portable way using strlen(). o changed backend/libpq/pqcomm.c. It uses atexit() to unlink an Unix domain socket when postmaster exits. Again unfortunately SunOS does not have it. Looking for an alternative, I found a general exiting callback manager called on_exitpg() has already existed in PostgreSQL! So I replaced atexit() with on_exitpg(). Tatsuo Ishii t-ishii@sra.co.jp ---------------------------------------------------------------- *** template/.similar.orig Wed Feb 25 10:00:55 1998 --- template/.similar Wed Feb 25 10:03:02 1998 *************** *** 20,22 **** --- 20,26 ---- powerpc-unknown-linux-gnu=linux-elf sparc-sun-solaris=sparc_solaris-gcc sparc-unknown-linux-gnu=linux-elf-sparc + sparc-sun-sunos4.1.3=sunos4-gcc + sparc-sun-sunos4.1.3_JL=sunos4-gcc + sparc-sun-sunos4.1.4=sunos4-gcc + sparc-sun-sunos4.1.4_JL=sunos4-gcc *** backend/libpq/pqcomm.c.orig Wed Feb 25 10:04:28 1998 --- backend/libpq/pqcomm.c Wed Feb 25 10:05:57 1998 *************** *** 645,651 **** if (family == AF_UNIX) { chmod(sock_path, 0777); ! atexit(do_unlink); } return (STATUS_OK); } --- 645,651 ---- if (family == AF_UNIX) { chmod(sock_path, 0777); ! on_exitpg(do_unlink,0); } return (STATUS_OK); } *** include/libpq/pqcomm.h.orig Wed Feb 25 10:07:25 1998 --- include/libpq/pqcomm.h Wed Feb 25 10:07:41 1998 *************** *** 34,41 **** /* Configure the UNIX socket address for the well known port. */ #define UNIXSOCK_PATH(sun,port) \ ! (sprintf((sun).sun_path, "/tmp/.s.PGSQL.%d", (port)) + \ ! offsetof(struct sockaddr_un, sun_path)) /* * We do this because sun_len is in BSD's struct, while others don't. * We never actually set BSD's sun_len, and I can't think of a --- 34,41 ---- /* Configure the UNIX socket address for the well known port. */ #define UNIXSOCK_PATH(sun,port) \ ! (sprintf((sun).sun_path, "/tmp/.s.PGSQL.%d", (port)), \ ! strlen((sun).sun_path)+ offsetof(struct sockaddr_un, sun_path)) /* * We do this because sun_len is in BSD's struct, while others don't. * We never actually set BSD's sun_len, and I can't think of a
>o changed backend/libpq/pqcomm.c. It uses atexit() to unlink an Unix >domain socket when postmaster exits. Again unfortunately SunOS does >not have it. Looking for an alternative, I found a general exiting >callback manager called on_exitpg() has already existed in PostgreSQL! >So I replaced atexit() with on_exitpg(). I found my previous patches for backend/libpq/pqcomm.c did not work if backend crashed accidentaly (yes, I found yet another SQL commands to crash the backend, but it's an another story). If the backend dies abnormally (SEGFALUT etc.) postmaster calls functions registered by on_exitpg(). This is ok. Problem is postmaster does not exits in that case and leaves the socket file with link counter 0! Too bad. It seems the only solution that would work with both SunOS and other platforms is let exitpg() call do_unlink() in backend/libpq/pqcomm.c (note that exitpg() is the only function calling exit()). included are patches for Feb 26 snapshot. I hope these are the last patches for SunOS port:-) P.S. I didn't see my patches to tempalte/.similar in Mar 1 snapshot. Should I repost the patches? --- Tatsuo Ishii t-ishii@sra.co.jp ---------------------------------------------------------------------- *** backend/libpq/pqcomm.c.orig Fri Feb 27 14:07:52 1998 --- backend/libpq/pqcomm.c Fri Feb 27 14:08:50 1998 *************** *** 564,571 **** * Shutdown routine for backend connection * If a Unix socket is used for communication, explicitly close it. */ ! static void ! do_unlink() { if (sock_path[0]) unlink(sock_path); --- 564,571 ---- * Shutdown routine for backend connection * If a Unix socket is used for communication, explicitly close it. */ ! void ! StreamDoUnlink() { if (sock_path[0]) unlink(sock_path); *************** *** 645,651 **** if (family == AF_UNIX) { chmod(sock_path, 0777); - atexit(do_unlink); } return (STATUS_OK); } --- 645,650 ---- *** backend/storage/ipc/ipc.c.orig Fri Feb 27 14:09:12 1998 --- backend/storage/ipc/ipc.c Fri Feb 27 14:09:26 1998 *************** *** 136,141 **** --- 136,142 ---- for (i = onexit_index - 1; i >= 0; --i) (*onexit_list[i].function) (code, onexit_list[i].arg); + StreamDoUnlink(); exit(code); }
On Mon, 2 Mar 1998 t-ishii@sra.co.jp wrote: > > included are patches for Feb 26 snapshot. I hope these are the last > patches for SunOS port:-) Ack! > P.S. I didn't see my patches to tempalte/.similar in Mar 1 > snapshot. Should I repost the patches? > --- > Tatsuo Ishii > t-ishii@sra.co.jp > ---------------------------------------------------------------------- > *** backend/libpq/pqcomm.c.orig Fri Feb 27 14:07:52 1998 > --- backend/libpq/pqcomm.c Fri Feb 27 14:08:50 1998 > *************** > *** 564,571 **** > * Shutdown routine for backend connection > * If a Unix socket is used for communication, explicitly close it. > */ > ! static void > ! do_unlink() > { > if (sock_path[0]) > unlink(sock_path); > --- 564,571 ---- > * Shutdown routine for backend connection > * If a Unix socket is used for communication, explicitly close it. > */ > ! void > ! StreamDoUnlink() What breaks by renaming this function?? :( Marc G. Fournier Systems Administrator @ hub.org primary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org
>> ---------------------------------------------------------------------- >> *** backend/libpq/pqcomm.c.orig Fri Feb 27 14:07:52 1998 >> --- backend/libpq/pqcomm.c Fri Feb 27 14:08:50 1998 >> *************** >> *** 564,571 **** >> * Shutdown routine for backend connection >> * If a Unix socket is used for communication, explicitly close it. >> */ >> ! static void >> ! do_unlink() >> { >> if (sock_path[0]) >> unlink(sock_path); >> --- 564,571 ---- >> * Shutdown routine for backend connection >> * If a Unix socket is used for communication, explicitly close it. >> */ >> ! void >> ! StreamDoUnlink() > > > What breaks by renaming this function?? :( Nothing. do_unlink() is called only from line 648 and the line was deleted by my patches. Just removing "static" from the declaration of do_unlink might be ok. However function name "do_unlink" would be too general IMHO. -- Tatsuo Ishii t-ishii@sra.co.jp
The Hermit Hacker wrote: > On Mon, 2 Mar 1998 t-ishii@sra.co.jp wrote: > > > > > included are patches for Feb 26 snapshot. I hope these are the last > > patches for SunOS port:-) > > Ack! > > > P.S. I didn't see my patches to tempalte/.similar in Mar 1 > > snapshot. Should I repost the patches? > > --- > > Tatsuo Ishii > > t-ishii@sra.co.jp > > ---------------------------------------------------------------------- > > *** backend/libpq/pqcomm.c.orig Fri Feb 27 14:07:52 1998 > > --- backend/libpq/pqcomm.c Fri Feb 27 14:08:50 1998 > > *************** > > *** 564,571 **** > > * Shutdown routine for backend connection > > * If a Unix socket is used for communication, explicitly close it. > > */ > > ! static void > > ! do_unlink() > > { > > if (sock_path[0]) > > unlink(sock_path); > > --- 564,571 ---- > > * Shutdown routine for backend connection > > * If a Unix socket is used for communication, explicitly close it. > > */ > > ! void > > ! StreamDoUnlink() > > What breaks by renaming this function?? :( On my linux box: ... gcc -I../../include -I../../backend -I/usr/include/ncurses -I/usr/include/readline -O2 -m486 -Wall -Wmissing-prototypes -I.. -c pqcomm.c -o pqcomm.o pqcomm.c:569: warning: no previous prototype for `StreamDoUnlink' ... ipc.c: In function `exitpg': ipc.c:139: warning: implicit declaration of function `StreamDoUnlink' Seems to run OK though. Would like to get the warnings out of there for a release. I didn't pay attention to Tatsuo's earlier patches since they didn't directly affect my platform; where would these things be declared? Can Tatsuo come up with a few more patches to fix this up? It would be worth waiting a few hours on the release to get SunOS in there (the docs claim it is supported :) - Tom
On Mon, 2 Mar 1998, Thomas G. Lockhart wrote: > gcc -I../../include -I../../backend -I/usr/include/ncurses > -I/usr/include/readline -O2 -m486 -Wall -Wmissing-prototypes -I.. -c > pqcomm.c -o pqcomm.o > pqcomm.c:569: warning: no previous prototype for `StreamDoUnlink' > ... > ipc.c: In function `exitpg': > ipc.c:139: warning: implicit declaration of function `StreamDoUnlink' > > Seems to run OK though. Would like to get the warnings out of there for a > release. I didn't pay attention to Tatsuo's earlier patches since they didn't > directly affect my platform; where would these things be declared? Can Tatsuo > come up with a few more patches to fix this up? It would be worth waiting a few > hours on the release to get SunOS in there (the docs claim it is supported :) What he said... :) Tatsuo...could you send me a clean patch that fixes the above and includes the last patch? I'll end up scrutinizing it in the morning, most likely, before applying it, but it would be nice to get it in before I bundle things... As long as I'm confident that it won't affect anything or anyone else :) Marc G. Fournier Systems Administrator @ hub.org primary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org
>On Mon, 2 Mar 1998, Thomas G. Lockhart wrote: > >> gcc -I../../include -I../../backend -I/usr/include/ncurses >> -I/usr/include/readline -O2 -m486 -Wall -Wmissing-prototypes -I.. -c >> pqcomm.c -o pqcomm.o >> pqcomm.c:569: warning: no previous prototype for `StreamDoUnlink' >> ... >> ipc.c: In function `exitpg': >> ipc.c:139: warning: implicit declaration of function `StreamDoUnlink' >> >> Seems to run OK though. Would like to get the warnings out of there for a >> release. I didn't pay attention to Tatsuo's earlier patches since they didn't >> directly affect my platform; where would these things be declared? Can Tatsuo >> come up with a few more patches to fix this up? It would be worth waiting a few >> hours on the release to get SunOS in there (the docs claim it is supported :) > > What he said... :) > > Tatsuo...could you send me a clean patch that fixes the above and >includes the last patch? I'll end up scrutinizing it in the morning, most >likely, before applying it, but it would be nice to get it in before I >bundle things... > > As long as I'm confident that it won't affect anything or anyone >else :) Please give me an hour. I will work on this now. -- Tatsuo Ishii t-ishii@sra.co.jp
>>On Mon, 2 Mar 1998, Thomas G. Lockhart wrote: >> >>> gcc -I../../include -I../../backend -I/usr/include/ncurses >>> -I/usr/include/readline -O2 -m486 -Wall -Wmissing-prototypes -I.. -c >>> pqcomm.c -o pqcomm.o >>> pqcomm.c:569: warning: no previous prototype for `StreamDoUnlink' >>> ... >>> ipc.c: In function `exitpg': >>> ipc.c:139: warning: implicit declaration of function `StreamDoUnlink' >>> >>> Seems to run OK though. Would like to get the warnings out of there for a >>> release. I didn't pay attention to Tatsuo's earlier patches since they didn't >>> directly affect my platform; where would these things be declared? Can Tatsuo >>> come up with a few more patches to fix this up? It would be worth waiting a few >>> hours on the release to get SunOS in there (the docs claim it is supported :) >> >> What he said... :) >> >> Tatsuo...could you send me a clean patch that fixes the above and >>includes the last patch? I'll end up scrutinizing it in the morning, most >>likely, before applying it, but it would be nice to get it in before I >>bundle things... >> >> As long as I'm confident that it won't affect anything or anyone >>else :) > >Please give me an hour. I will work on this now. Here it is. (Mar 1 snapshot seems to lack the pacthes to include/libpq/pqcomm.h. so I included that also.) ----------------------------- cut here ----------------------------- *** include/libpq/libpq.h.orig Thu Feb 26 17:02:21 1998 --- include/libpq/libpq.h Mon Mar 2 13:16:06 1998 *************** *** 282,286 **** --- 282,287 ---- extern int StreamServerPort(char *hostName, short portName, int *fdP); extern int StreamConnection(int server_fd, Port *port); extern void StreamClose(int sock); + extern void StreamDoUnlink(void); #endif /* LIBPQ_H */ *** include/libpq/pqcomm.h.orig Mon Mar 2 13:55:19 1998 --- include/libpq/pqcomm.h Mon Mar 2 13:57:10 1998 *************** *** 35,42 **** /* Configure the UNIX socket address for the well known port. */ #define UNIXSOCK_PATH(sun,port) \ ! (sprintf((sun).sun_path, "/tmp/.s.PGSQL.%d", (port)) + \ ! offsetof(struct sockaddr_un, sun_path)) /* * We do this because sun_len is in BSD's struct, while others don't. * We never actually set BSD's sun_len, and I can't think of a --- 35,43 ---- /* Configure the UNIX socket address for the well known port. */ #define UNIXSOCK_PATH(sun,port) \ ! (sprintf((sun).sun_path, "/tmp/.s.PGSQL.%d", (port)), \ ! strlen((sun).sun_path)+ offsetof(struct sockaddr_un, sun_path)) ! /* * We do this because sun_len is in BSD's struct, while others don't. * We never actually set BSD's sun_len, and I can't think of a *** backend/libpq/pqcomm.c.orig Thu Feb 26 17:01:05 1998 --- backend/libpq/pqcomm.c Mon Mar 2 13:16:06 1998 *************** *** 564,571 **** * Shutdown routine for backend connection * If a Unix socket is used for communication, explicitly close it. */ ! static void ! do_unlink() { if (sock_path[0]) unlink(sock_path); --- 564,571 ---- * Shutdown routine for backend connection * If a Unix socket is used for communication, explicitly close it. */ ! void ! StreamDoUnlink() { if (sock_path[0]) unlink(sock_path); *************** *** 645,651 **** if (family == AF_UNIX) { chmod(sock_path, 0777); - atexit(do_unlink); } return (STATUS_OK); } --- 645,650 ---- *** backend/storage/ipc/ipc.c.orig Thu Feb 26 17:01:36 1998 --- backend/storage/ipc/ipc.c Mon Mar 2 13:16:06 1998 *************** *** 38,43 **** --- 38,44 ---- #include <sys/sem.h> #include <sys/shm.h> #include "utils/memutils.h" + #include "libpq/libpq.h" #if defined(sparc_solaris) #include <string.h> *************** *** 136,141 **** --- 137,143 ---- for (i = onexit_index - 1; i >= 0; --i) (*onexit_list[i].function) (code, onexit_list[i].arg); + StreamDoUnlink(); exit(code); }
On Mon, 2 Mar 1998 t-ishii@sra.co.jp wrote: > Here it is. > (Mar 1 snapshot seems to lack the pacthes to include/libpq/pqcomm.h. so > I included that also.) Okay, applied...I'll test it under solaris tomorrow before I package things up, and FreeBSD tonight...everyone else should do a run though quickly also... Marc G. Fournier Systems Administrator @ hub.org primary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org