Thread: libecpg (8.0 and CVS) hits a gcc bug on powerpc and amd64 (crash)
NB: The crossposting is actually suggested on the mailing list description: "If you've found a bug in PostgreSQL, please send it both to this list and also to the pgsql-patches list." Amd64 and powerpc share an ABI design which enables variable passing in registers and which makes va_list a complex datatype. Due to gcc not implementing a part of the C9x standard (http://mail-index.netbsd.org/tech-toolchain/2003/11/11/0011.html) taking a pointer to a va_list will fail on these platforms. The corresponding bug (http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8262) is closed, so the gcc developers seem not to plan to change this. So I propose the following hack (feel free to improve it!) to go into mainline. Since only powerpc and amd64 are affected (and the old method does not work at all here at all) the patch should not be controversial. Christof PS: I don't know about other architectures. I can't believe that only ppc and amd64 pass parameters in registers. PPS: I could not test it on amd64, but all docs indicate the same behaviour.
Christof Petig schrieb:
> So I propose the following hack (feel free to improve it!) to go into
> mainline. Since only powerpc and amd64 are affected (and the old method
> does not work at all here at all) the patch should not be controversial.
I failed to mention that call by reference is the default for va_list on
these platforms (that's actually the cause of the bug).
Christof
Attachment
Re: [INTERFACES] libecpg (8.0 and CVS) hits a gcc bug on powerpc and amd64 (crash)
From
Michael Meskes
Date:
Patch committed. Thanks. Michael -- Michael Meskes Email: Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) ICQ: 179140304, AIM/Yahoo: michaelmeskes, Jabber: meskes@jabber.org Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL!
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Michael Meskes schrieb: > Patch committed. Thanks. I have to admit that I got the case of the preprocessor symbol on amd64 wrong. __AMD64__ is not defined, __amd64__ is. Sorry, corrected patch attached Christof -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.1 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFCsZCmng+R+0ucfO0RAoABAKDKnokBW+tgXMKgeEMecpZsFXRFTQCeKeBu YsdWP056cqxKtNAx5Gh/Wis= =b2Og -----END PGP SIGNATURE----- Index: src/interfaces/ecpg/ecpglib/execute.c =================================================================== RCS file: /projects/cvsroot/pgsql/src/interfaces/ecpg/ecpglib/execute.c,v retrieving revision 1.40 diff -u -u -r1.40 execute.c --- src/interfaces/ecpg/ecpglib/execute.c 2 Jun 2005 12:35:11 -0000 1.40 +++ src/interfaces/ecpg/ecpglib/execute.c 16 Jun 2005 14:35:09 -0000 @@ -70,7 +70,7 @@ return res; } -#if defined(__GNUC__) && (defined (__powerpc__) || defined(__AMD64__) || defined(__x86_64__)) +#if defined(__GNUC__) && (defined (__powerpc__) || defined(__amd64__) || defined(__x86_64__)) #define APREF ap #else #define APREF *ap @@ -178,7 +178,7 @@ if (!(var = (struct variable *) ECPGalloc(sizeof(struct variable), lineno))) return false; -#if defined(__GNUC__) && (defined (__powerpc__) || defined(__AMD64__) || defined(__x86_64__)) +#if defined(__GNUC__) && (defined (__powerpc__) || defined(__amd64__) || defined(__x86_64__)) ECPGget_variable(ap, type, var, true); #else ECPGget_variable(&ap, type, var, true); Index: src/interfaces/ecpg/ecpglib/extern.h =================================================================== RCS file: /projects/cvsroot/pgsql/src/interfaces/ecpg/ecpglib/extern.h,v retrieving revision 1.13 diff -u -u -r1.13 extern.h --- src/interfaces/ecpg/ecpglib/extern.h 2 Jun 2005 12:35:11 -0000 1.13 +++ src/interfaces/ecpg/ecpglib/extern.h 16 Jun 2005 14:35:09 -0000 @@ -125,7 +125,7 @@ bool ECPGstore_result(const PGresult *results, int act_field, const struct statement * stmt, struct variable * var); bool ECPGstore_input(const int, const bool, const struct variable *, const char **, bool *); -#if defined(__GNUC__) && (defined (__powerpc__) || defined(__AMD64__) || defined(__x86_64__)) +#if defined(__GNUC__) && (defined (__powerpc__) || defined(__amd64__) || defined(__x86_64__)) // work around a gcc/ABI bug with va_lists on ppc+amd64 void ECPGget_variable(va_list, enum ECPGttype, struct variable *, bool); #else
Patch applied. Thanks. --------------------------------------------------------------------------- Christof Petig wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Michael Meskes schrieb: > > Patch committed. Thanks. > > I have to admit that I got the case of the preprocessor symbol on amd64 > wrong. __AMD64__ is not defined, __amd64__ is. > > Sorry, corrected patch attached > Christof > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.1 (GNU/Linux) > Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org > > iD8DBQFCsZCmng+R+0ucfO0RAoABAKDKnokBW+tgXMKgeEMecpZsFXRFTQCeKeBu > YsdWP056cqxKtNAx5Gh/Wis= > =b2Og > -----END PGP SIGNATURE----- [ text/x-patch is unsupported, treating like TEXT/PLAIN ] > Index: src/interfaces/ecpg/ecpglib/execute.c > =================================================================== > RCS file: /projects/cvsroot/pgsql/src/interfaces/ecpg/ecpglib/execute.c,v > retrieving revision 1.40 > diff -u -u -r1.40 execute.c > --- src/interfaces/ecpg/ecpglib/execute.c 2 Jun 2005 12:35:11 -0000 1.40 > +++ src/interfaces/ecpg/ecpglib/execute.c 16 Jun 2005 14:35:09 -0000 > @@ -70,7 +70,7 @@ > return res; > } > > -#if defined(__GNUC__) && (defined (__powerpc__) || defined(__AMD64__) || defined(__x86_64__)) > +#if defined(__GNUC__) && (defined (__powerpc__) || defined(__amd64__) || defined(__x86_64__)) > #define APREF ap > #else > #define APREF *ap > @@ -178,7 +178,7 @@ > if (!(var = (struct variable *) ECPGalloc(sizeof(struct variable), lineno))) > return false; > > -#if defined(__GNUC__) && (defined (__powerpc__) || defined(__AMD64__) || defined(__x86_64__)) > +#if defined(__GNUC__) && (defined (__powerpc__) || defined(__amd64__) || defined(__x86_64__)) > ECPGget_variable(ap, type, var, true); > #else > ECPGget_variable(&ap, type, var, true); > Index: src/interfaces/ecpg/ecpglib/extern.h > =================================================================== > RCS file: /projects/cvsroot/pgsql/src/interfaces/ecpg/ecpglib/extern.h,v > retrieving revision 1.13 > diff -u -u -r1.13 extern.h > --- src/interfaces/ecpg/ecpglib/extern.h 2 Jun 2005 12:35:11 -0000 1.13 > +++ src/interfaces/ecpg/ecpglib/extern.h 16 Jun 2005 14:35:09 -0000 > @@ -125,7 +125,7 @@ > bool ECPGstore_result(const PGresult *results, int act_field, > const struct statement * stmt, struct variable * var); > bool ECPGstore_input(const int, const bool, const struct variable *, const char **, bool *); > -#if defined(__GNUC__) && (defined (__powerpc__) || defined(__AMD64__) || defined(__x86_64__)) > +#if defined(__GNUC__) && (defined (__powerpc__) || defined(__amd64__) || defined(__x86_64__)) > // work around a gcc/ABI bug with va_lists on ppc+amd64 > void ECPGget_variable(va_list, enum ECPGttype, struct variable *, bool); > #else > > ---------------------------(end of broadcast)--------------------------- > TIP 6: Have you searched our list archives? > > http://archives.postgresql.org -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073