Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111)
On Thursday 15 January 2009 13:20:55, rob1w...@aol.com wrote: > While building I found one "-Werror": > cc1: warnings being treated as errors > ../../gdb-cvs/gdb/procfs.c: In function 'proc_set_watchpoint': > ../../gdb-cvs/gdb/procfs.c:2942: error: dereferencing pointer 'pwatch' > does break strict-aliasing rules Nasty. Even more, the current code assumes that alignof (arg.watch) == alignof (prwatch_t)). Did you consider fixing this, and posting a patch at gdb-patches@ ? [...] If you look at the log file, does it look like many failures are related to \r\n patterns being misdetected? Passing --debug to RUNTESTFLAGS will make it easier to spot. I had some troubles related to that when I last ran the testsuite on that version of Solaris. See here: http://sourceware.org/ml/gdb/2008-08/msg00144.html I still don't know what the proper fix is. > I like to see the "# of unexpected failures" closer to zero. Someone has to put effort into making that happen. -- Pedro Alves ___ bug-gdb mailing list bug-gdb@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gdb
Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111)
Did you consider fixing this, and posting a patch at gdb-patches@ ? Not for a moment. All my spare time is devoted to gcc, OpenSolaris and a few other things ... There are other people better qualified than I am to ensure that GDB and Binutils is fixed _CORRECTLY_ for the OpenSolaris Platform. I would not want to implement fixes to have someone else come along and need to fix my fixes. These should be fixed by a better qualified person than me. I only have 6 months experience with OpenSolaris and only needed to build gdb to attempt to fix something else, not to adopt another GNU project to devote more time to. I volunteer (way) over 10 hours a week to the GNU projects so I feel no guilt about not taking on any more work. :) . I like to see the "# of unexpected failures" closer to zero. Someone has to put effort into making that happen. Yes, it would be nice. If you look at the log file, does it look like many failures are ... I can enclose the screen output during the check I made a few days ago but I can not find the '.log' file. See enclosure. Thanks for accepting my report, Rob -Original Message- From: Pedro Alves To: bug-gdb@gnu.org Cc: rob1w...@aol.com Sent: Fri, 16 Jan 2009 6:39 am Subject: Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111) On Thursday 15 January 2009 13:20:55, rob1w...@aol.com wrote: While building I found one "-Werror": cc1: warnings being treated as errors ../../gdb-cvs/gdb/procfs.c: In function 'proc_set_watchpoint': ../../gdb-cvs/gdb/procfs.c:2942: error: dereferencing pointer 'pwatch' does break strict-aliasing rules Nasty. Even more, the current code assumes that alignof (arg.watch) == alignof (prwatch_t)). Did you consider fixing this, and posting a patch at gdb-patches@ ? [...] If you look at the log file, does it look like many failures are related to \r\n patterns being misdetected? Passing --debug to RUNTESTFLAGS will make it easier to spot. I had some troubles related to that when I last ran the testsuite on that version of Solaris. See here: http://sourceware.org/ml/gdb/2008-08/msg00144.html I still don't know what the proper fix is. I like to see the "# of unexpected failures" closer to zero. Someone has to put effort into making that happen. -- Pedro Alves gdb_check_log.txt.gz Description: GNU Zip compressed data ___ bug-gdb mailing list bug-gdb@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gdb
Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111)
On Sunday 18 January 2009 11:25:59, rob1w...@aol.com wrote: > > Did you consider fixing this, and posting a patch at gdb-patches@ ? > Not for a moment. All my spare time is devoted to gcc, OpenSolaris and > a few other things ... > > There are other people better qualified than I am to ensure that GDB > and Binutils is fixed _CORRECTLY_ for > the OpenSolaris Platform. I would not want to implement fixes to have > someone else come along and need > to fix my fixes. These should be fixed by a better qualified person > than me. > > I only have 6 months experience with OpenSolaris and only needed to > build gdb to attempt to fix something > else, not to adopt another GNU project to devote more time to. I > volunteer (way) over 10 hours a week to > the GNU projects so I feel no guilt about not taking on any more work. > :) . Hey, calm down, I wasn't asking you to become solaris gdb maintainer. :-) To fix this, all it takes it do something like adjusting the affected functions to use char buffers and memcpy, instead of playing tricks with pointers. A bit of care has to be taken due to the ifdefery in place, but nothing major. You're in a good position to fix it, because, well, you have the compiler handy. If not taking care of it, is there somewhere where I can download binaries of a recent enough gcc that triggers those warnings? A bugzilla PR with the warning/error log would also be nice. > I can enclose the screen output during the check I made > a few days ago but I can not find the '.log' file. See enclosure. It's build-dir/gdb/testsuite/gdb.log. If you run a test with something like: make check RUNTESTFLAGS="--debug -v -v foo_test.exp" you'll find much more debug info in a build-dir/gdb/testsuite/dbg.log file. -- Pedro Alves ___ bug-gdb mailing list bug-gdb@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gdb
Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111)
-Original Message- From: Pedro Alves To: bug-gdb@gnu.org Cc: rob1w...@aol.com Sent: Sun, 18 Jan 2009 10:18 am Subject: Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111) On Sunday 18 January 2009 11:25:59, rob1w...@aol.com wrote: > Did you consider fixing this, and posting a patch at gdb-patches@ ? Not for a moment. All my spare time is devoted to gcc, OpenSolaris and a few other things ... There are other people better qualified than I am to ensure that GDB and Binutils is fixed _CORRECTLY_ ... ... I only have 6 months experience with OpenSolaris and only needed to build gdb to attempt to fix something else, not to adopt another GNU project to devote more time to. I volunteer (way) over 10 hours a week to ... :) . Hey, calm down, I wasn't asking you to become solaris gdb maintainer. :-) I meant fixing all 1600 Bugs. To fix this, all it takes it do something like adjusting the affected functions to use char buffers and memcpy, instead of playing tricks with pointers. A bit of care has to be taken due to the ifdefery in place, but nothing major. So this takes me as little time as possible ... Do you have a patch ? You're in a good position to fix it, because, well, you have the compiler handy. If not taking care of it, is there somewhere where I can download binaries of a recent enough gcc that triggers those warnings? I can do it (within a week) if I only need to apply the patch and recompile. There are no binaries (for gcc) yet. There are OpenSolaris Operating System binaries but you also need about a few days work to ensure that you can get gcc 4.4.0 compiled on OpenSolaris (as it will not "out-of-the-box"). So I can do it correctly, for free, and very quickly --- Pick two. If you have a patch I can get it done by next week. A bugzilla PR with the warning/error log would also be nice. I can enclose the screen output during the check I made a few days ago but I can not find the '.log' file. See enclosure. It's build-dir/gdb/testsuite/gdb.log. If you run a test with something like: make check RUNTESTFLAGS="--debug -v -v foo_test.exp" you'll find much more debug info in a build-dir/gdb/testsuite/dbg.log file. When I rebuild and recheck I will use those options, compress the log and mail it to you _regardless_ of it's size -- is that OK ? You might request "FREE" daily tests of your program here: http://build.alchar.org/ They do this for binutils and gcc maybe they can add gdb too. Labrat http://build.alchar.org/cgi/showPkg?pkg=binutils_incr Thanks, Rob ___ bug-gdb mailing list bug-gdb@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gdb
Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111)
From: Pedro Alves To: bug-gdb@gnu.org Sent: Sun, 18 Jan 2009 10:18 am Subject: Re: 1600 "unexpected failures" in gdb-weekly-CVS-6.8.50.20090113.tar.bz2 (on OpenSolaris 2008.11 using gcc version 4.4.0 20090111) > Did you consider fixing this, and posting a patch at gdb-patches@ ? > ... > To fix this, all it takes it do something like adjusting the affected functions > to use char buffers and memcpy, instead of playing tricks with pointers. > A bit of care has to be taken due to the ifdefery in place, but nothing major. > ... > Pedro Alves Here is my "fast/correct" recommendation. I may make a better recommendation in future. # if ((__GNUC__==4) && (__GNUC_MINOR__>=4)) gdb_siginfo_t sinfo; #else /* Use char array to avoid alignment issues. */ char sinfo[sizeof (gdb_siginfo_t)]; #endif Rationale: Older versions of gcc _might_ have had problems that you needed to "hack around" to fix. Newer versions _must_ not. The "newest" versions of gcc either: 1. Does not have these issues. 2. You need to file a Bug Report over at gcc's own bugzilla (and stop hiding bugs, so rule 1 will be true). Are you satisfied with this fix ? It is not intended to be a 'copout'. Rob ___ bug-gdb mailing list bug-gdb@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gdb