From diy-linux-dev@diy-linux.org Sat May 5 11:26:53 2007 From: diy-linux-dev@diy-linux.org (Ken Dyke) Date: Sat, 5 May 2007 05:26:53 -0600 Subject: chroot coreutils test error In-Reply-To: <20070406010203.GA28161@eyo32.local> References: <20070322181056.GD2225@hyper> <20070322231322.GA24301@hyper> <20070324052255.GF24301@hyper> <20070327225658.GA3845@hyper> <20070406010203.GA28161@eyo32.local> Message-ID: <20070505112653.GA11224@hyper> On Fri, Apr 06, 2007 at 11:02:03AM +1000, Greg Schafer (gschafer@zip.com.au) wrote: > Ken, sorry for late reply. Don't know what to think of this because it works > for me and most other folks I believe. But having said that, the Coreutils > testsuite can be very picky if you are doing something slightly different. > Have you tried running the test in question verbosely as per the > instructions in the Coreutils README ? Yes. I even set up strace so I could run that on it. I get nothing more then permission denied. If I skip the tests in coreutils and bash and continue to build the rest of the chroot phase I can then go back and rebuild those packages with the testsuites and they pass just fine. Then once I have a bootable system I boot into that and run the same scripts over again. The coreutils and bash pass their tests the first time through. I implemented my scripts before taking a look at yours. Needless to say they are different but being bash not that different. I hope to have a system freed up in the near future so that I can interatively isolate what is causing my problem. I image it is something trivial and hope to learn more from the process then finding the actual bug. Thank you, -- I reason and act, therefore, ken_i_m Chief Gadgeteer, Elegant Innovations Founder, Bozeman Linux Users Group Founder, Helena Linux Users Group (406) 581-0495 From diy-linux-dev@diy-linux.org Tue May 15 01:11:28 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Tue, 15 May 2007 11:11:28 +1000 Subject: GCC-4.2 Message-ID: <20070515011128.GA4893@eyo32.local> Hi Guys, GCC-4.2 is now released. I've just plugged it into the Refbuild along with latest HJL binutils and built a base system with Glibc-2.5. The current Refbuild already includes an old GCC-4.2 snapshot so most of the work is already done. Everything appears mostly OK but the Glibc build spews out an enormous amount of warnings of the type: warning: C99 inline functions are not supported; using GNU89 I don't think the warnings are as scary as they seem because I believe Glibc actually wants the older inline semantics. Anyhoo, I'll have to read up some more on this and see if I can figure out a way to get rid of the warnings. Here are some likely relevant links: http://gcc.gnu.org/ml/gcc-patches/2007-03/msg00702.html http://gcc.gnu.org/ml/gcc-patches/2007-03/msg01418.html http://sourceware.org/ml/libc-hacker/2007-03/msg00026.html http://gcc.gnu.org/gcc-4.2/changes.html As well as the math failures mentioned a few months back, there are some new failures that could be related to the above: # grep "\*\*\*.*Error" glibc-2.5.log make[2]: *** [/temptools/src/glibc-build/math/test-double.out] Error 1 make[2]: *** [/temptools/src/glibc-build/math/test-ldouble.out] Error 1 make[2]: *** [/temptools/src/glibc-build/math/test-ildoubl.out] Error 1 make[1]: *** [math/tests] Error 2 make[2]: *** [/temptools/src/glibc-build/nptl/tst-initializers1.o] Error 1 make[2]: *** [/temptools/src/glibc-build/nptl/tst-initializers1-c99.o] Error 1 make[2]: *** [/temptools/src/glibc-build/nptl/tst-initializers1-gnu99.o] Error 1 make[1]: *** [nptl/tests] Error 2 make: *** [check] Error 2 It may well turn out that Glibc CVS trunk is a better match with GCC-4.2.0. I'll try to find some time and test it out. Here are the GCC testsuite results: http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg00695.html Regards Greg From diy-linux-dev@diy-linux.org Tue May 15 14:06:35 2007 From: diy-linux-dev@diy-linux.org (Charles Johnston) Date: Tue, 15 May 2007 08:06:35 -0600 Subject: GCC-4.2 In-Reply-To: <20070515011128.GA4893@eyo32.local> References: <20070515011128.GA4893@eyo32.local> Message-ID: <4649BE6B.1060700@infoplatter.com> Greg Schafer wrote: > > It may well turn out that Glibc CVS trunk is a better match with GCC-4.2.0. > I'll try to find some time and test it out. > It appears that the Glibc 2.6 release is imminent. http://sourceware.org/cgi-bin/cvsweb.cgi/libc/version.h.diff?cvsroot=glibc&r1=1.72&r2=1.73 Charles From diy-linux-dev@diy-linux.org Fri May 18 03:08:38 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Fri, 18 May 2007 13:08:38 +1000 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <4649BE6B.1060700@infoplatter.com> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> Message-ID: <20070518030838.GA24769@eyo32.local> On Tue, May 15, 2007 at 08:06:35AM -0600, Charles Johnston wrote: > It appears that the Glibc 2.6 release is imminent. Indeed. Tarballs are now up. I'll start some testing. Regards Greg From diy-linux-dev@diy-linux.org Fri May 18 06:08:12 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Fri, 18 May 2007 16:08:12 +1000 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <20070518030838.GA24769@eyo32.local> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> <20070518030838.GA24769@eyo32.local> Message-ID: <20070518060812.GA25511@eyo32.local> On Fri, May 18, 2007 at 01:08:38PM +1000, Greg Schafer wrote: > I'll start some testing. Ok, first problem for x86 is Glibc-2.6 doesn't support i386 anymore. Lowest supported config is apparently i486. We'll have to change the temptools Glibc to compile with -march=i486 (chroot Glibc already uses -march=i686 so it's sweet): http://sourceware.org/ml/libc-alpha/2007-01/msg00060.html http://sourceware.org/ml/libc-alpha/2007-02/msg00008.html Second problem is coreutils-6.9 fails to compile. Glibc-2.6 has apparently gained futimens() and it conflicts with the Coreutils version. I tried hacking Coreutils to make it build but gave up after half an hour of frustration. Instead, I just remmed out the futimens declaration in sys/stat.h :-/ Someone needs to fix Coreutils.. I've sent upstream a note: http://lists.gnu.org/archive/html/bug-coreutils/2007-05/msg00158.html Glibc-2.6 tests for GCC-4.2 and adds `-fgnu89-inline' to CFLAGS which gets rid of the earlier mentioned warnings. There are still testsuite errors but slightly better than earlier: # grep "\*\*\*.*Error" glibc-2.6.log make[2]: *** [/temptools/src/glibc-build/math/test-double.out] Error 1 make[2]: *** [/temptools/src/glibc-build/math/test-ldouble.out] Error 1 make[2]: *** [/temptools/src/glibc-build/math/test-ildoubl.out] Error 1 make[1]: *** [math/tests] Error 2 make[2]: *** [/temptools/src/glibc-build/posix/tst-vfork3.out] Error 1 make[1]: *** [posix/tests] Error 2 make: *** [check] Error 2 More later... Regards Greg From diy-linux-dev@diy-linux.org Fri May 18 08:33:24 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Fri, 18 May 2007 18:33:24 +1000 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <20070518060812.GA25511@eyo32.local> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> <20070518030838.GA24769@eyo32.local> <20070518060812.GA25511@eyo32.local> Message-ID: <20070518083324.GA21559@eyo32.local> On Fri, May 18, 2007 at 04:08:12PM +1000, Greg Schafer wrote: > Second problem is coreutils-6.9 fails to compile. As just posted on the coreutils list, this sed will make it build: sed -i.bak 's/futimens/gl_&/' src/{copy,touch}.c lib/utimens.c Regards Greg From diy-linux-dev@diy-linux.org Fri May 18 11:48:50 2007 From: diy-linux-dev@diy-linux.org (=?ISO-8859-1?Q?J=FCrg?= Billeter) Date: Fri, 18 May 2007 13:48:50 +0200 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <20070518083324.GA21559@eyo32.local> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> <20070518030838.GA24769@eyo32.local> <20070518060812.GA25511@eyo32.local> <20070518083324.GA21559@eyo32.local> Message-ID: <1179488930.4362.12.camel@juerg-pd.bitron.ch> On Fre, 2007-05-18 at 18:33 +1000, Greg Schafer wrote: > On Fri, May 18, 2007 at 04:08:12PM +1000, Greg Schafer wrote: >=20 > > Second problem is coreutils-6.9 fails to compile. >=20 > As just posted on the coreutils list, this sed will make it build: >=20 > sed -i.bak 's/futimens/gl_&/' src/{copy,touch}.c lib/utimens.c Thanks for the heads-up. lib/utimens.h is also affected, this sed seems to work fine: sed -i.bak 's/futimens/gl_&/' src/{copy,touch}.c lib/utimens.{c,h} The same problem with gzip. The following sed should work, still testing: sed -i 's/futimens/gl_&/' gzip.c lib/utimens.{c,h} Regards, J=C3=BCrg From diy-linux-dev@diy-linux.org Fri May 18 13:53:42 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Fri, 18 May 2007 23:53:42 +1000 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <1179488930.4362.12.camel@juerg-pd.bitron.ch> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> <20070518030838.GA24769@eyo32.local> <20070518060812.GA25511@eyo32.local> <20070518083324.GA21559@eyo32.local> <1179488930.4362.12.camel@juerg-pd.bitron.ch> Message-ID: <20070518135342.GA11148@eyo32.local> On Fri, May 18, 2007 at 01:48:50PM +0200, J=FCrg Billeter wrote: > Thanks for the heads-up. lib/utimens.h is also affected, this sed seems > to work fine: >=20 > sed -i.bak 's/futimens/gl_&/' src/{copy,touch}.c lib/utimens.{c,h} Ughh, you're right. I'm an idiot. I was testing on a system with a hacked sys/stat.h :-( Thanks for the correction. > The same problem with gzip. The following sed should work, still > testing: >=20 > sed -i 's/futimens/gl_&/' gzip.c lib/utimens.{c,h} Likewise. Regards Greg From diy-linux-dev@diy-linux.org Sat May 19 18:49:41 2007 From: diy-linux-dev@diy-linux.org (Chris Staub) Date: Sat, 19 May 2007 14:49:41 -0400 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <1179488930.4362.12.camel@juerg-pd.bitron.ch> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> <20070518030838.GA24769@eyo32.local> <20070518060812.GA25511@eyo32.local> <20070518083324.GA21559@eyo32.local> <1179488930.4362.12.camel@juerg-pd.bitron.ch> Message-ID: <464F46C5.4040301@beaker67.com> J=C3=BCrg Billeter wrote: >=20 > Thanks for the heads-up. lib/utimens.h is also affected, this sed see= ms > to work fine: >=20 > sed -i.bak 's/futimens/gl_&/' src/{copy,touch}.c lib/utimens.{c,h} >=20 > The same problem with gzip. The following sed should work, still > testing: >=20 > sed -i 's/futimens/gl_&/' gzip.c lib/utimens.{c,h} >=20 > Regards, >=20 > J=C3=BCrg >=20 Tar has the same problem. Needs this... sed -i 's/futimens/gl_&/' lib/utimens.{c,h} src/misc.c From diy-linux-dev@diy-linux.org Mon May 21 05:07:03 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Mon, 21 May 2007 15:07:03 +1000 Subject: Glibc-2.6 (was Re: GCC-4.2) In-Reply-To: <20070518060812.GA25511@eyo32.local> References: <20070515011128.GA4893@eyo32.local> <4649BE6B.1060700@infoplatter.com> <20070518030838.GA24769@eyo32.local> <20070518060812.GA25511@eyo32.local> Message-ID: <20070521050703.GA2022@eyo32.local> On Fri, May 18, 2007 at 04:08:12PM +1000, Greg Schafer wrote: > There are still testsuite errors but slightly better than earlier: > > # grep "\*\*\*.*Error" glibc-2.6.log > make[2]: *** [/temptools/src/glibc-build/math/test-double.out] Error 1 This one appears to be sensitive to the -march switch and/or -O opt level used: -march=i486 -O2 OK -march=i686 -O3 FAIL -march=native -O3 OK (equivalent to k8 on this box) More info in this thread: http://sourceware.org/ml/libc-alpha/2007-05/msg00068.html > make[2]: *** [/temptools/src/glibc-build/math/test-ldouble.out] Error 1 > make[2]: *** [/temptools/src/glibc-build/math/test-ildoubl.out] Error 1 These 2 are triggered by -O3 opt level. They appear relatively minor. I've posted a patch here: http://sourceware.org/ml/libc-alpha/2007-05/msg00072.html > make[2]: *** [/temptools/src/glibc-build/posix/tst-vfork3.out] Error 1 This one is due to missing /bin/echo essential symlink. It was very obvious once I straced it. The test unsets PATH on purpose which means `echo' is not found in $TT_PFX/bin. Regards Greg From diy-linux-dev@diy-linux.org Mon May 21 23:54:01 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Tue, 22 May 2007 09:54:01 +1000 Subject: Coreutils-6.9 Message-ID: <20070521235401.GA23255@eyo32.local> Hi Guys, Some folks are asking why I haven't yet updated to latest Coreutils-6.9. It's because I noted there were some changes in there that affected fchownat() and our use of the fakeroot program that I needed to investigate. You might recall that fakeroot stopped working on glibc-2.4 and above due it not wrapping the new *at() calls in glibc. This is the main reason for the current workaround of adding ac_cv_func_openat=no to configure. The other reason was due to a nasty bug in Glibc-2.4 (noted in the refbuild). This bug was supposed to be taken care of (ie: worked around) by the coreutils-6.9 autofoo. http://lists.gnu.org/archive/html/bug-coreutils/2006-12/msg00246.html Unfortunately, that fix was bogus because it results in Coreutils-6.9 always using its own replacement fchownat and avoiding the one in glibc, even on glibc 2.5 and 2.6: http://lists.gnu.org/archive/html/bug-coreutils/2007-05/msg00171.html The interesting side effect of this bug is that the fakeroot version used in the Refbuild now mostly works again :-) ie: even without the ac_cv_func_openat=no workaround. I say "mostly" because I can't convince myself that all bases are covered (openat, mkdirat, etc), tho' a full system build with a fakeroot-using pacman doesn't exhibit any obvious errors. Therefore, I will upgrade coreutils in the Refbuild but will keep the current hack for fakeroot safety. When the next Coreutils release occurs, this issue will need revisiting and I'll look at upgrading to the latest "unstable" fakeroot program which supposedly supports all the new *at() calls (I haven't tested it yet). Let me know if I've missed anythng. Regards Greg From diy-linux-dev@diy-linux.org Wed May 23 01:02:07 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Wed, 23 May 2007 11:02:07 +1000 Subject: Status update Message-ID: <20070523010207.GA18037@eyo32.local> Hi Guys, Ok, the Refbuild has been updated with the Glibc-2.6 and GCC-4.2 combo. It seems reasonable at this point, but a lot more testing is needed. I have tested this combo on x86 only! Sorry, I just don't have the bandwidth to test x86_64 and ppc at the moment.. but I'll hopefully get to them eventually. It looks like an unpleasant bug snuck into Glibc-2.6 just before release: http://sourceware.org/ml/libc-hacker/2007-05/msg00015.html https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=239344 The ABI part of that issue shouldn't really affect us build-everything-from-source folks, but the malloc debug thing might be worth switching off for performance reasons. Wonder if we'll see a 2.6.1? We never did see a point release from the 2.5 branch.. BTW, I've put up a new gsbuild tarball. Any issues with gsbuild should go directly to me, not the list. But feel free to post to the list any experiences with the new toolchain combo, including troubles with stuff beyond the base. Regards Greg From diy-linux-dev@diy-linux.org Thu May 24 01:37:24 2007 From: diy-linux-dev@diy-linux.org (Greg Schafer) Date: Thu, 24 May 2007 11:37:24 +1000 Subject: Status update In-Reply-To: <20070523010207.GA18037@eyo32.local> References: <20070523010207.GA18037@eyo32.local> Message-ID: <20070524013724.GA14633@eyo32.local> On Wed, May 23, 2007 at 11:02:07AM +1000, Greg Schafer wrote: > It looks like an unpleasant bug snuck into Glibc-2.6 just before release: > > http://sourceware.org/ml/libc-hacker/2007-05/msg00015.html > https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=239344 > > The ABI part of that issue shouldn't really affect us > build-everything-from-source folks, but the malloc debug thing might be > worth switching off for performance reasons. This should be good enough for now: sed -i.bak '/MALLOC_DEBUG/d' malloc/Makefile > Wonder if we'll see a 2.6.1? We never did see a point release from the 2.5 > branch.. We might be lucky: http://sourceware.org/ml/libc-ports/2007-05/msg00045.html Regards Greg