[gentoo-dev] glibc build problem in make check

[gentoo-dev] glibc build problem in make check

[Pete Gavin]

Hi,

I tried building glibc-2.2.1 again, this time under linux-2.4.0.11-r3,
to see if the problem was related to a reiserfs problem in the
kernel. I got the same error that I got under linux-2.4.1-pre8. This
happened under the make check stage of the build. Do you know anything
about this, Achim?


GCONV_PATH=/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/iconvdata LC_ALL=C   /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/elf/ld-linux.so.2 --library-path /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/work/glibc-2.2.1/buildhere/math:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/elf:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/dlfcn:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/nss:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buidhere/nis:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/rt:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/resolv:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/crypt:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/linuxthreads /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs  > /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs.out
Timed out: killed the child process
make[2]: *** [/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs.out] Error 1
make[2]: Leaving directory `/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/io'
make[1]: *** [io/tests] Error 2
make[1]: Leaving directory `/tmp/portage/glibc-2.2.1/work/glibc-2.2.1'
make: *** [check] Error 2

Pete
<pete@gentoo.org>

Re: [gentoo-dev] glibc build problem in make check

[Pete Gavin]

On Tue, Feb 06, 2001 at 02:23:11PM -0700, Pete Gavin wrote:
> Hi,
> 
> I tried building glibc-2.2.1 again, this time under linux-2.4.0.11-r3,
> to see if the problem was related to a reiserfs problem in the
> kernel. I got the same error that I got under linux-2.4.1-pre8. This
> happened under the make check stage of the build. Do you know anything
> about this, Achim?
> 
> 
[snip]

I just reran make check in
/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere, and I didn't get
that problem. It didn't happen when I ran make check manually under
linux-2.4.1-pre8. I'm thinking the problem is one of 3 things: a)
there's some kind of problem w/ the ebuild script, or b) it just
doesn't happen all the time, or c) for one reason or another, make
check doesn't bother to run the lfs check the second time around. I
haven't checked to see if c) it the case yet, but I'm pretty sure
that's not it. I don't see how the ebuild script would cause that to
happen either, so I'm guessing it just doesn't always happen.

Pete

Re: [gentoo-dev] glibc build problem in make check

[drobbins]

On Tue, Feb 06, 2001 at 02:23:11PM -0700, Pete Gavin wrote:
> Hi,
> 
> I tried building glibc-2.2.1 again, this time under linux-2.4.0.11-r3,
> to see if the problem was related to a reiserfs problem in the
> kernel. I got the same error that I got under linux-2.4.1-pre8. This
> happened under the make check stage of the build. Do you know anything
> about this, Achim?
> 
> 
> GCONV_PATH=/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/iconvdata LC_ALL=C   /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/elf/ld-linux.so.2 --library-path /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/work/glibc-2.2.1/buildhere/math:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/elf:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/dlfcn:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/nss:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buidhere/nis:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/rt:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/resolv:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/crypt:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/linuxthreads /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs  > /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs.out
> Timed out: killed the child process
> make[2]: *** [/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs.out] Error 1
> make[2]: Leaving directory `/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/io'
> make[1]: *** [io/tests] Error 2
> make[1]: Leaving directory `/tmp/portage/glibc-2.2.1/work/glibc-2.2.1'
> make: *** [check] Error 2

I did a google search, and found that another person had that same problem
but with kernel 2.2.17.  I don't think it's a kernel issue.  It's either
a problem with glibc or binutils.  In a response, Andreas of SuSE says:

<quote>
Update to newer binutils,  2.10.1 should be ok.

Andreas
-- 
Andreas Jaeger
SuSE Labs aj@suse.de
private aj@arthur.inka.de
http://www.suse.de/~aj
</quote>

So maybe we're dealing with a buggy binutils?  It's happened before.

-- 
Daniel Robbins					<drobbins@gentoo.org>
President/CEO					http://www.gentoo.org 
Gentoo Technologies, Inc.			

Re: [gentoo-dev] glibc build problem in make check

[Pete Gavin]

On Tue, Feb 06, 2001 at 02:42:18PM -0700, drobbins@gentoo.org wrote:
> I did a google search, and found that another person had that same problem
> but with kernel 2.2.17.  I don't think it's a kernel issue.  It's either
> a problem with glibc or binutils.  In a response, Andreas of SuSE says:
> 
> <quote>
> Update to newer binutils,  2.10.1 should be ok.
> 

Yeah, I've got binutils 2.10.1, and gcc-2.95.2. Maybe the bug still
exists in the binutils 2.10.1, or there's a problem in the reiserfs
lfs code, or something like that.

Pete

Re: [gentoo-dev] glibc build problem in make check

[drobbins]

On Tue, Feb 06, 2001 at 05:32:13PM -0700, Pete Gavin wrote:
> On Tue, Feb 06, 2001 at 02:42:18PM -0700, drobbins@gentoo.org wrote:
> > I did a google search, and found that another person had that same problem
> > but with kernel 2.2.17.  I don't think it's a kernel issue.  It's either
> > a problem with glibc or binutils.  In a response, Andreas of SuSE says:
> > 
> > <quote>
> > Update to newer binutils,  2.10.1 should be ok.
> > 
> 
> Yeah, I've got binutils 2.10.1, and gcc-2.95.2. Maybe the bug still
> exists in the binutils 2.10.1, or there's a problem in the reiserfs
> lfs code, or something like that.

I doubt it's a ReiserFS problem -- that error is generated by the glibc
test skeleton.

-- 
Daniel Robbins					<drobbins@gentoo.org>
President/CEO					http://www.gentoo.org 
Gentoo Technologies, Inc.			

Re: [gentoo-dev] glibc build problem in make check

[Achim Gottinger]

Pete Gavin wrote:

> On Tue, Feb 06, 2001 at 02:23:11PM -0700, Pete Gavin wrote:
> > Hi,
> >
> > I tried building glibc-2.2.1 again, this time under linux-2.4.0.11-r3,
> > to see if the problem was related to a reiserfs problem in the
> > kernel. I got the same error that I got under linux-2.4.1-pre8. This
> > happened under the make check stage of the build. Do you know anything
> > about this, Achim?
> >
> >
> [snip]
>
> I just reran make check in
> /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere, and I didn't get
> that problem. It didn't happen when I ran make check manually under
> linux-2.4.1-pre8. I'm thinking the problem is one of 3 things: a)
> there's some kind of problem w/ the ebuild script, or b) it just
> doesn't happen all the time, or c) for one reason or another, make
> check doesn't bother to run the lfs check the second time around. I
> haven't checked to see if c) it the case yet, but I'm pretty sure
> that's not it. I don't see how the ebuild script would cause that to
> happen either, so I'm guessing it just doesn't always happen.
>

Try to reproduce the error with DMA turned off for your disks.
Again. Here this only happend once under very heavy load and I build glibc
every two days or so.

achim~

>
> Pete
>
> _______________________________________________
> gentoo-dev mailing list
> gentoo-dev@gentoo.org
> http://www.gentoo.org/mailman/listinfo/gentoo-dev

Re: [gentoo-dev] glibc build problem in make check

[Achim Gottinger]

Pete Gavin wrote:

> Hi,
>
> I tried building glibc-2.2.1 again, this time under linux-2.4.0.11-r3,
> to see if the problem was related to a reiserfs problem in the
> kernel. I got the same error that I got under linux-2.4.1-pre8. This
> happened under the make check stage of the build. Do you know anything
> about this, Achim?
>
> GCONV_PATH=/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/iconvdata LC_ALL=C   /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/elf/ld-linux.so.2 --library-path /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/work/glibc-2.2.1/buildhere/math:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/elf:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/dlfcn:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/nss:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buidhere/nis:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/rt:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/resolv:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/crypt:/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/linuxthreads /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs  > /tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs.out
> Timed out: killed the child process
> make[2]: *** [/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/buildhere/io/test-lfs.out] Error 1
> make[2]: Leaving directory `/tmp/portage/glibc-2.2.1/work/glibc-2.2.1/io'
> make[1]: *** [io/tests] Error 2
> make[1]: Leaving directory `/tmp/portage/glibc-2.2.1/work/glibc-2.2.1'
> make: *** [check] Error 2

Hi Pete,

I just updated glibc to 2.2.2 and that error now occures every time.
I simply increased the timeout in io/test-lfs.c and now it works. :-/
I posted that problem to the glibc-bug newsgroup.

achim~

>
>
> Pete
> <pete@gentoo.org>
>
> _______________________________________________
> gentoo-dev mailing list
> gentoo-dev@gentoo.org
> http://www.gentoo.org/mailman/listinfo/gentoo-dev

Re: [gentoo-dev] glibc build problem in make check

[drobbins]

On Sat, Feb 17, 2001 at 03:17:29AM +0100, Achim Gottinger wrote:

> Hi Pete,
> 
> I just updated glibc to 2.2.2 and that error now occures every time.
> I simply increased the timeout in io/test-lfs.c and now it works. :-/
> I posted that problem to the glibc-bug newsgroup.
> 
> achim~

Awesome work.  Because we're on the "cutting edge", oftentimes it's more likely
that we'll discover bugs before anyone else does.  Thanks for doing the right
thing and sending the bug report upstream to glibc-bug.  And it's great to be
working with a team that can resolve thorny problems :)

Best Regards,

-- 
Daniel Robbins					<drobbins@gentoo.org>
President/CEO					http://www.gentoo.org 
Gentoo Technologies, Inc.			

Re: [gentoo-dev] glibc build problem in make check

[Achim Gottinger]

drobbins@gentoo.org wrote:

> On Sat, Feb 17, 2001 at 03:17:29AM +0100, Achim Gottinger wrote:
>
> > Hi Pete,
> >
> > I just updated glibc to 2.2.2 and that error now occures every time.
> > I simply increased the timeout in io/test-lfs.c and now it works. :-/
> > I posted that problem to the glibc-bug newsgroup.
> >
> > achim~
>
> Awesome work.  Because we're on the "cutting edge", oftentimes it's more likely
> that we'll discover bugs before anyone else does.  Thanks for doing the right
> thing and sending the bug report upstream to glibc-bug.  And it's great to be
> working with a team that can resolve thorny problems :)

Thank you for the flowers.

Here is the reply I got:

----------------------------------------------------------------
> increase the timeout value.

If this is necessary something is wrong with your system.  The test is
doing almost no work.  It terminates almost instantly.  I think you
might want to look at the problems.

--
---------------.                          ,-.   1325 Chesapeake Terrace
Ulrich Drepper  \    ,-------------------'   \  Sunnyvale, CA 94089 USA
Red Hat          `--' drepper at redhat.com   `------------------------
------------------------------------------------------------------


So I looked into test-lfs.c and found that it tries to write to a tmp file in
/tmp.
So I created a ext2 partition and mounted that to /tmp.
After that test-lfs runs in under a second. With /tmp on reiserfs it takes 27
sec's.
So my first explanation that it is a reiserfs-low latency problem seems to be
correct.
I will update my kernel now to a more recent version and hope that the problem is
fixed.
Since Suse 7.1 ships with 2.4/reiserfs I think this problem is solved.

achim~

>
>
> Best Regards,
>
> --
> Daniel Robbins                                  <drobbins@gentoo.org>
> President/CEO                                   http://www.gentoo.org
> Gentoo Technologies, Inc.
>
> _______________________________________________
> gentoo-dev mailing list
> gentoo-dev@gentoo.org
> http://www.gentoo.org/mailman/listinfo/gentoo-dev

Re: [gentoo-dev] glibc build problem in make check

[Achim Gottinger]

>
> Since Suse 7.1 ships with 2.4/reiserfs I think this problem is solved.

Well it seems to be still unsolved.
I tested the lates kernel 2.4.1-ac17 with and without nfs and lvm patches all with
the same result.
I posted a message to the reiserfs mailinglist.

achim~

>
>
> achim~
>
> >
> >
> > Best Regards,
> >
> > --
> > Daniel Robbins                                  <drobbins@gentoo.org>
> > President/CEO                                   http://www.gentoo.org
> > Gentoo Technologies, Inc.
> >
> > _______________________________________________
> > gentoo-dev mailing list
> > gentoo-dev@gentoo.org
> > http://www.gentoo.org/mailman/listinfo/gentoo-dev
>
> _______________________________________________
> gentoo-dev mailing list
> gentoo-dev@gentoo.org
> http://www.gentoo.org/mailman/listinfo/gentoo-dev

Re: [gentoo-dev] glibc build problem in make check

[Achim Gottinger]

Achim Gottinger wrote:

> >
> > Since Suse 7.1 ships with 2.4/reiserfs I think this problem is solved.
>
> Well it seems to be still unsolved.
> I tested the lates kernel 2.4.1-ac17 with and without nfs and lvm patches all with
> the same result.
> I posted a message to the reiserfs mailinglist.
>

Here are the answers

Well, this is known weakness of reiserfs in its current state: it creates very
big holes in file very poorly.

Thanks,
vs

This isn't a bug, its a design flaw.  reiserfs creates a little slower than
ext2 does, and this test is really benchmarking the hole creation speed.
We don't do it quickly, but the holes do get created.

-chris

-------------------
I added a patch to the lates glibc-2.2.2 package that increases the timeout in
test-lfs so it should compile.
Since there are other problems with glibc-2.2.2/binutils-2.1.10.0.7 I will update
glibc-2.2.1 too.

achim~

Re: [gentoo-dev] glibc build problem in make check

[drobbins]

On Mon, Feb 19, 2001 at 06:28:14AM +0100, Achim Gottinger wrote:
> Achim Gottinger wrote:
> 
> > >
> > > Since Suse 7.1 ships with 2.4/reiserfs I think this problem is solved.
> >
> > Well it seems to be still unsolved.
> > I tested the lates kernel 2.4.1-ac17 with and without nfs and lvm patches all with
> > the same result.
> > I posted a message to the reiserfs mailinglist.
> >
> 
> Here are the answers
> 
> Well, this is known weakness of reiserfs in its current state: it creates very
> big holes in file very poorly.
> 
> Thanks,
> vs
> 
> This isn't a bug, its a design flaw.  reiserfs creates a little slower than
> ext2 does, and this test is really benchmarking the hole creation speed.
> We don't do it quickly, but the holes do get created.
> 
> -chris
> 
> -------------------
> I added a patch to the lates glibc-2.2.2 package that increases the timeout in
> test-lfs so it should compile.
> Since there are other problems with glibc-2.2.2/binutils-2.1.10.0.7 I will update
> glibc-2.2.1 too.

OK, thanks Achim.  Make sure to forward this information to Ulrich Drepper so that
maybe he can add a fix to the next version of glibc.

Best Regards,

-- 
Daniel Robbins					<drobbins@gentoo.org>
President/CEO					http://www.gentoo.org 
Gentoo Technologies, Inc.			

Re: [gentoo-dev] glibc build problem in make check

[Achim Gottinger]

drobbins@gentoo.org wrote:

> On Mon, Feb 19, 2001 at 06:28:14AM +0100, Achim Gottinger wrote:
> > Achim Gottinger wrote:
> >
> > > >
> > > > Since Suse 7.1 ships with 2.4/reiserfs I think this problem is solved.
> > >
> > > Well it seems to be still unsolved.
> > > I tested the lates kernel 2.4.1-ac17 with and without nfs and lvm patches all with
> > > the same result.
> > > I posted a message to the reiserfs mailinglist.
> > >
> >
> > Here are the answers
> >
> > Well, this is known weakness of reiserfs in its current state: it creates very
> > big holes in file very poorly.
> >
> > Thanks,
> > vs
> >
> > This isn't a bug, its a design flaw.  reiserfs creates a little slower than
> > ext2 does, and this test is really benchmarking the hole creation speed.
> > We don't do it quickly, but the holes do get created.
> >
> > -chris
> >
> > -------------------
> > I added a patch to the lates glibc-2.2.2 package that increases the timeout in
> > test-lfs so it should compile.
> > Since there are other problems with glibc-2.2.2/binutils-2.1.10.0.7 I will update
> > glibc-2.2.1 too.
>
> OK, thanks Achim.  Make sure to forward this information to Ulrich Drepper so that
> maybe he can add a fix to the next version of glibc.
>

I did this but it seems Ulrich is not interested. :-/

achim~

>
> Best Regards,
>
> --
> Daniel Robbins                                  <drobbins@gentoo.org>
> President/CEO                                   http://www.gentoo.org
> Gentoo Technologies, Inc.
>
> _______________________________________________
> gentoo-dev mailing list
> gentoo-dev@gentoo.org
> http://www.gentoo.org/mailman/listinfo/gentoo-dev