My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
When I set up mock 0.6.12 on my laptop running FC6 with all updates and run the same script to build the RPM, it builds just fine.
The target distribution for the mock build is FC6.
SELinux is disabled on the SL 4.4 box.
When I run mock with '--debug', there is a lot of information, but I have no idea what to look for or how to continue to troubleshoot this.
Can anyone offer any suggestions?
Tony
On Fri, 23 Feb 2007, Anthony Joseph Seward wrote:
My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
What rpm are you trying to build?
When I set up mock 0.6.12 on my laptop running FC6 with all updates and run the same script to build the RPM, it builds just fine.
The target distribution for the mock build is FC6.
SELinux is disabled on the SL 4.4 box.
When I run mock with '--debug', there is a lot of information, but I have no idea what to look for or how to continue to troubleshoot this.
Can anyone offer any suggestions?
Tony
-- Fedora-buildsys-list mailing list Fedora-buildsys-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-buildsys-list
-Connie Sieh
On Fri, 2007-02-23 at 17:42 -0600, Connie Sieh wrote:
On Fri, 23 Feb 2007, Anthony Joseph Seward wrote:
My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
What rpm are you trying to build?
An in-house application for work.
When I set up mock 0.6.12 on my laptop running FC6 with all updates and run the same script to build the RPM, it builds just fine.
The target distribution for the mock build is FC6.
SELinux is disabled on the SL 4.4 box.
When I run mock with '--debug', there is a lot of information, but I have no idea what to look for or how to continue to troubleshoot this.
Can anyone offer any suggestions?
Tony
-- Fedora-buildsys-list mailing list Fedora-buildsys-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-buildsys-list
-Connie Sieh
-- Fedora-buildsys-list mailing list Fedora-buildsys-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-buildsys-list
Anthony Joseph Seward wrote:
On Fri, 2007-02-23 at 17:42 -0600, Connie Sieh wrote:
On Fri, 23 Feb 2007, Anthony Joseph Seward wrote:
My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
What rpm are you trying to build?
An in-house application for work.
if you post your .src.rpm online somewhere, someone might be able to help debug the issue.
On Sat, 2007-02-24 at 00:13 +0000, Karanbir Singh wrote:
Anthony Joseph Seward wrote:
On Fri, 2007-02-23 at 17:42 -0600, Connie Sieh wrote:
On Fri, 23 Feb 2007, Anthony Joseph Seward wrote:
My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
What rpm are you trying to build?
An in-house application for work.
if you post your .src.rpm online somewhere, someone might be able to help debug the issue.
I don't think the .src.rpm is the issue.
Again, using the same version of mock, the RPM will build when mock is hosted on an FC 6 system but not when running SL 4.4.
Tony
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Anthony Joseph Seward wrote:
On Sat, 2007-02-24 at 00:13 +0000, Karanbir Singh wrote:
Anthony Joseph Seward wrote:
On Fri, 2007-02-23 at 17:42 -0600, Connie Sieh wrote:
On Fri, 23 Feb 2007, Anthony Joseph Seward wrote:
My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
What rpm are you trying to build?
An in-house application for work.
if you post your .src.rpm online somewhere, someone might be able to help debug the issue.
I don't think the .src.rpm is the issue.
Again, using the same version of mock, the RPM will build when mock is hosted on an FC 6 system but not when running SL 4.4.
Tony
Hmmm. That sounds like the version of gfortran on SL4.4 has a problem. What's the version of the fortran package (gcc4-gfortran or gcc-gfortran)?
Also, if you run the build by hand on SL4.4, do you still get the internal compiler error? If not, then there's something going on in the environment set up by mock. If you do, then you have a toolchain issue with gfortran and will need to take it up with your distro vendor :).
Clark
On Mon, 2007-02-26 at 09:17 -0600, Clark Williams wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Anthony Joseph Seward wrote:
On Sat, 2007-02-24 at 00:13 +0000, Karanbir Singh wrote:
Anthony Joseph Seward wrote:
On Fri, 2007-02-23 at 17:42 -0600, Connie Sieh wrote:
On Fri, 23 Feb 2007, Anthony Joseph Seward wrote:
My situation is this: my build system is running Scientific Linux 4.4 (basically RHEL 4) with mock 0.6.12. It fails to build a particular RPM when trying to compile a fortran file. The error issued is
mkscrlib.f:381: internal compiler error: Illegal instruction
What rpm are you trying to build?
An in-house application for work.
if you post your .src.rpm online somewhere, someone might be able to help debug the issue.
I don't think the .src.rpm is the issue.
Again, using the same version of mock, the RPM will build when mock is hosted on an FC 6 system but not when running SL 4.4.
Tony
Hmmm. That sounds like the version of gfortran on SL4.4 has a problem. What's the version of the fortran package (gcc4-gfortran or gcc-gfortran)?
Also, if you run the build by hand on SL4.4, do you still get the internal compiler error? If not, then there's something going on in the environment set up by mock. If you do, then you have a toolchain issue with gfortran and will need to take it up with your distro vendor :).
Here are the build attempts:
Target: SL 4.4 (gcc 3.4, g77) Native build: OK Mock build: FC 6 host: OK SL 4.4 host: OK
Target: FC 6 (gcc 4.1, gfortran) Native build: OK Mock build: FC 6 host: OK SL 4.4 host: FAIL: internal compiler error using gfortran
This leads me to believe that there is a problem with mock, but I have no idea where to look.
Tony
On Mon, Feb 26, 2007 at 09:58:54AM -0700, Anthony Joseph Seward wrote:
Here are the build attempts:
Target: SL 4.4 (gcc 3.4, g77) Native build: OK Mock build: FC 6 host: OK SL 4.4 host: OK
Target: FC 6 (gcc 4.1, gfortran) Native build: OK Mock build: FC 6 host: OK SL 4.4 host: FAIL: internal compiler error using gfortran
This leads me to believe that there is a problem with mock, but I have no idea where to look.
I tend to doubt that it would be a mock problem. It is more likely a problem where the libc/other_random_component on SL4.4 is expecting a kernel feature that is present on SL4.4 but not FC6. I ran into something similar where (IIRC) < FC3 host could not build packages for >=FC6 target because the FC6 libc was expecting a certain kernel feature not enabled on FC3.
This, afaict, is the only hole in the mock process: that we still rely on the host kernel. If the target build is relying on some kernel feature that can change from kernel to kernel, you are stuck. Something that cannot be avoided, though.
You might instrument the build to get an strace of what is going on when you get the internal compiler error. -- Michael
On Mon, 2007-02-26 at 11:40 -0600, Michael E Brown wrote:
On Mon, Feb 26, 2007 at 09:58:54AM -0700, Anthony Joseph Seward wrote:
Here are the build attempts:
Target: SL 4.4 (gcc 3.4, g77) Native build: OK Mock build: FC 6 host: OK SL 4.4 host: OK
Target: FC 6 (gcc 4.1, gfortran) Native build: OK Mock build: FC 6 host: OK SL 4.4 host: FAIL: internal compiler error using gfortran
This leads me to believe that there is a problem with mock, but I have no idea where to look.
I tend to doubt that it would be a mock problem. It is more likely a problem where the libc/other_random_component on SL4.4 is expecting a kernel feature that is present on SL4.4 but not FC6. I ran into something similar where (IIRC) < FC3 host could not build packages for
=FC6 target because the FC6 libc was expecting a certain kernel
feature not enabled on FC3.
This, afaict, is the only hole in the mock process: that we still rely on the host kernel. If the target build is relying on some kernel feature that can change from kernel to kernel, you are stuck. Something that cannot be avoided, though.
You might instrument the build to get an strace of what is going on when you get the internal compiler error. -- Michael
Thanks. I'll try that later this week.
Tony
On Mon, 2007-02-26 at 11:40 -0600, Michael E Brown wrote: <snip>
I tend to doubt that it would be a mock problem. It is more likely a problem where the libc/other_random_component on SL4.4 is expecting a kernel feature that is present on SL4.4 but not FC6. I ran into something similar where (IIRC) < FC3 host could not build packages for
=FC6 target because the FC6 libc was expecting a certain kernel
feature not enabled on FC3.
This, afaict, is the only hole in the mock process: that we still rely on the host kernel. If the target build is relying on some kernel feature that can change from kernel to kernel, you are stuck. Something that cannot be avoided, though.
You might instrument the build to get an strace of what is going on when you get the internal compiler error. -- Michael
-- Fedora-buildsys-list mailing list Fedora-buildsys-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-buildsys-list
I did builds in both environments with 'strace -eall -e verbose=all'. Most of the result is Greek to me. The only thing that I saw was differences in GCC's garbage collection parameters. I tried making those equal on both platforms. I know that is not supposed to affect code generation, but I thought it was worth a try.
I've attached the relevant logs, if anyone can read Greek.
mkscrlib.f.log-02 is the log when SL 4.4 is the build system
mkscrlib.f.log-03 is the log when FC 6 is the build system
Thanks, Tony
buildsys@lists.fedoraproject.org
-
Anthony Joseph Seward -
Clark Williams -
Connie Sieh -
Karanbir Singh -
Michael E Brown