R fails with version `GOMP_4.0' not found

[vbraun]

Compiling R with gcc 4.9.2 can fail because of OpenMP library version mismatch. This is on the "snapperkob" build bot:

/home/buildbot/build/sage/snapperkob/sage_git/build/local/var/tmp/sage/build/r-3.1.1.p0/src/bin/exec/R: /usr/lib/x86_64-linux-gnu/libgomp.so.1: version `GOMP_4.0' not found (required by /home/buildbot/build/sage/snapperkob/sage_git/build/local/var/tmp/sage/build/r-3.1.1.p0/src/lib/libR.so)
make[6]: *** [all] Error 1
make[6]: Leaving directory `/home/buildbot/build/sage/snapperkob/sage_git/build/local/var/tmp/sage/build/r-3.1.1.p0/src/src/library/tools'
make[5]: *** [R] Error 1
make[5]: Leaving directory `/home/buildbot/build/sage/snapperkob/sage_git/build/local/var/tmp/sage/build/r-3.1.1.p0/src/src/library'
make[4]: *** [R] Error 1
make[4]: Leaving directory `/home/buildbot/build/sage/snapperkob/sage_git/build/local/var/tmp/sage/build/r-3.1.1.p0/src/src'
make[3]: *** [R] Error 1
make[3]: Leaving directory `/home/buildbot/build/sage/snapperkob/sage_git/build/local/var/tmp/sage/build/r-3.1.1.p0/src'
Error building R.

Component: packages: standard

Author: Volker Braun

Branch: d872869

Reviewer: François Bissey

Issue created by migration from https://trac.sagemath.org/ticket/17349

[vbraun]

Branch: u/vbraun/r_fails_with_versiongomp_4_0not_found

[vbraun]

Commit: e3d33ca

[vbraun]

New commits:

e3d33ca

Disable openmp in binary builds

[vbraun]

Author: Volker Braun

[kcrisman]

comment:4

Hmm, does that mean that anytime anyone uploads a binary for use that one should use SAGE_FAT_BINARY? Or is this necessary in general for any old Sage build?

[vbraun]

comment:5

If you want to share binaries then you should set SAGE_FAT_BINARY=yes unless you know that the processor and os matches.

[kcrisman]

comment:6

Luckily, that is nearly the case with Mac binaries. But I'll keep it in mind.

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

d872869

Always use out ATLAS

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from e3d33ca to d872869

[kiwifb]

comment:8

lgtm

[vbraun]

Reviewer: François Bissey

[vbraun]

Changed branch from u/vbraun/r_fails_with_versiongomp_4_0not_found to d872869

[dimpase]

Changed commit from d872869 to none

[dimpase]

comment:11

On Ubuntu 14.04 ldd will tell you that something is fishy even if Sage (6.8.beta2) builds and passes ptestlong

$ ldd R/lib/libR.so
R/lib/libR.so: /usr/lib/x86_64-linux-gnu/libgomp.so.1: version `GOMP_4.0' not found (required by R/lib/libR.so)
        linux-vdso.so.1 =>  (0x00007fff066ae000)
        libf77blas.so.3 => not found
        libatlas.so.3 => not found
        libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007f8b89e3a000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f8b89b34000)
        libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00007f8b898f8000)
        libreadline.so.6 => /lib/x86_64-linux-gnu/libreadline.so.6 (0x00007f8b896b1000)
        libbz2.so.1 => /lib/x86_64-linux-gnu/libbz2.so.1 (0x00007f8b894a1000)
        libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00007f8b89288000)
        librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f8b8907f000)
        libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f8b88e7b000)
        libgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007f8b88c6c000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f8b88a4d000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f8b88688000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f8b8a7aa000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f8b88472000)
        libtinfo.so.5 => /lib/x86_64-linux-gnu/libtinfo.so.5 (0x00007f8b88248000)

Sure this does not look like a real bug, but still...

see also https://groups.google.com/d/msg/sage-devel/X_dVGq5gWoo/FiSnIPzO8FEJ

[kiwifb]

comment:12

Replying to @dimpase:

On Ubuntu 14.04 ldd will tell you that something is fishy even if Sage (6.8.beta2) builds and passes ptestlong

$ ldd R/lib/libR.so
R/lib/libR.so: /usr/lib/x86_64-linux-gnu/libgomp.so.1: version `GOMP_4.0' not found (required by R/lib/libR.so)
        linux-vdso.so.1 =>  (0x00007fff066ae000)
        libf77blas.so.3 => not found
        libatlas.so.3 => not found
        libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007f8b89e3a000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f8b89b34000)
        libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00007f8b898f8000)
        libreadline.so.6 => /lib/x86_64-linux-gnu/libreadline.so.6 (0x00007f8b896b1000)
        libbz2.so.1 => /lib/x86_64-linux-gnu/libbz2.so.1 (0x00007f8b894a1000)
        libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00007f8b89288000)
        librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f8b8907f000)
        libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f8b88e7b000)
        libgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007f8b88c6c000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f8b88a4d000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f8b88688000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f8b8a7aa000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f8b88472000)
        libtinfo.so.5 => /lib/x86_64-linux-gnu/libtinfo.so.5 (0x00007f8b88248000)

Sure this does not look like a real bug, but still...

see also https://groups.google.com/d/msg/sage-devel/X_dVGq5gWoo/FiSnIPzO8FEJ

Have you gone to a sage shell (./sage -sh) before running ldd? It looks suspiciously like LD_LIBRARY_PATH is not set.

[dimpase]

comment:13

Replying to @kiwifb:

Have you gone to a sage shell (./sage -sh) before running ldd? It looks suspiciously like LD_LIBRARY_PATH is not set.

Oops, indeed, you're right. But this means that I can only build Sage on this system using Sage's gcc rather than the system gcc, as libopenmp comes from Sage, and it seems that gcc is the only Sage package it could come from... (Otherwise R won't build without setting SAGE_FAT_BINARY=yes)

[kiwifb]

comment:14

Yes that must means that your system gcc has been deemed inadequate and sage has been built with its own gcc. Is this a problem?

[dimpase]

comment:15

Replying to @kiwifb:

Yes that must means that your system gcc has been deemed inadequate and sage has been built with its own gcc. Is this a problem?

Actually, it's all OK; the system gcc uses system's libopenmp, while Sage's gcc needs a newer version, coming from gcc's spkg. Sorry for barking up the wrong tree.