Julia: Pkg.build fails with LoadError: Unable to open libLLVM!

Created on 18 Oct 2018 · 25Comments · Source: JuliaLang/julia

Using Julia 0.7.0 on RHEL 7.5. Julia was installed from julia-epel-7 repo referenced in the documentation (https://copr.fedorainfracloud.org/coprs/nalimilan/julia/). Building any package fails with:

ERROR: LoadError: LoadError: Unable to open libLLVM!

llvm6-libs.x86_64-6.0.0-4.el7 package was installed from the julia repo. This rpm installs libLLVM to /usr/lib64/llvm6/lib/. For some reason the build process does not find the lib. I worked around the issue by creating a symlink in /usr/lib64:

ln -s /usr/lib64/llvm6/lib/libLLVM-6.0.0.so /usr/lib64/libLLVM.so

With this symlink in place, packages build successfully.

Julia versioninfo():

Julia Version 0.7.0
Commit a4cb80f3ed (2018-08-08 06:46 UTC)
Platform Info:
  OS: Linux (x86_64-redhat-linux)
  CPU: Intel(R) Xeon(R) CPU E7-8880 v3 @ 2.30GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, haswell)

build log:

ERROR: LoadError: LoadError: Unable to open libLLVM!
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] (::getfield(BinaryProvider, Symbol("#open_libllvm#114")))() at /bbkinghome/mleary/.julia/packages/BinaryProvider/1nGWd/src/PlatformNames.jl:650
 [3] detect_cxx11_string_abi() at /bbkinghome/mleary/.julia/packages/BinaryProvider/1nGWd/src/PlatformNames.jl:653
 [4] detect_compiler_abi() at /bbkinghome/mleary/.julia/packages/BinaryProvider/1nGWd/src/PlatformNames.jl:666
 [5] top-level scope at none:0
 [6] include at ./boot.jl:317 [inlined]
 [7] include_relative(::Module, ::String) at ./loading.jl:1038
 [8] include at ./sysimg.jl:29 [inlined]
 [9] include(::String) at /bbkinghome/mleary/.julia/packages/BinaryProvider/1nGWd/src/BinaryProvider.jl:1
 [10] top-level scope at none:0
 [11] top-level scope at none:2
in expression starting at /bbkinghome/mleary/.julia/packages/BinaryProvider/1nGWd/src/PlatformNames.jl:683
in expression starting at /bbkinghome/mleary/.julia/packages/BinaryProvider/1nGWd/src/BinaryProvider.jl:12
ERROR: LoadError: Failed to precompile BinaryProvider [b99e7846-7c00-51b0-8f62-c81ae34c0232] to /bbkinghome/mleary/.julia/compiled/v0.7/BinaryProvider/ek6VZ.ji.
Stacktrace:
 [1] include(::String) at ./client.jl:398
 [2] top-level scope at none:0
in expression starting at /bbkinghome/mleary/.julia/packages/Rmath/Py9gH/deps/build.jl:1

julia-epel-7.repo

[julia-epel-7]
name=Copr repo for julia owned by nalimilan
baseurl=https://copr-be.cloud.fedoraproject.org/results/nalimilan/julia/epel-7-$basearch/
enabled=True
gpgcheck=True
gpgkey=https://copr-be.cloud.fedoraproject.org/results/nalimilan/julia/pubkey.gpg
skip_if_unavailable=True

Source

markleary

👍1

Most helpful comment

I'm having the same issue on macOS Mojave, 10.14 with an anaconda setup and Julia v1.0.1

I create a new environment, install julia via conda install -c conda-forge julia

In order to access Julia via jupyter notebooks I do the following:

using Pkg Pkg.add("IJulia")

All goes fab,but then then opening jupyter notebook and selecting Julia yields the following:

Blosc not installed properly, run Pkg.build("ZMQ"), restart Julia and try again

Hoping for a successful resolution of the above, I run:

using Pkg Pkg.build("ZMQ")

Which slaps me back with the error that we are all experiencing here:

Error: Error buildingZMQ: │ ERROR: LoadError: LoadError: Unable to open libLLVM!

Is there anything wrong with the process I'm following? It's the first time I'm trying to get Julia to work with Jupyter Notebook.

Julia info

Julia Version 1.0.1 Platform Info: OS: macOS (x86_64-apple-darwin16.7.0) CPU: Intel(R) Core(TM) i7-6920HQ CPU @ 2.90GHz WORD_SIZE: 64 LIBM: libopenlibm LLVM: libLLVM-6.0.0 (ORCJIT, skylake)

I also have such problem!

MaxBalashov on 29 Oct 2018

👍11

All 25 comments

AFAIK this is fixed with 1.0.1. Why don't you update?

nalimilan on 18 Oct 2018

@nalimilan Even on 1.0.1, I run into a similar problem (ERROR: LoadError: LoadError: Unable to open libLLVM!) when attempting to build ZMQ and MBedTLS, as part of the IJulia build. Could you kindly point me to the fix?

egnha on 25 Oct 2018

👍3

I'm having the same issue as @egnha, though I'm on Ubuntu 18.

moppymopperson on 25 Oct 2018

👍1

@egnha I can't reproduce on Fedora 28 nor on Centos 7. What OS and what exact version of the package are you using?

Cc: @cdluminate for Debian/Ubuntu.

nalimilan on 26 Oct 2018

@nalimilan Cannot reproduce on both amd64 and ppc64el. Tested with julia version Debian ⛬ julia/1.0.1-2

cdluminate on 27 Oct 2018

Maybe something wrong with the shared lib / symlink /usr/lib/x86_64-linux-gnu/julia/libLLVM.so ?

These files / symlinks are included in the official binary tarball.

~/b/julia-1.0.0 ❯❯❯ find | rg LLVM
./lib/julia/libLLVM-6.0.so
./lib/julia/libLLVM.so
./lib/julia/libLLVM-6.0.0.so

cdluminate on 27 Oct 2018

I'm having the same issue on macOS Mojave, 10.14 with an anaconda setup and Julia v1.0.1

I create a new environment, install julia via conda install -c conda-forge julia

In order to access Julia via jupyter notebooks I do the following:

using Pkg Pkg.add("IJulia")

All goes fab,but then then opening jupyter notebook and selecting Julia yields the following:

Blosc not installed properly, run Pkg.build("ZMQ"), restart Julia and try again

Hoping for a successful resolution of the above, I run:

using Pkg Pkg.build("ZMQ")

Which slaps me back with the error that we are all experiencing here:

Error: Error buildingZMQ: │ ERROR: LoadError: LoadError: Unable to open libLLVM!

Is there anything wrong with the process I'm following? It's the first time I'm trying to get Julia to work with Jupyter Notebook.

Julia info

Julia Version 1.0.1 Platform Info: OS: macOS (x86_64-apple-darwin16.7.0) CPU: Intel(R) Core(TM) i7-6920HQ CPU @ 2.90GHz WORD_SIZE: 64 LIBM: libopenlibm LLVM: libLLVM-6.0.0 (ORCJIT, skylake)

nazywamsiepawel on 29 Oct 2018

😕3 👍1

I'm having the same issue on macOS Mojave, 10.14 with an anaconda setup and Julia v1.0.1

I create a new environment, install julia via conda install -c conda-forge julia

In order to access Julia via jupyter notebooks I do the following:

using Pkg Pkg.add("IJulia")

All goes fab,but then then opening jupyter notebook and selecting Julia yields the following:

Blosc not installed properly, run Pkg.build("ZMQ"), restart Julia and try again

Hoping for a successful resolution of the above, I run:

using Pkg Pkg.build("ZMQ")

Which slaps me back with the error that we are all experiencing here:

Error: Error buildingZMQ: │ ERROR: LoadError: LoadError: Unable to open libLLVM!

Is there anything wrong with the process I'm following? It's the first time I'm trying to get Julia to work with Jupyter Notebook.

Julia info

Julia Version 1.0.1 Platform Info: OS: macOS (x86_64-apple-darwin16.7.0) CPU: Intel(R) Core(TM) i7-6920HQ CPU @ 2.90GHz WORD_SIZE: 64 LIBM: libopenlibm LLVM: libLLVM-6.0.0 (ORCJIT, skylake)

I also have such problem!

MaxBalashov on 29 Oct 2018

👍11

@nazywamsiepawel Exactly the same problem for me !

quentinlouis on 1 Nov 2018

👍1

@nazywamsiepawel I am having same issue, Mojave 10.14, installing Julia via anaconda and trying to use jupyter. However, I was able to to get Julia working with juypter notebooks by avoiding Anaconda altogether. Using homebrew to install Julia, then running

using IJulia 
notebook()

worked for me. However I this is only the notebooks and I still want to be able to use JuypterLab

edit: fixed it by adding ENV["JUPYTER"]="jupyter"

rnevils on 1 Nov 2018

👍1

I'm having the exact same Issue when I tried to use IJulia via Anaconda. But problem solved when I switched to JuliaPro.

liushuya7 on 1 Nov 2018

same issue on ubuntu 18.04 when julia is installed via anaconda.

corrected with:
ln -s /usr/lib/llvm-6.0/lib/libLLVM-6.0.0.so /usr/lib/libLLVM.so

zlewko on 4 Nov 2018

👍7 🎉3

Same issue for me on Ubuntu 16.04, conda 4.5.11, Julia 1.0.1

mgei on 11 Nov 2018

same issue on ubuntu 18.04 when julia is installed via anaconda.

corrected with:
ln -s /usr/lib/llvm-6.0/lib/libLLVM-6.0.0.so /usr/lib/libLLVM.so

Yeah but I guess the solution differs in Mac. Can somebody show me where I can find libLLVM-6.0.0 in Mac system? I performed the same installation steps of the same version of Julia in a same Mac system with @nazywamsiepawel and I even do not know where to find that library.

GabrielEdogawa on 14 Nov 2018

👍2

Confirming the same issue on Ubuntu 18.10, Julia 1.0.2, conda 4.5.11.

Also resolved with linking:
ln -s /usr/lib/llvm-5.0/lib/libLLVM-5.0.0.so /usr/lib/libLLVM.so

versipellis on 19 Nov 2018

Confirming the same issue on Ubuntu 18.10, Julia 1.0.3, conda 4.5.12.

Fixed by softlinking as above.

mbudisic on 2 Jan 2019

👍1

Same here, ln -s fixed. Obviously the llvm's library path may slightly differ on different linux distros (and which package one uses to provide it) or other operatiing systems. For my Ubuntu 18.10 setup, it happened to be provided by a few packages including libjulia1.

Similarly fixed it with tracking the file down and making the libLLVM available in the more generic library path:

$ dpkg -S libLLVM.so 
llvm-6.0-dev: /usr/lib/llvm-6.0/lib/libLLVM.so
libjulia1: /usr/lib/x86_64-linux-gnu/julia/libLLVM.so
$ sudo ln -s /usr/lib/x86_64-linux-gnu/julia/libLLVM.so /usr/lib/libLLVM.so
$

The install requiring it, (Pkg.add("IJulia") in my case) worked smoothly after.

yuvilio on 6 Jan 2019

🎉1

Maybe this issue should be merge with (or vice versa) https://github.com/JuliaPackaging/BinaryProvider.jl/issues/130 ?

It also seems to be discussed here: https://github.com/conda-forge/julia-feedstock/issues/60

krinsman on 17 Jan 2019

Since this is an issue with the conda-forge julia package, the solution might be found in the current build script for that package, found here:

https://github.com/conda-forge/julia-feedstock/blob/master/recipe/build.sh

In particular, note that the package does not install the conda-forge llvm package for Julia to use by default, and that the following flag is used in the make: USE_SYSTEM_LLVM=0.

krinsman on 17 Jan 2019

Same issue on OpenSUSE 15 with julia 1.0.3 from the science repo, here we need ln -s /usr/lib64/libLLVM.so.5 /usr/lib64/libLLVM.so.
An alternative without root access is to link to a user accessible directory on LD_LIBRARY_PATH:
$ mkdir $HOME/lib64 $ ln -s /usr/lib64/libLLVM.so.5 $HOME/lib64/libLLVM.so $ export LD_LIBRARY_PATH=$HOME/lib64/.$LD_LIBRARY_PATH
(I had some error trying this out for the inital version of this post)

j-fu on 4 Apr 2019

🎉1

Same issue on OpenSUSE 15 with julia 1.0.3 from the science repo, here we need ln -s /usr/lib64/libLLVM.so.5 /usr/lib64/libLLVM.so.
An alternative without root access is to link to a user accessible directory on LD_LIBRARY_PATH:
$ mkdir $HOME/lib64
$ ln -s /usr/lib64/libLLVM.so.5 $HOME/lib64/libLLVM.so
$ export LD_LIBRARY_PATH=$HOME/lib64/.$LD_LIBRARY_PATH

Update (thanks to our sysadmin): OpenSuse 15 has a package llvm-devel which creates the link.
So this might be an issue of package dependencies to be addressed in the distro.

j-fu on 5 Apr 2019

❤1

Can you file an issue against the OpenSUSE repo (and post the link here)?

nalimilan on 5 Apr 2019

Same issue on OpenSUSE 15 with julia 1.0.3 from the science repo, here we need ln -s /usr/lib64/libLLVM.so.5 /usr/lib64/libLLVM.so.
An alternative without root access is to link to a user accessible directory on LD_LIBRARY_PATH:
$ mkdir $HOME/lib64
$ ln -s /usr/lib64/libLLVM.so.5 $HOME/lib64/libLLVM.so
$ export LD_LIBRARY_PATH=$HOME/lib64/.$LD_LIBRARY_PATH
Update (thanks to our sysadmin): OpenSuse 15 has a package llvm-devel which creates the link.
So this might be an issue of package dependencies to be addressed in the distro.
Update: Similar discussion is already running on https://bugzilla.opensuse.org/show_bug.cgi?id=1114692 .
I put in my 2cts in there and gave the link to the discussion here.

j-fu on 5 Apr 2019

Thanks. Actually it looks like the problem is just that OpenSUSE should use USE_LLVM_SHLIB=1 instead of USE_LLVM_SHLIB=0 in the juliabuildopts file.

nalimilan on 5 Apr 2019

Thanks for the hint. Over at OpenSuse they tried it and it worked: https://bugzilla.opensuse.org/show_bug.cgi?id=1114692#c7
On Tumbleweed I was able to check this - it is there now.

j-fu on 7 Oct 2019

Was this page helpful?

0 / 5 - 0 ratings