Calling Julia from Lua

Question

This example code previously appeared on a blog posting by Aditya, with one minor adjustments. I tried to run it on Debian 10 (buster), but it failed with the following error.

texlua julia.lua
ERROR: could not load library "/usr/lib/x86_64-linux-gnu/../bin/../lib/x86_64-linux-gnu/julia/sys.so"
/usr/lib/x86_64-linux-gnu/../bin/../lib/x86_64-linux-gnu/julia/sys.so: cannot open shared object file: No such file or directory

I tried this with both the default Julia version 1.0.3+dfsg-4, as well as the current unstable version 1.1.1+dfsg-1. In both cases this produced the error shown above. I'm also using a backport of TeX Live 2019 from Debian unstable (Debian 10/buster just has the prerelease), with LuaTeX 1.10.0.

Two different people report this code working for them on Arch Linux. One of them is Aditya.

To quote him from chat:

On my system, the library that should be loaded is libjulia.so and is located at /usr/lib/libjulia.so

On my Debian system, julia/sys.so is installed, but I'm not sure why it's not being found.

dlocate julia/sys.so
libjulia1: /usr/lib/x86_64-linux-gnu/julia/sys.so

The code follows.

local ffi = require("ffi")
local JULIA = ffi.load("julia", true)
ffi.cdef [[
void jl_init__threading(void);
typedef struct _jl_value_t jl_value_t;
jl_value_t jl_eval_string(const char);
]]
JULIA.jl_init__threading()
code = [[
x = [1 2 3]'
A = [1 0 1; 0 1 1; 1 1 0]
y = x'Ax
print(y[1])
]]
JULIA.jl_eval_string(code)

Answer 1

To function, Julia needs the system runtime sys.jl. The Julia interpreter ships with a precompiled version of this runtime, which is dumped into sys.so. I think that the resolution of the path to sys.so can somehow be influenced by setting environment variables, but I was unable to find out how. Another alternative is to point the jl_init function to the path of the system image by using jl_init_with_image.

local ffi = require("ffi")
local JULIA = ffi.load("julia", true)

ffi.cdef [[
void jl_init_with_image__threading(const char *julia_bindir,
                                   const char *image_relative_path);
typedef struct _jl_value_t jl_value_t;
jl_value_t *jl_eval_string(const char*);
]]

JULIA.jl_init_with_image__threading("/usr/lib/x86_64-linux-gnu/julia/", "sys.so")

code = [[
x = [1 2 3]'
A = [1 0 1; 0 1 1; 1 1 0]

y = x'*A*x

print(y[1])
]]

JULIA.jl_eval_string(code)

I saved this file as /tmp/julia/test.lua and ran this in a Docker container using the following commands:

user@host:~$ sudo docker run -v /usr/local/texlive/2019/:/usr/local/texlive/2019/:ro -v /tmp/julia/:/tmp/julia/ -it --rm debian:buster
root@9903c6e0ca52:/# export PATH=/usr/local/texlive/2019/bin/x86_64-linux/:$PATH
root@9903c6e0ca52:/# apt-get update
[...]
root@9903c6e0ca52:/# apt-get install --no-install-recommends julia libjulia-dev
[...]
root@9903c6e0ca52:/# texlua /tmp/julia/test.lua
23root@9903c6e0ca52:/#

The 23 before root@9903c6e0ca52 on the last line is the output from Julia. This is the same as I get from an interactive Julia session:

julia> x = [1 2 3]'
3×1 LinearAlgebra.Adjoint{Int64,Array{Int64,2}}:
 1
 2
 3

julia> A = [1 0 1; 0 1 1; 1 1 0]
3×3 Array{Int64,2}:
 1  0  1
 0  1  1
 1  1  0

julia> y = x'*A*x
1×1 Array{Int64,2}:
 23

julia> print(y[1])
23