JuliaCall segmentation fault

General Usage
1

I’ve got a Julia module (which contains compiled Fortran code) which is called in Python using JuliaCall (Guide · PythonCall & JuliaCall), for example within a Python script:

from juliacall import Main as jl
jl.seval(using MyModule.jl)

When I then call this in a Python function:


def mypythonfunction():
    Output = jl.MyModule.calc_something(args)

I get an instant segmentation fault (11), any ideas?

Edit: A note, this fails even when not using the compiled Fortran code and calling straight Julia code.

2

Update, this is also the same using PyJulia. As soon as I import the module in the global scope and try call it within a function it instantly segfaults.

I am at a loss!

3

Can you post a MWE please.

4

Sure, my Julia module looks like this:

module myModule
    include("constants.jl")
end

where constants.jl is a simple dictionary:

molMassDict=Dict(
    "La"=> 162.9085,
    "Ce"=>172.1180,
    )

I then call it in Python using:

#!/usr/bin/env python3

from juliacall import Main as jl
jl.seval("using myModule")


def test_julia_import():
    print(jl.myModule.molMassDict)

test_julia_import()

Interestingly this MWE works no problem. The problem arises when I try to do this inside a Django (I’m not sure if you’re familiar with Django but it’s a Web Dev framework) view. A Django view is similarly a Python function, for example:

def django_view(request):
    test_julia()

if I call test_julia() within this, it instantly segfaults:

signal (11): Segmentation fault
in expression starting at none:0
Allocations: 6828442 (Pool: 6825828; Big: 2614); GC: 8

Seems like it’s Django specific?

5

@cjdoris this is possibly related?

6

In case anyone comes here looking for answers, I’ve ended up moving the Julia code to it’s own microservice. The communication between Python and Julia is handled via requests (requests · PyPI) on the Python side and Genie on the Julia side.