Memory allocated when creating Python objects is never automatically freed

[dmaioli]

Affects: PythonCall

Describe the bug
I am trying to use PythonCall.jl to call the Python interface to CP-SAT, which is a popular constraint programming library from Google's OR-Tools toolkit. The problem is that the memory that gets allocated when creating a Python object from that interface is never automatically freed.

A function createmodel() is used to create the object model, which goes out of scope after the function terminates. Each time the function is executed, the memory gets fuller, until eventually the Julia process gets killed. The expected behavior is that at some point before Julia gets killed, the memory that was allocated for objects that are out of scope should get released.

Here are the steps to reproduce a MWE, assuming the environment already has PythonCall.jl and CondaPkg.jl:

Add OR-Tools to the environment and import it:

using PythonCall, CondaPkg
] conda pip_add ortools
cp_model = pyimport("ortools.sat.python.cp_model")

Define the function that will create and initialize a CpModel object:

createmodel() = let 
  model = cp_model.CpModel() # create model
  variables = [model.new_int_var(0,1, "x_" * string(i)) for i in 1:100] # add variables to the model
  model.add_allowed_assignments(variables, [rand(0:1, 100) for _ in 1:200000]) # add constraints over those variables
  return nothing
end

Repeatedly call createmodel() and observe that the memory usage keeps going up until the Julia process gets killed.

createmodel()
createmodel()
createmodel()
# ... if you keep calling createmodel(), eventually the memory gets full

Your system
Please provide detailed information about your system:

• Operating System: Fedora Linux 43

• Python version: 3.14.3

• versioninfo()

  Julia Version 1.12.3
  Commit 966d0af0fdf (2025-12-15 11:20 UTC)
  Build Info:
    Official https://julialang.org release
  Platform Info:
    OS: Linux (x86_64-linux-gnu)
    CPU: 8 × 11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz
    WORD_SIZE: 64
    LLVM: libLLVM-18.1.7 (ORCJIT, tigerlake)
    GC: Built with stock GC
  Threads: 1 default, 1 interactive, 1 GC (on 8 virtual cores)
  

•   import Pkg; Pkg.status()

  Status `~/MWECPSAT/Project.toml`
    [992eb4ea] CondaPkg v0.2.34
    [6099a3de] PythonCall v0.9.31
  

•   import CondaPkg; CondaPkg.status()

  Environment
    /home/dmaioli/MWECPSAT/.CondaPkg/.pixi/envs/default
  Pip Packages
    ortools v9.15.6755
  

Additional context

• Calling PythonCall.pydel!(model) and/or for v in variables PythonCall.pydel!(v) end inside the body of the function does not have any visible effect
• Executing GC.gc() after the function calls has the effect that at least part of the allocated memory gets freed, but sometimes this does not happen if it is executed only once.
• Calling the equivalent code directly from Python works as expected, i.e. the memory gets freed after the function terminates. Here is the Python version:

  from ortools.sat.python import cp_model
  import random

  def createmodel():
    model = cp_model.CpModel()
    variables = [model.new_int_var(0,1, f"x_{i}") for i in range(100)]
    model.add_allowed_assignments(variables, [[random.randint(0,1) for _ in range(100)] for _ in range(200000)])
    return None

  createmodel()
  createmodel()
  createmodel()
  # ... if you keep calling createmodel(), the memory never gets full