memory leak in Python 3.13 free-threaded build

I spent some time trying to debug a memory leak caused by calling plotly.express.line within a Dash live update callback under a Python 3.13.3t (free-threaded). Setting PYTHON_GIL=0 or PYTHON_GIL=1 appears to make no difference.

By no means have I located the root cause of the issue (not yet anyway), but I've been able to at least narrow it down to at least one problematic line of code in the BaseFigure class

https://github.com/plotly/plotly.py/blob/main/plotly/basedatatypes.py#L581

If I run the example code shown below (which is HIGHLY distilled from the originating Dash app I was trying to debug), the memory footprint seems to steadily climb with each iteration. If I comment out the offending line within BaseFigure (shown above), the memory footprint remains stable

#!/usr/bin/env -S python3 -u

import sys
import time
from plotly.basedatatypes import BaseFigure

gil = sys._is_gil_enabled()
print(f'GIL enabled = {gil}')

while True:
    BaseFigure()
    BaseFigure()
    BaseFigure()
    time.sleep(.01)

For long running applications like the one described earlier, slow and steady memory consumption issues are causing some rather significant issues.

System details

OS: docker://python:3.12-slim-bookworm
Arch: aarch64
Python version: uv python install 3.13.3t
Plotly version: 6.0.1

Commencez par exécuter la boucle BaseFigure réduite avec Python 3.13.3t et observez son empreinte mémoire. Inspectez plotly/basedatatypes.py autour de la ligne 581, puis comparez le comportement lorsque cette ligne est commentée. Le travail est terminé lorsque la cause est identifiée et qu’il est confirmé que la création répétée de BaseFigure n’entraîne plus d’augmentation continue de la mémoire.