Segmentation Fault When Importing matplotlib.pyplot After mitsuba

[JY251]

Description:

I encountered a segmentation fault when importing matplotlib.pyplot after importing mitsuba. The issue occurs at the line where matplotlib.pyplot is imported.

Steps to Reproduce:

1. Install mitsuba and matplotlib in the same Python environment.
2. Run the following script:

   print("OK1")
   import mitsuba as mi
   print("OK2")
   import matplotlib.pyplot as plt  # Segmentation Fault Here
   print("OK3")

Expected Behavior:

The script should run without any segmentation faults, and all print statements (OK1, OK2, OK3) should be displayed.

Actual Behavior:

A segmentation fault occurs at the line where matplotlib.pyplot is imported.

Environment:

• Python version: [3.10.12]
• Operating System: [Ubuntu 22.04.3 LTS]
• mitsuba version: [3.6.2]
• matplotlib version: [3.10.0]

Solution:

Reordering the imports resolved the issue. Specifically, importing matplotlib.pyplot before mitsuba prevents the segmentation fault:

import matplotlib.pyplot as plt  # Import matplotlib first
print("OK1")
import mitsuba as mi
print("OK2")

[merlinND]

Hello @JY251,

I suspect this has something to do with the libstdc++ / libc++ incompatibility in exception handling.
Did you build Mitsuba yourself using Clang? If so, when importing Mitsuba first, the libc++ exception handlers are used. But Matplotlib throws libstdc++ format exceptions, which causes a segfault when handled.

The workaround I can think of:

• Build Mitsuba with GCC (which uses libstdc++)
• Downgrade Matplotlib to an earlier version which doesn't throw such an exception
• Remove the source of the exception (only possible if the exception is not meant to be thrown; it might actually be part of the normal functioning of Matplotlib).