Segfault on M3 when loading specific Python subpackage

[williamjsdavis]

Affects: PythonCall

Describe the bug

On a M3 Mac I can produce a segfault when loading a specific subpackage, simpeg.potential_fields from version 0.22.0 of the simpeg package (https://github.com/simpeg/simpeg). When I import just 'simpeg', there is no segfault. Furthermore, other subpackages do not produce segfaults when loaded, e.g., simpeg.maps.

Here are the commands used to reproduce the segfault. (The output of some steps has been shortened; I will include a full output in a reply.)

(@v1.10) pkg> activate --temp
  Activating new project at `/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS`

julia> versioninfo()
Julia Version 1.10.4
Commit 48d4fd48430 (2024-06-04 10:41 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 12 × Apple M3 Pro
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, apple-m1)
Threads: 1 default, 0 interactive, 1 GC (on 6 virtual cores)
Environment:
  JULIA_PKG_USE_CLI_GIT = 1

(jl_tI92dS) pkg> add CondaPkg PythonCall
   Resolving package versions...
    Updating `/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/Project.toml`
  [992eb4ea] + CondaPkg v0.2.22
  [6099a3de] + PythonCall v0.9.20
...

julia> using CondaPkg

(jl_wZlac1) pkg> conda add simpeg
    CondaPkg Found dependencies: /var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/CondaPkg.toml
    CondaPkg Found dependencies: /Users/williamdavis/.julia/packages/PythonCall/S5MOg/CondaPkg.toml
    CondaPkg Resolving changes
             + python
             + simpeg
    CondaPkg Creating environment
...
  + simpeg              0.22.0  pyhd8ed1ab_0          conda-forge     Cached
...
Transaction finished

To activate this environment, use:

    micromamba activate /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env

julia> using PythonCall
    CondaPkg Found dependencies: /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/CondaPkg.toml
    CondaPkg Found dependencies: /Users/williamdavis/.julia/packages/PythonCall/S5MOg/CondaPkg.toml
    CondaPkg Removing environment
             │ /Users/williamdavis/.julia/artifacts/98aba21a6802e06a5d2e7c48633d7dc1c3d4a749/bin/micromamba
             │ -r /Users/williamdavis/.julia/scratchspaces/0b3b1443-0f03-428d-bdfb-f27f9c1191ea/root
             │ remove
             │ -y
             │ -p /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env
             └ --all
...
    CondaPkg Creating environment
             │ /Users/williamdavis/.julia/artifacts/98aba21a6802e06a5d2e7c48633d7dc1c3d4a749/bin/micromamba
             │ -r /Users/williamdavis/.julia/scratchspaces/0b3b1443-0f03-428d-bdfb-f27f9c1191ea/root
             │ create
             │ -y
             │ -p /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env
             │ --override-channels
             │ --no-channel-priority
             │ python[version='>=3.8,<4',channel='conda-forge',build='*cpython*']
             │ simpeg[version='*']
             └ -c conda-forge
...

julia> pyimport("simpeg")
Python: <module 'simpeg' from '/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env/lib/python3.12/site-packages/simpeg/__init__.py'>

julia> pyimport("simpeg.maps")
Python: <module 'simpeg.maps' from '/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env/lib/python3.12/site-packages/simpeg/maps/__init__.py'>

julia> pyimport("simpeg.potential_fields")

[50910] signal (11.2): Segmentation fault: 11
in expression starting at REPL[16]:1
_ZN4llvm17PMTopLevelManager16addImmutablePassEPNS_13ImmutablePassE at /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env/lib/libLLVM-14.dylib (unknown line)
Allocations: 10142116 (Pool: 10132215; Big: 9901); GC: 16
zsh: segmentation fault  julia

Your system

Running on a MacBook Pro M3, with macOS Sonoma 14.5 (23F79).

Output of Base.versioninfo()

Julia Version 1.10.4
Commit 48d4fd48430 (2024-06-04 10:41 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 12 × Apple M3 Pro
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, apple-m1)
Threads: 1 default, 0 interactive, 1 GC (on 6 virtual cores)
Environment:
  JULIA_PKG_USE_CLI_GIT = 1

Output of Pkg.status()

Status `/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/Project.toml`
  [992eb4ea] CondaPkg v0.2.22
  [6099a3de] PythonCall v0.9.20

Output of CondaPkg.status()

CondaPkg Status /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/CondaPkg.toml
Environment
  /private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_tI92dS/.CondaPkg/env
Packages
  simpeg v0.22.0

Additional context

• Running Python (from .CondaPkg/env) and importing simpeg.potential_fields directly does not cause a segfault, see example below.
• This segfault does not occur with simpeg version 0.21, and the version 0.21->0.22 changed the package name from SimPEG to simpeg.
• This bug is similar to issue #315, except this bug is specific to a subpackage

(@v1.10) pkg> activate --temp
...

(jl_yP78nT) pkg> add CondaPkg PythonCall

julia> using CondaPkg

(jl_yP78nT) pkg> conda add simpeg=0.21
...

julia> using PythonCall
...

julia> pyimport("SimPEG")
Python: <module 'SimPEG' from '/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_yP78nT/.CondaPkg/env/lib/python3.12/site-packages/SimPEG/__init__.py'>

julia> pyimport("SimPEG.maps")
Python: <module 'SimPEG.maps' from '/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_yP78nT/.CondaPkg/env/lib/python3.12/site-packages/SimPEG/maps.py'>

julia> pyimport("SimPEG.potential_fields")
Python: <module 'SimPEG.potential_fields' from '/private/var/folders/8q/1dg0hln57_xgmstm42jkdxpr0000gn/T/jl_yP78nT/.CondaPkg/env/lib/python3.12/site-packages/SimPEG/potential_fields/__init__.py'>

[williamjsdavis]

Full REPL output (800 lines long): https://pastebin.com/RDB30aY0