[t-miller]

An exception during Python OSM wrapper construction leaves m_valid true while m_pyobj is still null. The destructor then dereferences that null object while invalidating _pyosmium_data, turning a Python exception into a native crash.

Set m_valid only after construction and assignment succeed. Add a subprocess regression covering Node, Way and Relation constructors interrupted by KeyboardInterrupt, MemoryError and RuntimeError, including cleanup after the exception is caught.

Fixes #323.

Validation on CPython 3.14.8 / macOS ARM64:

• New regression: all nine cases crash against the released 4.3.1 native wheel; all nine pass against a locally built 4.3.1 sdist with the identical header fix.
• test_object_construction.py plus this checkout's test_file_processor.py: 24 passed against that patched native build.
• Separate application matrix: the same fix removes all 20 construction-related native crashes across 29 subprocess cases; ordinary scan and SQLite cache tests pass.

The native validation build is the 4.3.1 release with this header change, rather than a fresh build of all other current-master source. No signal handlers are suppressed or replaced by the fix.

[lonvia]

Thanks for finding the issue and proposing a fix. The Python code in a string in the test is a bit unusual and hard to read. I played around with the multiprocessing module and came up with a version in #326 that is closer to a simple function call. Closing here in favour of the other.