Snapshot startup filesystem encoding on Linux - #9689

Summary

• Initialize the Linux command-line locale with CPython 3.15's environment precedence and PEP 538 coercion. Keep embedded startup non-mutating; the executable explicitly opts into the unsafe, single-threaded process-startup contract.
• Snapshot filesystem encoding in interpreter configuration. Share the fixed configuration with subinterpreters, keep Python codec objects local to each interpreter, and preserve filesystem behavior after later LC_CTYPE changes.
• Reuse native UTF-8, ASCII and Latin-1 codecs. GNU Linux uses an owned locale snapshot for wider-codec bootstrap, then switches to the initialized codec registry.
• Make filesystem decoding fallible throughout startup and OS-facing callers. Preserve Python exceptions, retain accepted sockets until address conversion succeeds, and close received ancillary descriptors if conversion fails.
• Stop using frozen filesystem encoding for the internal current-locale conversion helper.

This addresses the runtime cause behind the Linux test_utf8_mode.UTF8ModeTests.test_cmd_line expected failure. No test code or decorators are changed because a RustPython runtime pass is not yet available.

CPython reference

Compared against CPython v3.15.0:

• Locale preconfiguration
• PEP 538 startup coercion
• Native locale conversion
• Filesystem codec initialization

Validation

• Passed cargo check -p rustpython --lib --bin rustpython.
• Passed strict Clippy: cargo clippy -p rustpython-host_env -p rustpython-vm -p rustpython-stdlib -p rustpython --lib --bin rustpython -- -D warnings.
• Passed no-host VM compilation: cargo check -p rustpython-vm --no-default-features --features compiler.
• Normal commit hooks passed.
• An out-of-tree, bounded host-helper probe passed 40 decode/encode roundtrips across C, C.UTF-8, ISO-8859-1, ISO-8859-5 and GB2312, including embedded NUL and malformed/incomplete input. Snapshot stability after changing LC_CTYPE and typed encoding errors passed.
• Fifteen host-helper startup cases matched CPython 3.15.0 output and warning stderr exactly, including environment precedence, ignored Python controls, coercion warnings, invalid locale settings and LC_NUMERIC behavior.
• Unmodified CPython 3.15.0 test_cmd_line and test_filesystemencoding reference cases passed.
• Separate source review completed. No repository test additions, skips or expected-failure removals.

The host-helper probe does not execute RustPython's VM or interpreter. Local RustPython runtime and aggregate workspace/C-API suites were not run because no verified runtime guard is available. This draft needs hosted CI and runtime review. Only the GNU Linux Rust target is installed; Windows/WASI call sites were source-reviewed but not cross-compiled.

Scope and integration

Non-Linux filesystem behavior is preserved. Non-GNU Linux supports existing native fast codecs and retains its previous UTF-8 behavior for wider CODESETs. This does not add arbitrary embedded filesystem-codec overrides, Windows legacy mbcs support, complete raw -m parsing, or non-UTF-8 getpath configuration strings.

VirtualMachine::fsdecode and dependent Rust path-result helpers now return PyResult; embedding callers must propagate errors. check_initialization lets embedded callers inspect startup failures without diagnostics or process exit.

This overlaps #9688 (native script filenames). A disposable composition of current main 9d657b02e4, #9688 and this patch passed cargo check -p rustpython --lib --bin rustpython after resolving the cant_open_file overlap and propagating four additional fsdecode results in src/lib.rs and crates/vm/src/vm/python_run.rs. Those merge-time adjustments are prepared separately; neither standalone branch includes the other PR.

AI assistance

This is maintainer-directed Python 3.15 upgrade work. OpenAI Codex assisted with investigation, implementation, source review and validation. Model version was not exposed. No human runtime validation is claimed.