ShareableList has two data corruption bugs, both rooted in the same design flaw: C-style null-terminated storage semantics applied to length-delimited Python types.

Bug 1 — UTF-8 underallocation (#145261, #88336): Slot allocation uses len(item) (character count) instead of byte count for str items. Multi-byte UTF-8 strings overflow their allocated slot and corrupt adjacent data.

Bug 2 — Null stripping (#106939, #96779): The back-transform lambdas call rstrip(b'\x00') to remove struct padding, but this also strips legitimate trailing null bytes from user data.

Approach

Store actual byte lengths in the format metadata (separate from the allocated slot sizes used for struct.pack_into), and use those exact lengths during retrieval instead of relying on null-termination. This makes both bugs go away with a single conceptual change.

Specifically:

• __init__: compute slot allocation using len(item.encode('utf-8')) for str items
• __init__: build _stored_formats list with actual byte lengths, write those to packing metadata
• __setitem__: separate pack_format (allocated slot size) from new_format (actual byte length in metadata)
• _back_transforms_mapping: remove rstrip(b'\x00')

Cross-version compatibility

• Old writer → new reader: The old code stores allocated-length values in format metadata (e.g. 8s for a 5-byte string in an 8-byte slot). New code reads that 8s format and returns 8 bytes including padding nulls. This is the same behavior as the current release — no regression.
• New writer → old reader: New code stores actual byte length (e.g. 5s). Old code reads 5s, gets 5 bytes, then rstrip(b'\x00') — harmless unless the data actually ends in nulls, which is the existing bug.
• Neither direction crashes. The shared memory layout (offsets, block sizes) is unchanged.

Tests

• test_shared_memory_ShareableList_trailing_nulls: bytes with trailing nulls, str with trailing nulls, all-null bytes, empty bytes, no-null bytes, cross-process read via name=
• test_shared_memory_ShareableList_multibyte_utf8: 1-byte (ASCII), 2-byte (é), 3-byte (中), and 4-byte (𐀀) UTF-8 sequences with cross-process verification
• Updated existing sl.format assertions to reflect actual byte lengths

Prior work

This PR consolidates the approaches from #144559 (@aisk, null-stripping fix) and #145266 (@zetzschest, both fixes). Both PRs are open with zero reviews. The fixes belong together because they share the same root cause and the same solution mechanism (stored byte lengths in format metadata).

• Issue: Copying bytes object to shared memory list truncates trailing zeros #106939

📚 Documentation preview 📚: https://cpython-previews--145488.org.readthedocs.build/