Summary

Longhand segfaults (EXC_BAD_ACCESS / SIGSEGV) on Python 3.14 when installed with chromadb >= 1.0. The crash happens in chromadb_rust_bindings.abi3.so during tokio worker thread operations — a NULL pointer dereference inside the PyO3 bindings.

Environment

• macOS 26.4.1 (ARM64, Apple M4)
• Python 3.14.2
• chromadb 1.5.7
• longhand 0.5.5 (also reproduced with 0.3.0)

Repro

pip install longhand
longhand stats   # → Segmentation fault (exit 139)

Faulting stack (condensed)

Thread 31 Crashed:
chromadb_rust_bindings.abi3.so + 25456296
chromadb_rust_bindings.abi3.so + 25443892
...
→ KERN_INVALID_ADDRESS at 0x0

Root cause

ChromaDB 1.x rewrote internals in Rust and ships PyO3-based chromadb_rust_bindings. Python 3.14 changed internals that PyO3 bindings are sensitive to, and chromadb hasn't shipped a 3.14-compatible build yet. ChromaDB 0.x (pre-Rust) works fine on Python 3.14.

Longhand's pyproject.toml declares chromadb>=0.5.0 with no upper bound, so pip resolves to 1.5.7 on a fresh 3.14 install, hitting the crash immediately.

Workaround (until fixed)

pip install 'chromadb<1.0'
rm -rf ~/.longhand/chroma   # 1.x on-disk format incompatible with 0.x
longhand ingest --force     # rebuild vector index

Fix

Constrain chromadb conditionally on Python version in pyproject.toml:

"chromadb>=0.5.0; python_version<'3.14'",
"chromadb>=0.5.0,<1.0; python_version>='3.14'",

This way 3.12/3.13 users are unaffected (keep 1.x) and 3.14 users get 0.x which works. Once chromadb ships a 3.14-compatible 1.x wheel, we can relax the constraint.

Will ship as 0.5.6.