Alpha 4: independent directory reuse

Alpha 4 extends guarded directory-report reuse to independent directories with at least 32 explicit test files each. Groups may be interleaved. Smaller groups and parent directories that contain one another retain stock discovery; the existing method, hook, option and metadata guards still apply.

Three fresh full pytest runs on Ubuntu 24.04, CPython 3.14.8 and pytest 9.1.1 measured:

Synthetic workload	Stock pytest	Boorst enabled	Time change
384 explicit files across four directories	2.251 s	0.851 s	62.2% less

Raw CI measurements record exact ordered-ID, phase and exit-code parity, 380 directory reuses and zero native ID calls. This is a Python discovery optimization. Synthetic results do not predict overall project CI savings. The README also records a separate macOS ARM64 universal-wheel trial.

Try it from a project with uv and its test dependencies already installed in a virtual environment:

uv run --no-sync https://raw.githubusercontent.com/dnikolayev/pytest-boorst/v0.1.0a4/scripts/try.py

Append your usual pytest arguments. The launcher installs only Boorst and enables it for that pytest process, preserving existing Python, pytest, dependencies, configuration and lockfiles. Ordinary pytest stays disabled unless explicitly enabled. Native wheels cover Linux x86-64, macOS ARM64 and Windows x86-64; the universal Python wheel retains supported directory reuse and uses stock parameter IDs. Windows uses stock directory discovery.

Compatibility and measurements describe the guarded scope. CI validation passed all 10 jobs, including installed native/universal trials, PowerShell, fallback, fixture/error parity and performance gates. The released source is a96b1c6; its tree matches the tested PR head. wheels.html provides direct wheel links with SHA256 hashes.