Skip to content
Open
No due date
•Last updated Nov 16, 2026

Minor. Library depth: the typing surface lowers and reach-based pruning lands. Contents: the typing surface (collections.abc in the import table, TypeVar handling, then unions/generics/forward references erased at lower time), reach-based pruning behind a recorded root-fence decision, the strong receipt from #137 (colorama or shellingham compiled from source), the pure-dep IR cache, and pilot 2 (tomli or iniconfig from PyPI). Exit: a pilot 2 target compiles from PyPI end to end and matches CPython, the census delta on the library corpus, 0 open issues at close.

The pyscriptc-ndarray M1 perf receipt (#496)

Workload: 1M float64 elements, ten iterations of an elementwise multiply and an add plus a whole-array sum. Measured on the VPS host (2026-11-13) with uv run python -m benchmarks.ndarray.receipt: compiled 0.169s, CPython numpy 0.442s, CPython pure-Python list floor 1.267s. The compiled program is 2.62x faster than numpy and 7.51x faster than the list floor on the same operation. The receipt script exits 1 when the compiled program does not clear both bars, so the number is reproducible; it is a receipt, not a CI gate.

50% complete

Issues