Evidence for the capability maps
Reproduction material for the [probe], [rustdoc:…] and [api:…] provenance markers in the
four maps in the parent directory (arrow-rust.md, datafusion-rust.md,
supporting-rust-libraries.md, python-libraries.md).
| Directory | Covers |
|---|---|
rust/ | arrow 59.3.0 / datafusion 55.1.0 and the §3.3 supporting crates: =-pinned manifests, committed lockfiles, rustdoc-JSON generation scripts and their logs, eleven probe programs and their captured outputs, and checkmap.py (the mechanical check the maps claim) |
python/ | the Python boundary libraries: pinned requirement sets for the platform (3.14) and parity (3.13) interpreters, apidump.py/apiq.py (the rustdoc-JSON analogue), three probe programs |
arrow_probe.rs | Arrow probes 1–5 (IPC metadata round-trip, null/NaN/-0.0 distinctness, totalOrder sort, batch-split byte equality, alignment sensitivity), still quoted by arrow-rust.md; run it as a src/bin/ target of the rust/ scratch project |
Regenerate with just evidence-regen (see rust/README.md and python/README.md for the manual
steps). The extracted fact profiles (build/facts/*) are derived output and are not committed;
the lockfiles are what make the extraction reproducible.
Note on pinning: every crate is =-pinned individually and deliberately. Pinning only the umbrella
(arrow = "=59.3.0") does not pin the family — sub-crates depend on ^, so they float upward,
and cargo tree -d does not flag it because a mixed family is not a duplicate. cargo xtask family-check exists because of this.