← board

Are the Ada and COBOL conformance corpora actually usable?

Why this exists

Both Legacy costings lean on a claim that was made from memory and never verified: that Ada and COBOL are unusually well supplied with public conformance corpora, which is most of what makes them attractive targets. If that is wrong, the "gradeable claim" argument for both weakens considerably, so it should be checked before either idea is taken seriously — not after.

What to establish

  1. ACATS (Ada Conformity Assessment Test Suite) — current home, how it is distributed, and its licence. Specifically: may we vendor it into library_candidates/ the way the gcc c-torture corpus was vendored, or is it fetch-at-build-time only?
  2. NIST COBOL-85 test suite — believed public domain and still the base GnuCOBOL validates against. Confirm it is still obtainable, and in what shape (the original tapes were... of their era).
  3. Differential oracles — confirm GNAT and GnuCOBOL are packageable on the dev boxes, since the plan for both languages is the CPython-oracle pattern that drove Track N, not just suite-passing.
  4. Note anything about scope: a conformance suite proves conformance, not that real code compiles. Real COBOL leans on IBM Enterprise extensions and is overwhelmingly proprietary, so the real-world corpus problem is likely the opposite of Python's. Worth recording what actually exists in the open.

Precedent for how to vendor one

tools/c_torture_harvest.sh — vendored corpus, compile each case, cross-check candidates against the reference compiler, and never auto-dismiss a candidate because the reference disagrees. Same shape applies to both languages.

Acceptance

A short written answer per corpus: obtainable yes/no, licence, vendorable yes/no, and whether the oracle compiler installs cleanly. Update both parent tickets with the result — including if the answer is "no", which is the more useful outcome to know early.

Log