eyeprolog

Neumerkel conformity sources and cache

EyeProlog checks Ulrich Neumerkel’s ISO/WG17 conformity material with dynamic case inventories. Downloaded sources live in Git-ignored .cache/neumerkel/. They remain outside Git, source release archives and the npm package.

npm test and node test/run-neumerkel.mjs reuse a complete cache without contacting upstream. If a source is missing, the manifest is invalid, or a source no longer matches its recorded SHA-256 hash and byte count, the runner downloads all eight sources before checking them. A failed download leaves the previous snapshot intact. The first run needs network access.

The eight TU Wien sources are:

  1. conformity_testing — Part 1 syntax/reader/writer matrix;
  2. number_chars_cont_quad.pl — number_chars/2 continuation corpus;
  3. variable_names_quad.pl — variable_names/1 corpus;
  4. dif — dif/2 comparison table;
  5. length_quad.pl — length/2 corpus;
  6. phrase_quad.pl — phrase/2,3 corpus;
  7. prologue_quad.pl — Prolog Prologue working-draft corpus;
  8. cleanup — setup_call_cleanup/3 examples.

Every run executes the cases discovered from its source bytes. npm run sync refreshes all eight sources from current upstream, checks every discovered case and updates NEUMERKEL-LATEST.md. New upstream rows therefore enter the next refreshed run automatically. The syntax extractor reads TU Wien’s labelled Codex column and checks the page’s declared total so hand-edited HTML cannot silently reduce coverage.

Commands

npm test
npm run sync
npm run neumerkel
node test/run-neumerkel.mjs --cached
node test/run-neumerkel.mjs --cached --update-report
node test/run-neumerkel.mjs --cached --verify-report

npm run neumerkel forces a fresh download and requires the tracked report to match. --live also forces a fresh download when using the runner directly. --cached requires a complete, matching cache and never downloads anything. npm test -- --offline skips the eight upstream suites entirely, including WG17 syntax; a warm normal run includes them without needing upstream access. There is no clean command: use npm run sync to refresh the cache explicitly.

Tracked evidence and releases

The stable report records the upstream inventory successfully checked by EyeProlog. A normal test run warns if it differs from the tracked report; --verify-report makes that difference an error. --cached --update-report records the exact snapshot already checked, without fetching again. Review and commit intentional report changes.

The version lifecycle regenerates library and book files, runs npm run sync once, then runs the full npm test against that same snapshot. It generates and stages the conformance reports before the release commit. Releases thus check current upstream without fetching twice, while ordinary tests reuse the cache. A cached run alone does not establish that upstream is unchanged.

Exact source bytes, SHA-256 hashes, fetch timestamps, HTTP validators and machine-readable results stay under .cache/neumerkel/. Cached reads use the saved content type so the text encoding matches the live download. Tracked reports omit volatile fetch metadata, keeping repeated runs stable.

The reviewed WG17 syntax regression corpus is an additional check on selected outcomes. Every discovered syntax case still has to satisfy its upstream Codex expectation; a reviewed outcome never replaces that assertion.