rand-nth vec'd its argument, so (rand-nth nil) hit index-out-of-bounds through the empty vector where the reference's (nth coll (rand-int (count coll))) returns nil (and a set throws) — the last genuinely-fixable row in the suite baseline; rand-nth is fully clean now. Corpus/README counts updated to the current ~3570 rows, the run-corpus regression floor raised from 2730 to 3390 to match current parity, and the stale traceability line dropped. |
||
|---|---|---|
| .. | ||
| certify.clj | ||
| known-divergences.edn | ||
| profile.edn | ||
| README.md | ||
| regen-corpus.clj | ||
| SPEC.md | ||
Conformance: certifying the corpus against reference Clojure
See SPEC.md for the full host-neutral language-spec contract: the corpus schema, conformance levels, the feature profile, and how to host jolt on a new runtime. This README covers the certification tooling specifically.
The corpus (test/chez/corpus.edn) is jolt's host-neutral behavioral suite — one
row per case: {:suite :label :expected :actual}, where :actual is a Clojure
source expression and :expected its result (or :throws). The runtime harness
(host/chez/run-corpus.ss, invoked by make corpus) replays it on Chez and
compares by value-equality.
Every :expected is sourced from reference JVM Clojure, so the corpus is both a
regression suite and a specification certified against Clojure rather than
against its authors' beliefs. This directory holds the certification tooling that
closes that gap.
What's here
-
certify.clj— runs every corpus row's:actualand:expectedthrough reference JVM Clojure (each in a freshusernamespace, output/stdin sunk, a 5s per-case watchdog) and compares with Clojure's=. It buckets each row:certified/certified-throws— jolt's:expectedmatches real Clojuredivergent— both evaluate but jolt's:expecteddisagrees with Clojurethrows-mismatch— jolt and Clojure disagree on whether it throwsjvm-error—:actualisn't runnable on vanilla Clojure (host-coupled / jolt-specific) — informational, not certifiableread-error/timeout— won't read on the JVM reader, or ran too long
-
known-divergences.edn— every current divergence, classified. Most are deliberate jolt-specific or host-model deltas (see:legend): the all-double numeric model, snapshot-heap concurrency, the no-JVM host model, jolt reader features, the jolt printer, intentional strictness. A few are genuine:bugentries with a tracked bead. These categories become the:featuresflags in conformance inc3.
make certify is the gate wrapper. It skips cleanly when clojure (JVM) is not
installed; otherwise it runs certify.clj and fails the build on a NEW
(unclassified) divergence or a stale allowlist entry. Flaky entries (JVM
result is timing-dependent, e.g. future-cancel) are tolerated either way.
Running
make certify # the gate wrapper (skips if clojure absent)
clojure -M test/conformance/certify.clj # gate directly (exit≠0 on new/stale)
clojure -M test/conformance/certify.clj test/chez/corpus.edn --edn /tmp/report.edn # full machine-readable report
Current state
Of ~2740 vanilla-certifiable rows, >2730 match reference Clojure exactly; the handful of divergences are all classified (deliberate deltas plus a few tracked bugs). The corpus is trustworthy as a spec, with the host-specific deltas made explicit rather than hidden.
Adding / changing cases
When you add corpus rows or change behavior, re-run the certifier. A NEW divergence
means either a real bug (file it, tag the allowlist entry :bug + :bead) or a
deliberate delta (classify it). A stale entry means a divergence was fixed — remove
it from known-divergences.edn.