8bea1abe12
Records were a jrec holding an alist of (kw . val) conses: ~113B/node, built fresh per construction, field reads a list scan. Replace that with a shared per-type descriptor (tag + field keywords + an eq?-keyed keyword->index table) plus a flat per-instance value vector and an extension map for any non-field keys assoc'd on (jolt-nil when there are none). Construction now allocates one vector instead of a cons chain and a field read is an index lookup. binary-trees construction allocation drops 2.085GB -> 1.19GB. That alone barely moved binary-trees wall-time: profiling showed the read loop, not allocation, dominates, and the read loop's own allocation came from (nil? l) lowering to (jolt-invoke (var-deref "clojure.core" "nil?") l), which conses its args every call. Add nil?/some? to the backend native-op table so they inline to jolt-nil?/jolt-some? (and drop the truthy wrapper, like the other predicates). check-tree's read loop goes from 1.476GB allocated to zero; binary-trees 18.9x -> 9.7x vs JVM. The remaining gap is the field-read dispatch chain (jolt-c3mw). Two JVM divergences fixed along the way, both certified: - dissoc of a declared field downgrades a record to a plain map (was kept as a record); an extension key still drops cleanly. - map->R keeps extension keys (was dropping anything outside the declared basis). 16 new corpus rows pin assoc/dissoc/count/keys/seq/=/hash/extension-field behavior against JVM Clojure. Co-authored-by: Yogthos <yogthos@gmail.com> |
||
|---|---|---|
| .. | ||
| build-app | ||
| bench-chez.ss | ||
| corpus.edn | ||
| directlink-test.ss | ||
| ffi-binding-test.ss | ||
| inline-test.ss | ||
| numeric-test.ss | ||
| README.md | ||
| transient-test.ss | ||
| unit.edn | ||
| values-test.ss | ||
Chez test harness
The correctness gate for jolt. Pure Chez (+ Clojure for the JVM oracle).
Correctness is judged against the JVM-sourced conformance spec; the spec itself
lives in test/conformance/ (see its SPEC.md). Run the whole gate with make test from the repo root.
The spec corpus
corpus.edn is the contract: ~2920 rows {:suite :label :expected :actual}, with
:expected sourced from reference JVM Clojure by test/conformance/regen-corpus.clj.
It is frozen (the canonical source) — add or change cases here, then re-source the
answers with regen-corpus.clj and re-certify with test/conformance/certify.clj.
The gate runners (host/chez/)
-
run-corpus.ss— runs every corpus case through the spine (read → analyze → IR → emit → eval, all on Chez), comparing each result by value-equality against the JVM:expected. Aknown-failallowlist covers cases jolt can't match because Chez has no JVM host (Java classes, arrays,BigDecimal, opaque host-object printers, …); the gate fails only on a NEW divergence or if the pass count drops below the floor.chez --script host/chez/run-corpus.ss JOLT_CORPUS_LIMIT=200 … # every-Nth stride, fast iteration JOLT_CHEZ_ZJ_FLOOR=N … # override the floor (default 2678) -
run-unit.ss— host-specific unit cases (test/chez/unit.edn) that aren't in the JVM-portable corpus: dot-forms, java statics, io, reader, walk, vars/namespaces, refs. Each:expris evaluated in-process and its printed value compared to a baked:expected(:throwsasserts a raise). -
selfcheck.sh— self-host fixpoint:bootstrap.ssrebuild byte-equals the checked-in seed (host/chez/seed/). -
smoke.sh— realbin/joltc -eCLI smoke.
Other Chez tests
values-test.ss— the value model (nil/truthiness/collections).make values.bench-chez.ss— compute bench through the pipeline (opt-in; not in the gate).
All runners assume chez on PATH.