jolt/test/integration/clojure-test-suite-test.janet
Yogthos 920fafe032 test: ~11x faster gate — ctx snapshot/fork + parallel suite battery
The gate spent almost all its time rebuilding identical contexts: init is
~50 ms interpreted / ~900 ms compiled (tier loading, analyzer build, macro
recompilation), and both the conformance harness and the spec harness built a
fresh ctx PER CASE — 269 cases x 3 modes for conformance alone (~285 s), and
~1500 spec cases (~90 s). The suite battery additionally ran its 234 worker
subprocesses sequentially (~100 s incl. 5 x 6 s timeout files).

- api: snapshot/fork — marshal a fully-built ctx once (reverse-lookup dicts
  from root-env, built at module load before any ctx exists), unmarshal cheap
  fully-isolated deep copies (~2 ms). A fork shares nothing mutable with its
  siblings, so per-case isolation is preserved exactly.
- conformance: one init per mode + fork per case (self-host pre-builds the
  analyzer before snapshotting). 285 s -> 5 s, same 269x3 results.
- spec harness: jeval/run-spec/expect-throws fork from one lazy module-level
  snapshot. Spec sweep ~90 s -> 9 s, all pass.
- clojure-test-suite: run the per-file worker subprocesses through a
  token-channel worker pool (default 4, JOLT_SUITE_WORKERS to override,
  capped at 8). 17.5 s with counts identical to the sequential run
  (4046/520/116, 5 timeouts).

Full gate wall-clock: ~8 min -> 43 s, everything green (conformance 269x3,
fallback-zero, fixpoint, self-host, sci, staged, suite >= 4034/67, all
specs+unit).
2026-06-10 09:11:58 -04:00

160 lines
8 KiB
Text

# clojure-test-suite conformance: runs the external, cross-dialect
# clojure-test-suite (jank-lang fork) against Jolt and asserts the number of
# passing per-function test files stays at/above a baseline. The suite is a git
# submodule at vendor/clojure-test-suite (CI checks it out via submodules:
# recursive). The test SKIPS cleanly only if the submodule isn't initialized
# (run `git submodule update --init`).
#
# Each suite file is a `clojure.test` namespace (one per clojure.core/string
# function). A minimal clojure.test + portability shim (test/support/clojure_test.clj)
# lets Jolt load them; `when-var-exists` auto-skips fns Jolt doesn't implement.
#
# Files are run in a one-shot worker subprocess (test/integration/suite-worker.janet)
# under a wall-clock deadline. A few suite tests build infinite sequences that an
# uncompilable/eager path can't truncate and so HANG rather than fail; the
# deadline contains them — a timed-out file contributes nothing, no skip-list.
(def suite-dir "vendor/clojure-test-suite/test/clojure")
# Baseline: assertions Jolt currently passes across the suite. Raise as Jolt
# improves so a regression (previously-passing assertion breaking) is caught.
# Lowered 3915 -> 3913 when futures landed: `realized?`/realized_qmark.cljc has a
# `(when-var-exists future ...)` block that was skipped while `future` was
# unresolved. With futures implemented the block now runs, but it depends on JVM
# `Thread/sleep` (jolt has no JVM interop) and on `future-cancel` interrupting a
# running thread (Janet OS threads can't be interrupted), so `(deref (future
# (sleep 1)))` re-raises the unresolved-`Thread/sleep` error — a documented
# platform gap, not a regression in any previously-working behavior.
# Raised 3913 -> 3916 with the staged-bootstrap kernel tier (ns-restore-on-throw
# + faithful subvec coercion), then 3916 -> 3919 moving juxt/every-pred/some-fn to
# Clojure (the canonical defs are more correct than the prior Janet ones). Raised
# 3919 -> 3926 preserving nil map values (jolt-c7h): a nil value is a present key,
# which several suite tests assert. Runs read 3927 consistently, occasionally 3926
# when a timeout-prone test (of the 9 that can time out) doesn't finish; floor at
# the consistent-minus-one 3926.
# Raised 3971 -> 3981 with Option A full laziness (jolt-fng): transformers return
# lazy seqs, lazy interleave stops timing out, and the lazy-seq nil-element +
# non-seqable-input fixes (case/seq/reverse/empty? over nil-first lazy seqs;
# lazy-from throws on non-seqable like Clojure) recovered + extended the suite.
# clean files 45 -> 66 (Option A makes seq?/vector? results match Clojure across
# many cross-dialect files). Stable across runs.
# Raised 3981 -> 4004 migrating 7 lazy seq fns to the Clojure overlay (40-lazy
# tier): the canonical CLJ versions add coverage (e.g. distinct value-equality).
# Raised 4004 -> 4034 / clean 66 -> 67 porting partition-all + repeatedly to the
# overlay, which required fixing two leniencies (a char is not callable; take
# validates its count) — correct beyond those fns, so the suite rose broadly.
(def baseline-pass 4034)
# A file is "clean" when it ran with zero failures AND zero errors.
(def baseline-clean-files 67)
# Per-file wall-clock budget (seconds). Normal files finish in well under 1s, so
# this normally only fires on genuinely-infinite-sequence hangs. It's an env var
# (JOLT_SUITE_TIMEOUT) so CI — whose runners are slower than a dev machine — can
# give slow-but-finite files generous headroom without timing them out (which
# would drop total-pass below the baseline and flake CI red). Default 6 locally.
(def per-file-timeout
(let [e (os/getenv "JOLT_SUITE_TIMEOUT")]
(or (and e (scan-number e)) 6)))
(defn- walk [dir acc]
(each e (os/dir dir)
(def p (string dir "/" e))
(case ((os/stat p) :mode)
:directory (walk p acc)
:file (when (and (string/has-suffix? ".cljc" p)
(not (string/has-suffix? "portability.cljc" p)))
(array/push acc p))))
acc)
# Run one file in a worker subprocess; return its "pass fail error" stdout, or
# nil if it exceeded the deadline (hang) or crashed.
(defn- run-file [path]
(def proc (os/spawn ["janet" "test/integration/suite-worker.janet" path] :p {:out :pipe}))
(def out (proc :out))
(var data nil)
(def ok
(try
(ev/with-deadline per-file-timeout
(set data (ev/read out 0x10000)) # workers print a single short line
(os/proc-wait proc)
true)
([err] false)))
(when (not ok)
(protect (os/proc-kill proc true))
(protect (ev/with-deadline 2 (os/proc-wait proc))))
(protect (:close out))
(if (and ok data) (string data) nil))
(defn- parse-counts [s]
# Find the "@@COUNTS p f e" sentinel line (a test body may have printed other
# lines to stdout, e.g. with-out-str tests).
(var result nil)
(each line (string/split "\n" s)
(when (string/has-prefix? "@@COUNTS " line)
(let [parts (string/split " " (string/trim line))]
(when (= 4 (length parts))
(set result [(scan-number (parts 1)) (scan-number (parts 2)) (scan-number (parts 3))])))))
result)
(if (not (os/stat suite-dir))
(print "clojure-test-suite: vendor/clojure-test-suite not initialized — skipped (run: git submodule update --init)")
(do
(def progress? (os/getenv "SUITE_PROGRESS"))
(def files (sort (walk suite-dir @[])))
(var total-pass 0)
(var total-fail 0)
(var total-error 0)
(var clean-files 0)
(var ran-files 0)
(var timeouts 0)
(def worst @[])
# Worker pool: each file is already its own subprocess (isolation + hang
# containment), so concurrency only changes wall-clock. A token-channel
# semaphore caps the live workers; os/spawn / ev/read / os/proc-wait are
# event-loop aware so the fibers genuinely interleave. Totals are
# order-independent. Default 4 keeps per-file wall-clock deadlines honest
# on small CI runners; override with JOLT_SUITE_WORKERS.
(def nworkers
(min 8 (max 1 (or (scan-number (or (os/getenv "JOLT_SUITE_WORKERS") ""))
(os/cpu-count) 4))))
(def sem (ev/chan nworkers))
(def done (ev/chan (length files)))
(each path files
(ev/spawn
(ev/give sem :tok) # acquire (blocks when pool is full)
(def out (run-file path))
(ev/take sem) # release
(ev/give done [path out])))
(for _ 0 (length files)
(def [path out] (ev/take done))
(def rel (string/slice path (+ 1 (length suite-dir))))
(def counts (and out (parse-counts out)))
(when progress?
(eprintf " %s%s" rel (cond (nil? out) " TIMEOUT" (nil? counts) " (no counts)" ""))
(eflush))
(cond
(nil? out) (++ timeouts)
(nil? counts) nil
(let [[pn fn* en] counts]
(++ ran-files)
(+= total-pass pn)
(+= total-fail fn*)
(+= total-error en)
(when (and (= 0 fn*) (= 0 en) (> pn 0)) (++ clean-files))
(when (> (+ fn* en) 0) (array/push worst [(+ fn* en) rel pn fn* en])))))
(def total (+ total-pass total-fail total-error))
(printf "\nclojure-test-suite: %d files ran (%d timed out), %d assertions — %d pass / %d fail / %d error"
ran-files timeouts total total-pass total-fail total-error)
(printf "\n clean files (0 fail/error, >0 pass): %d" clean-files)
(sort-by (fn [x] (- (x 0))) worst)
(when (> (length worst) 0)
(print " top files by fail+error:")
(each w (slice worst 0 (min 15 (length worst)))
(printf " %-40s pass=%d fail=%d err=%d" (w 1) (w 2) (w 3) (w 4))))
(assert (>= total-pass baseline-pass)
(string/format "regression: total-pass %d < baseline %d" total-pass baseline-pass))
(assert (>= clean-files baseline-clean-files)
(string/format "regression: clean-files %d < baseline %d" clean-files baseline-clean-files))
(printf "\nclojure-test-suite: OK (>= %d pass, >= %d clean files)\n" baseline-pass baseline-clean-files)))