jolt/jolt-core/clojure/core/MIGRATION.md

7.5 KiB
Raw Blame History

clojure.core migration worklist (jolt-1j0)

Tracking the move of clojure.core from native Janet (src/jolt/core.janet, 4145 lines / 421 core-* fns) into the self-hosted Clojure overlay (jolt-core/clojure/core/). Goal: shrink the Janet seed to core-renames + genuinely host-coupled fns.

Phase 0 classification (heuristic — validate per batch)

Bucket Count Disposition
SEED (in compiler/core-renames) 73 stay in Janet (compiler emits core-X directly)
MACRO (in core-macro-names) 44 Phase 3
HOST-coupled (atoms/vars/meta/proxy/transient/arrays/futures/ns/io) 80 Phase 4 (where feasible) / stay
LAZY-coupled 28 Phase 5
MOVABLE pure-eager (candidates) 193 Phase 2

Counts are heuristic (name + body markers); the MOVABLE list still has some host/lazy leakage (e.g. transient assoc!/conj!, doall/dorun, chunk-*, deliver) to filter out as each batch is actually moved.

Key finding: after removing SEED + HOST, the self-hosted compiler (jolt-core/jolt/{ir,analyzer}.clj) uses no additional clojure.core fns beyond the kernel tier (second/peek/subvec/mapv/update) plus host primitives (atom/swap!/reset!). So Phase 1 (compiler-dep kernel tier) is essentially already complete — to verify, not build.

Performance baseline (test/bench/core-bench.janet, compile mode, min of 5, ms)

bench ms
fib 128
seq-pipe 88
reduce 391
into-vec 194
map-build 681
map-read 6
str-join 244
hof 604
TOTAL 2336

Re-run after each phase; watch for regressions as fns move from native Janet to self-hosted Clojure (interpreted/compiled, slower than native primitives).

Per-batch workflow + gate (every migration step)

  1. Canonical Clojure def in the overlay tier; remove the Janet core-X defn + its core-bindings entry (confirm leaf first: only defn+binding refs).
  2. Add regression tests for each moved fn — spec cases (test/spec/*-spec.janet, interpret) and, for any fn whose behavior is subtle or was buggy, a case in the 3-mode conformance set (test/integration/conformance-test.janet).
  3. Gate: conformance ×3 modes · clojure-test-suite ≥ baseline · stage2==stage3 fixpoint · fib compiled-fast · core-bench A/B under identical load (the absolute number is load-sensitive — compare batch-vs-prior back to back).

If a moved fn surfaces a latent bug (e.g. nthrest's nil-vs-() result, the if-let/when-let else-scope leak), fix it to match Clojure and add a regression test, rather than preserving the bug.

MOVABLE candidates (Phase 2 worklist, 193)

Eduction NaN? abs aclone alength ancestors array-map array-seq assoc! associative? bean bigdec bigint biginteger boolean boolean? booleans byte bytes bytes? cat char char-escape-string char-name-string char? chars chunk chunk-append chunk-buffer chunk-cons chunk-first chunk-next chunk-rest chunked-seq? class clojure-version comparator compare-and-set! completing conj! counted? decimal? deliver denominator derive descendants destructure disj disj! dissoc! distinct? doall dorun double? doubles drop-last eduction empty ensure-reduced enumeration-seq ex-cause ex-data ex-info ex-info? ex-message find float? floats force halt-when hash-combine hash-map hash-ordered-coll hash-set hash-unordered-coll ident? ifn? indexed? infinite? inst-ms inst? integer? ints isa? iterator-seq key keyword keyword-identical? list* list? longs macrofy map-entry? memfn munge nat-int? neg-int? not-any? not-every? nthnext nthrest numerator numeric= object? parents persistent! pop pop! pos-int? pr prefers println-str prn-str promise qualified-ident? qualified-keyword? qualified-symbol? rand rand-nth random-sample ratio? rational? rationalize re-groups re-matcher record? reduce-kv reduced reduced? reductions replace replicate resolve reversible? rseq rsubseq run! seq-to-map-for-destructuring seque set set? short shorts shuffle simple-ident? simple-keyword? simple-symbol? some-search sort sort-by sorted-map sorted-map-by sorted-map? sorted-set sorted-set-by sorted-set? special-symbol? split-at split-with str-join str-replace-all str-replace-first str-split subseq supers symbol tagged-literal tagged-literal? take-last test transduce unchecked-add unchecked-byte unchecked-char unchecked-dec unchecked-divide-int unchecked-double unchecked-float unchecked-inc unchecked-int unchecked-multiply unchecked-negate unchecked-remainder-int unchecked-short unchecked-subtract undefined? underive uri? uuid? val vector volatile! volatile? xml-seq

HOST-coupled (Phase 4 / stay, 80)

add-watch aget alter-meta! alter-var-root aset aset-boolean aset-byte aset-char aset-double aset-float aset-int aset-long aset-short atom atom? avoid-method-too-large boolean-array bounded-count byte-array char-array construct-proxy copy-core-var copy-var delay? deref double-array float-array future-call future-cancel future-cancelled? future-done? future? get-proxy-class get-validator init-proxy int-array intern into-array long-array make-array make-delay meta namespace namespace-munge new-var ns-name object-array pop-thread-bindings prefer-method print-dup print-method print-str proxy-call-with-super proxy-mappings proxy-super push-thread-bindings reader-conditional reader-conditional? remove-watch reset! reset-meta! reset-vals! set-validator! short-array swap! swap-vals! thread-first thread-last to-array to-array-2d transient transient? update-proxy var-dynamic? var-get var-set var? vary-meta vreset! vswap! with-meta

LAZY-coupled (Phase 5, 28)

concat cycle dedupe distinct flatten interleave interpose iterate keep keep-indexed line-seq macro-names map-indexed mapcat partition partition-all partition-by rand-int random-uuid realized? repeat repeatedly seqable? sequence sequential? take-nth trampoline tree-seq unreduced

Phase 3 (macros) — status & findings

20 macros moved to the overlay: 19 user-facing in 30-macros.clj, plus when in a new 00-syntax.clj tier loaded before the kernel (interpreted defmacros, so the macros exist before any code that uses them compiles).

Macro-authoring toolkit for jolt (learned the hard way):

  • single-template hygiene: auto-gensym foo#
  • shared explicit fresh symbol: (symbol (str (gensym))) — a bare (gensym) in a macro body returns a Janet symbol the destructurer rejects
  • let-rebinding: splice binding pairs into a TEMPLATE vector ([~a ~b ~@pairs]), not a pre-built pvec value — core-let wants a tuple form
  • build sub-forms via templates, never cons/list (those make plists the evaluator can't run as a form)
  • Jolt defmacro is single-arity — use & rest/destructuring
  • syntax-tier macros may use only special forms + core-renames seed primitives

Performance wall (the hot macros stay in Janet for now): the load-order story works, but moving the hot fundamental control macros (and/or/cond/ when-not) regressed the battery — as interpreted overlay defmacros they expand slower than native Janet, and since they appear in nearly every form the cumulative overhead tipped a heavy suite file over the 6 s per-file timeout (3930 -> 3911, +1 timeout). They are correct (conformance 228×3, all edge cases), but reverted. Moving and/or/cond/when-not/case/doseq/declare/cond->/->/->> needs a fast (compiled) macro-expansion path, not interpreted defmacros.

Deferred: defn/defn-/fn/let/loop (fundamental + same speed concern), the type machinery (defrecord/defprotocol/extend-*/reify/proxy/definterface → Phase 4), lazy-seq/lazy-cat (→ Phase 5).