jolt/host/chez/natives-meta.ss
Yogthos bfa2cbf49d Small maps preserve insertion order
jolt maps were HAMTs with hash iteration order; Clojure keeps small maps as
PersistentArrayMap (insertion order), converting to PersistentHashMap past a
threshold. Map literals, array-map, assoc, into/transient, merge, zipmap,
select-keys, update-keys/vals, frequencies and group-by now iterate in insertion
order for <=8 entries, matching the JVM. hash-map and >8-entry maps stay hash
order; sets stay hash order.

The pmap record gains an order field (the insertion-order key list, or #f once
hashed); the HAMT still backs the values so equality/hash/lookup are unchanged.
pmap-fold visits an array-mode map last-to-first so the runtime's cons-accumulate
idiom reconstructs insertion order without touching its many call sites, and
hash-mode output stays byte-identical; pmap-fold-fwd visits in order for the few
sites that build a value directly. Transient maps track insertion order and
promote to hash past max(8, source-count), matching TransientArrayMap.

The hash-map native-op retargets to a hash-order builder so (hash-map ...) stays
hash-ordered while {...} literals are ordered; syntax-quote builds maps via the
hash builder (Clojure expands `{...} to apply hash-map). The core overlay map
builders seed from {} instead of (hash-map) to keep order.

Threshold is 8 for any key (the keyword exception in newer Clojure isn't in
1.12.5). honeysql now passes 832/0/0; 19 JVM-certified corpus rows added.
2026-06-27 05:48:17 -04:00

149 lines
7.4 KiB
Scheme

;; metadata — meta / with-meta. Chez values don't
;; carry metadata, so collections use an identity-keyed side-table: with-meta
;; returns a fresh COPY of the value (new identity) and records its meta there, so
;; the original is unchanged (Clojure's immutable-with-meta) and a copy made by a
;; later op (conj/assoc) drops the meta. Symbols carry meta in their own field.
;; meta on a non-metadatable value (number/string/keyword) is nil.
;;
;; Loaded after records.ss (jrec) + collections/seq/values (the ctors it copies).
;; Weak so a collection's metadata is reclaimed with the collection — collection
;; ops (conj/assoc/into) carry meta forward onto fresh values, so a strong table
;; would retain every meta-bearing intermediate.
(define meta-table (make-weak-eq-hashtable))
(define (jolt-meta x)
(cond
((symbol-t? x) (let ((m (symbol-t-meta x))) (if (jolt-nil? m) jolt-nil m)))
;; a var's meta is {:ns :name} (derived from the cell) + any def-time user
;; meta from rt.ss's var-meta-table.
((var-cell? x)
(let ((user (hashtable-ref var-meta-table x #f)))
(jolt-assoc (if user user (jolt-hash-map))
jolt-kw-var-ns (var-cell-ns x)
jolt-kw-var-name (var-cell-name x))))
((or (pvec? x) (pmap? x) (pset? x) (cseq? x) (empty-list-t? x) (jolt-lazyseq? x) (jrec? x) (jreify? x) (procedure? x))
(hashtable-ref meta-table x jolt-nil))
(else jolt-nil)))
;; fresh-identity copy of a metadatable value (so attaching meta doesn't mutate
;; the original). cseq/procedure can't be copied meaningfully — keyed in place.
(define (meta-copy x)
(cond
((pvec? x) (make-pvec (pvec-v x) (pvec-ent x)))
((pmap? x) (make-pmap (pmap-root x) (pmap-cnt x) (pmap-order x)))
((pset? x) (make-pset (pset-m x)))
((jrec? x) (make-jrec (jrec-desc x) (jrec-vec-copy (jrec-vals x)) (jrec-ext x)))
;; a reify shares its (read-only) method table + protos but gets a fresh
;; identity, so attaching meta leaves the original's meta untouched. Every
;; Clojure reify implements IObj.
((jreify? x) (make-jreify (jreify-methods x) (jreify-protos x)))
;; () is a shared singleton — a fresh instance keeps meta off every other ().
((empty-list-t? x) (fresh-empty-list))
;; a list/seq node gets a fresh identity too (Clojure's PersistentList is
;; immutable — (with-meta a-list m) returns a NEW list). Keying meta on the
;; original mutated it, so (with-meta xs {:k xs}) built a self-referential
;; cycle that loops *print-meta* printing.
((cseq? x) (make-cseq (cseq-head x) (cseq-tail x) (cseq-forced? x)
(cseq-list? x) (cseq-cvec x) (cseq-ci x)))
((jolt-lazyseq? x) (make-jolt-lazyseq (jolt-lazyseq-thunk x) (jolt-lazyseq-val x)
(jolt-lazyseq-realized? x)))
(else x))) ; procedure
(define (jolt-with-meta x m)
(cond
((symbol-t? x) (make-symbol-t (symbol-t-ns x) (symbol-t-name x) m))
((or (pvec? x) (pmap? x) (pset? x) (cseq? x) (empty-list-t? x) (jolt-lazyseq? x) (jrec? x) (jreify? x) (procedure? x))
(let ((c (meta-copy x)))
(if (jolt-nil? m) (hashtable-delete! meta-table c) (hashtable-set! meta-table c m))
c))
(else (error #f "with-meta: value does not support metadata" x))))
(def-var! "clojure.core" "meta" jolt-meta)
(def-var! "clojure.core" "with-meta" jolt-with-meta)
;; Carry SRC's collection metadata onto DST (a freshly-built collection of the
;; same kind), as Clojure's ops do — each new collection threads its receiver's
;; meta() forward. Returns DST. The size check is the fast path: programs that
;; never attach collection metadata pay one O(1) check per op, no lookup.
(define (meta-carry src dst)
(if (fx=? 0 (hashtable-size meta-table))
dst
(let ((m (hashtable-ref meta-table src #f)))
(if m
;; never attach to the shared () singleton — use a fresh instance
(let ((d (if (empty-list-t? dst) (fresh-empty-list) dst)))
(hashtable-set! meta-table d m) d)
dst))))
;; (type x) — Clojure's (or (:type (meta x)) (class x)). With no JVM classes the
;; "class" is a host taxonomy: a record yields its ns-qualified class-name SYMBOL
;; (user.TyR), everything else a keyword (:number/:vector/:seq/…).
;; MUST be total — a non-record value
;; falling through to a crash would read as a divergence, not the right keyword.
;; Forward refs (jolt-lazyseq?, the sorted-htable / wrapper predicates) all bind by
;; call time (every host .ss loads before any user expr runs).
(define ty-kw-type (keyword #f "type")) ; the :type meta key
(define ty-kw-jtype (keyword "jolt" "type")) ; tagged-map discriminator (ex-info)
(define ty-number (keyword #f "number"))
(define ty-string (keyword #f "string"))
(define ty-keyword (keyword #f "keyword"))
(define ty-symbol (keyword #f "symbol"))
(define ty-boolean (keyword #f "boolean"))
(define ty-char (keyword #f "char"))
(define ty-vector (keyword #f "vector"))
(define ty-map (keyword #f "map"))
(define ty-set (keyword #f "set"))
(define ty-seq (keyword #f "seq"))
(define ty-fn (keyword #f "fn"))
(define ty-atom (keyword "jolt" "atom"))
(define ty-volatile (keyword "jolt" "volatile"))
(define ty-regex (keyword "jolt" "regex"))
(define ty-var (keyword "jolt" "var"))
(define ty-transient (keyword "jolt" "transient"))
(define ty-uuid (keyword "jolt" "uuid"))
(define ty-sorted-set (keyword "jolt" "sorted-set"))
(define ty-object (keyword #f "object"))
(define (jolt-type x)
(let* ((m (jolt-meta x))
(override (if (jolt-nil? m) jolt-nil (jolt-get m ty-kw-type jolt-nil))))
(cond
((not (jolt-nil? override)) override) ; :type meta wins
;; record -> its ns-qualified class-name STRING (= (class x)). jolt models
;; classes as strings, so (symbol (str (type r))) is NOT (type r) — as on the
;; JVM where type is a Class, not a Symbol.
((jrec? x) (jrec-tag x))
((jolt-nil? x) jolt-nil)
((boolean? x) ty-boolean)
((number? x) ty-number)
((string? x) ty-string)
((keyword? x) ty-keyword)
((symbol-t? x) ty-symbol)
((char? x) ty-char)
;; host wrappers — keyed by their :jolt/* tags (checked before the
;; collection arms; none of these are pvec/pmap/pset).
((jolt-atom? x) ty-atom)
((jvol? x) ty-volatile)
((jolt-regex? x) ty-regex)
((var-cell? x) ty-var)
((jolt-transient? x) ty-transient)
((juuid? x) ty-uuid)
((htable-sorted-set? x) ty-sorted-set)
((htable-sorted-map? x) ty-map)
;; collections — pvec INCLUDES map entries (:vector).
((pvec? x) ty-vector)
((pmap? x) ; a :jolt/type-tagged map (ex-info) -> its tag
(let ((t (jolt-get x ty-kw-jtype jolt-nil))) (if (jolt-nil? t) ty-map t)))
((pset? x) ty-set)
((or (cseq? x) (empty-list-t? x) (jolt-lazyseq? x)) ty-seq)
((procedure? x) ty-fn)
(else ty-object))))
;; jolt-type is the keyword TAXONOMY (:string/:set/:jolt/inst/…) — jolt's native
;; value model, with no JVM in it. print-method/print-dup dispatch on it (via
;; __type-tag). The PUBLIC clojure.core/type is Clojure's (or (:type meta) (class
;; x)) — a JVM class — but that mapping belongs to the java host layer (host-class.ss
;; rebinds `type` next to `class`), so this core layer stays JVM-free.
(def-var! "clojure.core" "__type-tag" jolt-type)
(def-var! "clojure.core" "type" jolt-type)