Transients: mutable backing instead of copy-on-write
The Chez port had landed transients as copy-on-write — each conj!/assoc!/etc.
rebuilt the whole persistent collection. Semantics were right but a transient
vector was O(n^2) to build (the persistent vector is a flat array, so every
conj! copied it); maps/sets were ~O(n log n) since the HAMT only path-copies.
This restores the Janet host's approach: true mutable backing, snapshot once on
persistent!.
vec : a growable Scheme vector (capacity + fill count); conj!/pop! amortized
O(1), persistent! hands off the buffer (exact fit) or trims once.
map : a Chez hashtable keyed by key-hash/jolt= (value equality, nil-safe);
persistent! folds it into a pmap.
set : a Chez hashtable; persistent! folds into a pset.
cow : fallback for anything else (e.g. a sorted coll) keeps the old
copy-on-write path, preserving jolt's superset.
get/count/contains?/nth see through each representation. Building a 400k vector
went from minutes (quadratic) to ~50ms (linear). assoc! keeps the variadic
dangling-key nil-pad on both vectors and maps. test/chez/transient-test.ss pins
the invariants and the linear-time property; wired in as `make transient`.
This commit is contained in:
Yogthos
parent
6f433a1b3c
commit
fe3fdf6b9c
4 changed files with 210 additions and 44 deletions
|
|
@ -37,7 +37,8 @@
|
|||
;; multimethod; the readable fallback renders it directly).
|
||||
;; forward refs to transients.ss (loaded later) — resolved at call time.
|
||||
((jolt-transient? x)
|
||||
(if (pvec? (jolt-transient-coll x)) "#<transient vector>" "#<transient map>"))
|
||||
(case (jolt-transient-kind x)
|
||||
((vec) "#<transient vector>") ((set) "#<transient set>") (else "#<transient map>")))
|
||||
((pvec? x)
|
||||
(let ((acc '()))
|
||||
(let loop ((i (fx- (pvec-count x) 1)))
|
||||
|
|
|
|||
|
|
@ -1,31 +1,103 @@
|
|||
;; transients (jolt-kl2l) — copy-on-write over the persistent collections.
|
||||
;; transients (jolt-kl2l) — mutable backing per collection kind, snapshotted to
|
||||
;; the immutable collection on persistent!. A faithful port of the Janet host's
|
||||
;; mutable transients: conj!/assoc!/dissoc!/disj!/pop! mutate in place (amortized
|
||||
;; O(1)); persistent! converts back to a pvec / pmap / pset once.
|
||||
;;
|
||||
;; A first cut: each transient op rebuilds the persistent collection (no in-place
|
||||
;; mutation perf), but the SEMANTICS match — the overlay's transient users (into,
|
||||
;; frequencies, group-by) get correct results. get/count/contains? are extended to
|
||||
;; see THROUGH a transient to its held collection (frequencies/group-by do
|
||||
;; (get tm k) on a transient map). vector? on a transient vector is false (it's a
|
||||
;; jolt-transient record, not a pvec), which group-by relies on. Loaded after
|
||||
;; collections.ss (the persistent ops it delegates to) and converters.ss.
|
||||
;; vec : a growable Scheme vector (capacity) + a fill count `n`. conj!/pop! are
|
||||
;; O(1) amortized — the old copy-on-write rebuilt the whole vector per op,
|
||||
;; so building an N-vector was O(N^2).
|
||||
;; map : a Chez hashtable keyed by key-hash / jolt= (value-equality, nil-safe —
|
||||
;; a jolt-nil key stores fine here, unlike a Janet table).
|
||||
;; set : a Chez hashtable of elements.
|
||||
;; cow : fallback for anything else (e.g. a sorted coll) — copy-on-write over
|
||||
;; the persistent ops, preserving jolt's superset of Clojure's transients.
|
||||
;;
|
||||
;; get/count/contains?/nth see THROUGH a transient (frequencies/group-by read a
|
||||
;; transient map; a transient is callable). vector? on a transient is false (it's
|
||||
;; this record, not a pvec), which group-by relies on. Loaded after collections.ss
|
||||
;; (persistent ops + key-hash) and converters.ss.
|
||||
|
||||
(define-record-type jolt-transient (fields (mutable coll) (mutable active))
|
||||
(nongenerative jolt-transient-v1))
|
||||
(define-record-type jolt-transient
|
||||
(fields kind (mutable buf) (mutable n) (mutable active))
|
||||
(nongenerative jolt-transient-v2))
|
||||
|
||||
(define (jolt-transient-new coll) (make-jolt-transient coll #t))
|
||||
(define tvec-min-cap 8)
|
||||
|
||||
(define (jolt-transient-new coll)
|
||||
(cond
|
||||
((pvec? coll)
|
||||
(let* ((v (pvec-v coll)) (cnt (vector-length v)) (cap (fxmax tvec-min-cap cnt))
|
||||
(buf (make-vector cap jolt-nil)))
|
||||
(let loop ((i 0)) (when (fx<? i cnt) (vector-set! buf i (vector-ref v i)) (loop (fx+ i 1))))
|
||||
(make-jolt-transient 'vec buf cnt #t)))
|
||||
((pmap? coll)
|
||||
(let ((ht (make-hashtable key-hash jolt=2)))
|
||||
(pmap-fold coll (lambda (k v acc) (hashtable-set! ht k v) acc) 0)
|
||||
(make-jolt-transient 'map ht 0 #t)))
|
||||
((pset? coll)
|
||||
(let ((ht (make-hashtable key-hash jolt=2)))
|
||||
(pset-fold coll (lambda (e acc) (hashtable-set! ht e #t) acc) 0)
|
||||
(make-jolt-transient 'set ht 0 #t)))
|
||||
(else (make-jolt-transient 'cow coll 0 #t))))
|
||||
|
||||
(define (jolt-trans-check t who)
|
||||
(unless (jolt-transient? t) (error #f (string-append who ": not a transient") t))
|
||||
(unless (jolt-transient-active t)
|
||||
(error #f (string-append who ": transient used after persistent!"))))
|
||||
(unless (jolt-transient-active t) (error #f (string-append who ": transient used after persistent!"))))
|
||||
|
||||
;; --- persistent! : snapshot back to the immutable collection -----------------
|
||||
(define (jolt-persistent! t)
|
||||
(jolt-trans-check t "persistent!")
|
||||
(jolt-transient-active-set! t #f)
|
||||
(jolt-transient-coll t))
|
||||
(case (jolt-transient-kind t)
|
||||
((vec)
|
||||
(let ((buf (jolt-transient-buf t)) (cnt (jolt-transient-n t)))
|
||||
;; exact fit: hand off the buffer (no other reference exists, and the
|
||||
;; transient is now inactive so it can't mutate it). Else trim to size —
|
||||
;; a pvec's backing length must equal its count.
|
||||
(if (fx=? cnt (vector-length buf))
|
||||
(make-pvec buf)
|
||||
(let ((out (make-vector cnt)))
|
||||
(let loop ((i 0))
|
||||
(if (fx<? i cnt) (begin (vector-set! out i (vector-ref buf i)) (loop (fx+ i 1)))
|
||||
(make-pvec out)))))))
|
||||
((map)
|
||||
(let ((ht (jolt-transient-buf t)) (m empty-pmap))
|
||||
(vector-for-each (lambda (k) (set! m (pmap-assoc m k (hashtable-ref ht k jolt-nil)))) (hashtable-keys ht))
|
||||
m))
|
||||
((set)
|
||||
(let ((ht (jolt-transient-buf t)) (s empty-pset))
|
||||
(vector-for-each (lambda (e) (set! s (pset-conj s e))) (hashtable-keys ht))
|
||||
s))
|
||||
(else (jolt-transient-buf t))))
|
||||
|
||||
;; (conj!) with no args -> a fresh transient vector (Clojure); (conj! coll) is the
|
||||
;; 1-arity transducer-completion arity, which returns coll as-is with NO transient
|
||||
;; check (JVM: conj! 1-arity is identity); otherwise mutate t.
|
||||
;; --- in-place mutation -------------------------------------------------------
|
||||
(define (tvec-ensure! t need) ; grow capacity to >= need by doubling
|
||||
(let ((buf (jolt-transient-buf t)))
|
||||
(when (fx>? need (vector-length buf))
|
||||
(let* ((ncap (let grow ((c (fxmax tvec-min-cap (vector-length buf)))) (if (fx>=? c need) c (grow (fx* 2 c)))))
|
||||
(nbuf (make-vector ncap jolt-nil)) (cnt (jolt-transient-n t)))
|
||||
(let loop ((i 0)) (when (fx<? i cnt) (vector-set! nbuf i (vector-ref buf i)) (loop (fx+ i 1))))
|
||||
(jolt-transient-buf-set! t nbuf)))))
|
||||
(define (tvec-conj1! t x)
|
||||
(let ((cnt (jolt-transient-n t)))
|
||||
(tvec-ensure! t (fx+ cnt 1))
|
||||
(vector-set! (jolt-transient-buf t) cnt x)
|
||||
(jolt-transient-n-set! t (fx+ cnt 1))))
|
||||
(define (tvec-assoc1! t i x)
|
||||
(let ((i (->idx i)) (cnt (jolt-transient-n t)))
|
||||
(cond ((and (fixnum? i) (fx>=? i 0) (fx<? i cnt)) (vector-set! (jolt-transient-buf t) i x))
|
||||
((and (fixnum? i) (fx=? i cnt)) (tvec-conj1! t x))
|
||||
(else (error #f "assoc!: index out of bounds")))))
|
||||
;; conj! onto a transient map: a [k v] pair (vector/map-entry) or a whole map.
|
||||
(define (tmap-conj-entry! t x)
|
||||
(cond
|
||||
((jolt-nil? x) #t)
|
||||
((pvec? x) (hashtable-set! (jolt-transient-buf t) (pvec-nth-d x 0 jolt-nil) (pvec-nth-d x 1 jolt-nil)))
|
||||
((pmap? x) (pmap-fold x (lambda (k v acc) (hashtable-set! (jolt-transient-buf t) k v) acc) 0))
|
||||
(else (error #f "conj!: a transient map takes a map entry or a map" x))))
|
||||
|
||||
;; (conj!) -> fresh transient vector; (conj! coll) -> the 1-arity transducer-
|
||||
;; completion identity (JVM: no transient check). (conj! t x ...) mutates t.
|
||||
(define (jolt-conj! . args)
|
||||
(cond
|
||||
((null? args) (jolt-transient-new (jolt-vector)))
|
||||
|
|
@ -33,55 +105,96 @@
|
|||
(else
|
||||
(let ((t (car args)) (xs (cdr args)))
|
||||
(jolt-trans-check t "conj!")
|
||||
(jolt-transient-coll-set! t (apply jolt-conj (jolt-transient-coll t) xs))
|
||||
(case (jolt-transient-kind t)
|
||||
((vec) (for-each (lambda (x) (tvec-conj1! t x)) xs))
|
||||
((set) (for-each (lambda (x) (hashtable-set! (jolt-transient-buf t) x #t)) xs))
|
||||
((map) (for-each (lambda (x) (tmap-conj-entry! t x)) xs))
|
||||
(else (jolt-transient-buf-set! t (apply jolt-conj (jolt-transient-buf t) xs))))
|
||||
t))))
|
||||
;; (assoc! t k v & kvs): variadic like Clojure (jolt-assoc already folds pairs).
|
||||
(define (jolt-assoc! t . kvs)
|
||||
|
||||
;; assoc! is variadic. JVM: a complete first key/val pair present (>=3 kvs) with a
|
||||
;; trailing lone key fills nil; a lone key alone (1 kv) is a wrong-arity throw.
|
||||
(define (assoc-pad kvs) (if (and (>= (length kvs) 3) (odd? (length kvs))) (append kvs (list jolt-nil)) kvs))
|
||||
(define (jolt-assoc! t . kvs0)
|
||||
(jolt-trans-check t "assoc!")
|
||||
;; JVM assoc! is variadic: once a complete first key/val pair is present (>=3
|
||||
;; kvs), a TRAILING lone key fills nil; a lone key alone (1 kv) is a wrong-arity
|
||||
;; throw (so don't pad it — the persistent assoc raises on odd args).
|
||||
(let ((kvs (if (and (>= (length kvs) 3) (odd? (length kvs)))
|
||||
(append kvs (list jolt-nil))
|
||||
kvs)))
|
||||
(jolt-transient-coll-set! t (apply jolt-assoc (jolt-transient-coll t) kvs)))
|
||||
(let ((kvs (assoc-pad kvs0)))
|
||||
(when (odd? (length kvs)) (error #f "assoc!: no value supplied for key"))
|
||||
(case (jolt-transient-kind t)
|
||||
((map) (let lp ((xs kvs)) (unless (null? xs) (hashtable-set! (jolt-transient-buf t) (car xs) (cadr xs)) (lp (cddr xs)))))
|
||||
((vec) (let lp ((xs kvs)) (unless (null? xs) (tvec-assoc1! t (car xs) (cadr xs)) (lp (cddr xs)))))
|
||||
(else (jolt-transient-buf-set! t (apply jolt-assoc (jolt-transient-buf t) kvs)))))
|
||||
t)
|
||||
(define (jolt-dissoc! t . ks)
|
||||
(jolt-trans-check t "dissoc!")
|
||||
(jolt-transient-coll-set! t (apply jolt-dissoc (jolt-transient-coll t) ks))
|
||||
(case (jolt-transient-kind t)
|
||||
((map) (for-each (lambda (k) (hashtable-delete! (jolt-transient-buf t) k)) ks))
|
||||
(else (jolt-transient-buf-set! t (apply jolt-dissoc (jolt-transient-buf t) ks))))
|
||||
t)
|
||||
(define (jolt-disj! t . xs)
|
||||
(jolt-trans-check t "disj!")
|
||||
(jolt-transient-coll-set! t (apply jolt-disj (jolt-transient-coll t) xs))
|
||||
(case (jolt-transient-kind t)
|
||||
((set) (for-each (lambda (x) (hashtable-delete! (jolt-transient-buf t) x)) xs))
|
||||
(else (jolt-transient-buf-set! t (apply jolt-disj (jolt-transient-buf t) xs))))
|
||||
t)
|
||||
(define (jolt-pop! t)
|
||||
(jolt-trans-check t "pop!")
|
||||
(jolt-transient-coll-set! t (jolt-pop (jolt-transient-coll t)))
|
||||
(case (jolt-transient-kind t)
|
||||
((vec) (let ((cnt (jolt-transient-n t)))
|
||||
(if (fx=? cnt 0) (error #f "pop!: can't pop empty transient vector")
|
||||
(jolt-transient-n-set! t (fx- cnt 1)))))
|
||||
(else (jolt-transient-buf-set! t (jolt-pop (jolt-transient-buf t)))))
|
||||
t)
|
||||
|
||||
;; persistent disj over sets (pset-disj already exists in collections.ss).
|
||||
(define (jolt-disj s . xs)
|
||||
(let loop ((s s) (xs xs)) (if (null? xs) s (loop (pset-disj s (car xs)) (cdr xs)))))
|
||||
|
||||
;; get / count / contains? see through a transient. Redefine the native procedures
|
||||
;; (captured first) so the existing emit lowerings (jolt-get/jolt-count/
|
||||
;; jolt-contains?) unwrap a transient before delegating; non-transients are
|
||||
;; untouched.
|
||||
(define (jolt-deref-transient x) (if (jolt-transient? x) (jolt-transient-coll x) x))
|
||||
;; --- see-through accessors ---------------------------------------------------
|
||||
(define (tvec-in-bounds? t i) (and (fixnum? i) (fx>=? i 0) (fx<? i (jolt-transient-n t))))
|
||||
(define (t-get t k d)
|
||||
(case (jolt-transient-kind t)
|
||||
((vec) (let ((i (->idx k))) (if (tvec-in-bounds? t i) (vector-ref (jolt-transient-buf t) i) d)))
|
||||
((map) (hashtable-ref (jolt-transient-buf t) k d))
|
||||
((set) (if (hashtable-contains? (jolt-transient-buf t) k) k d))
|
||||
(else (%prev-jolt-get (jolt-transient-buf t) k d))))
|
||||
(define (t-count t)
|
||||
(case (jolt-transient-kind t)
|
||||
((vec) (jolt-transient-n t))
|
||||
((map set) (hashtable-size (jolt-transient-buf t)))
|
||||
(else (%prev-jolt-count (jolt-transient-buf t)))))
|
||||
(define (t-contains? t k)
|
||||
(case (jolt-transient-kind t)
|
||||
((vec) (tvec-in-bounds? t (->idx k)))
|
||||
((map set) (hashtable-contains? (jolt-transient-buf t) k))
|
||||
(else (%prev-jolt-contains? (jolt-transient-buf t) k))))
|
||||
|
||||
;; Redefine the native get/count/contains?/nth (captured first) so the existing
|
||||
;; emit lowerings unwrap a transient; non-transients are untouched.
|
||||
(define %prev-jolt-get jolt-get)
|
||||
(set! jolt-get
|
||||
(case-lambda
|
||||
((coll k) (%prev-jolt-get (jolt-deref-transient coll) k))
|
||||
((coll k d) (%prev-jolt-get (jolt-deref-transient coll) k d))))
|
||||
((coll k) (if (jolt-transient? coll) (t-get coll k jolt-nil) (%prev-jolt-get coll k)))
|
||||
((coll k d) (if (jolt-transient? coll) (t-get coll k d) (%prev-jolt-get coll k d)))))
|
||||
(define %prev-jolt-count jolt-count)
|
||||
(set! jolt-count (lambda (coll) (%prev-jolt-count (jolt-deref-transient coll))))
|
||||
(set! jolt-count (lambda (coll) (if (jolt-transient? coll) (t-count coll) (%prev-jolt-count coll))))
|
||||
(define %prev-jolt-contains? jolt-contains?)
|
||||
(set! jolt-contains? (lambda (coll k) (%prev-jolt-contains? (jolt-deref-transient coll) k)))
|
||||
(set! jolt-contains? (lambda (coll k) (if (jolt-transient? coll) (t-contains? coll k) (%prev-jolt-contains? coll k))))
|
||||
(define %prev-jolt-nth jolt-nth)
|
||||
(set! jolt-nth
|
||||
(case-lambda
|
||||
((coll i) (%prev-jolt-nth (jolt-deref-transient coll) i))
|
||||
((coll i d) (%prev-jolt-nth (jolt-deref-transient coll) i d))))
|
||||
((coll i)
|
||||
(if (jolt-transient? coll)
|
||||
(if (eq? (jolt-transient-kind coll) 'vec)
|
||||
(let ((idx (->idx i)))
|
||||
(if (tvec-in-bounds? coll idx) (vector-ref (jolt-transient-buf coll) idx) (error 'nth "index out of bounds")))
|
||||
(%prev-jolt-nth (jolt-transient-buf coll) i))
|
||||
(%prev-jolt-nth coll i)))
|
||||
((coll i d)
|
||||
(if (jolt-transient? coll)
|
||||
(if (eq? (jolt-transient-kind coll) 'vec)
|
||||
(let ((idx (->idx i))) (if (tvec-in-bounds? coll idx) (vector-ref (jolt-transient-buf coll) idx) d))
|
||||
(%prev-jolt-nth (jolt-transient-buf coll) i d))
|
||||
(%prev-jolt-nth coll i d)))))
|
||||
|
||||
(def-var! "clojure.core" "transient" jolt-transient-new)
|
||||
(def-var! "clojure.core" "persistent!" jolt-persistent!)
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue