mirror/guile
mirror/guile
1
Fork 0
mirror of https://git.savannah.gnu.org/git/guile.git synced 2025-06-17 09:10:22 +02:00
Code Activity

Add `(ice-9 vlist)'.

Browse source 
* module/ice-9/vlist.scm, test-suite/tests/vlist.test,
  benchmark-suite/benchmarks/vlists.bm: New files.

* module/Makefile.am (ICE_9_SOURCES): Add `vlist.scm'.

* test-suite/Makefile.am (SCM_TESTS): Add `tests/vlist.test'.

* benchmark-suite/Makefile.am (SCM_BENCHMARKS): Add
  `benchmarks/vlists.bm'.

* doc/ref/api-compound.texi (VLists, VHashes): New nodes.
This commit is contained in:
Ludovic Courtès 2010-02-02 23:57:02 +01:00
parent 30a700c8c1
commit 22ec6a31ed
7 changed files with 1183 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

3
module/Makefile.am
View file

@ -225,7 +225,8 @@ ICE_9_SOURCES = \
ice-9/deprecated.scm \
ice-9/list.scm \
ice-9/serialize.scm \
ice-9/gds-server.scm
ice-9/gds-server.scm \
ice-9/vlist.scm
SRFI_SOURCES = \
srfi/srfi-1.scm \

492
module/ice-9/vlist.scm Normal file
View file

@ -0,0 +1,492 @@
;;; -*- mode: scheme; coding: utf-8; -*-
;;;
;;; Copyright (C) 2009, 2010 Free Software Foundation, Inc.
;;;
;;; This library is free software; you can redistribute it and/or
;;; modify it under the terms of the GNU Lesser General Public
;;; License as published by the Free Software Foundation; either
;;; version 3 of the License, or (at your option) any later version.
;;;
;;; This library is distributed in the hope that it will be useful,
;;; but WITHOUT ANY WARRANTY; without even the implied warranty of
;;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
;;; Lesser General Public License for more details.
;;;
;;; You should have received a copy of the GNU Lesser General Public
;;; License along with this library; if not, write to the Free Software
;;; Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
(define-module (ice-9 vlist)
#:use-module (srfi srfi-1)
#:use-module (srfi srfi-9)
#:use-module (srfi srfi-26)
#:export (vlist? vlist-cons vlist-head vlist-tail vlist-null?
vlist-null list->vlist vlist-ref vlist-drop vlist-take
vlist-length vlist-fold vlist-fold-right vlist-map
vlist-unfold vlist-unfold-right vlist-append
vlist-reverse vlist-filter vlist-delete vlist->list
vlist-for-each
block-growth-factor
vhash? vhash-cons vhash-consq vhash-consv
vhash-assoc vhash-assq vhash-assv
vhash-delete vhash-fold alist->vhash))
;;; Author: Ludovic Courtès <ludo@gnu.org>
;;;
;;; Commentary:
;;;
;;; This module provides an implementations of vlists, a functional list-like
;;; data structure described by Phil Bagwell in "Fast Functional Lists,
;;; Hash-Lists, Dequeues and Variable-Length Arrays", EPFL Technical Report,
;;; 2002.
;;;
;;; The idea is to store vlist elements in increasingly large contiguous blocks
;;; (implemented as vectors here). These blocks are linked to one another using
;;; a pointer to the next block (called `block-base' here) and an offset within
;;; that block (`block-offset' here). The size of these blocks form a geometric
;;; series with ratio `block-growth-factor'.
;;;
;;; In the best case (e.g., using a vlist returned by `list->vlist'),
;;; elements from the first half of an N-element vlist are accessed in O(1)
;;; (assuming `block-growth-factor' is 2), and `vlist-length' takes only
;;; O(ln(N)). Furthermore, the data structure improves data locality since
;;; vlist elements are adjacent, which plays well with caches.
;;;
;;; Code:
;;;
;;; VList Blocks and Block Descriptors.
;;;
(define block-growth-factor
(let ((f (make-fluid)))
(fluid-set! f 2)
f))
(define-syntax define-inline
;; Work around the lack of an inliner.
(syntax-rules ()
((_ (name formals ...) body ...)
(define-syntax name
(syntax-rules ()
((_ formals ...)
body ...))))))
(define-inline (make-block base offset size hash-tab?)
;; Return a block (and block descriptor) of SIZE elements pointing to BASE
;; at OFFSET. If HASH-TAB? is true, a "hash table" is also added.
;; Note: We use `next-free' instead of `last-used' as suggested by Bagwell.
;; XXX: We could improve locality here by having a single vector but currently
;; the extra arithmetic outweighs the benefits (!).
(vector (make-vector size)
base offset size 0
(and hash-tab? (make-vector size #f))))
(define-syntax define-block-accessor
(syntax-rules ()
((_ name index)
(define-inline (name block)
(vector-ref block index)))))
(define-block-accessor block-content 0)
(define-block-accessor block-base 1)
(define-block-accessor block-offset 2)
(define-block-accessor block-size 3)
(define-block-accessor block-next-free 4)
(define-block-accessor block-hash-table 5)
(define-inline (increment-block-next-free! block)
(vector-set! block 4
(+ (block-next-free block) 1)))
(define-inline (block-append! block value)
;; This is not thread-safe. To fix it, see Section 2.8 of the paper.
(let ((offset (block-next-free block)))
(increment-block-next-free! block)
(vector-set! (block-content block) offset value)
#t))
(define-inline (block-ref block offset)
(vector-ref (block-content block) offset))
(define-inline (block-ref* block offset)
(let ((v (block-ref block offset)))
(if (block-hash-table block)
(car v) ;; hide the vhash link
v)))
(define-inline (block-hash-table-ref block offset)
(vector-ref (block-hash-table block) offset))
(define-inline (block-hash-table-set! block offset value)
(vector-set! (block-hash-table block) offset value))
(define block-null
;; The null block.
(make-block #f 0 0 #f))
;;;
;;; VLists.
;;;
(define-record-type <vlist>
;; A vlist is just a base+offset pair pointing to a block.
;; XXX: Allocating a <vlist> record in addition to the block at each
;; `vlist-cons' call is inefficient. However, Bagwell's hack to avoid it
;; (Section 2.2) would require GC_ALL_INTERIOR_POINTERS, which would be a
;; performance hit for everyone.
(make-vlist base offset)
vlist?
(base vlist-base)
(offset vlist-offset))
(define vlist-null
;; The empty vlist.
(make-vlist block-null 0))
(define-inline (block-cons item vlist hash-tab?)
(let loop ((base (vlist-base vlist))
(offset (+ 1 (vlist-offset vlist))))
(if (and (< offset (block-size base))
(= offset (block-next-free base))
(block-append! base item))
(make-vlist base offset)
(let ((size (cond ((eq? base block-null) 1)
((< offset (block-size base))
;; new vlist head
1)
(else
(* (fluid-ref block-growth-factor)
(block-size base))))))
;; Prepend a new block pointing to BASE.
(loop (make-block base (- offset 1) size hash-tab?)
0)))))
(define (vlist-cons item vlist)
"Return a new vlist with @var{item} as its head and @var{vlist} as its
tail."
;; Note: Calling `vlist-cons' on a vhash will not do the right thing: it
;; doesn't box ITEM so that it can have the hidden "next" link used by
;; vhash items, and it passes `#f' as the HASH-TAB? argument to
;; `block-cons'. However, inserting all the checks here has an important
;; performance penalty, hence this choice.
(block-cons item vlist #f))
(define (vlist-head vlist)
"Return the head of @var{vlist}."
(let ((base (vlist-base vlist))
(offset (vlist-offset vlist)))
(block-ref* base offset)))
(define (vlist-tail vlist)
"Return the tail of @var{vlist}."
(let ((base (vlist-base vlist))
(offset (vlist-offset vlist)))
(if (> offset 0)
(make-vlist base (- offset 1))
(make-vlist (block-base base)
(block-offset base)))))
(define (vlist-null? vlist)
"Return true if @var{vlist} is empty."
(let ((base (vlist-base vlist)))
(and (not (block-base base))
(= 0 (block-size base)))))
;;;
;;; VList Utilities.
;;;
(define (list->vlist lst)
"Return a new vlist whose contents correspond to @var{lst}."
(vlist-reverse (fold vlist-cons vlist-null lst)))
(define (vlist-fold proc init vlist)
"Fold over @var{vlist}, calling @var{proc} for each element."
;; FIXME: Handle multiple lists.
(let loop ((base (vlist-base vlist))
(offset (vlist-offset vlist))
(result init))
(if (eq? base block-null)
result
(let* ((next (- offset 1))
(done? (< next 0)))
(loop (if done? (block-base base) base)
(if done? (block-offset base) next)
(proc (block-ref* base offset) result))))))
(define (vlist-fold-right proc init vlist)
"Fold over @var{vlist}, calling @var{proc} for each element, starting from
the last element."
(vlist-fold proc init (vlist-reverse vlist)))
(define (vlist-reverse vlist)
"Return a new @var{vlist} whose content are those of @var{vlist} in reverse
order."
(vlist-fold vlist-cons vlist-null vlist))
(define (vlist-map proc vlist)
"Map @var{proc} over the elements of @var{vlist} and return a new vlist."
(vlist-fold (lambda (item result)
(vlist-cons (proc item) result))
vlist-null
(vlist-reverse vlist)))
(define (vlist->list vlist)
"Return a new list whose contents match those of @var{vlist}."
(vlist-fold-right cons '() vlist))
(define (vlist-ref vlist index)
"Return the element at index @var{index} in @var{vlist}."
(let loop ((index index)
(base (vlist-base vlist))
(offset (vlist-offset vlist)))
(if (<= index offset)
(block-ref* base (- offset index))
(loop (- index offset 1)
(block-base base)
(block-offset base)))))
(define (vlist-drop vlist count)
"Return a new vlist that does not contain the @var{count} first elements of
@var{vlist}."
(let loop ((count count)
(base (vlist-base vlist))
(offset (vlist-offset vlist)))
(if (<= count offset)
(make-vlist base (- offset count))
(loop (- count offset 1)
(block-base base)
(block-offset base)))))
(define (vlist-take vlist count)
"Return a new vlist that contains only the @var{count} first elements of
@var{vlist}."
(let loop ((count count)
(vlist vlist)
(result vlist-null))
(if (= 0 count)
(vlist-reverse result)
(loop (- count 1)
(vlist-tail vlist)
(vlist-cons (vlist-head vlist) result)))))
(define (vlist-filter pred vlist)
"Return a new vlist containing all the elements from @var{vlist} that
satisfy @var{pred}."
(vlist-fold-right (lambda (e v)
(if (pred e)
(vlist-cons e v)
v))
vlist-null
vlist))
(define* (vlist-delete x vlist #:optional (equal? equal?))
"Return a new vlist corresponding to @var{vlist} without the elements
@var{equal?} to @var{x}."
(vlist-filter (lambda (e)
(not (equal? e x)))
vlist))
(define (vlist-length vlist)
"Return the length of @var{vlist}."
(let loop ((base (vlist-base vlist))
(len (vlist-offset vlist)))
(if (eq? base block-null)
len
(loop (block-base base)
(+ len 1 (block-offset base))))))
(define* (vlist-unfold p f g seed
#:optional (tail-gen (lambda (x) vlist-null)))
"Return a new vlist. See the description of SRFI-1 `unfold' for details."
(let uf ((seed seed))
(if (p seed)
(tail-gen seed)
(vlist-cons (f seed)
(uf (g seed))))))
(define* (vlist-unfold-right p f g seed #:optional (tail vlist-null))
"Return a new vlist. See the description of SRFI-1 `unfold-right' for
details."
(let uf ((seed seed) (lis tail))
(if (p seed)
lis
(uf (g seed) (vlist-cons (f seed) lis)))))
(define (vlist-append . vlists)
"Append the given lists."
(if (null? vlists)
vlist-null
(fold-right (lambda (vlist result)
(vlist-fold-right (lambda (e v)
(vlist-cons e v))
result
vlist))
vlist-null
vlists)))
(define (vlist-for-each proc vlist)
"Call @var{proc} on each element of @var{vlist}. The result is unspecified."
(vlist-fold (lambda (item x)
(proc item))
(if #f #f)
vlist))
;;;
;;; Hash Lists, aka. `VHash'.
;;;
;; Assume keys K1 and K2, H = hash(K1) = hash(K2), and two values V1 and V2
;; associated with K1 and K2, respectively. The resulting layout is a
;; follows:
;;
;; ,--------------------.
;; | ,-> (K1 . V1) ---. |
;; | | | |
;; | | (K2 . V2) <--' |
;; | | |
;; +-|------------------+
;; | | |
;; | | |
;; | `-- O <---------------H
;; | |
;; `--------------------'
;;
;; The bottom part is the "hash table" part of the vhash, as returned by
;; `block-hash-table'; the other half is the data part. O is the offset of
;; the first value associated with a key that hashes to H in the data part.
;; The (K1 . V1) pair has a "hidden" link to the (K2 . V2) pair; hiding the
;; link is handled by `block-ref'.
;; This API potentially requires users to repeat which hash function and which
;; equality predicate to use. This can lead to unpredictable results if they
;; are used in consistenly, e.g., between `vhash-cons' and `vhash-assoc', which
;; is undesirable, as argued in http://savannah.gnu.org/bugs/?22159 . OTOH, two
;; arguments can be made in favor of this API:
;;
;; - It's consistent with how alists are handled in SRFI-1.
;;
;; - In practice, users will probably consistenly use either the `q', the `v',
;; or the plain variant (`vlist-cons' and `vlist-assoc' without any optional
;; argument), i.e., they will rarely explicitly pass a hash function or
;; equality predicate.
(define (vhash? obj)
"Return true if @var{obj} is a hash list."
(and (vlist? obj)
(let ((base (vlist-base obj)))
(and base
(vector? (block-hash-table base))))))
(define* (vhash-cons key value vhash #:optional (hash hash))
"Return a new hash list based on @var{vhash} where @var{key} is associated
with @var{value}. Use @var{hash} to compute @var{key}'s hash."
(let* ((key+value (cons key value))
(entry (cons key+value #f))
(vlist (block-cons entry vhash #t))
(base (vlist-base vlist))
(khash (hash key (block-size base))))
(let ((o (block-hash-table-ref base khash)))
(if o (set-cdr! entry o)))
(block-hash-table-set! base khash
(vlist-offset vlist))
vlist))
(define vhash-consq (cut vhash-cons <> <> <> hashq))
(define vhash-consv (cut vhash-cons <> <> <> hashv))
(define-syntax make-vhash-assoc
;; This hack to make sure `vhash-assq' gets to use the `eq?' instruction
;; instead of calling the `eq?' subr.
(syntax-rules ()
((_ key vhash equal? hash)
(begin
(define khash
(let ((size (block-size (vlist-base vhash))))
(and (> size 0) (hash key size))))
(let loop ((base (vlist-base vhash))
(khash khash)
(offset (and khash
(block-hash-table-ref (vlist-base vhash)
khash)))
(max-offset (vlist-offset vhash)))
(let ((answer (and offset (block-ref base offset))))
(cond ((and (pair? answer)
(<= offset max-offset)
(let ((answer-key (caar answer)))
(equal? key answer-key)))
(car answer))
((and (pair? answer) (cdr answer))
=>
(lambda (next-offset)
(loop base khash next-offset max-offset)))
(else
(let ((next-base (block-base base)))
(and next-base
(> (block-size next-base) 0)
(let* ((khash (hash key (block-size next-base)))
(offset (block-hash-table-ref next-base khash)))
(loop next-base khash offset
(block-offset base)))))))))))))
(define* (vhash-assoc key vhash #:optional (equal? equal?) (hash hash))
"Return the first key/value pair from @var{vhash} whose key is equal to
@var{key} according to the @var{equal?} equality predicate."
(make-vhash-assoc key vhash equal? hash))
(define (vhash-assq key vhash)
"Return the first key/value pair from @var{vhash} whose key is @code{eq?} to
@var{key}."
(make-vhash-assoc key vhash eq? hashq))
(define (vhash-assv key vhash)
"Return the first key/value pair from @var{vhash} whose key is @code{eqv?} to
@var{key}."
(make-vhash-assoc key vhash eqv? hashv))
(define* (vhash-delete key vhash #:optional (equal? equal?) (hash hash))
"Remove all associations from @var{vhash} with @var{key}, comparing keys
with @var{equal?}."
(vlist-fold (lambda (k+v result)
(let ((k (car k+v))
(v (cdr k+v)))
(if (equal? k key)
result
(vhash-cons k v result))))
vlist-null
vhash))
(define vhash-delq (cut vhash-delete <> <> eq? hashq))
(define vhash-delv (cut vhash-delete <> <> eqv? hashv))
(define (vhash-fold proc seed vhash)
"Fold over the key/pair elements of @var{vhash}. For each pair call
@var{proc} as @code{(@var{proc} key value result)}."
(vlist-fold (lambda (key+value result)
(proc (car key+value) (cdr key+value)
result))
seed
vhash))
(define* (alist->vhash alist #:optional (hash hash))
"Return the vhash corresponding to @var{alist}, an association list."
(fold-right (lambda (pair result)
(vhash-cons (car pair) (cdr pair) result hash))
vlist-null
alist))
;;; vlist.scm ends here