3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
4 * 2000, 2001, 2002, 2003, 2004, 2005, by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * ...they shuffled back towards the rear of the line. 'No, not at the
13 * rear!' the slave-driver shouted. 'Three files up. And stay there...
16 /* This file contains pp ("push/pop") functions that
17 * execute the opcodes that make up a perl program. A typical pp function
18 * expects to find its arguments on the stack, and usually pushes its
19 * results onto the stack, hence the 'pp' terminology. Each OP structure
20 * contains a pointer to the relevant pp_foo() function.
22 * This particular file just contains pp_sort(), which is complex
23 * enough to merit its own file! See the other pp*.c files for the rest of
28 #define PERL_IN_PP_SORT_C
32 /* looks like 'small' is reserved word for WINCE (or somesuch)*/
36 #define sv_cmp_static Perl_sv_cmp
37 #define sv_cmp_locale_static Perl_sv_cmp_locale
39 #define dSORTHINTS SV *hintsv = GvSV(gv_fetchpv("sort::hints", GV_ADDMULTI, SVt_IV))
40 #define SORTHINTS (SvIOK(hintsv) ? ((I32)SvIV(hintsv)) : 0)
43 #define SMALLSORT (200)
47 * The mergesort implementation is by Peter M. Mcilroy <pmcilroy@lucent.com>.
49 * The original code was written in conjunction with BSD Computer Software
50 * Research Group at University of California, Berkeley.
52 * See also: "Optimistic Merge Sort" (SODA '92)
54 * The integration to Perl is by John P. Linderman <jpl@research.att.com>.
56 * The code can be distributed under the same terms as Perl itself.
61 typedef char * aptr; /* pointer for arithmetic on sizes */
62 typedef SV * gptr; /* pointers in our lists */
64 /* Binary merge internal sort, with a few special mods
65 ** for the special perl environment it now finds itself in.
67 ** Things that were once options have been hotwired
68 ** to values suitable for this use. In particular, we'll always
69 ** initialize looking for natural runs, we'll always produce stable
70 ** output, and we'll always do Peter McIlroy's binary merge.
73 /* Pointer types for arithmetic and storage and convenience casts */
75 #define APTR(P) ((aptr)(P))
76 #define GPTP(P) ((gptr *)(P))
77 #define GPPP(P) ((gptr **)(P))
80 /* byte offset from pointer P to (larger) pointer Q */
81 #define BYTEOFF(P, Q) (APTR(Q) - APTR(P))
83 #define PSIZE sizeof(gptr)
85 /* If PSIZE is power of 2, make PSHIFT that power, if that helps */
88 #define PNELEM(P, Q) (BYTEOFF(P,Q) >> (PSHIFT))
89 #define PNBYTE(N) ((N) << (PSHIFT))
90 #define PINDEX(P, N) (GPTP(APTR(P) + PNBYTE(N)))
92 /* Leave optimization to compiler */
93 #define PNELEM(P, Q) (GPTP(Q) - GPTP(P))
94 #define PNBYTE(N) ((N) * (PSIZE))
95 #define PINDEX(P, N) (GPTP(P) + (N))
98 /* Pointer into other corresponding to pointer into this */
99 #define POTHER(P, THIS, OTHER) GPTP(APTR(OTHER) + BYTEOFF(THIS,P))
101 #define FROMTOUPTO(src, dst, lim) do *dst++ = *src++; while(src<lim)
104 /* Runs are identified by a pointer in the auxilliary list.
105 ** The pointer is at the start of the list,
106 ** and it points to the start of the next list.
107 ** NEXT is used as an lvalue, too.
110 #define NEXT(P) (*GPPP(P))
113 /* PTHRESH is the minimum number of pairs with the same sense to justify
114 ** checking for a run and extending it. Note that PTHRESH counts PAIRS,
115 ** not just elements, so PTHRESH == 8 means a run of 16.
120 /* RTHRESH is the number of elements in a run that must compare low
121 ** to the low element from the opposing run before we justify
122 ** doing a binary rampup instead of single stepping.
123 ** In random input, N in a row low should only happen with
124 ** probability 2^(1-N), so we can risk that we are dealing
125 ** with orderly input without paying much when we aren't.
132 ** Overview of algorithm and variables.
133 ** The array of elements at list1 will be organized into runs of length 2,
134 ** or runs of length >= 2 * PTHRESH. We only try to form long runs when
135 ** PTHRESH adjacent pairs compare in the same way, suggesting overall order.
137 ** Unless otherwise specified, pair pointers address the first of two elements.
139 ** b and b+1 are a pair that compare with sense "sense".
140 ** b is the "bottom" of adjacent pairs that might form a longer run.
142 ** p2 parallels b in the list2 array, where runs are defined by
145 ** t represents the "top" of the adjacent pairs that might extend
146 ** the run beginning at b. Usually, t addresses a pair
147 ** that compares with opposite sense from (b,b+1).
148 ** However, it may also address a singleton element at the end of list1,
149 ** or it may be equal to "last", the first element beyond list1.
151 ** r addresses the Nth pair following b. If this would be beyond t,
152 ** we back it off to t. Only when r is less than t do we consider the
153 ** run long enough to consider checking.
155 ** q addresses a pair such that the pairs at b through q already form a run.
156 ** Often, q will equal b, indicating we only are sure of the pair itself.
157 ** However, a search on the previous cycle may have revealed a longer run,
158 ** so q may be greater than b.
160 ** p is used to work back from a candidate r, trying to reach q,
161 ** which would mean b through r would be a run. If we discover such a run,
162 ** we start q at r and try to push it further towards t.
163 ** If b through r is NOT a run, we detect the wrong order at (p-1,p).
164 ** In any event, after the check (if any), we have two main cases.
166 ** 1) Short run. b <= q < p <= r <= t.
167 ** b through q is a run (perhaps trivial)
168 ** q through p are uninteresting pairs
169 ** p through r is a run
171 ** 2) Long run. b < r <= q < t.
172 ** b through q is a run (of length >= 2 * PTHRESH)
174 ** Note that degenerate cases are not only possible, but likely.
175 ** For example, if the pair following b compares with opposite sense,
176 ** then b == q < p == r == t.
181 dynprep(pTHX_ gptr *list1, gptr *list2, size_t nmemb, SVCOMPARE_t cmp)
184 register gptr *b, *p, *q, *t, *p2;
185 register gptr c, *last, *r;
190 last = PINDEX(b, nmemb);
191 sense = (cmp(aTHX_ *b, *(b+1)) > 0);
192 for (p2 = list2; b < last; ) {
193 /* We just started, or just reversed sense.
194 ** Set t at end of pairs with the prevailing sense.
196 for (p = b+2, t = p; ++p < last; t = ++p) {
197 if ((cmp(aTHX_ *t, *p) > 0) != sense) break;
200 /* Having laid out the playing field, look for long runs */
202 p = r = b + (2 * PTHRESH);
203 if (r >= t) p = r = t; /* too short to care about */
205 while (((cmp(aTHX_ *(p-1), *p) > 0) == sense) &&
208 /* b through r is a (long) run.
209 ** Extend it as far as possible.
212 while (((p += 2) < t) &&
213 ((cmp(aTHX_ *(p-1), *p) > 0) == sense)) q = p;
214 r = p = q + 2; /* no simple pairs, no after-run */
217 if (q > b) { /* run of greater than 2 at b */
220 /* pick up singleton, if possible */
223 ((cmp(aTHX_ *(p-1), *p) > 0) == sense))
224 savep = r = p = q = last;
225 p2 = NEXT(p2) = p2 + (p - b); ++runs;
226 if (sense) while (b < --p) {
233 while (q < p) { /* simple pairs */
234 p2 = NEXT(p2) = p2 + 2; ++runs;
241 if (((b = p) == t) && ((t+1) == last)) {
242 NEXT(p2) = p2 + 1; ++runs;
253 /* The original merge sort, in use since 5.7, was as fast as, or faster than,
254 * qsort on many platforms, but slower than qsort, conspicuously so,
255 * on others. The most likely explanation was platform-specific
256 * differences in cache sizes and relative speeds.
258 * The quicksort divide-and-conquer algorithm guarantees that, as the
259 * problem is subdivided into smaller and smaller parts, the parts
260 * fit into smaller (and faster) caches. So it doesn't matter how
261 * many levels of cache exist, quicksort will "find" them, and,
262 * as long as smaller is faster, take advantage of them.
264 * By contrast, consider how the original mergesort algorithm worked.
265 * Suppose we have five runs (each typically of length 2 after dynprep).
274 * Adjacent pairs are merged in "grand sweeps" through the input.
275 * This means, on pass 1, the records in runs 1 and 2 aren't revisited until
276 * runs 3 and 4 are merged and the runs from run 5 have been copied.
277 * The only cache that matters is one large enough to hold *all* the input.
278 * On some platforms, this may be many times slower than smaller caches.
280 * The following pseudo-code uses the same basic merge algorithm,
281 * but in a divide-and-conquer way.
283 * # merge $runs runs at offset $offset of list $list1 into $list2.
284 * # all unmerged runs ($runs == 1) originate in list $base.
286 * my ($offset, $runs, $base, $list1, $list2) = @_;
289 * if ($list1 is $base) copy run to $list2
290 * return offset of end of list (or copy)
292 * $off2 = mgsort2($offset, $runs-($runs/2), $base, $list2, $list1)
293 * mgsort2($off2, $runs/2, $base, $list2, $list1)
294 * merge the adjacent runs at $offset of $list1 into $list2
295 * return the offset of the end of the merged runs
298 * mgsort2(0, $runs, $base, $aux, $base);
300 * For our 5 runs, the tree of calls looks like
309 * and the corresponding activity looks like
311 * copy runs 1 and 2 from base to aux
312 * merge runs 1 and 2 from aux to base
313 * (run 3 is where it belongs, no copy needed)
314 * merge runs 12 and 3 from base to aux
315 * (runs 4 and 5 are where they belong, no copy needed)
316 * merge runs 4 and 5 from base to aux
317 * merge runs 123 and 45 from aux to base
319 * Note that we merge runs 1 and 2 immediately after copying them,
320 * while they are still likely to be in fast cache. Similarly,
321 * run 3 is merged with run 12 while it still may be lingering in cache.
322 * This implementation should therefore enjoy much of the cache-friendly
323 * behavior that quicksort does. In addition, it does less copying
324 * than the original mergesort implementation (only runs 1 and 2 are copied)
325 * and the "balancing" of merges is better (merged runs comprise more nearly
326 * equal numbers of original runs).
328 * The actual cache-friendly implementation will use a pseudo-stack
329 * to avoid recursion, and will unroll processing of runs of length 2,
330 * but it is otherwise similar to the recursive implementation.
334 IV offset; /* offset of 1st of 2 runs at this level */
335 IV runs; /* how many runs must be combined into 1 */
336 } off_runs; /* pseudo-stack element */
340 cmp_desc(pTHX_ gptr a, gptr b)
342 return -PL_sort_RealCmp(aTHX_ a, b);
346 S_mergesortsv(pTHX_ gptr *base, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
350 register gptr *f1, *f2, *t, *b, *p;
354 gptr small[SMALLSORT];
356 off_runs stack[60], *stackp;
357 SVCOMPARE_t savecmp = 0;
359 if (nmemb <= 1) return; /* sorted trivially */
362 savecmp = PL_sort_RealCmp; /* Save current comparison routine, if any */
363 PL_sort_RealCmp = cmp; /* Put comparison routine where cmp_desc can find it */
367 if (nmemb <= SMALLSORT) aux = small; /* use stack for aux array */
368 else { Newx(aux,nmemb,gptr); } /* allocate auxilliary array */
371 stackp->runs = dynprep(aTHX_ base, aux, nmemb, cmp);
372 stackp->offset = offset = 0;
373 which[0] = which[2] = base;
376 /* On levels where both runs have be constructed (stackp->runs == 0),
377 * merge them, and note the offset of their end, in case the offset
378 * is needed at the next level up. Hop up a level, and,
379 * as long as stackp->runs is 0, keep merging.
381 IV runs = stackp->runs;
385 list1 = which[iwhich]; /* area where runs are now */
386 list2 = which[++iwhich]; /* area for merged runs */
388 register gptr *l1, *l2, *tp2;
389 offset = stackp->offset;
390 f1 = p1 = list1 + offset; /* start of first run */
391 p = tp2 = list2 + offset; /* where merged run will go */
392 t = NEXT(p); /* where first run ends */
393 f2 = l1 = POTHER(t, list2, list1); /* ... on the other side */
394 t = NEXT(t); /* where second runs ends */
395 l2 = POTHER(t, list2, list1); /* ... on the other side */
396 offset = PNELEM(list2, t);
397 while (f1 < l1 && f2 < l2) {
398 /* If head 1 is larger than head 2, find ALL the elements
399 ** in list 2 strictly less than head1, write them all,
400 ** then head 1. Then compare the new heads, and repeat,
401 ** until one or both lists are exhausted.
403 ** In all comparisons (after establishing
404 ** which head to merge) the item to merge
405 ** (at pointer q) is the first operand of
406 ** the comparison. When we want to know
407 ** if "q is strictly less than the other",
410 ** because stability demands that we treat equality
411 ** as high when q comes from l2, and as low when
412 ** q was from l1. So we ask the question by doing
413 ** cmp(q, other) <= sense
414 ** and make sense == 0 when equality should look low,
415 ** and -1 when equality should look high.
419 if (cmp(aTHX_ *f1, *f2) <= 0) {
420 q = f2; b = f1; t = l1;
423 q = f1; b = f2; t = l2;
430 ** Leave t at something strictly
431 ** greater than q (or at the end of the list),
432 ** and b at something strictly less than q.
434 for (i = 1, run = 0 ;;) {
435 if ((p = PINDEX(b, i)) >= t) {
437 if (((p = PINDEX(t, -1)) > b) &&
438 (cmp(aTHX_ *q, *p) <= sense))
442 } else if (cmp(aTHX_ *q, *p) <= sense) {
446 if (++run >= RTHRESH) i += i;
450 /* q is known to follow b and must be inserted before t.
451 ** Increment b, so the range of possibilities is [b,t).
452 ** Round binary split down, to favor early appearance.
453 ** Adjust b and t until q belongs just before t.
458 p = PINDEX(b, (PNELEM(b, t) - 1) / 2);
459 if (cmp(aTHX_ *q, *p) <= sense) {
465 /* Copy all the strictly low elements */
468 FROMTOUPTO(f2, tp2, t);
471 FROMTOUPTO(f1, tp2, t);
477 /* Run out remaining list */
479 if (f2 < l2) FROMTOUPTO(f2, tp2, l2);
480 } else FROMTOUPTO(f1, tp2, l1);
481 p1 = NEXT(p1) = POTHER(tp2, list2, list1);
483 if (--level == 0) goto done;
485 t = list1; list1 = list2; list2 = t; /* swap lists */
486 } while ((runs = stackp->runs) == 0);
490 stackp->runs = 0; /* current run will finish level */
491 /* While there are more than 2 runs remaining,
492 * turn them into exactly 2 runs (at the "other" level),
493 * each made up of approximately half the runs.
494 * Stack the second half for later processing,
495 * and set about producing the first half now.
500 stackp->offset = offset;
501 runs -= stackp->runs = runs / 2;
503 /* We must construct a single run from 1 or 2 runs.
504 * All the original runs are in which[0] == base.
505 * The run we construct must end up in which[level&1].
509 /* Constructing a single run from a single run.
510 * If it's where it belongs already, there's nothing to do.
511 * Otherwise, copy it to where it belongs.
512 * A run of 1 is either a singleton at level 0,
513 * or the second half of a split 3. In neither event
514 * is it necessary to set offset. It will be set by the merge
515 * that immediately follows.
517 if (iwhich) { /* Belongs in aux, currently in base */
518 f1 = b = PINDEX(base, offset); /* where list starts */
519 f2 = PINDEX(aux, offset); /* where list goes */
520 t = NEXT(f2); /* where list will end */
521 offset = PNELEM(aux, t); /* offset thereof */
522 t = PINDEX(base, offset); /* where it currently ends */
523 FROMTOUPTO(f1, f2, t); /* copy */
524 NEXT(b) = t; /* set up parallel pointer */
525 } else if (level == 0) goto done; /* single run at level 0 */
527 /* Constructing a single run from two runs.
528 * The merge code at the top will do that.
529 * We need only make sure the two runs are in the "other" array,
530 * so they'll end up in the correct array after the merge.
534 stackp->offset = offset;
535 stackp->runs = 0; /* take care of both runs, trigger merge */
536 if (!iwhich) { /* Merged runs belong in aux, copy 1st */
537 f1 = b = PINDEX(base, offset); /* where first run starts */
538 f2 = PINDEX(aux, offset); /* where it will be copied */
539 t = NEXT(f2); /* where first run will end */
540 offset = PNELEM(aux, t); /* offset thereof */
541 p = PINDEX(base, offset); /* end of first run */
542 t = NEXT(t); /* where second run will end */
543 t = PINDEX(base, PNELEM(aux, t)); /* where it now ends */
544 FROMTOUPTO(f1, f2, t); /* copy both runs */
545 NEXT(b) = p; /* paralled pointer for 1st */
546 NEXT(p) = t; /* ... and for second */
551 if (aux != small) Safefree(aux); /* free iff allocated */
553 PL_sort_RealCmp = savecmp; /* Restore current comparison routine, if any */
559 * The quicksort implementation was derived from source code contributed
562 * NOTE: this code was derived from Tom Horsley's qsort replacement
563 * and should not be confused with the original code.
566 /* Copyright (C) Tom Horsley, 1997. All rights reserved.
568 Permission granted to distribute under the same terms as perl which are
571 This program is free software; you can redistribute it and/or modify
572 it under the terms of either:
574 a) the GNU General Public License as published by the Free
575 Software Foundation; either version 1, or (at your option) any
578 b) the "Artistic License" which comes with this Kit.
580 Details on the perl license can be found in the perl source code which
581 may be located via the www.perl.com web page.
583 This is the most wonderfulest possible qsort I can come up with (and
584 still be mostly portable) My (limited) tests indicate it consistently
585 does about 20% fewer calls to compare than does the qsort in the Visual
586 C++ library, other vendors may vary.
588 Some of the ideas in here can be found in "Algorithms" by Sedgewick,
589 others I invented myself (or more likely re-invented since they seemed
590 pretty obvious once I watched the algorithm operate for a while).
592 Most of this code was written while watching the Marlins sweep the Giants
593 in the 1997 National League Playoffs - no Braves fans allowed to use this
594 code (just kidding :-).
596 I realize that if I wanted to be true to the perl tradition, the only
597 comment in this file would be something like:
599 ...they shuffled back towards the rear of the line. 'No, not at the
600 rear!' the slave-driver shouted. 'Three files up. And stay there...
602 However, I really needed to violate that tradition just so I could keep
603 track of what happens myself, not to mention some poor fool trying to
604 understand this years from now :-).
607 /* ********************************************************** Configuration */
609 #ifndef QSORT_ORDER_GUESS
610 #define QSORT_ORDER_GUESS 2 /* Select doubling version of the netBSD trick */
613 /* QSORT_MAX_STACK is the largest number of partitions that can be stacked up for
614 future processing - a good max upper bound is log base 2 of memory size
615 (32 on 32 bit machines, 64 on 64 bit machines, etc). In reality can
616 safely be smaller than that since the program is taking up some space and
617 most operating systems only let you grab some subset of contiguous
618 memory (not to mention that you are normally sorting data larger than
619 1 byte element size :-).
621 #ifndef QSORT_MAX_STACK
622 #define QSORT_MAX_STACK 32
625 /* QSORT_BREAK_EVEN is the size of the largest partition we should insertion sort.
626 Anything bigger and we use qsort. If you make this too small, the qsort
627 will probably break (or become less efficient), because it doesn't expect
628 the middle element of a partition to be the same as the right or left -
629 you have been warned).
631 #ifndef QSORT_BREAK_EVEN
632 #define QSORT_BREAK_EVEN 6
635 /* QSORT_PLAY_SAFE is the size of the largest partition we're willing
636 to go quadratic on. We innoculate larger partitions against
637 quadratic behavior by shuffling them before sorting. This is not
638 an absolute guarantee of non-quadratic behavior, but it would take
639 staggeringly bad luck to pick extreme elements as the pivot
640 from randomized data.
642 #ifndef QSORT_PLAY_SAFE
643 #define QSORT_PLAY_SAFE 255
646 /* ************************************************************* Data Types */
648 /* hold left and right index values of a partition waiting to be sorted (the
649 partition includes both left and right - right is NOT one past the end or
652 struct partition_stack_entry {
655 #ifdef QSORT_ORDER_GUESS
656 int qsort_break_even;
660 /* ******************************************************* Shorthand Macros */
662 /* Note that these macros will be used from inside the qsort function where
663 we happen to know that the variable 'elt_size' contains the size of an
664 array element and the variable 'temp' points to enough space to hold a
665 temp element and the variable 'array' points to the array being sorted
666 and 'compare' is the pointer to the compare routine.
668 Also note that there are very many highly architecture specific ways
669 these might be sped up, but this is simply the most generally portable
670 code I could think of.
673 /* Return < 0 == 0 or > 0 as the value of elt1 is < elt2, == elt2, > elt2
675 #define qsort_cmp(elt1, elt2) \
676 ((*compare)(aTHX_ array[elt1], array[elt2]))
678 #ifdef QSORT_ORDER_GUESS
679 #define QSORT_NOTICE_SWAP swapped++;
681 #define QSORT_NOTICE_SWAP
684 /* swaps contents of array elements elt1, elt2.
686 #define qsort_swap(elt1, elt2) \
689 temp = array[elt1]; \
690 array[elt1] = array[elt2]; \
691 array[elt2] = temp; \
694 /* rotate contents of elt1, elt2, elt3 such that elt1 gets elt2, elt2 gets
695 elt3 and elt3 gets elt1.
697 #define qsort_rotate(elt1, elt2, elt3) \
700 temp = array[elt1]; \
701 array[elt1] = array[elt2]; \
702 array[elt2] = array[elt3]; \
703 array[elt3] = temp; \
706 /* ************************************************************ Debug stuff */
713 return; /* good place to set a breakpoint */
716 #define qsort_assert(t) (void)( (t) || (break_here(), 0) )
723 int (*compare)(const void * elt1, const void * elt2),
724 int pc_left, int pc_right, int u_left, int u_right)
728 qsort_assert(pc_left <= pc_right);
729 qsort_assert(u_right < pc_left);
730 qsort_assert(pc_right < u_left);
731 for (i = u_right + 1; i < pc_left; ++i) {
732 qsort_assert(qsort_cmp(i, pc_left) < 0);
734 for (i = pc_left; i < pc_right; ++i) {
735 qsort_assert(qsort_cmp(i, pc_right) == 0);
737 for (i = pc_right + 1; i < u_left; ++i) {
738 qsort_assert(qsort_cmp(pc_right, i) < 0);
742 #define qsort_all_asserts(PC_LEFT, PC_RIGHT, U_LEFT, U_RIGHT) \
743 doqsort_all_asserts(array, num_elts, elt_size, compare, \
744 PC_LEFT, PC_RIGHT, U_LEFT, U_RIGHT)
748 #define qsort_assert(t) ((void)0)
750 #define qsort_all_asserts(PC_LEFT, PC_RIGHT, U_LEFT, U_RIGHT) ((void)0)
754 /* ****************************************************************** qsort */
756 STATIC void /* the standard unstable (u) quicksort (qsort) */
757 S_qsortsvu(pTHX_ SV ** array, size_t num_elts, SVCOMPARE_t compare)
761 struct partition_stack_entry partition_stack[QSORT_MAX_STACK];
762 int next_stack_entry = 0;
766 #ifdef QSORT_ORDER_GUESS
767 int qsort_break_even;
771 /* Make sure we actually have work to do.
777 /* Innoculate large partitions against quadratic behavior */
778 if (num_elts > QSORT_PLAY_SAFE) {
780 register SV ** const q = array;
781 for (n = num_elts; n > 1; ) {
782 register const size_t j = (size_t)(n-- * Drand01());
789 /* Setup the initial partition definition and fall into the sorting loop
792 part_right = (int)(num_elts - 1);
793 #ifdef QSORT_ORDER_GUESS
794 qsort_break_even = QSORT_BREAK_EVEN;
796 #define qsort_break_even QSORT_BREAK_EVEN
799 if ((part_right - part_left) >= qsort_break_even) {
800 /* OK, this is gonna get hairy, so lets try to document all the
801 concepts and abbreviations and variables and what they keep
804 pc: pivot chunk - the set of array elements we accumulate in the
805 middle of the partition, all equal in value to the original
806 pivot element selected. The pc is defined by:
808 pc_left - the leftmost array index of the pc
809 pc_right - the rightmost array index of the pc
811 we start with pc_left == pc_right and only one element
812 in the pivot chunk (but it can grow during the scan).
814 u: uncompared elements - the set of elements in the partition
815 we have not yet compared to the pivot value. There are two
816 uncompared sets during the scan - one to the left of the pc
817 and one to the right.
819 u_right - the rightmost index of the left side's uncompared set
820 u_left - the leftmost index of the right side's uncompared set
822 The leftmost index of the left sides's uncompared set
823 doesn't need its own variable because it is always defined
824 by the leftmost edge of the whole partition (part_left). The
825 same goes for the rightmost edge of the right partition
828 We know there are no uncompared elements on the left once we
829 get u_right < part_left and no uncompared elements on the
830 right once u_left > part_right. When both these conditions
831 are met, we have completed the scan of the partition.
833 Any elements which are between the pivot chunk and the
834 uncompared elements should be less than the pivot value on
835 the left side and greater than the pivot value on the right
836 side (in fact, the goal of the whole algorithm is to arrange
837 for that to be true and make the groups of less-than and
838 greater-then elements into new partitions to sort again).
840 As you marvel at the complexity of the code and wonder why it
841 has to be so confusing. Consider some of the things this level
844 Once I do a compare, I squeeze every ounce of juice out of it. I
845 never do compare calls I don't have to do, and I certainly never
848 I also never swap any elements unless I can prove there is a
849 good reason. Many sort algorithms will swap a known value with
850 an uncompared value just to get things in the right place (or
851 avoid complexity :-), but that uncompared value, once it gets
852 compared, may then have to be swapped again. A lot of the
853 complexity of this code is due to the fact that it never swaps
854 anything except compared values, and it only swaps them when the
855 compare shows they are out of position.
857 int pc_left, pc_right;
862 pc_left = ((part_left + part_right) / 2);
864 u_right = pc_left - 1;
865 u_left = pc_right + 1;
867 /* Qsort works best when the pivot value is also the median value
868 in the partition (unfortunately you can't find the median value
869 without first sorting :-), so to give the algorithm a helping
870 hand, we pick 3 elements and sort them and use the median value
871 of that tiny set as the pivot value.
873 Some versions of qsort like to use the left middle and right as
874 the 3 elements to sort so they can insure the ends of the
875 partition will contain values which will stop the scan in the
876 compare loop, but when you have to call an arbitrarily complex
877 routine to do a compare, its really better to just keep track of
878 array index values to know when you hit the edge of the
879 partition and avoid the extra compare. An even better reason to
880 avoid using a compare call is the fact that you can drop off the
881 edge of the array if someone foolishly provides you with an
882 unstable compare function that doesn't always provide consistent
885 So, since it is simpler for us to compare the three adjacent
886 elements in the middle of the partition, those are the ones we
887 pick here (conveniently pointed at by u_right, pc_left, and
888 u_left). The values of the left, center, and right elements
889 are refered to as l c and r in the following comments.
892 #ifdef QSORT_ORDER_GUESS
895 s = qsort_cmp(u_right, pc_left);
898 s = qsort_cmp(pc_left, u_left);
899 /* if l < c, c < r - already in order - nothing to do */
901 /* l < c, c == r - already in order, pc grows */
903 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
905 /* l < c, c > r - need to know more */
906 s = qsort_cmp(u_right, u_left);
908 /* l < c, c > r, l < r - swap c & r to get ordered */
909 qsort_swap(pc_left, u_left);
910 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
912 /* l < c, c > r, l == r - swap c&r, grow pc */
913 qsort_swap(pc_left, u_left);
915 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
917 /* l < c, c > r, l > r - make lcr into rlc to get ordered */
918 qsort_rotate(pc_left, u_right, u_left);
919 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
924 s = qsort_cmp(pc_left, u_left);
926 /* l == c, c < r - already in order, grow pc */
928 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
930 /* l == c, c == r - already in order, grow pc both ways */
933 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
935 /* l == c, c > r - swap l & r, grow pc */
936 qsort_swap(u_right, u_left);
938 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
942 s = qsort_cmp(pc_left, u_left);
944 /* l > c, c < r - need to know more */
945 s = qsort_cmp(u_right, u_left);
947 /* l > c, c < r, l < r - swap l & c to get ordered */
948 qsort_swap(u_right, pc_left);
949 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
951 /* l > c, c < r, l == r - swap l & c, grow pc */
952 qsort_swap(u_right, pc_left);
954 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
956 /* l > c, c < r, l > r - rotate lcr into crl to order */
957 qsort_rotate(u_right, pc_left, u_left);
958 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
961 /* l > c, c == r - swap ends, grow pc */
962 qsort_swap(u_right, u_left);
964 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
966 /* l > c, c > r - swap ends to get in order */
967 qsort_swap(u_right, u_left);
968 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
971 /* We now know the 3 middle elements have been compared and
972 arranged in the desired order, so we can shrink the uncompared
977 qsort_all_asserts(pc_left, pc_right, u_left, u_right);
979 /* The above massive nested if was the simple part :-). We now have
980 the middle 3 elements ordered and we need to scan through the
981 uncompared sets on either side, swapping elements that are on
982 the wrong side or simply shuffling equal elements around to get
983 all equal elements into the pivot chunk.
987 int still_work_on_left;
988 int still_work_on_right;
990 /* Scan the uncompared values on the left. If I find a value
991 equal to the pivot value, move it over so it is adjacent to
992 the pivot chunk and expand the pivot chunk. If I find a value
993 less than the pivot value, then just leave it - its already
994 on the correct side of the partition. If I find a greater
995 value, then stop the scan.
997 while ((still_work_on_left = (u_right >= part_left))) {
998 s = qsort_cmp(u_right, pc_left);
1001 } else if (s == 0) {
1003 if (pc_left != u_right) {
1004 qsort_swap(u_right, pc_left);
1010 qsort_assert(u_right < pc_left);
1011 qsort_assert(pc_left <= pc_right);
1012 qsort_assert(qsort_cmp(u_right + 1, pc_left) <= 0);
1013 qsort_assert(qsort_cmp(pc_left, pc_right) == 0);
1016 /* Do a mirror image scan of uncompared values on the right
1018 while ((still_work_on_right = (u_left <= part_right))) {
1019 s = qsort_cmp(pc_right, u_left);
1022 } else if (s == 0) {
1024 if (pc_right != u_left) {
1025 qsort_swap(pc_right, u_left);
1031 qsort_assert(u_left > pc_right);
1032 qsort_assert(pc_left <= pc_right);
1033 qsort_assert(qsort_cmp(pc_right, u_left - 1) <= 0);
1034 qsort_assert(qsort_cmp(pc_left, pc_right) == 0);
1037 if (still_work_on_left) {
1038 /* I know I have a value on the left side which needs to be
1039 on the right side, but I need to know more to decide
1040 exactly the best thing to do with it.
1042 if (still_work_on_right) {
1043 /* I know I have values on both side which are out of
1044 position. This is a big win because I kill two birds
1045 with one swap (so to speak). I can advance the
1046 uncompared pointers on both sides after swapping both
1047 of them into the right place.
1049 qsort_swap(u_right, u_left);
1052 qsort_all_asserts(pc_left, pc_right, u_left, u_right);
1054 /* I have an out of position value on the left, but the
1055 right is fully scanned, so I "slide" the pivot chunk
1056 and any less-than values left one to make room for the
1057 greater value over on the right. If the out of position
1058 value is immediately adjacent to the pivot chunk (there
1059 are no less-than values), I can do that with a swap,
1060 otherwise, I have to rotate one of the less than values
1061 into the former position of the out of position value
1062 and the right end of the pivot chunk into the left end
1066 if (pc_left == u_right) {
1067 qsort_swap(u_right, pc_right);
1068 qsort_all_asserts(pc_left, pc_right-1, u_left, u_right-1);
1070 qsort_rotate(u_right, pc_left, pc_right);
1071 qsort_all_asserts(pc_left, pc_right-1, u_left, u_right-1);
1076 } else if (still_work_on_right) {
1077 /* Mirror image of complex case above: I have an out of
1078 position value on the right, but the left is fully
1079 scanned, so I need to shuffle things around to make room
1080 for the right value on the left.
1083 if (pc_right == u_left) {
1084 qsort_swap(u_left, pc_left);
1085 qsort_all_asserts(pc_left+1, pc_right, u_left+1, u_right);
1087 qsort_rotate(pc_right, pc_left, u_left);
1088 qsort_all_asserts(pc_left+1, pc_right, u_left+1, u_right);
1093 /* No more scanning required on either side of partition,
1094 break out of loop and figure out next set of partitions
1100 /* The elements in the pivot chunk are now in the right place. They
1101 will never move or be compared again. All I have to do is decide
1102 what to do with the stuff to the left and right of the pivot
1105 Notes on the QSORT_ORDER_GUESS ifdef code:
1107 1. If I just built these partitions without swapping any (or
1108 very many) elements, there is a chance that the elements are
1109 already ordered properly (being properly ordered will
1110 certainly result in no swapping, but the converse can't be
1113 2. A (properly written) insertion sort will run faster on
1114 already ordered data than qsort will.
1116 3. Perhaps there is some way to make a good guess about
1117 switching to an insertion sort earlier than partition size 6
1118 (for instance - we could save the partition size on the stack
1119 and increase the size each time we find we didn't swap, thus
1120 switching to insertion sort earlier for partitions with a
1121 history of not swapping).
1123 4. Naturally, if I just switch right away, it will make
1124 artificial benchmarks with pure ascending (or descending)
1125 data look really good, but is that a good reason in general?
1129 #ifdef QSORT_ORDER_GUESS
1131 #if QSORT_ORDER_GUESS == 1
1132 qsort_break_even = (part_right - part_left) + 1;
1134 #if QSORT_ORDER_GUESS == 2
1135 qsort_break_even *= 2;
1137 #if QSORT_ORDER_GUESS == 3
1138 const int prev_break = qsort_break_even;
1139 qsort_break_even *= qsort_break_even;
1140 if (qsort_break_even < prev_break) {
1141 qsort_break_even = (part_right - part_left) + 1;
1145 qsort_break_even = QSORT_BREAK_EVEN;
1149 if (part_left < pc_left) {
1150 /* There are elements on the left which need more processing.
1151 Check the right as well before deciding what to do.
1153 if (pc_right < part_right) {
1154 /* We have two partitions to be sorted. Stack the biggest one
1155 and process the smallest one on the next iteration. This
1156 minimizes the stack height by insuring that any additional
1157 stack entries must come from the smallest partition which
1158 (because it is smallest) will have the fewest
1159 opportunities to generate additional stack entries.
1161 if ((part_right - pc_right) > (pc_left - part_left)) {
1162 /* stack the right partition, process the left */
1163 partition_stack[next_stack_entry].left = pc_right + 1;
1164 partition_stack[next_stack_entry].right = part_right;
1165 #ifdef QSORT_ORDER_GUESS
1166 partition_stack[next_stack_entry].qsort_break_even = qsort_break_even;
1168 part_right = pc_left - 1;
1170 /* stack the left partition, process the right */
1171 partition_stack[next_stack_entry].left = part_left;
1172 partition_stack[next_stack_entry].right = pc_left - 1;
1173 #ifdef QSORT_ORDER_GUESS
1174 partition_stack[next_stack_entry].qsort_break_even = qsort_break_even;
1176 part_left = pc_right + 1;
1178 qsort_assert(next_stack_entry < QSORT_MAX_STACK);
1181 /* The elements on the left are the only remaining elements
1182 that need sorting, arrange for them to be processed as the
1185 part_right = pc_left - 1;
1187 } else if (pc_right < part_right) {
1188 /* There is only one chunk on the right to be sorted, make it
1189 the new partition and loop back around.
1191 part_left = pc_right + 1;
1193 /* This whole partition wound up in the pivot chunk, so
1194 we need to get a new partition off the stack.
1196 if (next_stack_entry == 0) {
1197 /* the stack is empty - we are done */
1201 part_left = partition_stack[next_stack_entry].left;
1202 part_right = partition_stack[next_stack_entry].right;
1203 #ifdef QSORT_ORDER_GUESS
1204 qsort_break_even = partition_stack[next_stack_entry].qsort_break_even;
1208 /* This partition is too small to fool with qsort complexity, just
1209 do an ordinary insertion sort to minimize overhead.
1212 /* Assume 1st element is in right place already, and start checking
1213 at 2nd element to see where it should be inserted.
1215 for (i = part_left + 1; i <= part_right; ++i) {
1217 /* Scan (backwards - just in case 'i' is already in right place)
1218 through the elements already sorted to see if the ith element
1219 belongs ahead of one of them.
1221 for (j = i - 1; j >= part_left; --j) {
1222 if (qsort_cmp(i, j) >= 0) {
1223 /* i belongs right after j
1230 /* Looks like we really need to move some things
1234 for (k = i - 1; k >= j; --k)
1235 array[k + 1] = array[k];
1240 /* That partition is now sorted, grab the next one, or get out
1241 of the loop if there aren't any more.
1244 if (next_stack_entry == 0) {
1245 /* the stack is empty - we are done */
1249 part_left = partition_stack[next_stack_entry].left;
1250 part_right = partition_stack[next_stack_entry].right;
1251 #ifdef QSORT_ORDER_GUESS
1252 qsort_break_even = partition_stack[next_stack_entry].qsort_break_even;
1257 /* Believe it or not, the array is sorted at this point! */
1260 /* Stabilize what is, presumably, an otherwise unstable sort method.
1261 * We do that by allocating (or having on hand) an array of pointers
1262 * that is the same size as the original array of elements to be sorted.
1263 * We initialize this parallel array with the addresses of the original
1264 * array elements. This indirection can make you crazy.
1265 * Some pictures can help. After initializing, we have
1269 * | | --------------> | | ------> first element to be sorted
1271 * | | --------------> | | ------> second element to be sorted
1273 * | | --------------> | | ------> third element to be sorted
1277 * | | --------------> | | ------> n-1st element to be sorted
1279 * | | --------------> | | ------> n-th element to be sorted
1282 * During the sort phase, we leave the elements of list1 where they are,
1283 * and sort the pointers in the indirect array in the same order determined
1284 * by the original comparison routine on the elements pointed to.
1285 * Because we don't move the elements of list1 around through
1286 * this phase, we can break ties on elements that compare equal
1287 * using their address in the list1 array, ensuring stabilty.
1288 * This leaves us with something looking like
1292 * | | --+ +---> | | ------> first element to be sorted
1294 * | | --|-------|---> | | ------> second element to be sorted
1296 * | | --|-------+ +-> | | ------> third element to be sorted
1299 * +----+ | | | | +----+
1300 * | | ---|-+ | +--> | | ------> n-1st element to be sorted
1302 * | | ---+ +----> | | ------> n-th element to be sorted
1305 * where the i-th element of the indirect array points to the element
1306 * that should be i-th in the sorted array. After the sort phase,
1307 * we have to put the elements of list1 into the places
1308 * dictated by the indirect array.
1313 cmpindir(pTHX_ gptr a, gptr b)
1315 gptr * const ap = (gptr *)a;
1316 gptr * const bp = (gptr *)b;
1317 const I32 sense = PL_sort_RealCmp(aTHX_ *ap, *bp);
1321 return (ap > bp) ? 1 : ((ap < bp) ? -1 : 0);
1325 cmpindir_desc(pTHX_ gptr a, gptr b)
1327 gptr * const ap = (gptr *)a;
1328 gptr * const bp = (gptr *)b;
1329 const I32 sense = PL_sort_RealCmp(aTHX_ *ap, *bp);
1331 /* Reverse the default */
1334 /* But don't reverse the stability test. */
1335 return (ap > bp) ? 1 : ((ap < bp) ? -1 : 0);
1340 S_qsortsv(pTHX_ gptr *list1, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1345 if (SORTHINTS & HINT_SORT_STABLE) {
1346 register gptr **pp, *q;
1347 register size_t n, j, i;
1348 gptr *small[SMALLSORT], **indir, tmp;
1349 SVCOMPARE_t savecmp;
1350 if (nmemb <= 1) return; /* sorted trivially */
1352 /* Small arrays can use the stack, big ones must be allocated */
1353 if (nmemb <= SMALLSORT) indir = small;
1354 else { Newx(indir, nmemb, gptr *); }
1356 /* Copy pointers to original array elements into indirect array */
1357 for (n = nmemb, pp = indir, q = list1; n--; ) *pp++ = q++;
1359 savecmp = PL_sort_RealCmp; /* Save current comparison routine, if any */
1360 PL_sort_RealCmp = cmp; /* Put comparison routine where cmpindir can find it */
1362 /* sort, with indirection */
1363 S_qsortsvu(aTHX_ (gptr *)indir, nmemb,
1364 flags ? cmpindir_desc : cmpindir);
1368 for (n = nmemb; n--; ) {
1369 /* Assert A: all elements of q with index > n are already
1370 * in place. This is vacuosly true at the start, and we
1371 * put element n where it belongs below (if it wasn't
1372 * already where it belonged). Assert B: we only move
1373 * elements that aren't where they belong,
1374 * so, by A, we never tamper with elements above n.
1376 j = pp[n] - q; /* This sets j so that q[j] is
1377 * at pp[n]. *pp[j] belongs in
1378 * q[j], by construction.
1380 if (n != j) { /* all's well if n == j */
1381 tmp = q[j]; /* save what's in q[j] */
1383 q[j] = *pp[j]; /* put *pp[j] where it belongs */
1384 i = pp[j] - q; /* the index in q of the element
1386 pp[j] = q + j; /* this is ok now */
1387 } while ((j = i) != n);
1388 /* There are only finitely many (nmemb) addresses
1390 * So we must eventually revisit an index we saw before.
1391 * Suppose the first revisited index is k != n.
1392 * An index is visited because something else belongs there.
1393 * If we visit k twice, then two different elements must
1394 * belong in the same place, which cannot be.
1395 * So j must get back to n, the loop terminates,
1396 * and we put the saved element where it belongs.
1398 q[n] = tmp; /* put what belongs into
1399 * the n-th element */
1403 /* free iff allocated */
1404 if (indir != small) { Safefree(indir); }
1405 /* restore prevailing comparison routine */
1406 PL_sort_RealCmp = savecmp;
1408 SVCOMPARE_t savecmp = PL_sort_RealCmp; /* Save current comparison routine, if any */
1409 PL_sort_RealCmp = cmp; /* Put comparison routine where cmp_desc can find it */
1411 S_qsortsvu(aTHX_ list1, nmemb, cmp);
1412 /* restore prevailing comparison routine */
1413 PL_sort_RealCmp = savecmp;
1415 S_qsortsvu(aTHX_ list1, nmemb, cmp);
1420 =head1 Array Manipulation Functions
1424 Sort an array. Here is an example:
1426 sortsv(AvARRAY(av), av_len(av)+1, Perl_sv_cmp_locale);
1428 See lib/sort.pm for details about controlling the sorting algorithm.
1434 Perl_sortsv(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp)
1436 void (*sortsvp)(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1439 const I32 hints = SORTHINTS;
1440 if (hints & HINT_SORT_QUICKSORT) {
1441 sortsvp = S_qsortsv;
1444 /* The default as of 5.8.0 is mergesort */
1445 sortsvp = S_mergesortsv;
1448 sortsvp(aTHX_ array, nmemb, cmp, 0);
1453 S_sortsv_desc(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp)
1455 void (*sortsvp)(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1458 const I32 hints = SORTHINTS;
1459 if (hints & HINT_SORT_QUICKSORT) {
1460 sortsvp = S_qsortsv;
1463 /* The default as of 5.8.0 is mergesort */
1464 sortsvp = S_mergesortsv;
1467 sortsvp(aTHX_ array, nmemb, cmp, 1);
1470 #define SvNSIOK(sv) ((SvFLAGS(sv) & SVf_NOK) || ((SvFLAGS(sv) & (SVf_IOK|SVf_IVisUV)) == SVf_IOK))
1471 #define SvSIOK(sv) ((SvFLAGS(sv) & (SVf_IOK|SVf_IVisUV)) == SVf_IOK)
1472 #define SvNSIV(sv) ( SvNOK(sv) ? SvNVX(sv) : ( SvSIOK(sv) ? SvIVX(sv) : sv_2nv(sv) ) )
1476 dVAR; dSP; dMARK; dORIGMARK;
1477 register SV **p1 = ORIGMARK+1, **p2;
1478 register I32 max, i;
1484 OP* const nextop = PL_op->op_next;
1485 I32 overloading = 0;
1486 bool hasargs = FALSE;
1489 const U8 priv = PL_op->op_private;
1490 const U8 flags = PL_op->op_flags;
1491 void (*sortsvp)(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp)
1495 if (gimme != G_ARRAY) {
1502 SAVEVPTR(PL_sortcop);
1503 if (flags & OPf_STACKED) {
1504 if (flags & OPf_SPECIAL) {
1505 OP *kid = cLISTOP->op_first->op_sibling; /* pass pushmark */
1506 kid = kUNOP->op_first; /* pass rv2gv */
1507 kid = kUNOP->op_first; /* pass leave */
1508 PL_sortcop = kid->op_next;
1509 stash = CopSTASH(PL_curcop);
1512 cv = sv_2cv(*++MARK, &stash, &gv, 0);
1513 if (cv && SvPOK(cv)) {
1514 const char * const proto = SvPV_nolen_const((SV*)cv);
1515 if (proto && strEQ(proto, "$$")) {
1519 if (!(cv && CvROOT(cv))) {
1520 if (cv && CvXSUB(cv)) {
1524 SV *tmpstr = sv_newmortal();
1525 gv_efullname3(tmpstr, gv, Nullch);
1526 DIE(aTHX_ "Undefined sort subroutine \"%"SVf"\" called",
1530 DIE(aTHX_ "Undefined subroutine in sort");
1535 PL_sortcop = (OP*)cv;
1537 PL_sortcop = CvSTART(cv);
1541 PL_sortcop = Nullop;
1542 stash = CopSTASH(PL_curcop);
1545 /* optimiser converts "@a = sort @a" to "sort \@a";
1546 * in case of tied @a, pessimise: push (@a) onto stack, then assign
1547 * result back to @a at the end of this function */
1548 if (priv & OPpSORT_INPLACE) {
1549 assert( MARK+1 == SP && *SP && SvTYPE(*SP) == SVt_PVAV);
1550 (void)POPMARK; /* remove mark associated with ex-OP_AASSIGN */
1552 max = AvFILL(av) + 1;
1553 if (SvMAGICAL(av)) {
1556 for (i=0; i < max; i++) {
1557 SV **svp = av_fetch(av, i, FALSE);
1558 *SP++ = (svp) ? *svp : Nullsv;
1563 Perl_croak(aTHX_ PL_no_modify);
1566 p1 = p2 = AvARRAY(av);
1575 if (priv & OPpSORT_DESCEND) {
1576 sortsvp = S_sortsv_desc;
1579 /* shuffle stack down, removing optional initial cv (p1!=p2), plus
1580 * any nulls; also stringify or converting to integer or number as
1581 * required any args */
1582 for (i=max; i > 0 ; i--) {
1583 if ((*p1 = *p2++)) { /* Weed out nulls. */
1586 if (priv & OPpSORT_NUMERIC) {
1587 if (priv & OPpSORT_INTEGER) {
1596 if (!SvNSIOK(*p1)) {
1602 if (all_SIVs && !SvSIOK(*p1))
1611 (void)sv_2pv_flags(*p1, 0,
1612 SV_GMAGIC|SV_CONST_RETURN);
1622 AvFILLp(av) = max-1;
1629 const bool oldcatch = CATCH_GET;
1635 PUSHSTACKi(PERLSI_SORT);
1636 if (!hasargs && !is_xsub) {
1637 SAVESPTR(PL_firstgv);
1638 SAVESPTR(PL_secondgv);
1639 SAVESPTR(PL_sortstash);
1640 PL_firstgv = gv_fetchpv("a", TRUE, SVt_PV);
1641 PL_secondgv = gv_fetchpv("b", TRUE, SVt_PV);
1642 PL_sortstash = stash;
1643 SAVESPTR(GvSV(PL_firstgv));
1644 SAVESPTR(GvSV(PL_secondgv));
1647 PUSHBLOCK(cx, CXt_NULL, PL_stack_base);
1648 if (!(flags & OPf_SPECIAL)) {
1649 cx->cx_type = CXt_SUB;
1650 cx->blk_gimme = G_SCALAR;
1653 AV* const padlist = CvPADLIST(cv);
1655 if (++CvDEPTH(cv) >= 2) {
1656 PERL_STACK_OVERFLOW_CHECK();
1657 pad_push(padlist, CvDEPTH(cv));
1660 PAD_SET_CUR_NOSAVE(padlist, CvDEPTH(cv));
1663 /* This is mostly copied from pp_entersub */
1664 AV *av = (AV*)PAD_SVl(0);
1666 cx->blk_sub.savearray = GvAV(PL_defgv);
1667 GvAV(PL_defgv) = (AV*)SvREFCNT_inc(av);
1668 CX_CURPAD_SAVE(cx->blk_sub);
1669 cx->blk_sub.argarray = av;
1674 cx->cx_type |= CXp_MULTICALL;
1677 sortsvp(aTHX_ start, max,
1678 is_xsub ? S_sortcv_xsub : hasargs ? S_sortcv_stacked : S_sortcv);
1680 if (!(flags & OPf_SPECIAL)) {
1685 POPBLOCK(cx,PL_curpm);
1686 PL_stack_sp = newsp;
1688 CATCH_SET(oldcatch);
1691 MEXTEND(SP, 20); /* Can't afford stack realloc on signal. */
1692 start = sorting_av ? AvARRAY(av) : ORIGMARK+1;
1693 sortsvp(aTHX_ start, max,
1694 (priv & OPpSORT_NUMERIC)
1695 ? ( ( ( priv & OPpSORT_INTEGER) || all_SIVs)
1696 ? ( overloading ? S_amagic_i_ncmp : S_sv_i_ncmp)
1697 : ( overloading ? S_amagic_ncmp : S_sv_ncmp ) )
1698 : ( IN_LOCALE_RUNTIME
1700 ? S_amagic_cmp_locale
1701 : sv_cmp_locale_static)
1702 : ( overloading ? S_amagic_cmp : sv_cmp_static)));
1704 if (priv & OPpSORT_REVERSE) {
1705 SV **q = start+max-1;
1707 SV * const tmp = *start;
1715 else if (av && !sorting_av) {
1716 /* simulate pp_aassign of tied AV */
1717 SV** const base = ORIGMARK+1;
1718 for (i=0; i < max; i++) {
1719 base[i] = newSVsv(base[i]);
1723 for (i=0; i < max; i++) {
1724 SV * const sv = base[i];
1725 SV ** const didstore = av_store(av, i, sv);
1733 PL_stack_sp = ORIGMARK + (sorting_av ? 0 : max);
1738 S_sortcv(pTHX_ SV *a, SV *b)
1741 const I32 oldsaveix = PL_savestack_ix;
1742 const I32 oldscopeix = PL_scopestack_ix;
1744 GvSV(PL_firstgv) = a;
1745 GvSV(PL_secondgv) = b;
1746 PL_stack_sp = PL_stack_base;
1749 if (PL_stack_sp != PL_stack_base + 1)
1750 Perl_croak(aTHX_ "Sort subroutine didn't return single value");
1751 if (!SvNIOKp(*PL_stack_sp))
1752 Perl_croak(aTHX_ "Sort subroutine didn't return a numeric value");
1753 result = SvIV(*PL_stack_sp);
1754 while (PL_scopestack_ix > oldscopeix) {
1757 leave_scope(oldsaveix);
1762 S_sortcv_stacked(pTHX_ SV *a, SV *b)
1765 const I32 oldsaveix = PL_savestack_ix;
1766 const I32 oldscopeix = PL_scopestack_ix;
1768 AV * const av = GvAV(PL_defgv);
1770 if (AvMAX(av) < 1) {
1771 SV** ary = AvALLOC(av);
1772 if (AvARRAY(av) != ary) {
1773 AvMAX(av) += AvARRAY(av) - AvALLOC(av);
1774 SvPV_set(av, (char*)ary);
1776 if (AvMAX(av) < 1) {
1779 SvPV_set(av, (char*)ary);
1786 PL_stack_sp = PL_stack_base;
1789 if (PL_stack_sp != PL_stack_base + 1)
1790 Perl_croak(aTHX_ "Sort subroutine didn't return single value");
1791 if (!SvNIOKp(*PL_stack_sp))
1792 Perl_croak(aTHX_ "Sort subroutine didn't return a numeric value");
1793 result = SvIV(*PL_stack_sp);
1794 while (PL_scopestack_ix > oldscopeix) {
1797 leave_scope(oldsaveix);
1802 S_sortcv_xsub(pTHX_ SV *a, SV *b)
1805 const I32 oldsaveix = PL_savestack_ix;
1806 const I32 oldscopeix = PL_scopestack_ix;
1807 CV * const cv=(CV*)PL_sortcop;
1816 (void)(*CvXSUB(cv))(aTHX_ cv);
1817 if (PL_stack_sp != PL_stack_base + 1)
1818 Perl_croak(aTHX_ "Sort subroutine didn't return single value");
1819 if (!SvNIOKp(*PL_stack_sp))
1820 Perl_croak(aTHX_ "Sort subroutine didn't return a numeric value");
1821 result = SvIV(*PL_stack_sp);
1822 while (PL_scopestack_ix > oldscopeix) {
1825 leave_scope(oldsaveix);
1831 S_sv_ncmp(pTHX_ SV *a, SV *b)
1833 const NV nv1 = SvNSIV(a);
1834 const NV nv2 = SvNSIV(b);
1835 return nv1 < nv2 ? -1 : nv1 > nv2 ? 1 : 0;
1839 S_sv_i_ncmp(pTHX_ SV *a, SV *b)
1841 const IV iv1 = SvIV(a);
1842 const IV iv2 = SvIV(b);
1843 return iv1 < iv2 ? -1 : iv1 > iv2 ? 1 : 0;
1846 #define tryCALL_AMAGICbin(left,right,meth) \
1847 (PL_amagic_generation && (SvAMAGIC(left)||SvAMAGIC(right))) \
1848 ? amagic_call(left, right, CAT2(meth,_amg), 0) \
1852 S_amagic_ncmp(pTHX_ register SV *a, register SV *b)
1854 SV * const tmpsv = tryCALL_AMAGICbin(a,b,ncmp);
1857 const I32 i = SvIVX(tmpsv);
1863 const NV d = SvNV(tmpsv);
1869 return S_sv_ncmp(aTHX_ a, b);
1873 S_amagic_i_ncmp(pTHX_ register SV *a, register SV *b)
1875 SV * const tmpsv = tryCALL_AMAGICbin(a,b,ncmp);
1878 const I32 i = SvIVX(tmpsv);
1884 const NV d = SvNV(tmpsv);
1890 return S_sv_i_ncmp(aTHX_ a, b);
1894 S_amagic_cmp(pTHX_ register SV *str1, register SV *str2)
1896 SV * const tmpsv = tryCALL_AMAGICbin(str1,str2,scmp);
1899 const I32 i = SvIVX(tmpsv);
1905 const NV d = SvNV(tmpsv);
1911 return sv_cmp(str1, str2);
1915 S_amagic_cmp_locale(pTHX_ register SV *str1, register SV *str2)
1917 SV * const tmpsv = tryCALL_AMAGICbin(str1,str2,scmp);
1920 const I32 i = SvIVX(tmpsv);
1926 const NV d = SvNV(tmpsv);
1932 return sv_cmp_locale(str1, str2);
1937 * c-indentation-style: bsd
1939 * indent-tabs-mode: t
1942 * ex: set ts=8 sts=4 sw=4 noet: