3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
4 * 2000, 2001, 2002, 2003, 2004, 2005, 2006, 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
40 #define SMALLSORT (200)
43 /* Flags for qsortsv and mergesortsv */
45 #define SORTf_STABLE 2
49 * The mergesort implementation is by Peter M. Mcilroy <pmcilroy@lucent.com>.
51 * The original code was written in conjunction with BSD Computer Software
52 * Research Group at University of California, Berkeley.
54 * See also: "Optimistic Merge Sort" (SODA '92)
56 * The integration to Perl is by John P. Linderman <jpl@research.att.com>.
58 * The code can be distributed under the same terms as Perl itself.
63 typedef char * aptr; /* pointer for arithmetic on sizes */
64 typedef SV * gptr; /* pointers in our lists */
66 /* Binary merge internal sort, with a few special mods
67 ** for the special perl environment it now finds itself in.
69 ** Things that were once options have been hotwired
70 ** to values suitable for this use. In particular, we'll always
71 ** initialize looking for natural runs, we'll always produce stable
72 ** output, and we'll always do Peter McIlroy's binary merge.
75 /* Pointer types for arithmetic and storage and convenience casts */
77 #define APTR(P) ((aptr)(P))
78 #define GPTP(P) ((gptr *)(P))
79 #define GPPP(P) ((gptr **)(P))
82 /* byte offset from pointer P to (larger) pointer Q */
83 #define BYTEOFF(P, Q) (APTR(Q) - APTR(P))
85 #define PSIZE sizeof(gptr)
87 /* If PSIZE is power of 2, make PSHIFT that power, if that helps */
90 #define PNELEM(P, Q) (BYTEOFF(P,Q) >> (PSHIFT))
91 #define PNBYTE(N) ((N) << (PSHIFT))
92 #define PINDEX(P, N) (GPTP(APTR(P) + PNBYTE(N)))
94 /* Leave optimization to compiler */
95 #define PNELEM(P, Q) (GPTP(Q) - GPTP(P))
96 #define PNBYTE(N) ((N) * (PSIZE))
97 #define PINDEX(P, N) (GPTP(P) + (N))
100 /* Pointer into other corresponding to pointer into this */
101 #define POTHER(P, THIS, OTHER) GPTP(APTR(OTHER) + BYTEOFF(THIS,P))
103 #define FROMTOUPTO(src, dst, lim) do *dst++ = *src++; while(src<lim)
106 /* Runs are identified by a pointer in the auxilliary list.
107 ** The pointer is at the start of the list,
108 ** and it points to the start of the next list.
109 ** NEXT is used as an lvalue, too.
112 #define NEXT(P) (*GPPP(P))
115 /* PTHRESH is the minimum number of pairs with the same sense to justify
116 ** checking for a run and extending it. Note that PTHRESH counts PAIRS,
117 ** not just elements, so PTHRESH == 8 means a run of 16.
122 /* RTHRESH is the number of elements in a run that must compare low
123 ** to the low element from the opposing run before we justify
124 ** doing a binary rampup instead of single stepping.
125 ** In random input, N in a row low should only happen with
126 ** probability 2^(1-N), so we can risk that we are dealing
127 ** with orderly input without paying much when we aren't.
134 ** Overview of algorithm and variables.
135 ** The array of elements at list1 will be organized into runs of length 2,
136 ** or runs of length >= 2 * PTHRESH. We only try to form long runs when
137 ** PTHRESH adjacent pairs compare in the same way, suggesting overall order.
139 ** Unless otherwise specified, pair pointers address the first of two elements.
141 ** b and b+1 are a pair that compare with sense "sense".
142 ** b is the "bottom" of adjacent pairs that might form a longer run.
144 ** p2 parallels b in the list2 array, where runs are defined by
147 ** t represents the "top" of the adjacent pairs that might extend
148 ** the run beginning at b. Usually, t addresses a pair
149 ** that compares with opposite sense from (b,b+1).
150 ** However, it may also address a singleton element at the end of list1,
151 ** or it may be equal to "last", the first element beyond list1.
153 ** r addresses the Nth pair following b. If this would be beyond t,
154 ** we back it off to t. Only when r is less than t do we consider the
155 ** run long enough to consider checking.
157 ** q addresses a pair such that the pairs at b through q already form a run.
158 ** Often, q will equal b, indicating we only are sure of the pair itself.
159 ** However, a search on the previous cycle may have revealed a longer run,
160 ** so q may be greater than b.
162 ** p is used to work back from a candidate r, trying to reach q,
163 ** which would mean b through r would be a run. If we discover such a run,
164 ** we start q at r and try to push it further towards t.
165 ** If b through r is NOT a run, we detect the wrong order at (p-1,p).
166 ** In any event, after the check (if any), we have two main cases.
168 ** 1) Short run. b <= q < p <= r <= t.
169 ** b through q is a run (perhaps trivial)
170 ** q through p are uninteresting pairs
171 ** p through r is a run
173 ** 2) Long run. b < r <= q < t.
174 ** b through q is a run (of length >= 2 * PTHRESH)
176 ** Note that degenerate cases are not only possible, but likely.
177 ** For example, if the pair following b compares with opposite sense,
178 ** then b == q < p == r == t.
183 dynprep(pTHX_ gptr *list1, gptr *list2, size_t nmemb, SVCOMPARE_t cmp)
186 register gptr *b, *p, *q, *t, *p2;
187 register gptr c, *last, *r;
192 last = PINDEX(b, nmemb);
193 sense = (cmp(aTHX_ *b, *(b+1)) > 0);
194 for (p2 = list2; b < last; ) {
195 /* We just started, or just reversed sense.
196 ** Set t at end of pairs with the prevailing sense.
198 for (p = b+2, t = p; ++p < last; t = ++p) {
199 if ((cmp(aTHX_ *t, *p) > 0) != sense) break;
202 /* Having laid out the playing field, look for long runs */
204 p = r = b + (2 * PTHRESH);
205 if (r >= t) p = r = t; /* too short to care about */
207 while (((cmp(aTHX_ *(p-1), *p) > 0) == sense) &&
210 /* b through r is a (long) run.
211 ** Extend it as far as possible.
214 while (((p += 2) < t) &&
215 ((cmp(aTHX_ *(p-1), *p) > 0) == sense)) q = p;
216 r = p = q + 2; /* no simple pairs, no after-run */
219 if (q > b) { /* run of greater than 2 at b */
222 /* pick up singleton, if possible */
225 ((cmp(aTHX_ *(p-1), *p) > 0) == sense))
226 savep = r = p = q = last;
227 p2 = NEXT(p2) = p2 + (p - b); ++runs;
228 if (sense) while (b < --p) {
235 while (q < p) { /* simple pairs */
236 p2 = NEXT(p2) = p2 + 2; ++runs;
243 if (((b = p) == t) && ((t+1) == last)) {
244 NEXT(p2) = p2 + 1; ++runs;
255 /* The original merge sort, in use since 5.7, was as fast as, or faster than,
256 * qsort on many platforms, but slower than qsort, conspicuously so,
257 * on others. The most likely explanation was platform-specific
258 * differences in cache sizes and relative speeds.
260 * The quicksort divide-and-conquer algorithm guarantees that, as the
261 * problem is subdivided into smaller and smaller parts, the parts
262 * fit into smaller (and faster) caches. So it doesn't matter how
263 * many levels of cache exist, quicksort will "find" them, and,
264 * as long as smaller is faster, take advantage of them.
266 * By contrast, consider how the original mergesort algorithm worked.
267 * Suppose we have five runs (each typically of length 2 after dynprep).
276 * Adjacent pairs are merged in "grand sweeps" through the input.
277 * This means, on pass 1, the records in runs 1 and 2 aren't revisited until
278 * runs 3 and 4 are merged and the runs from run 5 have been copied.
279 * The only cache that matters is one large enough to hold *all* the input.
280 * On some platforms, this may be many times slower than smaller caches.
282 * The following pseudo-code uses the same basic merge algorithm,
283 * but in a divide-and-conquer way.
285 * # merge $runs runs at offset $offset of list $list1 into $list2.
286 * # all unmerged runs ($runs == 1) originate in list $base.
288 * my ($offset, $runs, $base, $list1, $list2) = @_;
291 * if ($list1 is $base) copy run to $list2
292 * return offset of end of list (or copy)
294 * $off2 = mgsort2($offset, $runs-($runs/2), $base, $list2, $list1)
295 * mgsort2($off2, $runs/2, $base, $list2, $list1)
296 * merge the adjacent runs at $offset of $list1 into $list2
297 * return the offset of the end of the merged runs
300 * mgsort2(0, $runs, $base, $aux, $base);
302 * For our 5 runs, the tree of calls looks like
311 * and the corresponding activity looks like
313 * copy runs 1 and 2 from base to aux
314 * merge runs 1 and 2 from aux to base
315 * (run 3 is where it belongs, no copy needed)
316 * merge runs 12 and 3 from base to aux
317 * (runs 4 and 5 are where they belong, no copy needed)
318 * merge runs 4 and 5 from base to aux
319 * merge runs 123 and 45 from aux to base
321 * Note that we merge runs 1 and 2 immediately after copying them,
322 * while they are still likely to be in fast cache. Similarly,
323 * run 3 is merged with run 12 while it still may be lingering in cache.
324 * This implementation should therefore enjoy much of the cache-friendly
325 * behavior that quicksort does. In addition, it does less copying
326 * than the original mergesort implementation (only runs 1 and 2 are copied)
327 * and the "balancing" of merges is better (merged runs comprise more nearly
328 * equal numbers of original runs).
330 * The actual cache-friendly implementation will use a pseudo-stack
331 * to avoid recursion, and will unroll processing of runs of length 2,
332 * but it is otherwise similar to the recursive implementation.
336 IV offset; /* offset of 1st of 2 runs at this level */
337 IV runs; /* how many runs must be combined into 1 */
338 } off_runs; /* pseudo-stack element */
342 cmp_desc(pTHX_ gptr a, gptr b)
345 return -PL_sort_RealCmp(aTHX_ a, b);
349 S_mergesortsv(pTHX_ gptr *base, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
354 register gptr *f1, *f2, *t, *b, *p;
358 gptr small[SMALLSORT];
360 off_runs stack[60], *stackp;
361 SVCOMPARE_t savecmp = 0;
363 if (nmemb <= 1) return; /* sorted trivially */
366 savecmp = PL_sort_RealCmp; /* Save current comparison routine, if any */
367 PL_sort_RealCmp = cmp; /* Put comparison routine where cmp_desc can find it */
371 if (nmemb <= SMALLSORT) aux = small; /* use stack for aux array */
372 else { Newx(aux,nmemb,gptr); } /* allocate auxilliary array */
375 stackp->runs = dynprep(aTHX_ base, aux, nmemb, cmp);
376 stackp->offset = offset = 0;
377 which[0] = which[2] = base;
380 /* On levels where both runs have be constructed (stackp->runs == 0),
381 * merge them, and note the offset of their end, in case the offset
382 * is needed at the next level up. Hop up a level, and,
383 * as long as stackp->runs is 0, keep merging.
385 IV runs = stackp->runs;
389 list1 = which[iwhich]; /* area where runs are now */
390 list2 = which[++iwhich]; /* area for merged runs */
392 register gptr *l1, *l2, *tp2;
393 offset = stackp->offset;
394 f1 = p1 = list1 + offset; /* start of first run */
395 p = tp2 = list2 + offset; /* where merged run will go */
396 t = NEXT(p); /* where first run ends */
397 f2 = l1 = POTHER(t, list2, list1); /* ... on the other side */
398 t = NEXT(t); /* where second runs ends */
399 l2 = POTHER(t, list2, list1); /* ... on the other side */
400 offset = PNELEM(list2, t);
401 while (f1 < l1 && f2 < l2) {
402 /* If head 1 is larger than head 2, find ALL the elements
403 ** in list 2 strictly less than head1, write them all,
404 ** then head 1. Then compare the new heads, and repeat,
405 ** until one or both lists are exhausted.
407 ** In all comparisons (after establishing
408 ** which head to merge) the item to merge
409 ** (at pointer q) is the first operand of
410 ** the comparison. When we want to know
411 ** if "q is strictly less than the other",
414 ** because stability demands that we treat equality
415 ** as high when q comes from l2, and as low when
416 ** q was from l1. So we ask the question by doing
417 ** cmp(q, other) <= sense
418 ** and make sense == 0 when equality should look low,
419 ** and -1 when equality should look high.
423 if (cmp(aTHX_ *f1, *f2) <= 0) {
424 q = f2; b = f1; t = l1;
427 q = f1; b = f2; t = l2;
434 ** Leave t at something strictly
435 ** greater than q (or at the end of the list),
436 ** and b at something strictly less than q.
438 for (i = 1, run = 0 ;;) {
439 if ((p = PINDEX(b, i)) >= t) {
441 if (((p = PINDEX(t, -1)) > b) &&
442 (cmp(aTHX_ *q, *p) <= sense))
446 } else if (cmp(aTHX_ *q, *p) <= sense) {
450 if (++run >= RTHRESH) i += i;
454 /* q is known to follow b and must be inserted before t.
455 ** Increment b, so the range of possibilities is [b,t).
456 ** Round binary split down, to favor early appearance.
457 ** Adjust b and t until q belongs just before t.
462 p = PINDEX(b, (PNELEM(b, t) - 1) / 2);
463 if (cmp(aTHX_ *q, *p) <= sense) {
469 /* Copy all the strictly low elements */
472 FROMTOUPTO(f2, tp2, t);
475 FROMTOUPTO(f1, tp2, t);
481 /* Run out remaining list */
483 if (f2 < l2) FROMTOUPTO(f2, tp2, l2);
484 } else FROMTOUPTO(f1, tp2, l1);
485 p1 = NEXT(p1) = POTHER(tp2, list2, list1);
487 if (--level == 0) goto done;
489 t = list1; list1 = list2; list2 = t; /* swap lists */
490 } while ((runs = stackp->runs) == 0);
494 stackp->runs = 0; /* current run will finish level */
495 /* While there are more than 2 runs remaining,
496 * turn them into exactly 2 runs (at the "other" level),
497 * each made up of approximately half the runs.
498 * Stack the second half for later processing,
499 * and set about producing the first half now.
504 stackp->offset = offset;
505 runs -= stackp->runs = runs / 2;
507 /* We must construct a single run from 1 or 2 runs.
508 * All the original runs are in which[0] == base.
509 * The run we construct must end up in which[level&1].
513 /* Constructing a single run from a single run.
514 * If it's where it belongs already, there's nothing to do.
515 * Otherwise, copy it to where it belongs.
516 * A run of 1 is either a singleton at level 0,
517 * or the second half of a split 3. In neither event
518 * is it necessary to set offset. It will be set by the merge
519 * that immediately follows.
521 if (iwhich) { /* Belongs in aux, currently in base */
522 f1 = b = PINDEX(base, offset); /* where list starts */
523 f2 = PINDEX(aux, offset); /* where list goes */
524 t = NEXT(f2); /* where list will end */
525 offset = PNELEM(aux, t); /* offset thereof */
526 t = PINDEX(base, offset); /* where it currently ends */
527 FROMTOUPTO(f1, f2, t); /* copy */
528 NEXT(b) = t; /* set up parallel pointer */
529 } else if (level == 0) goto done; /* single run at level 0 */
531 /* Constructing a single run from two runs.
532 * The merge code at the top will do that.
533 * We need only make sure the two runs are in the "other" array,
534 * so they'll end up in the correct array after the merge.
538 stackp->offset = offset;
539 stackp->runs = 0; /* take care of both runs, trigger merge */
540 if (!iwhich) { /* Merged runs belong in aux, copy 1st */
541 f1 = b = PINDEX(base, offset); /* where first run starts */
542 f2 = PINDEX(aux, offset); /* where it will be copied */
543 t = NEXT(f2); /* where first run will end */
544 offset = PNELEM(aux, t); /* offset thereof */
545 p = PINDEX(base, offset); /* end of first run */
546 t = NEXT(t); /* where second run will end */
547 t = PINDEX(base, PNELEM(aux, t)); /* where it now ends */
548 FROMTOUPTO(f1, f2, t); /* copy both runs */
549 NEXT(b) = p; /* paralled pointer for 1st */
550 NEXT(p) = t; /* ... and for second */
555 if (aux != small) Safefree(aux); /* free iff allocated */
557 PL_sort_RealCmp = savecmp; /* Restore current comparison routine, if any */
563 * The quicksort implementation was derived from source code contributed
566 * NOTE: this code was derived from Tom Horsley's qsort replacement
567 * and should not be confused with the original code.
570 /* Copyright (C) Tom Horsley, 1997. All rights reserved.
572 Permission granted to distribute under the same terms as perl which are
575 This program is free software; you can redistribute it and/or modify
576 it under the terms of either:
578 a) the GNU General Public License as published by the Free
579 Software Foundation; either version 1, or (at your option) any
582 b) the "Artistic License" which comes with this Kit.
584 Details on the perl license can be found in the perl source code which
585 may be located via the www.perl.com web page.
587 This is the most wonderfulest possible qsort I can come up with (and
588 still be mostly portable) My (limited) tests indicate it consistently
589 does about 20% fewer calls to compare than does the qsort in the Visual
590 C++ library, other vendors may vary.
592 Some of the ideas in here can be found in "Algorithms" by Sedgewick,
593 others I invented myself (or more likely re-invented since they seemed
594 pretty obvious once I watched the algorithm operate for a while).
596 Most of this code was written while watching the Marlins sweep the Giants
597 in the 1997 National League Playoffs - no Braves fans allowed to use this
598 code (just kidding :-).
600 I realize that if I wanted to be true to the perl tradition, the only
601 comment in this file would be something like:
603 ...they shuffled back towards the rear of the line. 'No, not at the
604 rear!' the slave-driver shouted. 'Three files up. And stay there...
606 However, I really needed to violate that tradition just so I could keep
607 track of what happens myself, not to mention some poor fool trying to
608 understand this years from now :-).
611 /* ********************************************************** Configuration */
613 #ifndef QSORT_ORDER_GUESS
614 #define QSORT_ORDER_GUESS 2 /* Select doubling version of the netBSD trick */
617 /* QSORT_MAX_STACK is the largest number of partitions that can be stacked up for
618 future processing - a good max upper bound is log base 2 of memory size
619 (32 on 32 bit machines, 64 on 64 bit machines, etc). In reality can
620 safely be smaller than that since the program is taking up some space and
621 most operating systems only let you grab some subset of contiguous
622 memory (not to mention that you are normally sorting data larger than
623 1 byte element size :-).
625 #ifndef QSORT_MAX_STACK
626 #define QSORT_MAX_STACK 32
629 /* QSORT_BREAK_EVEN is the size of the largest partition we should insertion sort.
630 Anything bigger and we use qsort. If you make this too small, the qsort
631 will probably break (or become less efficient), because it doesn't expect
632 the middle element of a partition to be the same as the right or left -
633 you have been warned).
635 #ifndef QSORT_BREAK_EVEN
636 #define QSORT_BREAK_EVEN 6
639 /* QSORT_PLAY_SAFE is the size of the largest partition we're willing
640 to go quadratic on. We innoculate larger partitions against
641 quadratic behavior by shuffling them before sorting. This is not
642 an absolute guarantee of non-quadratic behavior, but it would take
643 staggeringly bad luck to pick extreme elements as the pivot
644 from randomized data.
646 #ifndef QSORT_PLAY_SAFE
647 #define QSORT_PLAY_SAFE 255
650 /* ************************************************************* Data Types */
652 /* hold left and right index values of a partition waiting to be sorted (the
653 partition includes both left and right - right is NOT one past the end or
656 struct partition_stack_entry {
659 #ifdef QSORT_ORDER_GUESS
660 int qsort_break_even;
664 /* ******************************************************* Shorthand Macros */
666 /* Note that these macros will be used from inside the qsort function where
667 we happen to know that the variable 'elt_size' contains the size of an
668 array element and the variable 'temp' points to enough space to hold a
669 temp element and the variable 'array' points to the array being sorted
670 and 'compare' is the pointer to the compare routine.
672 Also note that there are very many highly architecture specific ways
673 these might be sped up, but this is simply the most generally portable
674 code I could think of.
677 /* Return < 0 == 0 or > 0 as the value of elt1 is < elt2, == elt2, > elt2
679 #define qsort_cmp(elt1, elt2) \
680 ((*compare)(aTHX_ array[elt1], array[elt2]))
682 #ifdef QSORT_ORDER_GUESS
683 #define QSORT_NOTICE_SWAP swapped++;
685 #define QSORT_NOTICE_SWAP
688 /* swaps contents of array elements elt1, elt2.
690 #define qsort_swap(elt1, elt2) \
693 temp = array[elt1]; \
694 array[elt1] = array[elt2]; \
695 array[elt2] = temp; \
698 /* rotate contents of elt1, elt2, elt3 such that elt1 gets elt2, elt2 gets
699 elt3 and elt3 gets elt1.
701 #define qsort_rotate(elt1, elt2, elt3) \
704 temp = array[elt1]; \
705 array[elt1] = array[elt2]; \
706 array[elt2] = array[elt3]; \
707 array[elt3] = temp; \
710 /* ************************************************************ Debug stuff */
717 return; /* good place to set a breakpoint */
720 #define qsort_assert(t) (void)( (t) || (break_here(), 0) )
727 int (*compare)(const void * elt1, const void * elt2),
728 int pc_left, int pc_right, int u_left, int u_right)
732 qsort_assert(pc_left <= pc_right);
733 qsort_assert(u_right < pc_left);
734 qsort_assert(pc_right < u_left);
735 for (i = u_right + 1; i < pc_left; ++i) {
736 qsort_assert(qsort_cmp(i, pc_left) < 0);
738 for (i = pc_left; i < pc_right; ++i) {
739 qsort_assert(qsort_cmp(i, pc_right) == 0);
741 for (i = pc_right + 1; i < u_left; ++i) {
742 qsort_assert(qsort_cmp(pc_right, i) < 0);
746 #define qsort_all_asserts(PC_LEFT, PC_RIGHT, U_LEFT, U_RIGHT) \
747 doqsort_all_asserts(array, num_elts, elt_size, compare, \
748 PC_LEFT, PC_RIGHT, U_LEFT, U_RIGHT)
752 #define qsort_assert(t) ((void)0)
754 #define qsort_all_asserts(PC_LEFT, PC_RIGHT, U_LEFT, U_RIGHT) ((void)0)
758 /* ****************************************************************** qsort */
760 STATIC void /* the standard unstable (u) quicksort (qsort) */
761 S_qsortsvu(pTHX_ SV ** array, size_t num_elts, SVCOMPARE_t compare)
765 struct partition_stack_entry partition_stack[QSORT_MAX_STACK];
766 int next_stack_entry = 0;
770 #ifdef QSORT_ORDER_GUESS
771 int qsort_break_even;
775 /* Make sure we actually have work to do.
781 /* Innoculate large partitions against quadratic behavior */
782 if (num_elts > QSORT_PLAY_SAFE) {
784 register SV ** const q = array;
785 for (n = num_elts; n > 1; ) {
786 register const size_t j = (size_t)(n-- * Drand01());
793 /* Setup the initial partition definition and fall into the sorting loop
796 part_right = (int)(num_elts - 1);
797 #ifdef QSORT_ORDER_GUESS
798 qsort_break_even = QSORT_BREAK_EVEN;
800 #define qsort_break_even QSORT_BREAK_EVEN
803 if ((part_right - part_left) >= qsort_break_even) {
804 /* OK, this is gonna get hairy, so lets try to document all the
805 concepts and abbreviations and variables and what they keep
808 pc: pivot chunk - the set of array elements we accumulate in the
809 middle of the partition, all equal in value to the original
810 pivot element selected. The pc is defined by:
812 pc_left - the leftmost array index of the pc
813 pc_right - the rightmost array index of the pc
815 we start with pc_left == pc_right and only one element
816 in the pivot chunk (but it can grow during the scan).
818 u: uncompared elements - the set of elements in the partition
819 we have not yet compared to the pivot value. There are two
820 uncompared sets during the scan - one to the left of the pc
821 and one to the right.
823 u_right - the rightmost index of the left side's uncompared set
824 u_left - the leftmost index of the right side's uncompared set
826 The leftmost index of the left sides's uncompared set
827 doesn't need its own variable because it is always defined
828 by the leftmost edge of the whole partition (part_left). The
829 same goes for the rightmost edge of the right partition
832 We know there are no uncompared elements on the left once we
833 get u_right < part_left and no uncompared elements on the
834 right once u_left > part_right. When both these conditions
835 are met, we have completed the scan of the partition.
837 Any elements which are between the pivot chunk and the
838 uncompared elements should be less than the pivot value on
839 the left side and greater than the pivot value on the right
840 side (in fact, the goal of the whole algorithm is to arrange
841 for that to be true and make the groups of less-than and
842 greater-then elements into new partitions to sort again).
844 As you marvel at the complexity of the code and wonder why it
845 has to be so confusing. Consider some of the things this level
848 Once I do a compare, I squeeze every ounce of juice out of it. I
849 never do compare calls I don't have to do, and I certainly never
852 I also never swap any elements unless I can prove there is a
853 good reason. Many sort algorithms will swap a known value with
854 an uncompared value just to get things in the right place (or
855 avoid complexity :-), but that uncompared value, once it gets
856 compared, may then have to be swapped again. A lot of the
857 complexity of this code is due to the fact that it never swaps
858 anything except compared values, and it only swaps them when the
859 compare shows they are out of position.
861 int pc_left, pc_right;
866 pc_left = ((part_left + part_right) / 2);
868 u_right = pc_left - 1;
869 u_left = pc_right + 1;
871 /* Qsort works best when the pivot value is also the median value
872 in the partition (unfortunately you can't find the median value
873 without first sorting :-), so to give the algorithm a helping
874 hand, we pick 3 elements and sort them and use the median value
875 of that tiny set as the pivot value.
877 Some versions of qsort like to use the left middle and right as
878 the 3 elements to sort so they can insure the ends of the
879 partition will contain values which will stop the scan in the
880 compare loop, but when you have to call an arbitrarily complex
881 routine to do a compare, its really better to just keep track of
882 array index values to know when you hit the edge of the
883 partition and avoid the extra compare. An even better reason to
884 avoid using a compare call is the fact that you can drop off the
885 edge of the array if someone foolishly provides you with an
886 unstable compare function that doesn't always provide consistent
889 So, since it is simpler for us to compare the three adjacent
890 elements in the middle of the partition, those are the ones we
891 pick here (conveniently pointed at by u_right, pc_left, and
892 u_left). The values of the left, center, and right elements
893 are refered to as l c and r in the following comments.
896 #ifdef QSORT_ORDER_GUESS
899 s = qsort_cmp(u_right, pc_left);
902 s = qsort_cmp(pc_left, u_left);
903 /* if l < c, c < r - already in order - nothing to do */
905 /* l < c, c == r - already in order, pc grows */
907 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
909 /* l < c, c > r - need to know more */
910 s = qsort_cmp(u_right, u_left);
912 /* l < c, c > r, l < r - swap c & r to get ordered */
913 qsort_swap(pc_left, u_left);
914 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
916 /* l < c, c > r, l == r - swap c&r, grow pc */
917 qsort_swap(pc_left, u_left);
919 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
921 /* l < c, c > r, l > r - make lcr into rlc to get ordered */
922 qsort_rotate(pc_left, u_right, u_left);
923 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
928 s = qsort_cmp(pc_left, u_left);
930 /* l == c, c < r - already in order, grow pc */
932 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
934 /* l == c, c == r - already in order, grow pc both ways */
937 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
939 /* l == c, c > r - swap l & r, grow pc */
940 qsort_swap(u_right, u_left);
942 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
946 s = qsort_cmp(pc_left, u_left);
948 /* l > c, c < r - need to know more */
949 s = qsort_cmp(u_right, u_left);
951 /* l > c, c < r, l < r - swap l & c to get ordered */
952 qsort_swap(u_right, pc_left);
953 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
955 /* l > c, c < r, l == r - swap l & c, grow pc */
956 qsort_swap(u_right, pc_left);
958 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
960 /* l > c, c < r, l > r - rotate lcr into crl to order */
961 qsort_rotate(u_right, pc_left, u_left);
962 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
965 /* l > c, c == r - swap ends, grow pc */
966 qsort_swap(u_right, u_left);
968 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
970 /* l > c, c > r - swap ends to get in order */
971 qsort_swap(u_right, u_left);
972 qsort_all_asserts(pc_left, pc_right, u_left + 1, u_right - 1);
975 /* We now know the 3 middle elements have been compared and
976 arranged in the desired order, so we can shrink the uncompared
981 qsort_all_asserts(pc_left, pc_right, u_left, u_right);
983 /* The above massive nested if was the simple part :-). We now have
984 the middle 3 elements ordered and we need to scan through the
985 uncompared sets on either side, swapping elements that are on
986 the wrong side or simply shuffling equal elements around to get
987 all equal elements into the pivot chunk.
991 int still_work_on_left;
992 int still_work_on_right;
994 /* Scan the uncompared values on the left. If I find a value
995 equal to the pivot value, move it over so it is adjacent to
996 the pivot chunk and expand the pivot chunk. If I find a value
997 less than the pivot value, then just leave it - its already
998 on the correct side of the partition. If I find a greater
999 value, then stop the scan.
1001 while ((still_work_on_left = (u_right >= part_left))) {
1002 s = qsort_cmp(u_right, pc_left);
1005 } else if (s == 0) {
1007 if (pc_left != u_right) {
1008 qsort_swap(u_right, pc_left);
1014 qsort_assert(u_right < pc_left);
1015 qsort_assert(pc_left <= pc_right);
1016 qsort_assert(qsort_cmp(u_right + 1, pc_left) <= 0);
1017 qsort_assert(qsort_cmp(pc_left, pc_right) == 0);
1020 /* Do a mirror image scan of uncompared values on the right
1022 while ((still_work_on_right = (u_left <= part_right))) {
1023 s = qsort_cmp(pc_right, u_left);
1026 } else if (s == 0) {
1028 if (pc_right != u_left) {
1029 qsort_swap(pc_right, u_left);
1035 qsort_assert(u_left > pc_right);
1036 qsort_assert(pc_left <= pc_right);
1037 qsort_assert(qsort_cmp(pc_right, u_left - 1) <= 0);
1038 qsort_assert(qsort_cmp(pc_left, pc_right) == 0);
1041 if (still_work_on_left) {
1042 /* I know I have a value on the left side which needs to be
1043 on the right side, but I need to know more to decide
1044 exactly the best thing to do with it.
1046 if (still_work_on_right) {
1047 /* I know I have values on both side which are out of
1048 position. This is a big win because I kill two birds
1049 with one swap (so to speak). I can advance the
1050 uncompared pointers on both sides after swapping both
1051 of them into the right place.
1053 qsort_swap(u_right, u_left);
1056 qsort_all_asserts(pc_left, pc_right, u_left, u_right);
1058 /* I have an out of position value on the left, but the
1059 right is fully scanned, so I "slide" the pivot chunk
1060 and any less-than values left one to make room for the
1061 greater value over on the right. If the out of position
1062 value is immediately adjacent to the pivot chunk (there
1063 are no less-than values), I can do that with a swap,
1064 otherwise, I have to rotate one of the less than values
1065 into the former position of the out of position value
1066 and the right end of the pivot chunk into the left end
1070 if (pc_left == u_right) {
1071 qsort_swap(u_right, pc_right);
1072 qsort_all_asserts(pc_left, pc_right-1, u_left, u_right-1);
1074 qsort_rotate(u_right, pc_left, pc_right);
1075 qsort_all_asserts(pc_left, pc_right-1, u_left, u_right-1);
1080 } else if (still_work_on_right) {
1081 /* Mirror image of complex case above: I have an out of
1082 position value on the right, but the left is fully
1083 scanned, so I need to shuffle things around to make room
1084 for the right value on the left.
1087 if (pc_right == u_left) {
1088 qsort_swap(u_left, pc_left);
1089 qsort_all_asserts(pc_left+1, pc_right, u_left+1, u_right);
1091 qsort_rotate(pc_right, pc_left, u_left);
1092 qsort_all_asserts(pc_left+1, pc_right, u_left+1, u_right);
1097 /* No more scanning required on either side of partition,
1098 break out of loop and figure out next set of partitions
1104 /* The elements in the pivot chunk are now in the right place. They
1105 will never move or be compared again. All I have to do is decide
1106 what to do with the stuff to the left and right of the pivot
1109 Notes on the QSORT_ORDER_GUESS ifdef code:
1111 1. If I just built these partitions without swapping any (or
1112 very many) elements, there is a chance that the elements are
1113 already ordered properly (being properly ordered will
1114 certainly result in no swapping, but the converse can't be
1117 2. A (properly written) insertion sort will run faster on
1118 already ordered data than qsort will.
1120 3. Perhaps there is some way to make a good guess about
1121 switching to an insertion sort earlier than partition size 6
1122 (for instance - we could save the partition size on the stack
1123 and increase the size each time we find we didn't swap, thus
1124 switching to insertion sort earlier for partitions with a
1125 history of not swapping).
1127 4. Naturally, if I just switch right away, it will make
1128 artificial benchmarks with pure ascending (or descending)
1129 data look really good, but is that a good reason in general?
1133 #ifdef QSORT_ORDER_GUESS
1135 #if QSORT_ORDER_GUESS == 1
1136 qsort_break_even = (part_right - part_left) + 1;
1138 #if QSORT_ORDER_GUESS == 2
1139 qsort_break_even *= 2;
1141 #if QSORT_ORDER_GUESS == 3
1142 const int prev_break = qsort_break_even;
1143 qsort_break_even *= qsort_break_even;
1144 if (qsort_break_even < prev_break) {
1145 qsort_break_even = (part_right - part_left) + 1;
1149 qsort_break_even = QSORT_BREAK_EVEN;
1153 if (part_left < pc_left) {
1154 /* There are elements on the left which need more processing.
1155 Check the right as well before deciding what to do.
1157 if (pc_right < part_right) {
1158 /* We have two partitions to be sorted. Stack the biggest one
1159 and process the smallest one on the next iteration. This
1160 minimizes the stack height by insuring that any additional
1161 stack entries must come from the smallest partition which
1162 (because it is smallest) will have the fewest
1163 opportunities to generate additional stack entries.
1165 if ((part_right - pc_right) > (pc_left - part_left)) {
1166 /* stack the right partition, process the left */
1167 partition_stack[next_stack_entry].left = pc_right + 1;
1168 partition_stack[next_stack_entry].right = part_right;
1169 #ifdef QSORT_ORDER_GUESS
1170 partition_stack[next_stack_entry].qsort_break_even = qsort_break_even;
1172 part_right = pc_left - 1;
1174 /* stack the left partition, process the right */
1175 partition_stack[next_stack_entry].left = part_left;
1176 partition_stack[next_stack_entry].right = pc_left - 1;
1177 #ifdef QSORT_ORDER_GUESS
1178 partition_stack[next_stack_entry].qsort_break_even = qsort_break_even;
1180 part_left = pc_right + 1;
1182 qsort_assert(next_stack_entry < QSORT_MAX_STACK);
1185 /* The elements on the left are the only remaining elements
1186 that need sorting, arrange for them to be processed as the
1189 part_right = pc_left - 1;
1191 } else if (pc_right < part_right) {
1192 /* There is only one chunk on the right to be sorted, make it
1193 the new partition and loop back around.
1195 part_left = pc_right + 1;
1197 /* This whole partition wound up in the pivot chunk, so
1198 we need to get a new partition off the stack.
1200 if (next_stack_entry == 0) {
1201 /* the stack is empty - we are done */
1205 part_left = partition_stack[next_stack_entry].left;
1206 part_right = partition_stack[next_stack_entry].right;
1207 #ifdef QSORT_ORDER_GUESS
1208 qsort_break_even = partition_stack[next_stack_entry].qsort_break_even;
1212 /* This partition is too small to fool with qsort complexity, just
1213 do an ordinary insertion sort to minimize overhead.
1216 /* Assume 1st element is in right place already, and start checking
1217 at 2nd element to see where it should be inserted.
1219 for (i = part_left + 1; i <= part_right; ++i) {
1221 /* Scan (backwards - just in case 'i' is already in right place)
1222 through the elements already sorted to see if the ith element
1223 belongs ahead of one of them.
1225 for (j = i - 1; j >= part_left; --j) {
1226 if (qsort_cmp(i, j) >= 0) {
1227 /* i belongs right after j
1234 /* Looks like we really need to move some things
1238 for (k = i - 1; k >= j; --k)
1239 array[k + 1] = array[k];
1244 /* That partition is now sorted, grab the next one, or get out
1245 of the loop if there aren't any more.
1248 if (next_stack_entry == 0) {
1249 /* the stack is empty - we are done */
1253 part_left = partition_stack[next_stack_entry].left;
1254 part_right = partition_stack[next_stack_entry].right;
1255 #ifdef QSORT_ORDER_GUESS
1256 qsort_break_even = partition_stack[next_stack_entry].qsort_break_even;
1261 /* Believe it or not, the array is sorted at this point! */
1264 /* Stabilize what is, presumably, an otherwise unstable sort method.
1265 * We do that by allocating (or having on hand) an array of pointers
1266 * that is the same size as the original array of elements to be sorted.
1267 * We initialize this parallel array with the addresses of the original
1268 * array elements. This indirection can make you crazy.
1269 * Some pictures can help. After initializing, we have
1273 * | | --------------> | | ------> first element to be sorted
1275 * | | --------------> | | ------> second element to be sorted
1277 * | | --------------> | | ------> third element to be sorted
1281 * | | --------------> | | ------> n-1st element to be sorted
1283 * | | --------------> | | ------> n-th element to be sorted
1286 * During the sort phase, we leave the elements of list1 where they are,
1287 * and sort the pointers in the indirect array in the same order determined
1288 * by the original comparison routine on the elements pointed to.
1289 * Because we don't move the elements of list1 around through
1290 * this phase, we can break ties on elements that compare equal
1291 * using their address in the list1 array, ensuring stabilty.
1292 * This leaves us with something looking like
1296 * | | --+ +---> | | ------> first element to be sorted
1298 * | | --|-------|---> | | ------> second element to be sorted
1300 * | | --|-------+ +-> | | ------> third element to be sorted
1303 * +----+ | | | | +----+
1304 * | | ---|-+ | +--> | | ------> n-1st element to be sorted
1306 * | | ---+ +----> | | ------> n-th element to be sorted
1309 * where the i-th element of the indirect array points to the element
1310 * that should be i-th in the sorted array. After the sort phase,
1311 * we have to put the elements of list1 into the places
1312 * dictated by the indirect array.
1317 cmpindir(pTHX_ gptr a, gptr b)
1320 gptr * const ap = (gptr *)a;
1321 gptr * const bp = (gptr *)b;
1322 const I32 sense = PL_sort_RealCmp(aTHX_ *ap, *bp);
1326 return (ap > bp) ? 1 : ((ap < bp) ? -1 : 0);
1330 cmpindir_desc(pTHX_ gptr a, gptr b)
1333 gptr * const ap = (gptr *)a;
1334 gptr * const bp = (gptr *)b;
1335 const I32 sense = PL_sort_RealCmp(aTHX_ *ap, *bp);
1337 /* Reverse the default */
1340 /* But don't reverse the stability test. */
1341 return (ap > bp) ? 1 : ((ap < bp) ? -1 : 0);
1346 S_qsortsv(pTHX_ gptr *list1, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1349 if ((flags & SORTf_STABLE) != 0) {
1350 register gptr **pp, *q;
1351 register size_t n, j, i;
1352 gptr *small[SMALLSORT], **indir, tmp;
1353 SVCOMPARE_t savecmp;
1354 if (nmemb <= 1) return; /* sorted trivially */
1356 /* Small arrays can use the stack, big ones must be allocated */
1357 if (nmemb <= SMALLSORT) indir = small;
1358 else { Newx(indir, nmemb, gptr *); }
1360 /* Copy pointers to original array elements into indirect array */
1361 for (n = nmemb, pp = indir, q = list1; n--; ) *pp++ = q++;
1363 savecmp = PL_sort_RealCmp; /* Save current comparison routine, if any */
1364 PL_sort_RealCmp = cmp; /* Put comparison routine where cmpindir can find it */
1366 /* sort, with indirection */
1367 S_qsortsvu(aTHX_ (gptr *)indir, nmemb,
1368 ((flags & SORTf_DESC) != 0 ? cmpindir_desc : cmpindir));
1372 for (n = nmemb; n--; ) {
1373 /* Assert A: all elements of q with index > n are already
1374 * in place. This is vacuosly true at the start, and we
1375 * put element n where it belongs below (if it wasn't
1376 * already where it belonged). Assert B: we only move
1377 * elements that aren't where they belong,
1378 * so, by A, we never tamper with elements above n.
1380 j = pp[n] - q; /* This sets j so that q[j] is
1381 * at pp[n]. *pp[j] belongs in
1382 * q[j], by construction.
1384 if (n != j) { /* all's well if n == j */
1385 tmp = q[j]; /* save what's in q[j] */
1387 q[j] = *pp[j]; /* put *pp[j] where it belongs */
1388 i = pp[j] - q; /* the index in q of the element
1390 pp[j] = q + j; /* this is ok now */
1391 } while ((j = i) != n);
1392 /* There are only finitely many (nmemb) addresses
1394 * So we must eventually revisit an index we saw before.
1395 * Suppose the first revisited index is k != n.
1396 * An index is visited because something else belongs there.
1397 * If we visit k twice, then two different elements must
1398 * belong in the same place, which cannot be.
1399 * So j must get back to n, the loop terminates,
1400 * and we put the saved element where it belongs.
1402 q[n] = tmp; /* put what belongs into
1403 * the n-th element */
1407 /* free iff allocated */
1408 if (indir != small) { Safefree(indir); }
1409 /* restore prevailing comparison routine */
1410 PL_sort_RealCmp = savecmp;
1411 } else if ((flags & SORTf_DESC) != 0) {
1412 SVCOMPARE_t savecmp = PL_sort_RealCmp; /* Save current comparison routine, if any */
1413 PL_sort_RealCmp = cmp; /* Put comparison routine where cmp_desc can find it */
1415 S_qsortsvu(aTHX_ list1, nmemb, cmp);
1416 /* restore prevailing comparison routine */
1417 PL_sort_RealCmp = savecmp;
1419 S_qsortsvu(aTHX_ list1, nmemb, cmp);
1424 =head1 Array Manipulation Functions
1428 Sort an array. Here is an example:
1430 sortsv(AvARRAY(av), av_len(av)+1, Perl_sv_cmp_locale);
1432 Currently this always uses mergesort. See sortsv_flags for a more
1439 Perl_sortsv(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp)
1441 sortsv_flags(array, nmemb, cmp, 0);
1445 =for apidoc sortsv_flags
1447 Sort an array, with various options.
1452 Perl_sortsv_flags(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1454 void (*sortsvp)(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1455 = ((flags & SORTf_QSORT) != 0 ? S_qsortsv : S_mergesortsv);
1457 sortsvp(aTHX_ array, nmemb, cmp, flags);
1460 #define SvNSIOK(sv) ((SvFLAGS(sv) & SVf_NOK) || ((SvFLAGS(sv) & (SVf_IOK|SVf_IVisUV)) == SVf_IOK))
1461 #define SvSIOK(sv) ((SvFLAGS(sv) & (SVf_IOK|SVf_IVisUV)) == SVf_IOK)
1462 #define SvNSIV(sv) ( SvNOK(sv) ? SvNVX(sv) : ( SvSIOK(sv) ? SvIVX(sv) : sv_2nv(sv) ) )
1466 dVAR; dSP; dMARK; dORIGMARK;
1467 register SV **p1 = ORIGMARK+1, **p2;
1468 register I32 max, i;
1474 OP* const nextop = PL_op->op_next;
1475 I32 overloading = 0;
1476 bool hasargs = FALSE;
1479 const U8 priv = PL_op->op_private;
1480 const U8 flags = PL_op->op_flags;
1482 void (*sortsvp)(pTHX_ SV **array, size_t nmemb, SVCOMPARE_t cmp, U32 flags)
1483 = Perl_sortsv_flags;
1486 if ((priv & OPpSORT_DESCEND) != 0)
1487 sort_flags |= SORTf_DESC;
1488 if ((priv & OPpSORT_QSORT) != 0)
1489 sort_flags |= SORTf_QSORT;
1490 if ((priv & OPpSORT_STABLE) != 0)
1491 sort_flags |= SORTf_STABLE;
1493 if (gimme != G_ARRAY) {
1500 SAVEVPTR(PL_sortcop);
1501 if (flags & OPf_STACKED) {
1502 if (flags & OPf_SPECIAL) {
1503 OP *kid = cLISTOP->op_first->op_sibling; /* pass pushmark */
1504 kid = kUNOP->op_first; /* pass rv2gv */
1505 kid = kUNOP->op_first; /* pass leave */
1506 PL_sortcop = kid->op_next;
1507 stash = CopSTASH(PL_curcop);
1510 cv = sv_2cv(*++MARK, &stash, &gv, 0);
1511 if (cv && SvPOK(cv)) {
1512 const char * const proto = SvPV_nolen_const((SV*)cv);
1513 if (proto && strEQ(proto, "$$")) {
1517 if (!(cv && CvROOT(cv))) {
1518 if (cv && CvISXSUB(cv)) {
1522 SV *tmpstr = sv_newmortal();
1523 gv_efullname3(tmpstr, gv, NULL);
1524 DIE(aTHX_ "Undefined sort subroutine \"%"SVf"\" called",
1528 DIE(aTHX_ "Undefined subroutine in sort");
1533 PL_sortcop = (OP*)cv;
1535 PL_sortcop = CvSTART(cv);
1540 stash = CopSTASH(PL_curcop);
1543 /* optimiser converts "@a = sort @a" to "sort \@a";
1544 * in case of tied @a, pessimise: push (@a) onto stack, then assign
1545 * result back to @a at the end of this function */
1546 if (priv & OPpSORT_INPLACE) {
1547 assert( MARK+1 == SP && *SP && SvTYPE(*SP) == SVt_PVAV);
1548 (void)POPMARK; /* remove mark associated with ex-OP_AASSIGN */
1550 max = AvFILL(av) + 1;
1551 if (SvMAGICAL(av)) {
1554 for (i=0; i < max; i++) {
1555 SV **svp = av_fetch(av, i, FALSE);
1556 *SP++ = (svp) ? *svp : NULL;
1561 Perl_croak(aTHX_ PL_no_modify);
1564 p1 = p2 = AvARRAY(av);
1573 /* shuffle stack down, removing optional initial cv (p1!=p2), plus
1574 * any nulls; also stringify or converting to integer or number as
1575 * required any args */
1576 for (i=max; i > 0 ; i--) {
1577 if ((*p1 = *p2++)) { /* Weed out nulls. */
1580 if (priv & OPpSORT_NUMERIC) {
1581 if (priv & OPpSORT_INTEGER) {
1590 if (!SvNSIOK(*p1)) {
1596 if (all_SIVs && !SvSIOK(*p1))
1605 (void)sv_2pv_flags(*p1, 0,
1606 SV_GMAGIC|SV_CONST_RETURN);
1616 AvFILLp(av) = max-1;
1623 const bool oldcatch = CATCH_GET;
1629 PUSHSTACKi(PERLSI_SORT);
1630 if (!hasargs && !is_xsub) {
1631 SAVESPTR(PL_firstgv);
1632 SAVESPTR(PL_secondgv);
1633 SAVESPTR(PL_sortstash);
1634 PL_firstgv = gv_fetchpvs("a", GV_ADD|GV_NOTQUAL, SVt_PV);
1635 PL_secondgv = gv_fetchpvs("b", GV_ADD|GV_NOTQUAL, SVt_PV);
1636 PL_sortstash = stash;
1637 SAVESPTR(GvSV(PL_firstgv));
1638 SAVESPTR(GvSV(PL_secondgv));
1641 PUSHBLOCK(cx, CXt_NULL, PL_stack_base);
1642 if (!(flags & OPf_SPECIAL)) {
1643 cx->cx_type = CXt_SUB;
1644 cx->blk_gimme = G_SCALAR;
1647 AV* const padlist = CvPADLIST(cv);
1649 if (++CvDEPTH(cv) >= 2) {
1650 PERL_STACK_OVERFLOW_CHECK();
1651 pad_push(padlist, CvDEPTH(cv));
1654 PAD_SET_CUR_NOSAVE(padlist, CvDEPTH(cv));
1657 /* This is mostly copied from pp_entersub */
1658 AV * const av = (AV*)PAD_SVl(0);
1660 cx->blk_sub.savearray = GvAV(PL_defgv);
1661 GvAV(PL_defgv) = (AV*)SvREFCNT_inc_simple(av);
1662 CX_CURPAD_SAVE(cx->blk_sub);
1663 cx->blk_sub.argarray = av;
1668 cx->cx_type |= CXp_MULTICALL;
1671 sortsvp(aTHX_ start, max,
1672 (is_xsub ? S_sortcv_xsub : hasargs ? S_sortcv_stacked : S_sortcv),
1675 if (!(flags & OPf_SPECIAL)) {
1680 POPBLOCK(cx,PL_curpm);
1681 PL_stack_sp = newsp;
1683 CATCH_SET(oldcatch);
1686 MEXTEND(SP, 20); /* Can't afford stack realloc on signal. */
1687 start = sorting_av ? AvARRAY(av) : ORIGMARK+1;
1688 sortsvp(aTHX_ start, max,
1689 (priv & OPpSORT_NUMERIC)
1690 ? ( ( ( priv & OPpSORT_INTEGER) || all_SIVs)
1691 ? ( overloading ? S_amagic_i_ncmp : S_sv_i_ncmp)
1692 : ( overloading ? S_amagic_ncmp : S_sv_ncmp ) )
1693 : ( IN_LOCALE_RUNTIME
1695 ? S_amagic_cmp_locale
1696 : sv_cmp_locale_static)
1697 : ( overloading ? S_amagic_cmp : sv_cmp_static)),
1700 if ((priv & OPpSORT_REVERSE) != 0) {
1701 SV **q = start+max-1;
1703 SV * const tmp = *start;
1711 else if (av && !sorting_av) {
1712 /* simulate pp_aassign of tied AV */
1713 SV** const base = ORIGMARK+1;
1714 for (i=0; i < max; i++) {
1715 base[i] = newSVsv(base[i]);
1719 for (i=0; i < max; i++) {
1720 SV * const sv = base[i];
1721 SV ** const didstore = av_store(av, i, sv);
1729 PL_stack_sp = ORIGMARK + (sorting_av ? 0 : max);
1734 S_sortcv(pTHX_ SV *a, SV *b)
1737 const I32 oldsaveix = PL_savestack_ix;
1738 const I32 oldscopeix = PL_scopestack_ix;
1740 GvSV(PL_firstgv) = a;
1741 GvSV(PL_secondgv) = b;
1742 PL_stack_sp = PL_stack_base;
1745 if (PL_stack_sp != PL_stack_base + 1)
1746 Perl_croak(aTHX_ "Sort subroutine didn't return single value");
1747 if (!SvNIOKp(*PL_stack_sp))
1748 Perl_croak(aTHX_ "Sort subroutine didn't return a numeric value");
1749 result = SvIV(*PL_stack_sp);
1750 while (PL_scopestack_ix > oldscopeix) {
1753 leave_scope(oldsaveix);
1758 S_sortcv_stacked(pTHX_ SV *a, SV *b)
1761 const I32 oldsaveix = PL_savestack_ix;
1762 const I32 oldscopeix = PL_scopestack_ix;
1764 AV * const av = GvAV(PL_defgv);
1766 if (AvMAX(av) < 1) {
1767 SV** ary = AvALLOC(av);
1768 if (AvARRAY(av) != ary) {
1769 AvMAX(av) += AvARRAY(av) - AvALLOC(av);
1770 SvPV_set(av, (char*)ary);
1772 if (AvMAX(av) < 1) {
1775 SvPV_set(av, (char*)ary);
1782 PL_stack_sp = PL_stack_base;
1785 if (PL_stack_sp != PL_stack_base + 1)
1786 Perl_croak(aTHX_ "Sort subroutine didn't return single value");
1787 if (!SvNIOKp(*PL_stack_sp))
1788 Perl_croak(aTHX_ "Sort subroutine didn't return a numeric value");
1789 result = SvIV(*PL_stack_sp);
1790 while (PL_scopestack_ix > oldscopeix) {
1793 leave_scope(oldsaveix);
1798 S_sortcv_xsub(pTHX_ SV *a, SV *b)
1801 const I32 oldsaveix = PL_savestack_ix;
1802 const I32 oldscopeix = PL_scopestack_ix;
1803 CV * const cv=(CV*)PL_sortcop;
1812 (void)(*CvXSUB(cv))(aTHX_ cv);
1813 if (PL_stack_sp != PL_stack_base + 1)
1814 Perl_croak(aTHX_ "Sort subroutine didn't return single value");
1815 if (!SvNIOKp(*PL_stack_sp))
1816 Perl_croak(aTHX_ "Sort subroutine didn't return a numeric value");
1817 result = SvIV(*PL_stack_sp);
1818 while (PL_scopestack_ix > oldscopeix) {
1821 leave_scope(oldsaveix);
1827 S_sv_ncmp(pTHX_ SV *a, SV *b)
1829 const NV nv1 = SvNSIV(a);
1830 const NV nv2 = SvNSIV(b);
1831 return nv1 < nv2 ? -1 : nv1 > nv2 ? 1 : 0;
1835 S_sv_i_ncmp(pTHX_ SV *a, SV *b)
1837 const IV iv1 = SvIV(a);
1838 const IV iv2 = SvIV(b);
1839 return iv1 < iv2 ? -1 : iv1 > iv2 ? 1 : 0;
1842 #define tryCALL_AMAGICbin(left,right,meth) \
1843 (PL_amagic_generation && (SvAMAGIC(left)||SvAMAGIC(right))) \
1844 ? amagic_call(left, right, CAT2(meth,_amg), 0) \
1848 S_amagic_ncmp(pTHX_ register SV *a, register SV *b)
1851 SV * const tmpsv = tryCALL_AMAGICbin(a,b,ncmp);
1854 const I32 i = SvIVX(tmpsv);
1860 const NV d = SvNV(tmpsv);
1866 return S_sv_ncmp(aTHX_ a, b);
1870 S_amagic_i_ncmp(pTHX_ register SV *a, register SV *b)
1873 SV * const tmpsv = tryCALL_AMAGICbin(a,b,ncmp);
1876 const I32 i = SvIVX(tmpsv);
1882 const NV d = SvNV(tmpsv);
1888 return S_sv_i_ncmp(aTHX_ a, b);
1892 S_amagic_cmp(pTHX_ register SV *str1, register SV *str2)
1895 SV * const tmpsv = tryCALL_AMAGICbin(str1,str2,scmp);
1898 const I32 i = SvIVX(tmpsv);
1904 const NV d = SvNV(tmpsv);
1910 return sv_cmp(str1, str2);
1914 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: