commit 12c5bde7b2b0371050910b2446b9aaad5fc44573 from: Stefan Sperling date: Fri Oct 16 20:09:45 2020 UTC use mergesort(3) instead of qsort(3) in patience diff Fixes test failures on OpenBSD due to unstable results of qsort if two or more elements are equal. Add a compat copy of OpenBSD mergesort(3) for Linux. commit - f087e96869b35f67db93f1a9a3df95d0dc9da883 commit + 12c5bde7b2b0371050910b2446b9aaad5fc44573 blob - b16aa2e0d80e22365fd914d1ec0cefea62efdd90 blob + 75b18881a7b7b883ce7117d6641ca4f1f24c0d6e --- compat/include/stdlib.h +++ compat/include/stdlib.h @@ -15,5 +15,6 @@ const char * getprogname(void); void *reallocarray(void *, size_t, size_t); void *recallocarray(void *, size_t, size_t, size_t); +int mergesort(void *, size_t, size_t, int (*cmp)(const void *, const void *)); #endif blob - /dev/null blob + a5b0d88aa33784aedb4c207512de0089d2306b7d (mode 644) --- /dev/null +++ compat/merge.c @@ -0,0 +1,338 @@ +/* $OpenBSD: merge.c,v 1.10 2015/06/21 03:20:56 millert Exp $ */ +/*- + * Copyright (c) 1992, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Peter McIlroy. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +/* + * Hybrid exponential search/linear search merge sort with hybrid + * natural/pairwise first pass. Requires about .3% more comparisons + * for random data than LSMS with pairwise first pass alone. + * It works for objects as small as two bytes. + */ + +#define NATURAL +#define THRESHOLD 16 /* Best choice for natural merge cut-off. */ + +/* #define NATURAL to get hybrid natural merge. + * (The default is pairwise merging.) + */ + +#include + +#include +#include +#include + +static void setup(u_char *list1, u_char *list2, size_t n, size_t size, + int (*cmp)(const void *, const void *)); +static void insertionsort(u_char *a, size_t n, size_t size, + int (*cmp)(const void *, const void *)); + +#define ISIZE sizeof(int) +#define PSIZE sizeof(u_char *) +#define ICOPY_LIST(src, dst, last) \ + do \ + *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ + while(src < last) +#define ICOPY_ELT(src, dst, i) \ + do \ + *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ + while (i -= ISIZE) + +#define CCOPY_LIST(src, dst, last) \ + do \ + *dst++ = *src++; \ + while (src < last) +#define CCOPY_ELT(src, dst, i) \ + do \ + *dst++ = *src++; \ + while (i -= 1) + +/* + * Find the next possible pointer head. (Trickery for forcing an array + * to do double duty as a linked list when objects do not align with word + * boundaries. + */ +/* Assumption: PSIZE is a power of 2. */ +#define EVAL(p) (u_char **) \ + ((u_char *)0 + \ + (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))) + +/* + * Arguments are as for qsort. + */ +int +mergesort(void *base, size_t nmemb, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, sense; + int big, iflag; + u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; + u_char *list2, *list1, *p2, *p, *last, **p1; + + if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ + errno = EINVAL; + return (-1); + } + + if (nmemb == 0) + return (0); + + /* + * XXX + * Stupid subtraction for the Cray. + */ + iflag = 0; + if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) + iflag = 1; + + if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) + return (-1); + + list1 = base; + setup(list1, list2, nmemb, size, cmp); + last = list2 + nmemb * size; + i = big = 0; + while (*EVAL(list2) != last) { + l2 = list1; + p1 = EVAL(list1); + for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { + p2 = *EVAL(p2); + f1 = l2; + f2 = l1 = list1 + (p2 - list2); + if (p2 != last) + p2 = *EVAL(p2); + l2 = list1 + (p2 - list2); + while (f1 < l1 && f2 < l2) { + if ((*cmp)(f1, f2) <= 0) { + q = f2; + b = f1, t = l1; + sense = -1; + } else { + q = f1; + b = f2, t = l2; + sense = 0; + } + if (!big) { /* here i = 0 */ + while ((b += size) < t && cmp(q, b) >sense) + if (++i == 6) { + big = 1; + goto EXPONENTIAL; + } + } else { +EXPONENTIAL: for (i = size; ; i <<= 1) + if ((p = (b + i)) >= t) { + if ((p = t - size) > b && + (*cmp)(q, p) <= sense) + t = p; + else + b = p; + break; + } else if ((*cmp)(q, p) <= sense) { + t = p; + if (i == size) + big = 0; + goto FASTCASE; + } else + b = p; + while (t > b+size) { + i = (((t - b) / size) >> 1) * size; + if ((*cmp)(q, p = b + i) <= sense) + t = p; + else + b = p; + } + goto COPY; +FASTCASE: while (i > size) + if ((*cmp)(q, + p = b + (i >>= 1)) <= sense) + t = p; + else + b = p; +COPY: b = t; + } + i = size; + if (q == f1) { + if (iflag) { + ICOPY_LIST(f2, tp2, b); + ICOPY_ELT(f1, tp2, i); + } else { + CCOPY_LIST(f2, tp2, b); + CCOPY_ELT(f1, tp2, i); + } + } else { + if (iflag) { + ICOPY_LIST(f1, tp2, b); + ICOPY_ELT(f2, tp2, i); + } else { + CCOPY_LIST(f1, tp2, b); + CCOPY_ELT(f2, tp2, i); + } + } + } + if (f2 < l2) { + if (iflag) + ICOPY_LIST(f2, tp2, l2); + else + CCOPY_LIST(f2, tp2, l2); + } else if (f1 < l1) { + if (iflag) + ICOPY_LIST(f1, tp2, l1); + else + CCOPY_LIST(f1, tp2, l1); + } + *p1 = l2; + } + tp2 = list1; /* swap list1, list2 */ + list1 = list2; + list2 = tp2; + last = list2 + nmemb*size; + } + if (base == list2) { + memmove(list2, list1, nmemb*size); + list2 = list1; + } + free(list2); + return (0); +} + +#define swap(a, b) { \ + s = b; \ + i = size; \ + do { \ + tmp = *a; *a++ = *s; *s++ = tmp; \ + } while (--i); \ + a -= size; \ + } +#define reverse(bot, top) { \ + s = top; \ + do { \ + i = size; \ + do { \ + tmp = *bot; *bot++ = *s; *s++ = tmp; \ + } while (--i); \ + s -= size2; \ + } while(bot < s); \ +} + +/* + * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of + * increasing order, list2 in a corresponding linked list. Checks for runs + * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL + * is defined. Otherwise simple pairwise merging is used.) + */ +void +setup(u_char *list1, u_char *list2, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, length, size2, sense; + u_char tmp, *f1, *f2, *s, *l2, *last, *p2; + + size2 = size*2; + if (n <= 5) { + insertionsort(list1, n, size, cmp); + *EVAL(list2) = (u_char*) list2 + n*size; + return; + } + /* + * Avoid running pointers out of bounds; limit n to evens + * for simplicity. + */ + i = 4 + (n & 1); + insertionsort(list1 + (n - i) * size, i, size, cmp); + last = list1 + size * (n - i); + *EVAL(list2 + (last - list1)) = list2 + n * size; + +#ifdef NATURAL + p2 = list2; + f1 = list1; + sense = (cmp(f1, f1 + size) > 0); + for (; f1 < last; sense = !sense) { + length = 2; + /* Find pairs with same sense. */ + for (f2 = f1 + size2; f2 < last; f2 += size2) { + if ((cmp(f2, f2+ size) > 0) != sense) + break; + length += 2; + } + if (length < THRESHOLD) { /* Pairwise merge */ + do { + p2 = *EVAL(p2) = f1 + size2 - list1 + list2; + if (sense > 0) + swap (f1, f1 + size); + } while ((f1 += size2) < f2); + } else { /* Natural merge */ + l2 = f2; + for (f2 = f1 + size2; f2 < l2; f2 += size2) { + if ((cmp(f2-size, f2) > 0) != sense) { + p2 = *EVAL(p2) = f2 - list1 + list2; + if (sense > 0) + reverse(f1, f2-size); + f1 = f2; + } + } + if (sense > 0) + reverse (f1, f2-size); + f1 = f2; + if (f2 < last || cmp(f2 - size, f2) > 0) + p2 = *EVAL(p2) = f2 - list1 + list2; + else + p2 = *EVAL(p2) = list2 + n*size; + } + } +#else /* pairwise merge only. */ + for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { + p2 = *EVAL(p2) = p2 + size2; + if (cmp (f1, f1 + size) > 0) + swap(f1, f1 + size); + } +#endif /* NATURAL */ +} + +/* + * This is to avoid out-of-bounds addresses in sorting the + * last 4 elements. + */ +static void +insertionsort(u_char *a, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + u_char *ai, *s, *t, *u, tmp; + int i; + + for (ai = a+size; --n >= 1; ai += size) + for (t = ai; t > a; t -= size) { + u = t - size; + if (cmp(u, t) <= 0) + break; + swap(u, t); + } +} blob - 3f009fdeb7740bffe70cd5093c5ea5cce4d786f4 blob + 6f84afef060f64a2ce75d24c4001f5b7b59dc700 --- lib/GNUmakefile +++ lib/GNUmakefile @@ -11,7 +11,7 @@ SRCS = \ # Compat sources VPATH= $(CURDIR)/../compat -SRCS+= getprogname_linux.c reallocarray.c recallocarray.c +SRCS+= getprogname_linux.c reallocarray.c recallocarray.c merge.c CFLAGS+= -I$(CURDIR)/../compat/include OBJS = $(SRCS:.c=.o) blob - ccc539a9a6b47881667a6b8d2c1c7b0fd71fe10c blob + ce30d7edc9df8f722c837f476d478fa55ee9617b --- lib/diff_patience.c +++ lib/diff_patience.c @@ -46,7 +46,7 @@ struct atom_patience { (((struct atom_patience*)((ATOM)->root->current->algo_data))\ [diff_atom_idx((ATOM)->root->current, ATOM)]) -int diff_atoms_qsort_compar(const void *_a, const void *_b) +int diff_atoms_mergesort_compar(const void *_a, const void *_b) { const struct diff_atom *a = *(struct diff_atom**)_a; const struct diff_atom *b = *(struct diff_atom**)_b; @@ -54,7 +54,7 @@ int diff_atoms_qsort_compar(const void *_a, const void int rc = 0; /* If there's been an error (e.g. I/O error) in a previous compar, we - * have no way to abort the qsort but just report the rc and stop + * have no way to abort the mergesort but just report the rc and stop * comparing. Make sure to catch errors on either side. If atoms are * from more than one diff_data, make sure the error, if any, spreads * to all of them, so we can cut short all future comparisons. */ @@ -92,18 +92,21 @@ int diff_atoms_qsort_compar(const void *_a, const void } /* Sort an array of struct diff_atom* in-place. */ -static int diff_atoms_qsort(struct diff_atom *atoms[], +static int diff_atoms_sort(struct diff_atom *atoms[], size_t atoms_count) { - qsort(atoms, atoms_count, sizeof(struct diff_atom*), - diff_atoms_qsort_compar); + /* + * Use mergesort(3) bcause it provides stable ordering + * even if two or more elements are equal. + */ + mergesort(atoms, atoms_count, sizeof(struct diff_atom*), + diff_atoms_mergesort_compar); return atoms[0]->root->err; } static int -diff_atoms_mark_unique_in_both_by_qsort(struct diff_data *left, - struct diff_data *right, - unsigned int *unique_in_both_count_p) +diff_atoms_mark_unique_in_both(struct diff_data *left, struct diff_data *right, + unsigned int *unique_in_both_count_p) { struct diff_atom *a; struct diff_atom *b; @@ -129,7 +132,7 @@ diff_atoms_mark_unique_in_both_by_qsort(struct diff_da all_atoms[len++] = b; } - rc = diff_atoms_qsort(all_atoms, len); + rc = diff_atoms_sort(all_atoms, len); if (rc) goto free_and_exit; @@ -320,8 +323,7 @@ diff_algo_patience(const struct diff_algo_config *algo /* Find those lines that appear exactly once in 'left' and exactly once * in 'right'. */ - rc = diff_atoms_mark_unique_in_both_by_qsort(left, right, - &unique_in_both_count); + rc = diff_atoms_mark_unique_in_both(left, right, &unique_in_both_count); if (rc) goto free_and_exit;