2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2009 Gabor Kovesdan <gabor@FreeBSD.org>
5 * Copyright (C) 2012 Oleg Moskalenko <mom040267@gmail.com>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
35 #include <sys/types.h>
36 #include <sys/queue.h>
40 #if defined(SORT_THREADS)
43 #include <semaphore.h>
53 #include "radixsort.h"
55 unsigned long long free_memory = 1000000;
56 unsigned long long available_free_memory = 1000000;
60 const char *tmpdir = "/var/tmp";
61 const char *compress_program;
63 size_t max_open_files = 16;
66 * How much space we read from file at once
68 #define READ_CHUNK (4096)
71 * File reader structure
75 struct reader_buffer rb;
79 unsigned char *mmapaddr;
80 unsigned char *mmapptr;
90 * Structure to be used in file merge process.
94 struct file_reader *fr;
95 struct sort_list_item *si; /* current top line */
100 * List elements of "cleanable" files list.
102 struct CLEANABLE_FILE
105 LIST_ENTRY(CLEANABLE_FILE) files;
109 * List header of "cleanable" files list.
111 static LIST_HEAD(CLEANABLE_FILES,CLEANABLE_FILE) tmp_files;
114 * Semaphore to protect the tmp file list.
115 * We use semaphore here because it is signal-safe, according to POSIX.
116 * And semaphore does not require pthread library.
118 static sem_t tmp_files_sem;
120 static void mt_sort(struct sort_list *list,
121 int (*sort_func)(void *, size_t, size_t,
122 int (*)(const void *, const void *)), const char* fn);
125 * Init tmp files list
131 LIST_INIT(&tmp_files);
132 sem_init(&tmp_files_sem, 0, 1);
136 * Save name of a tmp file for signal cleanup
139 tmp_file_atexit(const char *tmp_file)
143 sem_wait(&tmp_files_sem);
144 struct CLEANABLE_FILE *item =
145 sort_malloc(sizeof(struct CLEANABLE_FILE));
146 item->fn = sort_strdup(tmp_file);
147 LIST_INSERT_HEAD(&tmp_files, item, files);
148 sem_post(&tmp_files_sem);
156 clear_tmp_files(void)
158 struct CLEANABLE_FILE *item;
160 sem_wait(&tmp_files_sem);
161 LIST_FOREACH(item,&tmp_files,files) {
162 if ((item) && (item->fn))
165 sem_post(&tmp_files_sem);
169 * Check whether a file is a temporary file
172 file_is_tmp(const char* fn)
174 struct CLEANABLE_FILE *item;
178 sem_wait(&tmp_files_sem);
179 LIST_FOREACH(item,&tmp_files,files) {
180 if ((item) && (item->fn))
181 if (strcmp(item->fn, fn) == 0) {
186 sem_post(&tmp_files_sem);
193 * Generate new temporary file name
196 new_tmp_file_name(void)
201 if (asprintf(&ret, "%s/.bsdsort.XXXXXXXXXX", tmpdir) == -1)
202 err(2, "asprintf()");
203 if ((fd = mkstemp(ret)) == -1)
207 tmp_file_atexit(ret);
212 * Initialize file list
215 file_list_init(struct file_list *fl, bool tmp)
227 * Add a file name to the list
230 file_list_add(struct file_list *fl, const char *fn, bool allocate)
234 if (fl->count >= fl->sz || (fl->fns == NULL)) {
235 fl->sz = (fl->sz) * 2 + 1;
236 fl->fns = sort_realloc(fl->fns, fl->sz *
239 fl->fns[fl->count] = allocate ? sort_strdup(fn) : fn;
245 * Populate file list from array of file names
248 file_list_populate(struct file_list *fl, int argc, char **argv, bool allocate)
254 for (i = 0; i < argc; i++)
255 file_list_add(fl, argv[i], allocate);
260 * Clean file list data and delete the files,
261 * if this is a list of temporary files
264 file_list_clean(struct file_list *fl)
271 for (i = 0; i < fl->count; i++) {
275 sort_free(fl->fns[i]);
292 sort_list_init(struct sort_list *l)
298 l->memsize = sizeof(struct sort_list);
304 * Add string to sort list
307 sort_list_add(struct sort_list *l, struct bwstring *str)
311 size_t indx = l->count;
313 if ((l->list == NULL) || (indx >= l->size)) {
314 size_t newsize = (l->size + 1) + 1024;
316 l->list = sort_realloc(l->list,
317 sizeof(struct sort_list_item*) * newsize);
318 l->memsize += (newsize - l->size) *
319 sizeof(struct sort_list_item*);
322 l->list[indx] = sort_list_item_alloc();
323 sort_list_item_set(l->list[indx], str);
324 l->memsize += sort_list_item_size(l->list[indx]);
330 * Clean sort list data
333 sort_list_clean(struct sort_list *l)
340 for (i = 0; i < l->count; i++) {
341 struct sort_list_item *item;
346 sort_list_item_clean(item);
356 l->memsize = sizeof(struct sort_list);
361 * Write sort list to file
364 sort_list_dump(struct sort_list *l, const char *fn)
370 f = openfile(fn, "w");
376 if (!(sort_opts_vals.uflag)) {
377 for (i = 0; i < l->count; ++i)
378 bwsfwrite(l->list[i]->str, f,
379 sort_opts_vals.zflag);
381 struct sort_list_item *last_printed_item = NULL;
382 struct sort_list_item *item;
383 for (i = 0; i < l->count; ++i) {
385 if ((last_printed_item == NULL) ||
386 list_coll(&last_printed_item, &item)) {
387 bwsfwrite(item->str, f, sort_opts_vals.zflag);
388 last_printed_item = item;
399 * Checks if the given file is sorted. Stops at the first disorder,
400 * prints the disordered line and returns 1.
403 check(const char *fn)
405 struct bwstring *s1, *s2, *s1disorder, *s2disorder;
406 struct file_reader *fr;
407 struct keys_array *ka1, *ka2;
409 size_t pos, posdisorder;
411 s1 = s2 = s1disorder = s2disorder = NULL;
414 fr = file_reader_init(fn);
425 s1 = file_reader_readline(fr);
429 ka1 = keys_array_alloc();
432 s2 = file_reader_readline(fr);
436 ka2 = keys_array_alloc();
442 bwsprintf(stdout, s2, "s1=<", ">");
443 bwsprintf(stdout, s1, "s2=<", ">");
445 int cmp = key_coll(ka2, ka1, 0);
447 printf("; cmp1=%d", cmp);
449 if (!cmp && sort_opts_vals.complex_sort &&
450 !(sort_opts_vals.uflag) && !(sort_opts_vals.sflag)) {
451 cmp = top_level_str_coll(s2, s1);
453 printf("; cmp2=%d", cmp);
458 if ((sort_opts_vals.uflag && (cmp <= 0)) || (cmp < 0)) {
459 if (!(sort_opts_vals.csilentflag)) {
460 s2disorder = bwsdup(s2);
463 s1disorder = bwsdup(s1);
471 clean_keys_array(s1, ka1);
479 s2 = file_reader_readline(fr);
483 ka2 = keys_array_alloc();
489 clean_keys_array(s1, ka1);
497 clean_keys_array(s2, ka2);
504 if ((fn == NULL) || (*fn == 0) || (strcmp(fn, "-") == 0)) {
506 s2 = file_reader_readline(fr);
513 file_reader_free(fr);
516 bws_disorder_warnx(s2disorder, fn, posdisorder);
518 bws_disorder_warnx(s1disorder, fn, posdisorder);
519 if (s1disorder != s2disorder)
534 * Opens a file. If the given filename is "-", stdout will be
538 openfile(const char *fn, const char *mode)
542 if (strcmp(fn, "-") == 0) {
543 return ((mode && mode[0] == 'r') ? stdin : stdout);
545 mode_t orig_file_mask = 0;
546 int is_tmp = file_is_tmp(fn);
548 if (is_tmp && (mode[0] == 'w'))
549 orig_file_mask = umask(S_IWGRP | S_IWOTH |
552 if (is_tmp && (compress_program != NULL)) {
556 cmdsz = strlen(fn) + 128;
557 cmd = sort_malloc(cmdsz);
562 snprintf(cmd, cmdsz - 1, "cat %s | %s -d",
563 fn, compress_program);
564 else if (mode[0] == 'w')
565 snprintf(cmd, cmdsz - 1, "%s > %s",
566 compress_program, fn);
568 err(2, "%s", getstr(7));
570 if ((file = popen(cmd, mode)) == NULL)
576 if ((file = fopen(fn, mode)) == NULL)
579 if (is_tmp && (mode[0] == 'w'))
580 umask(orig_file_mask);
590 closefile(FILE *f, const char *fn)
594 } else if (f == stdin) {
596 } else if (f == stdout) {
599 if (file_is_tmp(fn) && compress_program != NULL) {
608 * Reads a file into the internal buffer.
611 file_reader_init(const char *fsrc)
613 struct file_reader *ret;
618 ret = sort_calloc(1, sizeof(struct file_reader));
621 if (sort_opts_vals.zflag)
624 ret->fname = sort_strdup(fsrc);
626 if (strcmp(fsrc, "-") && (compress_program == NULL) && use_mmap) {
629 struct stat stat_buf;
634 flags = MAP_NOCORE | MAP_NOSYNC;
636 fd = open(fsrc, O_RDONLY);
640 if (fstat(fd, &stat_buf) < 0) {
645 sz = stat_buf.st_size;
647 #if defined(MAP_PREFAULT_READ)
648 flags |= MAP_PREFAULT_READ;
651 addr = mmap(NULL, sz, PROT_READ, flags, fd, 0);
652 if (addr == MAP_FAILED) {
658 ret->mmapaddr = addr;
660 ret->mmapptr = ret->mmapaddr;
665 if (ret->mmapaddr == NULL) {
666 ret->file = openfile(fsrc, "r");
667 if (ret->file == NULL)
670 if (strcmp(fsrc, "-")) {
671 ret->cbsz = READ_CHUNK;
672 ret->buffer = sort_malloc(ret->cbsz);
676 ret->bsz = fread(ret->buffer, 1, ret->cbsz, ret->file);
678 if (ferror(ret->file))
688 file_reader_readline(struct file_reader *fr)
690 struct bwstring *ret = NULL;
693 unsigned char *mmapend;
695 mmapend = fr->mmapaddr + fr->mmapsize;
696 if (fr->mmapptr >= mmapend)
699 unsigned char *strend;
702 sz = mmapend - fr->mmapptr;
703 strend = memchr(fr->mmapptr, fr->elsymb, sz);
705 if (strend == NULL) {
706 ret = bwscsbdup(fr->mmapptr, sz);
707 fr->mmapptr = mmapend;
709 ret = bwscsbdup(fr->mmapptr, strend -
711 fr->mmapptr = strend + 1;
715 } else if (fr->file != stdin) {
717 size_t bsz1, remsz, search_start;
723 if (fr->bsz > fr->strbeg)
724 remsz = fr->bsz - fr->strbeg;
726 /* line read cycle */
728 if (remsz > search_start)
729 strend = memchr(fr->buffer + fr->strbeg +
730 search_start, fr->elsymb, remsz -
740 if (fr->bsz != fr->cbsz)
742 err(2, "File read software error 1");
744 if (remsz > (READ_CHUNK >> 1)) {
745 search_start = fr->cbsz - fr->strbeg;
746 fr->cbsz += READ_CHUNK;
747 fr->buffer = sort_realloc(fr->buffer,
749 bsz1 = fread(fr->buffer + fr->bsz, 1,
750 READ_CHUNK, fr->file);
752 if (ferror(fr->file))
759 if (remsz > 0 && fr->strbeg>0)
760 bcopy(fr->buffer + fr->strbeg,
764 search_start = remsz;
765 bsz1 = fread(fr->buffer + remsz, 1,
766 fr->cbsz - remsz, fr->file);
768 if (ferror(fr->file))
772 fr->bsz = remsz + bsz1;
778 strend = fr->buffer + fr->bsz;
780 if ((fr->buffer + fr->strbeg <= strend) &&
781 (fr->strbeg < fr->bsz) && (remsz>0))
782 ret = bwscsbdup(fr->buffer + fr->strbeg, strend -
783 fr->buffer - fr->strbeg);
785 fr->strbeg = (strend - fr->buffer) + 1;
788 int delim = sort_opts_vals.zflag ? '\0' : '\n';
789 ssize_t len = getdelim(&fr->buffer, &fr->bsz, delim, fr->file);
795 if (len > 0 && fr->buffer[len - 1] == delim)
797 ret = bwscsbdup(fr->buffer, len);
804 file_reader_clean(struct file_reader *fr)
809 munmap(fr->mmapaddr, fr->mmapsize);
815 sort_free(fr->buffer);
818 if (fr->file != stdin)
819 closefile(fr->file, fr->fname);
822 sort_free(fr->fname);
824 memset(fr, 0, sizeof(struct file_reader));
829 file_reader_free(struct file_reader *fr)
833 file_reader_clean(fr);
839 procfile(const char *fsrc, struct sort_list *list, struct file_list *fl)
841 struct file_reader *fr;
843 fr = file_reader_init(fsrc);
847 /* file browse cycle */
849 struct bwstring *bws;
851 bws = file_reader_readline(fr);
856 sort_list_add(list, bws);
858 if (list->memsize >= available_free_memory) {
861 fn = new_tmp_file_name();
862 sort_list_to_file(list, fn);
863 file_list_add(fl, fn, false);
864 sort_list_clean(list);
868 file_reader_free(fr);
874 * Compare file headers. Files with EOF always go to the end of the list.
877 file_header_cmp(struct file_header *f1, struct file_header *f2)
883 if (f1->fr == NULL) {
884 return ((f2->fr == NULL) ? 0 : +1);
885 } else if (f2->fr == NULL)
890 ret = list_coll(&(f1->si), &(f2->si));
892 return ((f1->file_pos < f2->file_pos) ? -1 : +1);
899 * Allocate and init file header structure
902 file_header_init(struct file_header **fh, const char *fn, size_t file_pos)
906 struct bwstring *line;
908 *fh = sort_malloc(sizeof(struct file_header));
909 (*fh)->file_pos = file_pos;
910 (*fh)->fr = file_reader_init(fn);
911 if ((*fh)->fr == NULL) {
913 err(2, "%s", getstr(8));
915 line = file_reader_readline((*fh)->fr);
917 file_reader_free((*fh)->fr);
921 (*fh)->si = sort_list_item_alloc();
922 sort_list_item_set((*fh)->si, line);
931 file_header_close(struct file_header **fh)
936 file_reader_free((*fh)->fr);
940 sort_list_item_clean((*fh)->si);
941 sort_free((*fh)->si);
950 * Swap two array elements
953 file_header_swap(struct file_header **fh, size_t i1, size_t i2)
955 struct file_header *tmp;
962 /* heap algorithm ==>> */
965 * See heap sort algorithm
966 * "Raises" last element to its right place
969 file_header_heap_swim(struct file_header **fh, size_t indx)
975 parent_index = (indx - 1) >> 1;
977 if (file_header_cmp(fh[indx], fh[parent_index]) < 0) {
978 /* swap child and parent and continue */
979 file_header_swap(fh, indx, parent_index);
980 file_header_heap_swim(fh, parent_index);
986 * Sink the top element to its correct position
989 file_header_heap_sink(struct file_header **fh, size_t indx, size_t size)
991 size_t left_child_index;
992 size_t right_child_index;
994 left_child_index = indx + indx + 1;
995 right_child_index = left_child_index + 1;
997 if (left_child_index < size) {
998 size_t min_child_index;
1000 min_child_index = left_child_index;
1002 if ((right_child_index < size) &&
1003 (file_header_cmp(fh[left_child_index],
1004 fh[right_child_index]) > 0))
1005 min_child_index = right_child_index;
1006 if (file_header_cmp(fh[indx], fh[min_child_index]) > 0) {
1007 file_header_swap(fh, indx, min_child_index);
1008 file_header_heap_sink(fh, min_child_index, size);
1013 /* <<== heap algorithm */
1016 * Adds element to the "left" end
1019 file_header_list_rearrange_from_header(struct file_header **fh, size_t size)
1022 file_header_heap_sink(fh, 0, size);
1026 * Adds element to the "right" end
1029 file_header_list_push(struct file_header *f, struct file_header **fh, size_t size)
1033 file_header_heap_swim(fh, size - 1);
1038 struct bwstring *str;
1042 * Prints the current line of the file
1045 file_header_print(struct file_header *fh, FILE *f_out, struct last_printed *lp)
1048 if (fh && fh->fr && f_out && fh->si && fh->si->str) {
1049 if (sort_opts_vals.uflag) {
1050 if ((lp->str == NULL) || (str_list_coll(lp->str, &(fh->si)))) {
1051 bwsfwrite(fh->si->str, f_out, sort_opts_vals.zflag);
1054 lp->str = bwsdup(fh->si->str);
1057 bwsfwrite(fh->si->str, f_out, sort_opts_vals.zflag);
1065 file_header_read_next(struct file_header *fh)
1069 struct bwstring *tmp;
1071 tmp = file_reader_readline(fh->fr);
1073 file_reader_free(fh->fr);
1076 sort_list_item_clean(fh->si);
1082 fh->si = sort_list_item_alloc();
1083 sort_list_item_set(fh->si, tmp);
1089 * Merge array of "files headers"
1092 file_headers_merge(size_t fnum, struct file_header **fh, FILE *f_out)
1094 struct last_printed lp;
1097 memset(&lp, 0, sizeof(lp));
1100 * construct the initial sort structure
1102 for (i = 0; i < fnum; i++)
1103 file_header_list_push(fh[i], fh, i);
1105 while (fh[0]->fr) { /* unfinished files are always in front */
1106 /* output the smallest line: */
1107 file_header_print(fh[0], f_out, &lp);
1108 /* read a new line, if possible: */
1109 file_header_read_next(fh[0]);
1110 /* re-arrange the list: */
1111 file_header_list_rearrange_from_header(fh, fnum);
1119 * Merges the given files into the output file, which can be
1123 merge_files_array(size_t argc, const char **argv, const char *fn_out)
1126 if (argv && fn_out) {
1127 struct file_header **fh;
1131 f_out = openfile(fn_out, "w");
1136 fh = sort_malloc((argc + 1) * sizeof(struct file_header *));
1138 for (i = 0; i < argc; i++)
1139 file_header_init(fh + i, argv[i], (size_t) i);
1141 file_headers_merge(argc, fh, f_out);
1143 for (i = 0; i < argc; i++)
1144 file_header_close(fh + i);
1148 closefile(f_out, fn_out);
1153 * Shrinks the file list until its size smaller than max number of opened files
1156 shrink_file_list(struct file_list *fl)
1159 if ((fl == NULL) || (size_t) (fl->count) < max_open_files)
1162 struct file_list new_fl;
1165 file_list_init(&new_fl, true);
1166 while (indx < fl->count) {
1170 num = fl->count - indx;
1171 fnew = new_tmp_file_name();
1173 if ((size_t) num >= max_open_files)
1174 num = max_open_files - 1;
1175 merge_files_array(num, fl->fns + indx, fnew);
1179 for (i = 0; i < num; i++)
1180 unlink(fl->fns[indx + i]);
1182 file_list_add(&new_fl, fnew, false);
1185 fl->tmp = false; /* already taken care of */
1186 file_list_clean(fl);
1188 fl->count = new_fl.count;
1189 fl->fns = new_fl.fns;
1191 fl->tmp = new_fl.tmp;
1198 * Merge list of files
1201 merge_files(struct file_list *fl, const char *fn_out)
1205 while (shrink_file_list(fl));
1207 merge_files_array(fl->count, fl->fns, fn_out);
1212 get_sort_method_name(int sm)
1215 if (sm == SORT_MERGESORT)
1217 else if (sort_opts_vals.sort_method == SORT_RADIXSORT)
1219 else if (sort_opts_vals.sort_method == SORT_HEAPSORT)
1228 static int sort_qsort(void *list, size_t count, size_t elem_size,
1229 int (*cmp_func)(const void *, const void *))
1232 qsort(list, count, elem_size, cmp_func);
1237 * Sort list of lines and writes it to the file
1240 sort_list_to_file(struct sort_list *list, const char *outfile)
1242 struct sort_mods *sm = &(keys[0].sm);
1244 if (!(sm->Mflag) && !(sm->Rflag) && !(sm->Vflag) &&
1245 !(sm->gflag) && !(sm->hflag) && !(sm->nflag)) {
1246 if ((sort_opts_vals.sort_method == SORT_DEFAULT) && byte_sort)
1247 sort_opts_vals.sort_method = SORT_RADIXSORT;
1249 } else if (sort_opts_vals.sort_method == SORT_RADIXSORT)
1250 err(2, "%s", getstr(9));
1253 * to handle stable sort and the unique cases in the
1254 * right order, we need stable basic algorithm
1256 if (sort_opts_vals.sflag) {
1257 switch (sort_opts_vals.sort_method){
1258 case SORT_MERGESORT:
1260 case SORT_RADIXSORT:
1263 sort_opts_vals.sort_method = SORT_MERGESORT;
1266 errx(2, "%s", getstr(10));
1270 if (sort_opts_vals.sort_method == SORT_DEFAULT)
1271 sort_opts_vals.sort_method = DEFAULT_SORT_ALGORITHM;
1274 printf("sort_method=%s\n",
1275 get_sort_method_name(sort_opts_vals.sort_method));
1277 switch (sort_opts_vals.sort_method){
1278 case SORT_RADIXSORT:
1279 rxsort(list->list, list->count);
1280 sort_list_dump(list, outfile);
1282 case SORT_MERGESORT:
1283 mt_sort(list, mergesort, outfile);
1286 mt_sort(list, heapsort, outfile);
1289 mt_sort(list, sort_qsort, outfile);
1292 mt_sort(list, DEFAULT_SORT_FUNC, outfile);
1297 /******************* MT SORT ************************/
1299 #if defined(SORT_THREADS)
1300 /* semaphore to count threads */
1303 /* current system sort function */
1304 static int (*g_sort_func)(void *, size_t, size_t,
1305 int(*)(const void *, const void *));
1308 * Sort cycle thread (in multi-threaded mode)
1311 mt_sort_thread(void* arg)
1313 struct sort_list *list = arg;
1315 g_sort_func(list->list, list->count, sizeof(struct sort_list_item *),
1316 (int(*)(const void *, const void *)) list_coll);
1324 * Compare sub-lists. Empty sub-lists always go to the end of the list.
1327 sub_list_cmp(struct sort_list *l1, struct sort_list *l2)
1333 if (l1->count == 0) {
1334 return ((l2->count == 0) ? 0 : +1);
1335 } else if (l2->count == 0) {
1340 ret = list_coll(&(l1->list[0]), &(l2->list[0]));
1342 return ((l1->sub_list_pos < l2->sub_list_pos) ?
1350 * Swap two array elements
1353 sub_list_swap(struct sort_list **sl, size_t i1, size_t i2)
1355 struct sort_list *tmp;
1362 /* heap algorithm ==>> */
1365 * See heap sort algorithm
1366 * "Raises" last element to its right place
1369 sub_list_swim(struct sort_list **sl, size_t indx)
1373 size_t parent_index;
1375 parent_index = (indx - 1) >> 1;
1377 if (sub_list_cmp(sl[indx], sl[parent_index]) < 0) {
1378 /* swap child and parent and continue */
1379 sub_list_swap(sl, indx, parent_index);
1380 sub_list_swim(sl, parent_index);
1386 * Sink the top element to its correct position
1389 sub_list_sink(struct sort_list **sl, size_t indx, size_t size)
1391 size_t left_child_index;
1392 size_t right_child_index;
1394 left_child_index = indx + indx + 1;
1395 right_child_index = left_child_index + 1;
1397 if (left_child_index < size) {
1398 size_t min_child_index;
1400 min_child_index = left_child_index;
1402 if ((right_child_index < size) &&
1403 (sub_list_cmp(sl[left_child_index],
1404 sl[right_child_index]) > 0))
1405 min_child_index = right_child_index;
1406 if (sub_list_cmp(sl[indx], sl[min_child_index]) > 0) {
1407 sub_list_swap(sl, indx, min_child_index);
1408 sub_list_sink(sl, min_child_index, size);
1413 /* <<== heap algorithm */
1416 * Adds element to the "right" end
1419 sub_list_push(struct sort_list *s, struct sort_list **sl, size_t size)
1423 sub_list_swim(sl, size - 1);
1426 struct last_printed_item
1428 struct sort_list_item *item;
1432 * Prints the current line of the file
1435 sub_list_header_print(struct sort_list *sl, FILE *f_out,
1436 struct last_printed_item *lp)
1439 if (sl && sl->count && f_out && sl->list[0]->str) {
1440 if (sort_opts_vals.uflag) {
1441 if ((lp->item == NULL) || (list_coll(&(lp->item),
1443 bwsfwrite(sl->list[0]->str, f_out,
1444 sort_opts_vals.zflag);
1445 lp->item = sl->list[0];
1448 bwsfwrite(sl->list[0]->str, f_out,
1449 sort_opts_vals.zflag);
1457 sub_list_next(struct sort_list *sl)
1460 if (sl && sl->count) {
1467 * Merge sub-lists to a file
1470 merge_sub_lists(struct sort_list **sl, size_t n, FILE* f_out)
1472 struct last_printed_item lp;
1475 memset(&lp,0,sizeof(lp));
1477 /* construct the initial list: */
1478 for (i = 0; i < n; i++)
1479 sub_list_push(sl[i], sl, i);
1481 while (sl[0]->count) { /* unfinished lists are always in front */
1482 /* output the smallest line: */
1483 sub_list_header_print(sl[0], f_out, &lp);
1484 /* move to a new line, if possible: */
1485 sub_list_next(sl[0]);
1486 /* re-arrange the list: */
1487 sub_list_sink(sl, 0, n);
1492 * Merge sub-lists to a file
1495 merge_list_parts(struct sort_list **parts, size_t n, const char *fn)
1499 f_out = openfile(fn,"w");
1501 merge_sub_lists(parts, n, f_out);
1503 closefile(f_out, fn);
1506 #endif /* defined(SORT_THREADS) */
1508 * Multi-threaded sort algorithm "driver"
1511 mt_sort(struct sort_list *list,
1512 int(*sort_func)(void *, size_t, size_t, int(*)(const void *, const void *)),
1515 #if defined(SORT_THREADS)
1516 if (nthreads < 2 || list->count < MT_SORT_THRESHOLD) {
1517 size_t nthreads_save = nthreads;
1520 /* if single thread or small data, do simple sort */
1521 sort_func(list->list, list->count,
1522 sizeof(struct sort_list_item *),
1523 (int(*)(const void *, const void *)) list_coll);
1524 sort_list_dump(list, fn);
1525 #if defined(SORT_THREADS)
1526 nthreads = nthreads_save;
1528 /* multi-threaded sort */
1529 struct sort_list **parts;
1530 size_t avgsize, cstart, i;
1532 /* array of sub-lists */
1533 parts = sort_malloc(sizeof(struct sort_list*) * nthreads);
1535 avgsize = list->count / nthreads;
1537 /* set global system sort function */
1538 g_sort_func = sort_func;
1541 for (i = 0; i < nthreads; ++i) {
1544 parts[i] = sort_malloc(sizeof(struct sort_list));
1545 parts[i]->list = list->list + cstart;
1546 parts[i]->memsize = 0;
1547 parts[i]->sub_list_pos = i;
1549 sz = (i == nthreads - 1) ? list->count - cstart :
1552 parts[i]->count = sz;
1554 parts[i]->size = parts[i]->count;
1559 /* init threads counting semaphore */
1560 sem_init(&mtsem, 0, 0);
1563 for (i = 0; i < nthreads; ++i) {
1565 pthread_attr_t attr;
1567 pthread_attr_init(&attr);
1568 pthread_attr_setdetachstate(&attr, PTHREAD_DETACHED);
1571 int res = pthread_create(&pth, &attr,
1572 mt_sort_thread, parts[i]);
1576 if (errno == EAGAIN) {
1583 pthread_attr_destroy(&attr);
1586 /* wait for threads completion */
1587 for (i = 0; i < nthreads; ++i) {
1590 /* destroy the semaphore - we do not need it anymore */
1591 sem_destroy(&mtsem);
1593 /* merge sorted sub-lists to the file */
1594 merge_list_parts(parts, nthreads, fn);
1596 /* free sub-lists data */
1597 for (i = 0; i < nthreads; ++i) {
1598 sort_free(parts[i]);
1602 #endif /* defined(SORT_THREADS) */