Annotation of src/usr.bin/tsort/tsort.c, Revision 1.31
1.31 ! espie 1: /* $OpenBSD: tsort.c,v 1.30 2015/10/10 15:47:22 deraadt Exp $ */
1.15 espie 2: /* ex:ts=8 sw=4:
1.1 deraadt 3: *
1.19 espie 4: * Copyright (c) 1999-2004 Marc Espie <espie@openbsd.org>
1.1 deraadt 5: *
1.19 espie 6: * Permission to use, copy, modify, and distribute this software for any
7: * purpose with or without fee is hereby granted, provided that the above
8: * copyright notice and this permission notice appear in all copies.
9: *
10: * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11: * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12: * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13: * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14: * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15: * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16: * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
1.1 deraadt 17: */
18:
1.5 espie 19: #include <assert.h>
1.1 deraadt 20: #include <ctype.h>
21: #include <err.h>
1.5 espie 22: #include <limits.h>
23: #include <stddef.h>
1.1 deraadt 24: #include <stdio.h>
1.20 espie 25: #include <stdint.h>
1.1 deraadt 26: #include <stdlib.h>
27: #include <string.h>
1.5 espie 28: #include <sysexits.h>
1.1 deraadt 29: #include <unistd.h>
1.20 espie 30: #include <ohash.h>
1.1 deraadt 31:
1.15 espie 32: /* The complexity of topological sorting is O(e), where e is the
1.5 espie 33: * size of input. While reading input, vertices have to be identified,
34: * thus add the complexity of e keys retrieval among v keys using
1.15 espie 35: * an appropriate data structure. This program uses open double hashing
36: * for that purpose. See Knuth for the expected complexity of double
1.5 espie 37: * hashing (Brent variation should probably be used if v << e, as a user
38: * option).
39: *
40: * The algorithm used for longest cycle reporting is accurate, but somewhat
41: * expensive. It may need to build all free paths of the graph (a free
42: * path is a path that never goes twice through the same node), whose
1.15 espie 43: * number can be as high as O(2^e). Usually, the number of free paths is
1.5 espie 44: * much smaller though. This program's author does not believe that a
45: * significantly better worst-case complexity algorithm exists.
46: *
47: * In case of a hints file, the set of minimal nodes is maintained as a
48: * heap. The resulting complexity is O(e+v log v) for the worst case.
49: * The average should actually be near O(e).
50: *
1.8 espie 51: * If the hints file is incomplete, there is some extra complexity incurred
52: * by make_transparent, which does propagate order values to unmarked
53: * nodes. In the worst case, make_transparent is O(e u),
54: * where u is the number of originally unmarked nodes.
55: * In practice, it is much faster.
56: *
1.5 espie 57: * The simple topological sort algorithm detects cycles. This program
1.15 espie 58: * goes further, breaking cycles through the use of simple heuristics.
59: * Each cycle break checks the whole set of nodes, hence if c cycles break
1.5 espie 60: * are needed, this is an extra cost of O(c v).
1.1 deraadt 61: *
1.5 espie 62: * Possible heuristics are as follows:
63: * - break cycle at node with lowest number of predecessors (default case),
64: * - break longest cycle at node with lowest number of predecessors,
65: * - break cycle at next node from the hints file.
1.1 deraadt 66: *
1.5 espie 67: * Except for the hints file case, which sets an explicit constraint on
1.15 espie 68: * which cycle to break, those heuristics locally result in the smallest
1.5 espie 69: * number of broken edges.
1.1 deraadt 70: *
1.5 espie 71: * Those are admittedly greedy strategies, as is the selection of the next
72: * node from the hints file amongst equivalent candidates that is used for
73: * `stable' topological sorting.
1.1 deraadt 74: */
1.5 espie 75:
76: #ifdef __GNUC__
77: #define UNUSED __attribute__((unused))
78: #else
79: #define UNUSED
80: #endif
81:
82: struct node;
83:
84: /* The set of arcs from a given node is stored as a linked list. */
85: struct link {
86: struct link *next;
87: struct node *node;
88: };
89:
1.7 espie 90: #define NO_ORDER UINT_MAX
91:
1.5 espie 92: struct node {
1.15 espie 93: unsigned int refs; /* Number of arcs left, coming into this node.
94: * Note that nodes with a null count can't
1.5 espie 95: * be part of cycles. */
96: struct link *arcs; /* List of forward arcs. */
97:
98: unsigned int order; /* Order of nodes according to a hint file. */
99:
100: /* Cycle detection algorithms build a free path of nodes. */
101: struct node *from; /* Previous node in the current path. */
102:
103: unsigned int mark; /* Mark processed nodes in cycle discovery. */
104: struct link *traverse; /* Next link to traverse when backtracking. */
105: char k[1]; /* Name of this node. */
1.1 deraadt 106: };
107:
1.5 espie 108: #define HASH_START 9
109:
110: struct array {
111: unsigned int entries;
112: struct node **t;
113: };
114:
1.12 millert 115: static void nodes_init(struct ohash *);
116: static struct node *node_lookup(struct ohash *, const char *, const char *);
117: static void usage(void);
118: static struct node *new_node(const char *, const char *);
1.5 espie 119:
1.15 espie 120: static unsigned int read_pairs(FILE *, struct ohash *, int,
1.13 millert 121: const char *, unsigned int, int);
1.12 millert 122: static void split_nodes(struct ohash *, struct array *, struct array *);
123: static void make_transparent(struct ohash *);
124: static void insert_arc(struct node *, struct node *);
1.5 espie 125:
126: #ifdef DEBUG
1.12 millert 127: static void dump_node(struct node *);
128: static void dump_array(struct array *);
129: static void dump_hash(struct ohash *);
1.5 espie 130: #endif
1.15 espie 131: static unsigned int read_hints(FILE *, struct ohash *, int,
1.13 millert 132: const char *, unsigned int);
1.12 millert 133: static struct node *find_smallest_node(struct array *);
134: static struct node *find_good_cycle_break(struct array *);
135: static void print_cycle(struct array *);
136: static struct node *find_cycle_from(struct node *, struct array *);
137: static struct node *find_predecessor(struct array *, struct node *);
138: static unsigned int traverse_node(struct node *, unsigned int, struct array *);
139: static struct node *find_longest_cycle(struct array *, struct array *);
1.29 espie 140: static struct node *find_normal_cycle(struct array *, struct array *);
1.12 millert 141:
142: static void heap_down(struct array *, unsigned int);
143: static void heapify(struct array *, int);
144: static struct node *dequeue(struct array *);
145: static void enqueue(struct array *, struct node *);
1.5 espie 146:
1.1 deraadt 147:
1.5 espie 148:
1.23 espie 149: static void *hash_calloc(size_t, size_t, void *);
150: static void hash_free(void *, void *);
1.12 millert 151: static void* entry_alloc(size_t, void *);
1.24 deraadt 152: static void *ereallocarray(void *, size_t, size_t);
1.12 millert 153: static void *emem(void *);
1.5 espie 154: #define DEBUG_TRAVERSE 0
1.15 espie 155: static struct ohash_info node_info = {
1.23 espie 156: offsetof(struct node, k), NULL, hash_calloc, hash_free, entry_alloc };
1.29 espie 157: static void parse_args(int, char *[], struct ohash *);
158: static int tsort(struct ohash *);
159:
160: static int quiet_flag, long_flag,
161: warn_flag, hints_flag, verbose_flag;
1.5 espie 162:
163:
1.12 millert 164: int main(int, char *[]);
1.5 espie 165:
166: /***
1.15 espie 167: *** Memory handling.
1.5 espie 168: ***/
169:
170: static void *
1.14 espie 171: emem(void *p)
1.5 espie 172: {
173: if (p)
174: return p;
175: else
176: errx(EX_SOFTWARE, "Memory exhausted");
177: }
178:
179: static void *
1.23 espie 180: hash_calloc(size_t n, size_t s, void *u UNUSED)
1.5 espie 181: {
1.23 espie 182: return emem(calloc(n, s));
1.5 espie 183: }
184:
185: static void
1.23 espie 186: hash_free(void *p, void *u UNUSED)
1.5 espie 187: {
188: free(p);
189: }
190:
191: static void *
1.14 espie 192: entry_alloc(size_t s, void *u UNUSED)
1.5 espie 193: {
1.24 deraadt 194: return ereallocarray(NULL, 1, s);
1.5 espie 195: }
196:
197: static void *
1.24 deraadt 198: ereallocarray(void *p, size_t n, size_t s)
1.1 deraadt 199: {
1.24 deraadt 200: return emem(reallocarray(p, n, s));
1.5 espie 201: }
1.1 deraadt 202:
1.5 espie 203:
1.15 espie 204: /***
1.5 espie 205: *** Hash table.
206: ***/
207:
208: /* Inserting and finding nodes in the hash structure.
209: * We handle interval strings for efficiency wrt fgetln. */
210: static struct node *
1.14 espie 211: new_node(const char *start, const char *end)
1.5 espie 212: {
213: struct node *n;
1.1 deraadt 214:
1.5 espie 215: n = ohash_create_entry(&node_info, start, &end);
216: n->from = NULL;
217: n->arcs = NULL;
218: n->refs = 0;
219: n->mark = 0;
1.7 espie 220: n->order = NO_ORDER;
1.5 espie 221: n->traverse = NULL;
222: return n;
223: }
1.1 deraadt 224:
225:
1.15 espie 226: static void
1.14 espie 227: nodes_init(struct ohash *h)
1.5 espie 228: {
229: ohash_init(h, HASH_START, &node_info);
1.1 deraadt 230: }
231:
1.5 espie 232: static struct node *
1.14 espie 233: node_lookup(struct ohash *h, const char *start, const char *end)
1.1 deraadt 234: {
1.5 espie 235: unsigned int i;
236: struct node * n;
1.1 deraadt 237:
1.5 espie 238: i = ohash_qlookupi(h, start, &end);
1.1 deraadt 239:
1.5 espie 240: n = ohash_find(h, i);
241: if (n == NULL)
242: n = ohash_insert(h, i, new_node(start, end));
243: return n;
244: }
1.25 jasper 245:
1.5 espie 246: #ifdef DEBUG
247: static void
1.14 espie 248: dump_node(struct node *n)
1.5 espie 249: {
250: struct link *l;
251:
252: if (n->refs == 0)
1.1 deraadt 253: return;
1.8 espie 254: printf("%s (%u/%u): ", n->k, n->order, n->refs);
1.5 espie 255: for (l = n->arcs; l != NULL; l = l->next)
256: if (n->refs != 0)
1.8 espie 257: printf("%s(%u/%u) ", l->node->k, l->node->order, l->node->refs);
1.5 espie 258: putchar('\n');
259: }
1.1 deraadt 260:
1.5 espie 261: static void
1.14 espie 262: dump_array(struct array *a)
1.5 espie 263: {
264: unsigned int i;
1.1 deraadt 265:
1.5 espie 266: for (i = 0; i < a->entries; i++)
267: dump_node(a->t[i]);
268: }
1.25 jasper 269:
1.15 espie 270: static void
1.14 espie 271: dump_hash(struct ohash *h)
1.5 espie 272: {
273: unsigned int i;
274: struct node *n;
275:
276: for (n = ohash_first(h, &i); n != NULL; n = ohash_next(h, &i))
277: dump_node(n);
278: }
279: #endif
1.25 jasper 280:
1.5 espie 281: /***
282: *** Reading data.
283: ***/
284:
1.15 espie 285: static void
1.14 espie 286: insert_arc(struct node *a, struct node *b)
1.5 espie 287: {
288: struct link *l;
289:
290: /* Check that this arc is not already present. */
291: for (l = a->arcs; l != NULL; l = l->next) {
292: if (l->node == b)
1.1 deraadt 293: return;
1.5 espie 294: }
295: b->refs++;
1.24 deraadt 296: l = ereallocarray(NULL, 1, sizeof(struct link));
1.5 espie 297: l->node = b;
298: l->next = a->arcs;
299: a->arcs = l;
300: }
1.1 deraadt 301:
1.7 espie 302: static unsigned int
1.15 espie 303: read_pairs(FILE *f, struct ohash *h, int reverse, const char *name,
1.14 espie 304: unsigned int order, int hint)
1.5 espie 305: {
306: int toggle;
307: struct node *a;
308: size_t size;
309: char *str;
310:
311: toggle = 1;
312: a = NULL;
1.25 jasper 313:
1.5 espie 314: while ((str = fgetln(f, &size)) != NULL) {
315: char *sentinel;
316:
317: sentinel = str + size;
318: for (;;) {
319: char *e;
320:
1.22 deraadt 321: while (str < sentinel &&
322: isspace((unsigned char)*str))
1.5 espie 323: str++;
324: if (str == sentinel)
325: break;
1.22 deraadt 326: for (e = str;
327: e < sentinel && !isspace((unsigned char)*e); e++)
1.5 espie 328: continue;
329: if (toggle) {
330: a = node_lookup(h, str, e);
1.8 espie 331: if (a->order == NO_ORDER && hint)
1.7 espie 332: a->order = order++;
1.5 espie 333: } else {
334: struct node *b;
335:
336: b = node_lookup(h, str, e);
337: assert(a != NULL);
338: if (b != a) {
1.15 espie 339: if (reverse)
1.5 espie 340: insert_arc(b, a);
1.15 espie 341: else
1.5 espie 342: insert_arc(a, b);
343: }
344: }
345: toggle = !toggle;
346: str = e;
347: }
348: }
349: if (toggle == 0)
1.21 jmc 350: errx(EX_DATAERR, "odd number of node names in %s", name);
1.5 espie 351: if (!feof(f))
352: err(EX_IOERR, "error reading %s", name);
1.7 espie 353: return order;
1.5 espie 354: }
1.1 deraadt 355:
1.7 espie 356: static unsigned int
1.15 espie 357: read_hints(FILE *f, struct ohash *h, int quiet, const char *name,
1.14 espie 358: unsigned int order)
1.5 espie 359: {
360: char *str;
361: size_t size;
362:
363: while ((str = fgetln(f, &size)) != NULL) {
364: char *sentinel;
365:
366: sentinel = str + size;
367: for (;;) {
368: char *e;
369: struct node *a;
370:
1.22 deraadt 371: while (str < sentinel && isspace((unsigned char)*str))
1.5 espie 372: str++;
373: if (str == sentinel)
374: break;
1.22 deraadt 375: for (e = str;
376: e < sentinel && !isspace((unsigned char)*e); e++)
1.5 espie 377: continue;
378: a = node_lookup(h, str, e);
1.7 espie 379: if (a->order != NO_ORDER) {
1.6 espie 380: if (!quiet)
381: warnx(
1.5 espie 382: "duplicate node %s in hints file %s",
383: a->k, name);
1.7 espie 384: } else
385: a->order = order++;
1.5 espie 386: str = e;
387: }
388: }
1.7 espie 389: return order;
1.5 espie 390: }
391:
392:
393: /***
1.15 espie 394: *** Standard heap handling routines.
1.5 espie 395: ***/
396:
1.15 espie 397: static void
1.14 espie 398: heap_down(struct array *h, unsigned int i)
1.5 espie 399: {
400: unsigned int j;
401: struct node *swap;
402:
403: for (; (j=2*i+1) < h->entries; i = j) {
404: if (j+1 < h->entries && h->t[j+1]->order < h->t[j]->order)
405: j++;
406: if (h->t[i]->order <= h->t[j]->order)
407: break;
408: swap = h->t[i];
409: h->t[i] = h->t[j];
410: h->t[j] = swap;
1.1 deraadt 411: }
1.5 espie 412: }
413:
1.15 espie 414: static void
1.14 espie 415: heapify(struct array *h, int verbose)
1.5 espie 416: {
417: unsigned int i;
1.1 deraadt 418:
1.6 espie 419: for (i = h->entries; i != 0;) {
1.8 espie 420: if (h->t[--i]->order == NO_ORDER && verbose)
1.6 espie 421: warnx("node %s absent from hints file", h->t[i]->k);
422: heap_down(h, i);
423: }
1.5 espie 424: }
425:
426: #define DEQUEUE(h) ( hints_flag ? dequeue(h) : (h)->t[--(h)->entries] )
1.1 deraadt 427:
1.5 espie 428: static struct node *
1.14 espie 429: dequeue(struct array *h)
1.5 espie 430: {
431: struct node *n;
432:
433: if (h->entries == 0)
434: n = NULL;
435: else {
436: n = h->t[0];
437: if (--h->entries != 0) {
438: h->t[0] = h->t[h->entries];
439: heap_down(h, 0);
440: }
441: }
442: return n;
1.1 deraadt 443: }
1.25 jasper 444:
1.5 espie 445: #define ENQUEUE(h, n) do { \
446: if (hints_flag) \
447: enqueue((h), (n)); \
448: else \
449: (h)->t[(h)->entries++] = (n); \
450: } while(0);
451:
1.15 espie 452: static void
1.14 espie 453: enqueue(struct array *h, struct node *n)
1.5 espie 454: {
455: unsigned int i, j;
456: struct node *swap;
1.1 deraadt 457:
1.5 espie 458: h->t[h->entries++] = n;
459: for (i = h->entries-1; i > 0; i = j) {
460: j = (i-1)/2;
461: if (h->t[j]->order < h->t[i]->order)
462: break;
463: swap = h->t[j];
464: h->t[j] = h->t[i];
465: h->t[i] = swap;
466: }
467: }
1.1 deraadt 468:
1.8 espie 469: /* Nodes without order should not hinder direct dependencies.
1.15 espie 470: * Iterate until no nodes are left.
1.8 espie 471: */
472: static void
1.14 espie 473: make_transparent(struct ohash *hash)
1.8 espie 474: {
475: struct node *n;
476: unsigned int i;
477: struct link *l;
478: int adjusted;
479: int bad;
480: unsigned int min;
481:
482: /* first try to solve complete nodes */
483: do {
484: adjusted = 0;
485: bad = 0;
1.15 espie 486: for (n = ohash_first(hash, &i); n != NULL;
1.8 espie 487: n = ohash_next(hash, &i)) {
488: if (n->order == NO_ORDER) {
489: min = NO_ORDER;
490:
491: for (l = n->arcs; l != NULL; l = l->next) {
492: /* unsolved node -> delay resolution */
493: if (l->node->order == NO_ORDER) {
494: bad = 1;
495: break;
496: } else if (l->node->order < min)
497: min = l->node->order;
498: }
499: if (min < NO_ORDER && l == NULL) {
500: n->order = min;
501: adjusted = 1;
502: }
503: }
504: }
505:
506: } while (adjusted);
507:
508: /* then, if incomplete nodes are left, do them */
509: if (bad) do {
510: adjusted = 0;
1.15 espie 511: for (n = ohash_first(hash, &i); n != NULL;
1.8 espie 512: n = ohash_next(hash, &i))
513: if (n->order == NO_ORDER)
514: for (l = n->arcs; l != NULL; l = l->next)
515: if (l->node->order < n->order) {
516: n->order = l->node->order;
517: adjusted = 1;
518: }
519: } while (adjusted);
520: }
521:
1.5 espie 522:
523: /***
524: *** Search through hash array for nodes.
525: ***/
526:
527: /* Split nodes into unrefed nodes/live nodes. */
528: static void
1.14 espie 529: split_nodes(struct ohash *hash, struct array *heap, struct array *remaining)
1.1 deraadt 530: {
531:
1.5 espie 532: struct node *n;
533: unsigned int i;
534:
1.24 deraadt 535: heap->t = ereallocarray(NULL, ohash_entries(hash),
536: sizeof(struct node *));
537: remaining->t = ereallocarray(NULL, ohash_entries(hash),
538: sizeof(struct node *));
1.5 espie 539: heap->entries = 0;
540: remaining->entries = 0;
541:
542: for (n = ohash_first(hash, &i); n != NULL; n = ohash_next(hash, &i)) {
1.15 espie 543: if (n->refs == 0)
1.5 espie 544: heap->t[heap->entries++] = n;
545: else
1.9 espie 546: remaining->t[remaining->entries++] = n;
1.5 espie 547: }
1.1 deraadt 548: }
549:
1.5 espie 550: /* Good point to break a cycle: live node with as few refs as possible. */
551: static struct node *
1.14 espie 552: find_good_cycle_break(struct array *h)
1.5 espie 553: {
554: unsigned int i;
1.15 espie 555: unsigned int best;
556: struct node *u;
1.5 espie 557:
558: best = UINT_MAX;
559: u = NULL;
560:
561: assert(h->entries != 0);
562: for (i = 0; i < h->entries; i++) {
563: struct node *n = h->t[i];
1.25 jasper 564: /* No need to look further. */
1.5 espie 565: if (n->refs == 1)
566: return n;
567: if (n->refs != 0 && n->refs < best) {
568: best = n->refs;
569: u = n;
570: }
571: }
572: assert(u != NULL);
573: return u;
574: }
1.25 jasper 575:
1.5 espie 576: /* Retrieve the node with the smallest order. */
1.15 espie 577: static struct node *
1.14 espie 578: find_smallest_node(struct array *h)
1.5 espie 579: {
580: unsigned int i;
581: unsigned int best;
582: struct node *u;
583:
584: best = UINT_MAX;
585: u = NULL;
586:
587: assert(h->entries != 0);
588: for (i = 0; i < h->entries; i++) {
589: struct node *n = h->t[i];
590: if (n->refs != 0 && n->order < best) {
591: best = n->order;
592: u = n;
593: }
594: }
595: assert(u != NULL);
596: return u;
597: }
598:
599:
600: /***
601: *** Graph algorithms.
602: ***/
603:
1.15 espie 604: /* Explore the nodes reachable from i to find a cycle, store it in c.
1.10 espie 605: * This may fail. */
1.15 espie 606: static struct node *
1.14 espie 607: find_cycle_from(struct node *i, struct array *c)
1.5 espie 608: {
609: struct node *n;
610:
611: n = i;
612: /* XXX Previous cycle findings may have left this pointer non-null. */
613: i->from = NULL;
614:
615: for (;;) {
616: /* Note that all marks are reversed before this code exits. */
617: n->mark = 1;
1.15 espie 618: if (n->traverse)
1.5 espie 619: n->traverse = n->traverse->next;
620: else
621: n->traverse = n->arcs;
622: /* Skip over dead nodes. */
623: while (n->traverse && n->traverse->node->refs == 0)
624: n->traverse = n->traverse->next;
625: if (n->traverse) {
626: struct node *go = n->traverse->node;
627:
628: if (go->mark) {
1.10 espie 629: c->entries = 0;
630: for (; n != NULL && n != go; n = n->from) {
631: c->t[c->entries++] = n;
632: n->mark = 0;
1.1 deraadt 633: }
1.10 espie 634: for (; n != NULL; n = n->from)
635: n->mark = 0;
636: c->t[c->entries++] = go;
637: return go;
1.5 espie 638: } else {
639: go->from = n;
640: n = go;
1.1 deraadt 641: }
1.5 espie 642: } else {
643: n->mark = 0;
644: n = n->from;
1.15 espie 645: if (n == NULL)
1.10 espie 646: return NULL;
1.5 espie 647: }
648: }
649: }
1.1 deraadt 650:
1.5 espie 651: /* Find a live predecessor of node n. This is a slow routine, as it needs
652: * to go through the whole array, but it is not needed often.
653: */
654: static struct node *
1.14 espie 655: find_predecessor(struct array *a, struct node *n)
1.5 espie 656: {
657: unsigned int i;
658:
659: for (i = 0; i < a->entries; i++) {
660: struct node *m;
1.1 deraadt 661:
1.5 espie 662: m = a->t[i];
663: if (m->refs != 0) {
664: struct link *l;
665:
666: for (l = m->arcs; l != NULL; l = l->next)
667: if (l->node == n)
668: return m;
1.1 deraadt 669: }
1.5 espie 670: }
671: assert(1 == 0);
672: return NULL;
673: }
674:
675: /* Traverse all strongly connected components reachable from node n.
676: Start numbering them at o. Return the maximum order reached.
677: Update the largest cycle found so far.
678: */
1.15 espie 679: static unsigned int
1.14 espie 680: traverse_node(struct node *n, unsigned int o, struct array *c)
1.5 espie 681: {
682: unsigned int min, max;
683:
684: n->from = NULL;
685: min = o;
686: max = ++o;
687:
688: for (;;) {
689: n->mark = o;
690: if (DEBUG_TRAVERSE)
1.11 espie 691: printf("%s(%u) ", n->k, n->mark);
1.5 espie 692: /* Find next arc to explore. */
1.15 espie 693: if (n->traverse)
1.5 espie 694: n->traverse = n->traverse->next;
695: else
696: n->traverse = n->arcs;
697: /* Skip over dead nodes. */
698: while (n->traverse && n->traverse->node->refs == 0)
699: n->traverse = n->traverse->next;
700: /* If arc left. */
701: if (n->traverse) {
702: struct node *go;
703:
704: go = n->traverse->node;
1.15 espie 705: /* Optimisation: if go->mark < min, we already
1.5 espie 706: * visited this strongly-connected component in
707: * a previous pass. Hence, this can yield no new
708: * cycle. */
709:
710: /* Not part of the current path: go for it. */
711: if (go->mark == 0 || go->mark == min) {
712: go->from = n;
713: n = go;
714: o++;
715: if (o > max)
716: max = o;
717: /* Part of the current path: check cycle length. */
718: } else if (go->mark > min) {
719: if (DEBUG_TRAVERSE)
720: printf("%d\n", o - go->mark + 1);
721: if (o - go->mark + 1 > c->entries) {
722: struct node *t;
723: unsigned int i;
724:
725: c->entries = o - go->mark + 1;
726: i = 0;
727: c->t[i++] = go;
728: for (t = n; t != go; t = t->from)
729: c->t[i++] = t;
1.1 deraadt 730: }
1.5 espie 731: }
1.1 deraadt 732:
1.5 espie 733: /* No arc left: backtrack. */
734: } else {
735: n->mark = min;
736: n = n->from;
737: if (!n)
738: return max;
1.25 jasper 739: o--;
1.5 espie 740: }
1.1 deraadt 741: }
742: }
743:
1.5 espie 744: static void
1.14 espie 745: print_cycle(struct array *c)
1.5 espie 746: {
747: unsigned int i;
748:
749: /* Printing in reverse order, since cycle discoveries finds reverse
750: * edges. */
751: for (i = c->entries; i != 0;) {
752: i--;
753: warnx("%s", c->t[i]->k);
754: }
1.1 deraadt 755: }
756:
1.5 espie 757: static struct node *
1.14 espie 758: find_longest_cycle(struct array *h, struct array *c)
1.5 espie 759: {
760: unsigned int i;
761: unsigned int o;
762: unsigned int best;
763: struct node *n;
764: static int notfirst = 0;
765:
766: assert(h->entries != 0);
767:
768: /* No cycle found yet. */
769: c->entries = 0;
770:
771: /* Reset the set of marks, except the first time around. */
772: if (notfirst) {
1.15 espie 773: for (i = 0; i < h->entries; i++)
1.5 espie 774: h->t[i]->mark = 0;
775: } else
776: notfirst = 1;
777:
778: o = 0;
779:
780: /* Traverse the array. Each unmarked, live node heralds a
781: * new set of strongly connected components. */
782: for (i = 0; i < h->entries; i++) {
783: n = h->t[i];
784: if (n->refs != 0 && n->mark == 0) {
785: /* Each call to traverse_node uses a separate
786: * interval of numbers to mark nodes. */
787: o++;
788: o = traverse_node(n, o, c);
789: }
790: }
1.25 jasper 791:
1.5 espie 792: assert(c->entries != 0);
793: n = c->t[0];
794: best = n->refs;
795: for (i = 0; i < c->entries; i++) {
796: if (c->t[i]->refs < best) {
797: n = c->t[i];
798: best = n->refs;
799: }
800: }
801: return n;
802: }
1.1 deraadt 803:
1.29 espie 804: static struct node *
805: find_normal_cycle(struct array *h, struct array *c)
806: {
807: struct node *b, *n;
808:
809: if (hints_flag)
810: n = find_smallest_node(h);
811: else
812: n = find_good_cycle_break(h);
813: while ((b = find_cycle_from(n, c)) == NULL)
814: n = find_predecessor(h, n);
815: return b;
816: }
817:
1.5 espie 818:
819: #define plural(n) ((n) > 1 ? "s" : "")
1.1 deraadt 820:
1.29 espie 821: static void
822: parse_args(int argc, char *argv[], struct ohash *pairs)
1.5 espie 823: {
1.29 espie 824: int c;
1.7 espie 825: unsigned int order;
1.29 espie 826: int reverse_flag;
1.31 ! espie 827: char **files;
! 828: int i, j;
1.7 espie 829:
1.31 ! espie 830: i = 0;
1.5 espie 831:
1.15 espie 832: reverse_flag = quiet_flag = long_flag =
1.5 espie 833: warn_flag = hints_flag = verbose_flag = 0;
1.31 ! espie 834: /* argc is good enough, as we start at argv[1] */
! 835: files = ereallocarray(NULL, argc, sizeof (char *));
1.29 espie 836: while ((c = getopt(argc, argv, "h:flqrvw")) != -1) {
837: switch(c) {
1.31 ! espie 838: case 'h':
! 839: files[i++] = optarg;
1.29 espie 840: hints_flag = 1;
841: break;
842: /*FALLTHRU*/
843: case 'f':
844: hints_flag = 2;
845: break;
846: case 'l':
847: long_flag = 1;
848: break;
849: case 'q':
850: quiet_flag = 1;
851: break;
852: case 'r':
853: reverse_flag = 1;
854: break;
855: case 'v':
856: verbose_flag = 1;
857: break;
858: case 'w':
859: warn_flag = 1;
860: break;
861: default:
862: usage();
863: }
864: }
1.1 deraadt 865:
1.29 espie 866: argc -= optind;
867: argv += optind;
1.1 deraadt 868:
1.5 espie 869: switch(argc) {
1.31 ! espie 870: case 1:
! 871: files[i++] = argv[0];
! 872: break;
! 873: case 0:
! 874: break;
! 875: default:
! 876: usage();
! 877: }
! 878:
! 879: files[i] = NULL;
! 880:
! 881: nodes_init(pairs);
! 882: order = 0;
! 883:
! 884: for (j = 0; j != i-argc; j++) {
! 885: FILE *f;
! 886:
! 887: f = fopen(files[j], "r");
! 888: if (f == NULL)
! 889: err(EX_NOINPUT, "Can't open hint file %s", files[i]);
! 890: order = read_hints(f, pairs, quiet_flag, files[i], order);
! 891: fclose(f);
! 892: }
! 893: free(files);
! 894:
! 895: if (argc == 1) {
1.5 espie 896: FILE *f;
897:
898: f = fopen(argv[0], "r");
899: if (f == NULL)
1.27 espie 900: err(EX_NOINPUT, "Can't open file %s", argv[0]);
1.29 espie 901: order = read_pairs(f, pairs, reverse_flag, argv[0], order,
1.8 espie 902: hints_flag == 2);
1.5 espie 903: fclose(f);
1.31 ! espie 904: } else {
1.29 espie 905: order = read_pairs(stdin, pairs, reverse_flag, "stdin",
1.8 espie 906: order, hints_flag == 2);
1.5 espie 907: }
1.29 espie 908: }
1.1 deraadt 909:
1.29 espie 910: static int
911: tsort(struct ohash *pairs)
912: {
1.5 espie 913: struct array aux; /* Unrefed nodes/cycle reporting. */
914: struct array remaining;
915: unsigned int broken_arcs, broken_cycles;
916: unsigned int left;
917:
918: broken_arcs = 0;
919: broken_cycles = 0;
920:
1.8 espie 921: if (hints_flag)
1.29 espie 922: make_transparent(pairs);
923: split_nodes(pairs, &aux, &remaining);
924: ohash_delete(pairs);
1.5 espie 925:
926: if (hints_flag)
1.6 espie 927: heapify(&aux, verbose_flag);
1.5 espie 928:
929: left = remaining.entries + aux.entries;
930: while (left != 0) {
931:
932: /* Standard topological sort. */
933: while (aux.entries) {
934: struct link *l;
935: struct node *n;
936:
937: n = DEQUEUE(&aux);
938: printf("%s\n", n->k);
939: left--;
940: /* We can't free nodes, as we don't know which
941: * entry we can remove in the hash table. We
1.15 espie 942: * rely on refs == 0 to recognize live nodes.
1.5 espie 943: * Decrease ref count of live nodes, enter new
944: * candidates into the unrefed list. */
1.15 espie 945: for (l = n->arcs; l != NULL; l = l->next)
946: if (l->node->refs != 0 &&
1.5 espie 947: --l->node->refs == 0) {
948: ENQUEUE(&aux, l->node);
949: }
950: }
951: /* There are still cycles to break. */
952: if (left != 0) {
953: struct node *n;
954:
955: broken_cycles++;
956: /* XXX Simple cycle detection and long cycle
957: * detection are mutually exclusive. */
958:
1.29 espie 959: if (long_flag)
1.5 espie 960: n = find_longest_cycle(&remaining, &aux);
1.29 espie 961: else
962: n = find_normal_cycle(&remaining, &aux);
1.5 espie 963:
964: if (!quiet_flag) {
965: warnx("cycle in data");
966: print_cycle(&aux);
967: }
968:
969: if (verbose_flag)
970: warnx("%u edge%s broken", n->refs,
971: plural(n->refs));
972: broken_arcs += n->refs;
973: n->refs = 0;
974: /* Reinitialization, cycle reporting uses aux. */
975: aux.t[0] = n;
976: aux.entries = 1;
977: }
978: }
979: if (verbose_flag && broken_cycles != 0)
980: warnx("%u cycle%s broken, for a total of %u edge%s",
981: broken_cycles, plural(broken_cycles),
982: broken_arcs, plural(broken_arcs));
1.15 espie 983: if (warn_flag)
1.29 espie 984: return (broken_cycles < 256 ? broken_cycles : 255);
1.5 espie 985: else
1.29 espie 986: return (EX_OK);
987: }
988:
989: int
990: main(int argc, char *argv[])
991: {
992: struct ohash pairs;
1.30 deraadt 993:
994: if (pledge("stdio rpath", NULL) == -1)
995: err(1, "pledge");
1.29 espie 996:
997: parse_args(argc, argv, &pairs);
998: return tsort(&pairs);
1.1 deraadt 999: }
1000:
1.5 espie 1001:
1002: extern char *__progname;
1003:
1004: static void
1.16 deraadt 1005: usage(void)
1.1 deraadt 1006: {
1.18 espie 1007: fprintf(stderr, "Usage: %s [-flqrvw] [-h file] [file]\n", __progname);
1.5 espie 1008: exit(EX_USAGE);
1.1 deraadt 1009: }