Annotation of src/usr.bin/window/compress.c, Revision 1.3
1.3 ! downsj 1: /* $OpenBSD$ */
1.1 deraadt 2: /* $NetBSD: compress.c,v 1.3 1995/09/28 10:34:13 tls Exp $ */
3:
4: /*
5: * Copyright (c) 1989, 1993
6: * The Regents of the University of California. All rights reserved.
7: *
8: * This code is derived from software contributed to Berkeley by
9: * Edward Wang at The University of California, Berkeley.
10: *
11: * Redistribution and use in source and binary forms, with or without
12: * modification, are permitted provided that the following conditions
13: * are met:
14: * 1. Redistributions of source code must retain the above copyright
15: * notice, this list of conditions and the following disclaimer.
16: * 2. Redistributions in binary form must reproduce the above copyright
17: * notice, this list of conditions and the following disclaimer in the
18: * documentation and/or other materials provided with the distribution.
19: * 3. All advertising materials mentioning features or use of this software
20: * must display the following acknowledgement:
21: * This product includes software developed by the University of
22: * California, Berkeley and its contributors.
23: * 4. Neither the name of the University nor the names of its contributors
24: * may be used to endorse or promote products derived from this software
25: * without specific prior written permission.
26: *
27: * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30: * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37: * SUCH DAMAGE.
38: */
39:
40: #ifndef lint
41: #if 0
42: static char sccsid[] = "@(#)compress.c 8.1 (Berkeley) 6/6/93";
43: #else
1.3 ! downsj 44: static char rcsid[] = "$OpenBSD$";
1.1 deraadt 45: #endif
46: #endif /* not lint */
1.3 ! downsj 47:
! 48: #include <stdlib.h>
1.1 deraadt 49:
50: #include "ww.h"
51: #include "tt.h"
52:
53: /* special */
54: #include <stdio.h>
55: #include <fcntl.h>
56: int cc_trace = 0;
57: FILE *cc_trace_fp;
58:
59: /* tunable parameters */
60:
61: int cc_reverse = 1;
62: int cc_sort = 0;
63: int cc_chop = 0;
64:
65: int cc_token_max = 8; /* <= TOKEN_MAX */
66: int cc_token_min = 2; /* > tt.tt_put_token_cost */
67: int cc_npass0 = 1;
68: int cc_npass1 = 1;
69:
70: int cc_bufsize = 1024 * 3; /* XXX, or 80 * 24 * 2 */
71:
72: int cc_ntoken = 8192;
73:
74: #define cc_weight XXX
75: #ifndef cc_weight
76: int cc_weight = 0;
77: #endif
78:
79: #define TOKEN_MAX 16
80:
81: struct cc {
82: char string[TOKEN_MAX];
83: char length;
84: char flag;
85: #ifndef cc_weight
86: short weight;
87: #endif
88: long time; /* time last seen */
89: short bcount; /* count in this buffer */
90: short ccount; /* count in compression */
91: short places; /* places in the buffer */
92: short code; /* token code */
93: struct cc *qforw, *qback;
94: struct cc *hforw, **hback;
95: };
96:
97: short cc_thresholds[TOKEN_MAX + 1];
98: #define thresh(length) (cc_thresholds[length])
99: #define threshp(code, count, length) \
100: ((code) >= 0 || (short) (count) >= cc_thresholds[length])
101:
102: #ifndef cc_weight
103: short cc_wthresholds[TOKEN_MAX + 1];
104: #define wthresh(length) (cc_wthresholds[length])
105: #define wthreshp(weight, length) ((short) (weight) >= cc_wthresholds[length])
106: #else
107: #define wthreshp(weight, length) (0)
108: #endif
109:
110: #ifndef cc_weight
111: short cc_wlimits[TOKEN_MAX + 1];
112: #define wlimit(length) (cc_wlimits[length])
113: #endif
114:
115: #define put_token_score(length) ((length) - tt.tt_put_token_cost)
116:
117: int cc_score_adjustments[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
118: #define score_adjust(score, p) \
119: do { \
120: int length = (p)->length; \
121: int ccount = (p)->ccount; \
122: if (threshp((p)->code, ccount, length) || \
123: wthreshp((p)->weight, length)) /* XXX */ \
124: (score) -= length - tt.tt_put_token_cost; \
125: else \
126: (score) += cc_score_adjustments[length][ccount]; \
127: } while (0)
128:
129: int cc_initial_scores[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
130:
131: struct cc cc_q0a, cc_q0b, cc_q1a, cc_q1b;
132:
133: #define qinsert(p1, p2) \
134: do { \
135: register struct cc *forw = (p1)->qforw; \
136: register struct cc *back = (p1)->qback; \
137: back->qforw = forw; \
138: forw->qback = back; \
139: forw = (p2)->qforw; \
140: (p1)->qforw = forw; \
141: forw->qback = (p1); \
142: (p2)->qforw = (p1); \
143: (p1)->qback = (p2); \
144: } while (0)
145:
146: #define qinsertq(q, p) \
147: ((q)->qforw == (q) ? 0 : \
148: ((q)->qback->qforw = (p)->qforw, \
149: (p)->qforw->qback = (q)->qback, \
150: (q)->qforw->qback = (p), \
151: (p)->qforw = (q)->qforw, \
152: (q)->qforw = (q), \
153: (q)->qback = (q)))
154:
155: #define H (14)
156: #define HSIZE (1 << H)
157: #define hash(h, c) ((((h) >> H - 8 | (h) << 8) ^ (c)) & HSIZE - 1)
158:
159: char *cc_buffer;
160: struct cc **cc_output; /* the output array */
161: short *cc_places[TOKEN_MAX + 1];
162: short *cc_hashcodes; /* for computing hashcodes */
163: struct cc **cc_htab; /* the hash table */
164: struct cc **cc_tokens; /* holds all the active tokens */
165: struct cc_undo {
166: struct cc **pos;
167: struct cc *val;
168: } *cc_undo;
169:
170: long cc_time, cc_time0;
171:
172: char *cc_tt_ob, *cc_tt_obe;
173:
174: ccinit()
175: {
176: register i, j;
177: register struct cc *p;
178:
179: if (tt.tt_token_max > cc_token_max)
180: tt.tt_token_max = cc_token_max;
181: if (tt.tt_token_min < cc_token_min)
182: tt.tt_token_min = cc_token_min;
183: if (tt.tt_token_min > tt.tt_token_max) {
184: tt.tt_ntoken = 0;
185: return 0;
186: }
187: if (tt.tt_ntoken > cc_ntoken / 2) /* not likely */
188: tt.tt_ntoken = cc_ntoken / 2;
189: #define C(x) (sizeof (x) / sizeof *(x))
190: for (i = 0; i < C(cc_thresholds); i++) {
191: int h = i - tt.tt_put_token_cost;
192: if (h > 0)
193: cc_thresholds[i] =
194: (tt.tt_set_token_cost + 1 + h - 1) / h + 1;
195: else
196: cc_thresholds[i] = 0;
197: }
198: for (i = 0; i < C(cc_score_adjustments); i++) {
199: int t = cc_thresholds[i];
200: for (j = 0; j < C(*cc_score_adjustments); j++) {
201: if (j >= t)
202: cc_score_adjustments[i][j] =
203: - (i - tt.tt_put_token_cost);
204: else if (j < t - 1)
205: cc_score_adjustments[i][j] = 0;
206: else
207: /*
208: * cost now is
209: * length * (ccount + 1) a
210: * cost before was
211: * set-token-cost + length +
212: * ccount * put-token-cost b
213: * the score adjustment is (b - a)
214: */
215: cc_score_adjustments[i][j] =
216: tt.tt_set_token_cost + i +
217: j * tt.tt_put_token_cost -
218: i * (j + 1);
219: if (j >= t)
220: cc_initial_scores[i][j] = 0;
221: else
222: /*
223: * - (set-token-cost +
224: * (length - put-token-cost) -
225: * (length - put-token-cost) * ccount)
226: */
227: cc_initial_scores[i][j] =
228: - (tt.tt_set_token_cost +
229: (i - tt.tt_put_token_cost) -
230: (i - tt.tt_put_token_cost) * j);
231: }
232: }
233: #ifndef cc_weight
234: for (i = 1; i < C(cc_wthresholds); i++) {
235: cc_wthresholds[i] =
236: ((tt.tt_set_token_cost + tt.tt_put_token_cost) / i +
237: i / 5 + 1) *
238: cc_weight + 1;
239: cc_wlimits[i] = cc_wthresholds[i] + cc_weight;
240: }
241: #endif
242: #undef C
243: if ((cc_output = (struct cc **)
244: malloc((unsigned) cc_bufsize * sizeof *cc_output)) == 0)
245: goto nomem;
246: if ((cc_hashcodes = (short *)
247: malloc((unsigned) cc_bufsize * sizeof *cc_hashcodes)) == 0)
248: goto nomem;
249: if ((cc_htab = (struct cc **) malloc(HSIZE * sizeof *cc_htab)) == 0)
250: goto nomem;
251: if ((cc_tokens = (struct cc **)
252: malloc((unsigned)
253: (cc_ntoken + tt.tt_token_max - tt.tt_token_min + 1) *
254: sizeof *cc_tokens)) == 0)
255: goto nomem;
256: if ((cc_undo = (struct cc_undo *)
257: malloc((unsigned) cc_bufsize * sizeof *cc_undo)) == 0)
258: goto nomem;
259: for (i = tt.tt_token_min; i <= tt.tt_token_max; i++)
260: if ((cc_places[i] = (short *)
261: malloc((unsigned) cc_bufsize * sizeof **cc_places)) == 0)
262: goto nomem;
263: cc_q0a.qforw = cc_q0a.qback = &cc_q0a;
264: cc_q0b.qforw = cc_q0b.qback = &cc_q0b;
265: cc_q1a.qforw = cc_q1a.qback = &cc_q1a;
266: cc_q1b.qforw = cc_q1b.qback = &cc_q1b;
267: if ((p = (struct cc *) malloc((unsigned) cc_ntoken * sizeof *p)) == 0)
268: goto nomem;
269: for (i = 0; i < tt.tt_ntoken; i++) {
270: p->code = i;
271: p->time = -1;
272: p->qback = cc_q0a.qback;
273: p->qforw = &cc_q0a;
274: p->qback->qforw = p;
275: cc_q0a.qback = p;
276: p++;
277: }
278: for (; i < cc_ntoken; i++) {
279: p->code = -1;
280: p->time = -1;
281: p->qback = cc_q1a.qback;
282: p->qforw = &cc_q1a;
283: p->qback->qforw = p;
284: cc_q1a.qback = p;
285: p++;
286: }
287: cc_tt_ob = tt_ob;
288: cc_tt_obe = tt_obe;
289: if ((cc_buffer = malloc((unsigned) cc_bufsize)) == 0)
290: goto nomem;
291: return 0;
292: nomem:
293: wwerrno = WWE_NOMEM;
294: return -1;
295: }
296:
297: ccstart()
298: {
299: int ccflush();
300:
301: ttflush();
302: tt_obp = tt_ob = cc_buffer;
303: tt_obe = tt_ob + cc_bufsize;
304: tt.tt_flush = ccflush;
305: if (cc_trace) {
306: cc_trace_fp = fopen("window-trace", "a");
307: (void) fcntl(fileno(cc_trace_fp), F_SETFD, 1);
308: }
309: ccreset();
310: }
311:
312: ccreset()
313: {
314: register struct cc *p;
315:
316: bzero((char *) cc_htab, HSIZE * sizeof *cc_htab);
317: for (p = cc_q0a.qforw; p != &cc_q0a; p = p->qforw)
318: p->hback = 0;
319: for (p = cc_q1a.qforw; p != &cc_q1a; p = p->qforw)
320: p->hback = 0;
321: }
322:
323: ccend()
324: {
325:
326: ttflush();
327: tt_obp = tt_ob = cc_tt_ob;
328: tt_obe = cc_tt_obe;
329: tt.tt_flush = 0;
330: if (cc_trace_fp != NULL) {
331: (void) fclose(cc_trace_fp);
332: cc_trace_fp = NULL;
333: }
334: }
335:
336: ccflush()
337: {
338: int bufsize = tt_obp - tt_ob;
339: int n;
340:
341: if (tt_ob != cc_buffer)
342: abort();
343: if (cc_trace_fp != NULL) {
344: (void) fwrite(tt_ob, 1, bufsize, cc_trace_fp);
345: (void) putc(-1, cc_trace_fp);
346: }
347: tt.tt_flush = 0;
348: (*tt.tt_compress)(1);
349: if (bufsize < tt.tt_token_min) {
350: ttflush();
351: goto out;
352: }
353: tt_obp = tt_ob = cc_tt_ob;
354: tt_obe = cc_tt_obe;
355: cc_time0 = cc_time;
356: cc_time += bufsize;
357: n = cc_sweep_phase(cc_buffer, bufsize, cc_tokens);
358: cc_compress_phase(cc_output, bufsize, cc_tokens, n);
359: cc_output_phase(cc_buffer, cc_output, bufsize);
360: ttflush();
361: tt_obp = tt_ob = cc_buffer;
362: tt_obe = cc_buffer + cc_bufsize;
363: out:
364: (*tt.tt_compress)(0);
365: tt.tt_flush = ccflush;
366: }
367:
368: cc_sweep_phase(buffer, bufsize, tokens)
369: char *buffer;
370: struct cc **tokens;
371: {
372: register struct cc **pp = tokens;
373: register i, n;
374: #ifdef STATS
375: int nn, ii;
376: #endif
377:
378: #ifdef STATS
379: if (verbose >= 0)
380: time_begin();
381: if (verbose > 0)
382: printf("Sweep:");
383: #endif
384: cc_sweep0(buffer, bufsize, tt.tt_token_min - 1);
385: #ifdef STATS
386: ntoken_stat = 0;
387: nn = 0;
388: ii = 0;
389: #endif
390: for (i = tt.tt_token_min; i <= tt.tt_token_max; i++) {
391: #ifdef STATS
392: if (verbose > 0) {
393: if (ii > 7) {
394: printf("\n ");
395: ii = 0;
396: }
397: ii++;
398: printf(" (%d", i);
399: (void) fflush(stdout);
400: }
401: #endif
402: n = cc_sweep(buffer, bufsize, pp, i);
403: pp += n;
404: #ifdef STATS
405: if (verbose > 0) {
406: if (--n > 0) {
407: printf(" %d", n);
408: nn += n;
409: }
410: putchar(')');
411: }
412: #endif
413: }
414: qinsertq(&cc_q1b, &cc_q1a);
415: #ifdef STATS
416: if (verbose > 0)
417: printf("\n %d tokens, %d candidates\n",
418: ntoken_stat, nn);
419: if (verbose >= 0)
420: time_end();
421: #endif
422: return pp - tokens;
423: }
424:
425: cc_sweep0(buffer, n, length)
426: char *buffer;
427: {
428: register char *p;
429: register short *hc;
430: register i;
431: register short c;
432: register short pc = tt.tt_padc;
433:
434: /* n and length are at least 1 */
435: p = buffer++;
436: hc = cc_hashcodes;
437: i = n;
438: do {
439: if ((*hc++ = *p++) == pc)
440: hc[-1] = -1;
441: } while (--i);
442: while (--length) {
443: p = buffer++;
444: hc = cc_hashcodes;
445: for (i = n--; --i;) {
446: if ((c = *p++) == pc || *hc < 0)
447: c = -1;
448: else
449: c = hash(*hc, c);
450: *hc++ = c;
451: }
452: }
453: }
454:
455: cc_sweep(buffer, bufsize, tokens, length)
456: char *buffer;
457: struct cc **tokens;
458: register length;
459: {
460: register struct cc *p;
461: register char *cp;
462: register i;
463: short *hc;
464: short *places = cc_places[length];
465: struct cc **pp = tokens;
466: short threshold = thresh(length);
467: #ifndef cc_weight
468: short wthreshold = wthresh(length);
469: short limit = wlimit(length);
470: #endif
471: int time;
472: short pc = tt.tt_padc;
473:
474: i = length - 1;
475: bufsize -= i;
476: cp = buffer + i;
477: hc = cc_hashcodes;
478: time = cc_time0;
479: for (i = 0; i < bufsize; i++, time++) {
480: struct cc **h;
481:
482: {
483: register short *hc1 = hc;
484: register short c = *cp++;
485: register short hh;
486: if ((hh = *hc1) < 0 || c == pc) {
487: *hc1++ = -1;
488: hc = hc1;
489: continue;
490: }
491: h = cc_htab + (*hc1++ = hash(hh, c));
492: hc = hc1;
493: }
494: for (p = *h; p != 0; p = p->hforw)
495: if (p->length == (char) length) {
496: register char *p1 = p->string;
497: register char *p2 = cp - length;
498: register n = length;
499: do
500: if (*p1++ != *p2++)
501: goto fail;
502: while (--n);
503: break;
504: fail:;
505: }
506: if (p == 0) {
507: p = cc_q1a.qback;
508: if (p == &cc_q1a ||
509: p->time >= cc_time0 && p->length == (char) length)
510: continue;
511: if (p->hback != 0)
512: if ((*p->hback = p->hforw) != 0)
513: p->hforw->hback = p->hback;
514: {
515: register char *p1 = p->string;
516: register char *p2 = cp - length;
517: register n = length;
518: do
519: *p1++ = *p2++;
520: while (--n);
521: }
522: p->length = length;
523: #ifndef cc_weight
524: p->weight = cc_weight;
525: #endif
526: p->time = time;
527: p->bcount = 1;
528: p->ccount = 0;
529: p->flag = 0;
530: if ((p->hforw = *h) != 0)
531: p->hforw->hback = &p->hforw;
532: *h = p;
533: p->hback = h;
534: qinsert(p, &cc_q1a);
535: places[i] = -1;
536: p->places = i;
537: #ifdef STATS
538: ntoken_stat++;
539: #endif
540: } else if (p->time < cc_time0) {
541: #ifndef cc_weight
542: if ((p->weight += p->time - time) < 0)
543: p->weight = cc_weight;
544: else if ((p->weight += cc_weight) > limit)
545: p->weight = limit;
546: #endif
547: p->time = time;
548: p->bcount = 1;
549: p->ccount = 0;
550: if (p->code >= 0) {
551: p->flag = 1;
552: *pp++ = p;
553: } else
554: #ifndef cc_weight
555: if (p->weight >= wthreshold) {
556: p->flag = 1;
557: *pp++ = p;
558: qinsert(p, &cc_q1b);
559: } else
560: #endif
561: {
562: p->flag = 0;
563: qinsert(p, &cc_q1a);
564: }
565: places[i] = -1;
566: p->places = i;
567: #ifdef STATS
568: ntoken_stat++;
569: #endif
570: } else if (p->time + length > time) {
571: /*
572: * overlapping token, don't count as two and
573: * don't update time, but do adjust weight to offset
574: * the difference
575: */
576: #ifndef cc_weight
577: if (cc_weight != 0) { /* XXX */
578: p->weight += time - p->time;
579: if (!p->flag && p->weight >= wthreshold) {
580: p->flag = 1;
581: *pp++ = p;
582: qinsert(p, &cc_q1b);
583: }
584: }
585: #endif
586: places[i] = p->places;
587: p->places = i;
588: } else {
589: #ifndef cc_weight
590: if ((p->weight += p->time - time) < 0)
591: p->weight = cc_weight;
592: else if ((p->weight += cc_weight) > limit)
593: p->weight = limit;
594: #endif
595: p->time = time;
596: p->bcount++;
597: if (!p->flag &&
598: /* code must be < 0 if flag false here */
599: (p->bcount >= threshold
600: #ifndef cc_weight
601: || p->weight >= wthreshold
602: #endif
603: )) {
604: p->flag = 1;
605: *pp++ = p;
606: qinsert(p, &cc_q1b);
607: }
608: places[i] = p->places;
609: p->places = i;
610: }
611: }
612: if ((i = pp - tokens) > 0) {
613: *pp = 0;
614: if (cc_reverse)
615: cc_sweep_reverse(tokens, places);
616: if (cc_sort && i > 1) {
617: int cc_token_compare();
618: qsort((char *) tokens, i, sizeof *tokens,
619: cc_token_compare);
620: }
621: if (cc_chop) {
622: if ((i = i * cc_chop / 100) == 0)
623: i = 1;
624: tokens[i] = 0;
625: }
626: i++;
627: }
628: return i;
629: }
630:
631: cc_sweep_reverse(pp, places)
632: register struct cc **pp;
633: register short *places;
634: {
635: register struct cc *p;
636: register short front, back, t;
637:
638: while ((p = *pp++) != 0) {
639: back = -1;
640: t = p->places;
641: /* the list is never empty */
642: do {
643: front = places[t];
644: places[t] = back;
645: back = t;
646: } while ((t = front) >= 0);
647: p->places = back;
648: }
649: }
650:
651: cc_compress_phase(output, bufsize, tokens, ntoken)
652: struct cc **output;
653: struct cc **tokens;
654: {
655: register i;
656:
657: bzero((char *) output, bufsize * sizeof *output);
658: for (i = 0; i < cc_npass0; i++)
659: cc_compress_phase1(output, tokens, ntoken, 0);
660: for (i = 0; i < cc_npass1; i++)
661: cc_compress_phase1(output, tokens, ntoken, 1);
662: cc_compress_cleanup(output, bufsize);
663: }
664:
665: cc_compress_phase1(output, tokens, ntoken, flag)
666: register struct cc **output;
667: struct cc **tokens;
668: {
669: register struct cc **pp;
670: #ifdef STATS
671: register int i = 0;
672: int nt = 0, cc = 0, nc = 0;
673: #endif
674:
675: #ifdef STATS
676: if (verbose >= 0)
677: time_begin();
678: if (verbose > 0)
679: printf("Compress:");
680: #endif
681: pp = tokens;
682: while (pp < tokens + ntoken) {
683: #ifdef STATS
684: if (verbose > 0) {
685: ntoken_stat = 0;
686: ccount_stat = 0;
687: ncover_stat = 0;
688: if (i > 2) {
689: printf("\n ");
690: i = 0;
691: }
692: i++;
693: printf(" (%d", (*pp)->length);
694: (void) fflush(stdout);
695: }
696: #endif
697: pp += cc_compress(output, pp, flag);
698: #ifdef STATS
699: if (verbose > 0) {
700: printf(" %dt %du %dc)", ntoken_stat, ccount_stat,
701: ncover_stat);
702: nt += ntoken_stat;
703: cc += ccount_stat;
704: nc += ncover_stat;
705: }
706: #endif
707: }
708: #ifdef STATS
709: if (verbose > 0)
710: printf("\n total: (%dt %du %dc)\n", nt, cc, nc);
711: if (verbose >= 0)
712: time_end();
713: #endif
714: }
715:
716: cc_compress_cleanup(output, bufsize)
717: register struct cc **output;
718: {
719: register struct cc **end;
720:
721: /* the previous output phase may have been interrupted */
722: qinsertq(&cc_q0b, &cc_q0a);
723: for (end = output + bufsize; output < end;) {
724: register struct cc *p;
725: register length;
726: if ((p = *output) == 0) {
727: output++;
728: continue;
729: }
730: length = p->length;
731: if (!p->flag) {
732: } else if (p->code >= 0) {
733: qinsert(p, &cc_q0b);
734: p->flag = 0;
735: } else if (p->ccount == 0) {
736: *output = 0;
737: } else if (p->ccount >= thresh(length)
738: #ifndef cc_weight
739: || wthreshp(p->weight, length)
740: #endif
741: ) {
742: p->flag = 0;
743: } else {
744: p->ccount = 0;
745: *output = 0;
746: }
747: output += length;
748: }
749: }
750:
751: cc_compress(output, tokens, flag)
752: struct cc **output;
753: struct cc **tokens;
754: char flag;
755: {
756: struct cc **pp = tokens;
757: register struct cc *p = *pp++;
758: int length = p->length;
759: int threshold = thresh(length);
760: #ifndef cc_weight
761: short wthreshold = wthresh(length);
762: #endif
763: short *places = cc_places[length];
764: int *initial_scores = cc_initial_scores[length];
765: int initial_score0 = put_token_score(length);
766:
767: do {
768: int score;
769: register struct cc_undo *undop;
770: int ccount;
771: #ifdef STATS
772: int ncover;
773: #endif
774: int i;
775:
776: ccount = p->ccount;
777: if ((short) ccount >= p->bcount)
778: continue;
779: if (p->code >= 0 || ccount >= threshold)
780: score = 0;
781: #ifndef cc_weight
782: else if (p->weight >= wthreshold)
783: /* allow one fewer match than normal */
784: /* XXX, should adjust for ccount */
785: score = - tt.tt_set_token_cost;
786: #endif
787: else
788: score = initial_scores[ccount];
789: undop = cc_undo;
790: #ifdef STATS
791: ncover = 0;
792: #endif
793: for (i = p->places; i >= 0; i = places[i]) {
794: register struct cc **jp;
795: register struct cc *x;
796: register struct cc **ip = output + i;
797: register score0 = initial_score0;
798: struct cc **iip = ip + length;
799: struct cc_undo *undop1 = undop;
800:
801: if ((x = *(jp = ip)) != 0)
802: goto z;
803: while (--jp >= output)
804: if ((x = *jp) != 0) {
805: if (jp + x->length > ip)
806: goto z;
807: break;
808: }
809: jp = ip + 1;
810: while (jp < iip) {
811: if ((x = *jp) == 0) {
812: jp++;
813: continue;
814: }
815: z:
816: if (x == p)
817: goto undo;
818: #ifdef STATS
819: ncover++;
820: #endif
821: undop->pos = jp;
822: undop->val = x;
823: undop++;
824: *jp = 0;
825: x->ccount--;
826: score_adjust(score0, x);
827: if (score0 < 0 && flag)
828: goto undo;
829: jp += x->length;
830: }
831: undop->pos = ip;
832: undop->val = 0;
833: undop++;
834: *ip = p;
835: ccount++;
836: score += score0;
837: continue;
838: undo:
839: while (--undop >= undop1)
840: if (*undop->pos = x = undop->val)
841: x->ccount++;
842: undop++;
843: }
844: if (score > 0) {
845: #ifdef STATS
846: ccount_stat += ccount - p->ccount;
847: ntoken_stat++;
848: ncover_stat += ncover;
849: #endif
850: p->ccount = ccount;
851: } else {
852: register struct cc_undo *u = cc_undo;
853: while (--undop >= u) {
854: register struct cc *x;
855: if (*undop->pos = x = undop->val)
856: x->ccount++;
857: }
858: }
859: } while ((p = *pp++) != 0);
860: return pp - tokens;
861: }
862:
863: cc_output_phase(buffer, output, bufsize)
864: register char *buffer;
865: register struct cc **output;
866: register bufsize;
867: {
868: register i;
869: register struct cc *p, *p1;
870:
871: for (i = 0; i < bufsize;) {
872: if ((p = output[i]) == 0) {
873: ttputc(buffer[i]);
874: i++;
875: } else if (p->code >= 0) {
876: if (--p->ccount == 0)
877: qinsert(p, &cc_q0a);
878: (*tt.tt_put_token)(p->code, p->string, p->length);
879: wwntokuse++;
880: wwntoksave += put_token_score(p->length);
881: i += p->length;
882: } else if ((p1 = cc_q0a.qback) != &cc_q0a) {
883: p->code = p1->code;
884: p1->code = -1;
885: qinsert(p1, &cc_q1a);
886: if (--p->ccount == 0)
887: qinsert(p, &cc_q0a);
888: else
889: qinsert(p, &cc_q0b);
890: (*tt.tt_set_token)(p->code, p->string, p->length);
891: wwntokdef++;
892: wwntoksave -= tt.tt_set_token_cost;
893: i += p->length;
894: } else {
895: p->ccount--;
896: ttwrite(p->string, p->length);
897: wwntokbad++;
898: i += p->length;
899: }
900: }
901: wwntokc += bufsize;
902: }
903:
904: cc_token_compare(p1, p2)
905: struct cc **p1, **p2;
906: {
907: return (*p2)->bcount - (*p1)->bcount;
908: }