Annotation of parser3/src/lib/cord/cordxtra.c, revision 1.16
1.2 paf 1: /*
2: * Copyright (c) 1993-1994 by Xerox Corporation. All rights reserved.
3: *
4: * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
5: * OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
6: *
7: * Permission is hereby granted to use or copy this program
8: * for any purpose, provided the above notices are retained on all copies.
9: * Permission to modify the code and to distribute modified code is granted,
10: * provided the above notices are retained, and a notice that the code was
11: * modified is included with the above copyright notice.
12: *
13: * Author: Hans-J. Boehm (boehm@parc.xerox.com)
14: */
15: /*
16: * These are functions on cords that do not need to understand their
17: * implementation. They serve also serve as example client code for
18: * cord_basics.
19: */
20: /* Boehm, December 8, 1995 1:53 pm PST */
1.16 ! moko 21:
1.7 paf 22: #include "pa_config_includes.h"
1.16 ! moko 23: #include "cord.h"
! 24: #include "ec.h"
1.9 misha 25:
1.2 paf 26: # define I_HIDE_POINTERS /* So we get access to allocation lock. */
27: /* We use this for lazy file reading, */
28: /* so that we remain independent */
29: /* of the threads primitives. */
30: # include "gc.h"
31:
32: /* For now we assume that pointer reads and writes are atomic, */
33: /* i.e. another thread always sees the state before or after */
34: /* a write. This might be false on a Motorola M68K with */
35: /* pointers that are not 32-bit aligned. But there probably */
36: /* aren't too many threads packages running on those. */
37: # define ATOMIC_WRITE(x,y) (x) = (y)
38: # define ATOMIC_READ(x) (*(x))
39:
40: /* The standard says these are in stdio.h, but they aren't always: */
41: # ifndef SEEK_SET
42: # define SEEK_SET 0
43: # endif
44: # ifndef SEEK_END
45: # define SEEK_END 2
46: # endif
47:
48: # define BUFSZ 2048 /* Size of stack allocated buffers when */
49: /* we want large buffers. */
50:
51: typedef void (* oom_fn)(void);
52:
53: # define OUT_OF_MEMORY { if (CORD_oom_fn != (oom_fn) 0) (*CORD_oom_fn)(); \
54: ABORT("Out of memory\n"); }
55: # define ABORT(msg) { fprintf(stderr, "%s\n", msg); abort(); }
56:
57: CORD CORD_cat_char(CORD x, char c)
58: {
59: register char * string;
60:
61: if (c == '\0') return(CORD_cat(x, CORD_nul(1)));
62: string = GC_MALLOC_ATOMIC(2);
63: if (string == 0) OUT_OF_MEMORY;
64: string[0] = c;
65: string[1] = '\0';
66: return(CORD_cat_char_star(x, string, 1));
67: }
68:
69: CORD CORD_catn(int nargs, ...)
70: {
71: register CORD result = CORD_EMPTY;
72: va_list args;
73: register int i;
74:
75: va_start(args, nargs);
76: for (i = 0; i < nargs; i++) {
77: register CORD next = va_arg(args, CORD);
78: result = CORD_cat(result, next);
79: }
80: va_end(args);
81: return(result);
82: }
83:
84: typedef struct {
85: size_t len;
86: size_t count;
87: char * buf;
88: } CORD_fill_data;
89:
90: int CORD_fill_proc(char c, void * client_data)
91: {
92: register CORD_fill_data * d = (CORD_fill_data *)client_data;
93: register size_t count = d -> count;
94:
95: (d -> buf)[count] = c;
96: d -> count = ++count;
97: if (count >= d -> len) {
98: return(1);
99: } else {
100: return(0);
101: }
102: }
103:
104: int CORD_batched_fill_proc(const char* s, void * client_data)
105: {
106: register CORD_fill_data * d = (CORD_fill_data *)client_data;
107: register size_t count = d -> count;
108: register size_t max = d -> len;
109: register char * buf = d -> buf;
110: register const char* t = s;
111:
112: while((buf[count] = *t++) != '\0') {
113: count++;
114: if (count >= max) {
115: d -> count = count;
116: return(1);
117: }
118: }
119: d -> count = count;
120: return(0);
121: }
122:
123: /* Fill buf with len characters starting at i. */
124: /* Assumes len characters are available. */
125: void CORD_fill_buf(CORD x, size_t i, size_t len, char * buf)
126: {
127: CORD_fill_data fd;
128:
129: fd.len = len;
130: fd.buf = buf;
131: fd.count = 0;
132: (void)CORD_iter5(x, i, CORD_fill_proc, CORD_batched_fill_proc, &fd);
133: }
134:
135: int CORD_cmp(CORD x, CORD y)
136: {
137: CORD_pos xpos;
138: CORD_pos ypos;
139: register size_t avail, yavail;
140:
141: if (y == CORD_EMPTY) return(x != CORD_EMPTY);
142: if (x == CORD_EMPTY) return(-1);
1.15 misha 143: if (x == y) return (0);
1.2 paf 144: if (CORD_IS_STRING(y) && CORD_IS_STRING(x)) return(strcmp(x,y));
145: CORD_set_pos(xpos, x, 0);
146: CORD_set_pos(ypos, y, 0);
147: for(;;) {
148: if (!CORD_pos_valid(xpos)) {
149: if (CORD_pos_valid(ypos)) {
150: return(-1);
151: } else {
152: return(0);
153: }
154: }
155: if (!CORD_pos_valid(ypos)) {
156: return(1);
157: }
158: if ((avail = CORD_pos_chars_left(xpos)) <= 0
159: || (yavail = CORD_pos_chars_left(ypos)) <= 0) {
160: register char xcurrent = CORD_pos_fetch(xpos);
161: register char ycurrent = CORD_pos_fetch(ypos);
162: if (xcurrent != ycurrent) return(xcurrent - ycurrent);
163: CORD_next(xpos);
164: CORD_next(ypos);
165: } else {
166: /* process as many characters as we can */
167: register int result;
168:
169: if (avail > yavail) avail = yavail;
170: result = strncmp(CORD_pos_cur_char_addr(xpos),
171: CORD_pos_cur_char_addr(ypos), avail);
172: if (result != 0) return(result);
173: CORD_pos_advance(xpos, avail);
174: CORD_pos_advance(ypos, avail);
175: }
176: }
177: }
178:
179: int CORD_ncmp(CORD x, size_t x_start, CORD y, size_t y_start, size_t len)
180: {
181: CORD_pos xpos;
182: CORD_pos ypos;
183: register size_t count;
184: register long avail, yavail;
185:
186: CORD_set_pos(xpos, x, x_start);
187: CORD_set_pos(ypos, y, y_start);
188: for(count = 0; count < len;) {
189: if (!CORD_pos_valid(xpos)) {
190: if (CORD_pos_valid(ypos)) {
191: return(-1);
192: } else {
193: return(0);
194: }
195: }
196: if (!CORD_pos_valid(ypos)) {
197: return(1);
198: }
199: if ((avail = CORD_pos_chars_left(xpos)) <= 0
200: || (yavail = CORD_pos_chars_left(ypos)) <= 0) {
201: register char xcurrent = CORD_pos_fetch(xpos);
202: register char ycurrent = CORD_pos_fetch(ypos);
203: if (xcurrent != ycurrent) return(xcurrent - ycurrent);
204: CORD_next(xpos);
205: CORD_next(ypos);
206: count++;
207: } else {
208: /* process as many characters as we can */
209: register int result;
210:
211: if (avail > yavail) avail = yavail;
212: count += avail;
213: if (count > len) avail -= (count - len);
214: result = strncmp(CORD_pos_cur_char_addr(xpos),
215: CORD_pos_cur_char_addr(ypos), (size_t)avail);
216: if (result != 0) return(result);
217: CORD_pos_advance(xpos, (size_t)avail);
218: CORD_pos_advance(ypos, (size_t)avail);
219: }
220: }
221: return(0);
222: }
223:
1.13 misha 224: char * CORD_to_char_star(CORD x, size_t len)
1.2 paf 225: {
1.13 misha 226: char * result;
227: if(0 == len) len = CORD_len(x);
228: result = GC_MALLOC_ATOMIC(len + 1);
1.2 paf 229:
230: if (result == 0) OUT_OF_MEMORY;
231: CORD_fill_buf(x, 0, len, result);
232: result[len] = '\0';
233: return(result);
234: }
235:
236: CORD CORD_from_char_star(const char* s)
237: {
238: char * result;
239: size_t len = strlen(s);
240:
241: if (0 == len) return(CORD_EMPTY);
242: result = GC_MALLOC_ATOMIC(len + 1);
243: if (result == 0) OUT_OF_MEMORY;
244: memcpy(result, s, len+1);
245: return(result);
246: }
247:
1.13 misha 248: const char* CORD_to_const_char_star(CORD x, size_t len)
1.2 paf 249: {
250: if (x == 0) return("");
251: if (CORD_IS_STRING(x)) return((const char* )x);
1.13 misha 252: return(CORD_to_char_star(x, len));
1.2 paf 253: }
254:
255: char CORD_fetch(CORD x, size_t i)
256: {
257: CORD_pos xpos;
258:
259: CORD_set_pos(xpos, x, i);
260: if (!CORD_pos_valid(xpos)) ABORT("bad index?");
261: return(CORD_pos_fetch(xpos));
262: }
263:
264:
265: int CORD_put_proc(char c, void * client_data)
266: {
267: register FILE * f = (FILE *)client_data;
268:
269: return(putc(c, f) == EOF);
270: }
271:
272: int CORD_batched_put_proc(const char* s, void * client_data)
273: {
274: register FILE * f = (FILE *)client_data;
275:
276: return(fputs(s, f) == EOF);
277: }
278:
279:
280: int CORD_put(CORD x, FILE * f)
281: {
282: if (CORD_iter5(x, 0, CORD_put_proc, CORD_batched_put_proc, f)) {
283: return(EOF);
284: } else {
285: return(1);
286: }
287: }
288:
289: typedef struct {
290: size_t pos; /* Current position in the cord */
291: char target; /* Character we're looking for */
292: } chr_data;
293:
294: int CORD_chr_proc(char c, void * client_data)
295: {
296: register chr_data * d = (chr_data *)client_data;
297:
298: if (c == d -> target) return(1);
299: (d -> pos) ++;
300: return(0);
301: }
1.4 paf 302:
1.2 paf 303: int CORD_rchr_proc(char c, void * client_data)
304: {
305: register chr_data * d = (chr_data *)client_data;
306:
307: if (c == d -> target) return(1);
308: (d -> pos) --;
309: return(0);
310: }
311:
312: int CORD_batched_chr_proc(const char* s, void * client_data)
313: {
314: register chr_data * d = (chr_data *)client_data;
315: register char * occ = strchr(s, d -> target);
316:
317: if (occ == 0) {
318: d -> pos += strlen(s);
319: return(0);
320: } else {
321: d -> pos += occ - s;
322: return(1);
323: }
324: }
325:
326: size_t CORD_chr(CORD x, size_t i, int c)
327: {
328: chr_data d;
329:
330: d.pos = i;
331: d.target = c;
332: if (CORD_iter5(x, i, CORD_chr_proc, CORD_batched_chr_proc, &d)) {
333: return(d.pos);
334: } else {
335: return(CORD_NOT_FOUND);
336: }
337: }
338:
339: size_t CORD_rchr(CORD x, size_t i, int c)
340: {
341: chr_data d;
342:
343: d.pos = i;
344: d.target = c;
345: if (CORD_riter4(x, i, CORD_rchr_proc, &d)) {
346: return(d.pos);
347: } else {
348: return(CORD_NOT_FOUND);
349: }
350: }
351:
352: /* Find the first occurrence of s in x at position start or later. */
353: /* This uses an asymptotically poor algorithm, which should typically */
354: /* perform acceptably. We compare the first few characters directly, */
355: /* and call CORD_ncmp whenever there is a partial match. */
356: /* This has the advantage that we allocate very little, or not at all. */
357: /* It's very fast if there are few close misses. */
1.13 misha 358: size_t CORD_str(CORD x, size_t start, CORD s, size_t xlen)
1.2 paf 359: {
360: CORD_pos xpos;
361: size_t slen;
362: register size_t start_len;
363: const char* s_start;
364: unsigned long s_buf = 0; /* The first few characters of s */
365: unsigned long x_buf = 0; /* Start of candidate substring. */
366: /* Initialized only to make compilers */
367: /* happy. */
368: unsigned long mask = 0;
369: register size_t i;
370: register size_t match_pos;
371:
372: if (s == CORD_EMPTY) return(start);
373: if (CORD_IS_STRING(s)) {
374: s_start = s;
375: slen = strlen(s);
376: } else {
1.13 misha 377: s_start = CORD_to_char_star(CORD_substr(s, 0, sizeof(unsigned long), 0), 0);
1.2 paf 378: slen = CORD_len(s);
379: }
380: if (xlen < start || xlen - start < slen) return(CORD_NOT_FOUND);
381: start_len = slen;
382: if (start_len > sizeof(unsigned long)) start_len = sizeof(unsigned long);
383: CORD_set_pos(xpos, x, start);
384: for (i = 0; i < start_len; i++) {
385: mask <<= 8;
386: mask |= 0xff;
387: s_buf <<= 8;
1.3 paf 388: s_buf |= (unsigned char)s_start[i];
1.2 paf 389: x_buf <<= 8;
1.5 paf 390: x_buf |= (unsigned char)CORD_pos_fetch(xpos);
1.2 paf 391: CORD_next(xpos);
392: }
393: for (match_pos = start; ; match_pos++) {
394: if ((x_buf & mask) == s_buf) {
395: if (slen == start_len ||
396: CORD_ncmp(x, match_pos + start_len,
397: s, start_len, slen - start_len) == 0) {
398: return(match_pos);
399: }
400: }
401: if ( match_pos == xlen - slen ) {
402: return(CORD_NOT_FOUND);
403: }
404: x_buf <<= 8;
1.5 paf 405: x_buf |= (unsigned char)CORD_pos_fetch(xpos);
1.2 paf 406: CORD_next(xpos);
407: }
408: }
409:
410: void CORD_ec_flush_buf(CORD_ec x)
411: {
412: register size_t len = x[0].ec_bufptr - x[0].ec_buf;
413: char * s;
414:
415: if (len == 0) return;
416: s = GC_MALLOC_ATOMIC(len+1);
1.14 misha 417: if (s == 0) OUT_OF_MEMORY;
1.2 paf 418: memcpy(s, x[0].ec_buf, len);
419: s[len] = '\0';
420: x[0].ec_cord = CORD_cat_char_star(x[0].ec_cord, s, len);
421: x[0].ec_bufptr = x[0].ec_buf;
422: }
423:
424: void CORD_ec_append_cord(CORD_ec x, CORD s)
425: {
426: CORD_ec_flush_buf(x);
427: x[0].ec_cord = CORD_cat(x[0].ec_cord, s);
428: }
429:
430: /*ARGSUSED*/
431: char CORD_nul_func(size_t i, void * client_data)
432: {
433: return((char)(unsigned long)client_data);
1.4 paf 434: }
435:
1.9 misha 436: #ifdef CORD_CHARS_CACHE
1.11 misha 437: static char* cord_chars_cache[256][15]={0};
1.9 misha 438: #endif
439:
1.2 paf 440: CORD CORD_chars(char c, size_t i)
441: {
1.8 misha 442: if (i>0 && i<16 /* SHORT_LIMIT */) {
1.10 misha 443: register char* result;
1.9 misha 444: #ifdef CORD_CHARS_CACHE
1.11 misha 445: if(cord_chars_cache[(unsigned char)c][i])
446: return((CORD) cord_chars_cache[(unsigned char)c][i]);
1.9 misha 447: #endif
1.8 misha 448: result=GC_MALLOC_ATOMIC(i+1);
449: if(result==0) OUT_OF_MEMORY;
450: memset(result, c, i);
451: result[i] = '\0';
1.9 misha 452: #ifdef CORD_CHARS_CACHE
1.11 misha 453: cord_chars_cache[(unsigned char)c][i]=result;
1.9 misha 454: #endif
1.8 misha 455: return((CORD) result);
456: } else {
457: return(CORD_from_fn(CORD_nul_func, (void *)(unsigned long)c, i));
458: }
1.2 paf 459: }
460:
461: CORD CORD_from_file_eager(FILE * f)
462: {
463: register int c;
464: CORD_ec ecord;
465:
466: CORD_ec_init(ecord);
467: for(;;) {
468: c = getc(f);
469: if (c == 0) {
470: /* Append the right number of NULs */
471: /* Note that any string of NULs is rpresented in 4 words, */
472: /* independent of its length. */
473: register size_t count = 1;
474:
475: CORD_ec_flush_buf(ecord);
476: while ((c = getc(f)) == 0) count++;
477: ecord[0].ec_cord = CORD_cat(ecord[0].ec_cord, CORD_nul(count));
478: }
479: if (c == EOF) break;
480: CORD_ec_append(ecord, c);
481: }
482: (void) fclose(f);
483: return(CORD_balance(CORD_ec_to_cord(ecord)));
484: }
485:
486: /* The state maintained for a lazily read file consists primarily */
487: /* of a large direct-mapped cache of previously read values. */
488: /* We could rely more on stdio buffering. That would have 2 */
489: /* disadvantages: */
490: /* 1) Empirically, not all fseek implementations preserve the */
491: /* buffer whenever they could. */
492: /* 2) It would fail if 2 different sections of a long cord */
493: /* were being read alternately. */
494: /* We do use the stdio buffer for read ahead. */
495: /* To guarantee thread safety in the presence of atomic pointer */
496: /* writes, cache lines are always replaced, and never modified in */
497: /* place. */
498:
499: # define LOG_CACHE_SZ 14
500: # define CACHE_SZ (1 << LOG_CACHE_SZ)
501: # define LOG_LINE_SZ 9
502: # define LINE_SZ (1 << LOG_LINE_SZ)
503:
504: typedef struct {
505: size_t tag;
506: char data[LINE_SZ];
507: /* data[i%LINE_SZ] = ith char in file if tag = i/LINE_SZ */
508: } cache_line;
509:
510: typedef struct {
511: FILE * lf_file;
512: size_t lf_current; /* Current file pointer value */
513: cache_line * volatile lf_cache[CACHE_SZ/LINE_SZ];
514: } lf_state;
515:
516: # define MOD_CACHE_SZ(n) ((n) & (CACHE_SZ - 1))
517: # define DIV_CACHE_SZ(n) ((n) >> LOG_CACHE_SZ)
518: # define MOD_LINE_SZ(n) ((n) & (LINE_SZ - 1))
519: # define DIV_LINE_SZ(n) ((n) >> LOG_LINE_SZ)
520: # define LINE_START(n) ((n) & ~(LINE_SZ - 1))
521:
522: typedef struct {
523: lf_state * state;
524: size_t file_pos; /* Position of needed character. */
525: cache_line * new_cache;
526: } refill_data;
527:
528: /* Executed with allocation lock. */
529: static char refill_cache(client_data)
530: refill_data * client_data;
531: {
532: register lf_state * state = client_data -> state;
533: register size_t file_pos = client_data -> file_pos;
534: FILE *f = state -> lf_file;
535: size_t line_start = LINE_START(file_pos);
536: size_t line_no = DIV_LINE_SZ(MOD_CACHE_SZ(file_pos));
537: cache_line * new_cache = client_data -> new_cache;
538:
539: if (line_start != state -> lf_current
540: && fseek(f, line_start, SEEK_SET) != 0) {
541: ABORT("fseek failed");
542: }
543: if (fread(new_cache -> data, sizeof(char), LINE_SZ, f)
544: <= file_pos - line_start) {
545: ABORT("fread failed");
546: }
547: new_cache -> tag = DIV_LINE_SZ(file_pos);
548: /* Store barrier goes here. */
549: ATOMIC_WRITE(state -> lf_cache[line_no], new_cache);
550: state -> lf_current = line_start + LINE_SZ;
551: return(new_cache->data[MOD_LINE_SZ(file_pos)]);
552: }
553:
554: char CORD_lf_func(size_t i, void * client_data)
555: {
556: register lf_state * state = (lf_state *)client_data;
557: register cache_line * volatile * cl_addr =
558: &(state -> lf_cache[DIV_LINE_SZ(MOD_CACHE_SZ(i))]);
559: register cache_line * cl = (cache_line *)ATOMIC_READ(cl_addr);
560:
561: if (cl == 0 || cl -> tag != DIV_LINE_SZ(i)) {
562: /* Cache miss */
563: refill_data rd;
564:
565: rd.state = state;
566: rd.file_pos = i;
567: rd.new_cache = GC_NEW_ATOMIC(cache_line);
568: if (rd.new_cache == 0) OUT_OF_MEMORY;
569: return((char)(GC_word)
570: GC_call_with_alloc_lock((GC_fn_type) refill_cache, &rd));
571: }
572: return(cl -> data[MOD_LINE_SZ(i)]);
573: }
574:
575: /*ARGSUSED*/
576: void CORD_lf_close_proc(void * obj, void * client_data)
577: {
578: if (fclose(((lf_state *)obj) -> lf_file) != 0) {
579: ABORT("CORD_lf_close_proc: fclose failed");
580: }
581: }
582:
583: #ifndef PA_DEBUG_DISABLE_GC
584: CORD CORD_from_file_lazy_inner(FILE * f, size_t len)
585: {
586: register lf_state * state = GC_NEW(lf_state);
587: register int i;
588:
589: if (state == 0) OUT_OF_MEMORY;
590: if (len != 0) {
591: /* Dummy read to force buffer allocation. */
592: /* This greatly increases the probability */
593: /* of avoiding deadlock if buffer allocation */
594: /* is redirected to GC_malloc and the */
595: /* world is multithreaded. */
596: char buf[1];
597:
598: (void) fread(buf, 1, 1, f);
599: rewind(f);
600: }
601: state -> lf_file = f;
602: for (i = 0; i < CACHE_SZ/LINE_SZ; i++) {
603: state -> lf_cache[i] = 0;
604: }
605: state -> lf_current = 0;
606: GC_REGISTER_FINALIZER(state, CORD_lf_close_proc, 0, 0, 0);
607: return(CORD_from_fn(CORD_lf_func, state, len));
608: }
609:
610: CORD CORD_from_file_lazy(FILE * f)
611: {
612: register long len;
613:
614: if (fseek(f, 0l, SEEK_END) != 0) {
615: ABORT("Bad fd argument - fseek failed");
616: }
617: if ((len = ftell(f)) < 0) {
618: ABORT("Bad fd argument - ftell failed");
619: }
620: rewind(f);
621: return(CORD_from_file_lazy_inner(f, (size_t)len));
622: }
623:
624: # define LAZY_THRESHOLD (128*1024 + 1)
625:
626: CORD CORD_from_file(FILE * f)
627: {
628: register long len;
629:
630: if (fseek(f, 0l, SEEK_END) != 0) {
631: ABORT("Bad fd argument - fseek failed");
632: }
633: if ((len = ftell(f)) < 0) {
634: ABORT("Bad fd argument - ftell failed");
635: }
636: rewind(f);
637: if (len < LAZY_THRESHOLD) {
638: return(CORD_from_file_eager(f));
639: } else {
640: return(CORD_from_file_lazy_inner(f, (size_t)len));
641: }
642: }
643: #endif
E-mail:
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to cordxtra.c CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [parser3project] / parser3 / src / lib / cord