1 /*- 2 * Copyright (c) 1992, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 1992, 1993, 1994, 1995, 1996 5 * Keith Bostic. All rights reserved. 6 * 7 * See the LICENSE file for redistribution information. 8 */ 9 10 #include "config.h" 11 12 #include <sys/types.h> 13 #include <sys/queue.h> 14 #include <sys/time.h> 15 16 #include <bitstring.h> 17 #include <ctype.h> 18 #include <errno.h> 19 #include <limits.h> 20 #include <stdio.h> 21 #include <stdlib.h> 22 #include <string.h> 23 #include <unistd.h> 24 25 #include "../common/common.h" 26 #include "../vi/vi.h" 27 28 #define SUB_FIRST 0x01 /* The 'r' flag isn't reasonable. */ 29 #define SUB_MUSTSETR 0x02 /* The 'r' flag is required. */ 30 31 static int re_conv(SCR *, CHAR_T **, size_t *, int *); 32 static int re_cscope_conv(SCR *, CHAR_T **, size_t *, int *); 33 static int re_sub(SCR *, 34 CHAR_T *, CHAR_T **, size_t *, size_t *, regmatch_t [10]); 35 static int re_tag_conv(SCR *, CHAR_T **, size_t *, int *); 36 static int s(SCR *, EXCMD *, CHAR_T *, regex_t *, u_int); 37 38 /* 39 * ex_s -- 40 * [line [,line]] s[ubstitute] [[/;]pat[/;]/repl[/;] [cgr] [count] [#lp]] 41 * 42 * Substitute on lines matching a pattern. 43 * 44 * PUBLIC: int ex_s(SCR *, EXCMD *); 45 */ 46 int 47 ex_s(SCR *sp, EXCMD *cmdp) 48 { 49 regex_t *re; 50 size_t blen, len; 51 u_int flags; 52 int delim; 53 CHAR_T *bp, *p, *ptrn, *rep, *t; 54 55 /* 56 * Skip leading white space. 57 * 58 * !!! 59 * Historic vi allowed any non-alphanumeric to serve as the 60 * substitution command delimiter. 61 * 62 * !!! 63 * If the arguments are empty, it's the same as &, i.e. we 64 * repeat the last substitution. 65 */ 66 if (cmdp->argc == 0) 67 goto subagain; 68 for (p = cmdp->argv[0]->bp, 69 len = cmdp->argv[0]->len; len > 0; --len, ++p) { 70 if (!cmdskip(*p)) 71 break; 72 } 73 if (len == 0) 74 subagain: return (ex_subagain(sp, cmdp)); 75 76 delim = *p++; 77 if (is09azAZ(delim) || delim == '\\') 78 return (s(sp, cmdp, p, &sp->subre_c, SUB_MUSTSETR)); 79 80 /* 81 * !!! 82 * The full-blown substitute command reset the remembered 83 * state of the 'c' and 'g' suffices. 84 */ 85 sp->c_suffix = sp->g_suffix = 0; 86 87 /* 88 * Get the pattern string, toss escaping characters. 89 * 90 * !!! 91 * Historic vi accepted any of the following forms: 92 * 93 * :s/abc/def/ change "abc" to "def" 94 * :s/abc/def change "abc" to "def" 95 * :s/abc/ delete "abc" 96 * :s/abc delete "abc" 97 * 98 * QUOTING NOTE: 99 * 100 * Only toss an escaping character if it escapes a delimiter. 101 * This means that "s/A/\\\\f" replaces "A" with "\\f". It 102 * would be nice to be more regular, i.e. for each layer of 103 * escaping a single escaping character is removed, but that's 104 * not how the historic vi worked. 105 */ 106 for (ptrn = t = p;;) { 107 if (p[0] == '\0' || p[0] == delim) { 108 if (p[0] == delim) 109 ++p; 110 /* 111 * !!! 112 * Nul terminate the pattern string -- it's passed 113 * to regcomp which doesn't understand anything else. 114 */ 115 *t = '\0'; 116 break; 117 } 118 if (p[0] == '\\') { 119 if (p[1] == delim) 120 ++p; 121 else if (p[1] == '\\') 122 *t++ = *p++; 123 } 124 *t++ = *p++; 125 } 126 127 /* 128 * If the pattern string is empty, use the last RE (not just the 129 * last substitution RE). 130 */ 131 if (*ptrn == '\0') { 132 if (sp->re == NULL) { 133 ex_emsg(sp, NULL, EXM_NOPREVRE); 134 return (1); 135 } 136 137 /* Re-compile the RE if necessary. */ 138 if (!F_ISSET(sp, SC_RE_SEARCH) && 139 re_compile(sp, sp->re, sp->re_len, 140 NULL, NULL, &sp->re_c, RE_C_SEARCH)) 141 return (1); 142 flags = 0; 143 } else { 144 /* 145 * !!! 146 * Compile the RE. Historic practice is that substitutes set 147 * the search direction as well as both substitute and search 148 * RE's. We compile the RE twice, as we don't want to bother 149 * ref counting the pattern string and (opaque) structure. 150 */ 151 if (re_compile(sp, ptrn, t - ptrn, &sp->re, 152 &sp->re_len, &sp->re_c, RE_C_SEARCH)) 153 return (1); 154 if (re_compile(sp, ptrn, t - ptrn, &sp->subre, 155 &sp->subre_len, &sp->subre_c, RE_C_SUBST)) 156 return (1); 157 158 flags = SUB_FIRST; 159 sp->searchdir = FORWARD; 160 } 161 re = &sp->re_c; 162 163 /* 164 * Get the replacement string. 165 * 166 * The special character & (\& if O_MAGIC not set) matches the 167 * entire RE. No handling of & is required here, it's done by 168 * re_sub(). 169 * 170 * The special character ~ (\~ if O_MAGIC not set) inserts the 171 * previous replacement string into this replacement string. 172 * Count ~'s to figure out how much space we need. We could 173 * special case nonexistent last patterns or whether or not 174 * O_MAGIC is set, but it's probably not worth the effort. 175 * 176 * QUOTING NOTE: 177 * 178 * Only toss an escaping character if it escapes a delimiter or 179 * if O_MAGIC is set and it escapes a tilde. 180 * 181 * !!! 182 * If the entire replacement pattern is "%", then use the last 183 * replacement pattern. This semantic was added to vi in System 184 * V and then percolated elsewhere, presumably around the time 185 * that it was added to their version of ed(1). 186 */ 187 if (p[0] == '\0' || p[0] == delim) { 188 if (p[0] == delim) 189 ++p; 190 free(sp->repl); 191 sp->repl = NULL; 192 sp->repl_len = 0; 193 } else if (p[0] == '%' && (p[1] == '\0' || p[1] == delim)) 194 p += p[1] == delim ? 2 : 1; 195 else { 196 for (rep = p, len = 0; 197 p[0] != '\0' && p[0] != delim; ++p, ++len) 198 if (p[0] == '~') 199 len += sp->repl_len; 200 GET_SPACE_RETW(sp, bp, blen, len); 201 for (t = bp, len = 0, p = rep;;) { 202 if (p[0] == '\0' || p[0] == delim) { 203 if (p[0] == delim) 204 ++p; 205 break; 206 } 207 if (p[0] == '\\') { 208 if (p[1] == delim) 209 ++p; 210 else if (p[1] == '\\') { 211 *t++ = *p++; 212 ++len; 213 } else if (p[1] == '~') { 214 ++p; 215 if (!O_ISSET(sp, O_MAGIC)) 216 goto tilde; 217 } 218 } else if (p[0] == '~' && O_ISSET(sp, O_MAGIC)) { 219 tilde: ++p; 220 MEMCPY(t, sp->repl, sp->repl_len); 221 t += sp->repl_len; 222 len += sp->repl_len; 223 continue; 224 } 225 *t++ = *p++; 226 ++len; 227 } 228 if ((sp->repl_len = len) != 0) { 229 free(sp->repl); 230 MALLOC(sp, sp->repl, len * sizeof(CHAR_T)); 231 if (sp->repl == NULL) { 232 FREE_SPACEW(sp, bp, blen); 233 return (1); 234 } 235 MEMCPY(sp->repl, bp, len); 236 } 237 FREE_SPACEW(sp, bp, blen); 238 } 239 return (s(sp, cmdp, p, re, flags)); 240 } 241 242 /* 243 * ex_subagain -- 244 * [line [,line]] & [cgr] [count] [#lp]] 245 * 246 * Substitute using the last substitute RE and replacement pattern. 247 * 248 * PUBLIC: int ex_subagain(SCR *, EXCMD *); 249 */ 250 int 251 ex_subagain(SCR *sp, EXCMD *cmdp) 252 { 253 if (sp->subre == NULL) { 254 ex_emsg(sp, NULL, EXM_NOPREVRE); 255 return (1); 256 } 257 if (!F_ISSET(sp, SC_RE_SUBST) && 258 re_compile(sp, sp->subre, sp->subre_len, 259 NULL, NULL, &sp->subre_c, RE_C_SUBST)) 260 return (1); 261 return (s(sp, 262 cmdp, cmdp->argc ? cmdp->argv[0]->bp : NULL, &sp->subre_c, 0)); 263 } 264 265 /* 266 * ex_subtilde -- 267 * [line [,line]] ~ [cgr] [count] [#lp]] 268 * 269 * Substitute using the last RE and last substitute replacement pattern. 270 * 271 * PUBLIC: int ex_subtilde(SCR *, EXCMD *); 272 */ 273 int 274 ex_subtilde(SCR *sp, EXCMD *cmdp) 275 { 276 if (sp->re == NULL) { 277 ex_emsg(sp, NULL, EXM_NOPREVRE); 278 return (1); 279 } 280 if (!F_ISSET(sp, SC_RE_SEARCH) && re_compile(sp, sp->re, 281 sp->re_len, NULL, NULL, &sp->re_c, RE_C_SEARCH)) 282 return (1); 283 return (s(sp, 284 cmdp, cmdp->argc ? cmdp->argv[0]->bp : NULL, &sp->re_c, 0)); 285 } 286 287 /* 288 * s -- 289 * Do the substitution. This stuff is *really* tricky. There are lots of 290 * special cases, and general nastiness. Don't mess with it unless you're 291 * pretty confident. 292 * 293 * The nasty part of the substitution is what happens when the replacement 294 * string contains newlines. It's a bit tricky -- consider the information 295 * that has to be retained for "s/f\(o\)o/^M\1^M\1/". The solution here is 296 * to build a set of newline offsets which we use to break the line up later, 297 * when the replacement is done. Don't change it unless you're *damned* 298 * confident. 299 */ 300 #define NEEDNEWLINE(sp) do { \ 301 if (sp->newl_len == sp->newl_cnt) { \ 302 sp->newl_len += 25; \ 303 REALLOC(sp, sp->newl, size_t *, \ 304 sp->newl_len * sizeof(size_t)); \ 305 if (sp->newl == NULL) { \ 306 sp->newl_len = 0; \ 307 return (1); \ 308 } \ 309 } \ 310 } while (0) 311 312 #define BUILD(sp, l, len) do { \ 313 if (lbclen + (len) > lblen) { \ 314 lblen = p2roundup(MAX(lbclen + (len), 256)); \ 315 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \ 316 if (lb == NULL) { \ 317 lbclen = 0; \ 318 return (1); \ 319 } \ 320 } \ 321 MEMCPY(lb + lbclen, l, len); \ 322 lbclen += len; \ 323 } while (0) 324 325 #define NEEDSP(sp, len, pnt) do { \ 326 if (lbclen + (len) > lblen) { \ 327 lblen = p2roundup(MAX(lbclen + (len), 256)); \ 328 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \ 329 if (lb == NULL) { \ 330 lbclen = 0; \ 331 return (1); \ 332 } \ 333 pnt = lb + lbclen; \ 334 } \ 335 } while (0) 336 337 static int 338 s(SCR *sp, EXCMD *cmdp, CHAR_T *s, regex_t *re, u_int flags) 339 { 340 EVENT ev; 341 MARK from, to; 342 TEXTH tiq[] = {{ 0 }}; 343 recno_t elno, lno, slno; 344 u_long ul; 345 regmatch_t match[10]; 346 size_t blen, cnt, last, lbclen, lblen, len, llen; 347 size_t offset, saved_offset, scno; 348 int cflag, lflag, nflag, pflag, rflag; 349 int didsub, do_eol_match, eflags, empty_ok, eval; 350 int linechanged, matched, quit, rval; 351 CHAR_T *bp, *lb; 352 enum nresult nret; 353 354 NEEDFILE(sp, cmdp); 355 356 slno = sp->lno; 357 scno = sp->cno; 358 359 /* 360 * !!! 361 * Historically, the 'g' and 'c' suffices were always toggled as flags, 362 * so ":s/A/B/" was the same as ":s/A/B/ccgg". If O_EDCOMPATIBLE was 363 * not set, they were initialized to 0 for all substitute commands. If 364 * O_EDCOMPATIBLE was set, they were initialized to 0 only if the user 365 * specified substitute/replacement patterns (see ex_s()). 366 */ 367 if (!O_ISSET(sp, O_EDCOMPATIBLE)) 368 sp->c_suffix = sp->g_suffix = 0; 369 370 /* 371 * Historic vi permitted the '#', 'l' and 'p' options in vi mode, but 372 * it only displayed the last change. I'd disallow them, but they are 373 * useful in combination with the [v]global commands. In the current 374 * model the problem is combining them with the 'c' flag -- the screen 375 * would have to flip back and forth between the confirm screen and the 376 * ex print screen, which would be pretty awful. We do display all 377 * changes, though, for what that's worth. 378 * 379 * !!! 380 * Historic vi was fairly strict about the order of "options", the 381 * count, and "flags". I'm somewhat fuzzy on the difference between 382 * options and flags, anyway, so this is a simpler approach, and we 383 * just take it them in whatever order the user gives them. (The ex 384 * usage statement doesn't reflect this.) 385 */ 386 cflag = lflag = nflag = pflag = rflag = 0; 387 if (s == NULL) 388 goto noargs; 389 for (lno = OOBLNO; *s != '\0'; ++s) 390 switch (*s) { 391 case ' ': 392 case '\t': 393 continue; 394 case '+': 395 ++cmdp->flagoff; 396 break; 397 case '-': 398 --cmdp->flagoff; 399 break; 400 case '0': case '1': case '2': case '3': case '4': 401 case '5': case '6': case '7': case '8': case '9': 402 if (lno != OOBLNO) 403 goto usage; 404 errno = 0; 405 nret = nget_uslong(&ul, s, &s, 10); 406 lno = ul; 407 if (*s == '\0') /* Loop increment correction. */ 408 --s; 409 if (nret != NUM_OK) { 410 if (nret == NUM_OVER) 411 msgq(sp, M_ERR, "153|Count overflow"); 412 else if (nret == NUM_UNDER) 413 msgq(sp, M_ERR, "154|Count underflow"); 414 else 415 msgq(sp, M_SYSERR, NULL); 416 return (1); 417 } 418 /* 419 * In historic vi, the count was inclusive from the 420 * second address. 421 */ 422 cmdp->addr1.lno = cmdp->addr2.lno; 423 cmdp->addr2.lno += lno - 1; 424 if (!db_exist(sp, cmdp->addr2.lno) && 425 db_last(sp, &cmdp->addr2.lno)) 426 return (1); 427 break; 428 case '#': 429 nflag = 1; 430 break; 431 case 'c': 432 sp->c_suffix = !sp->c_suffix; 433 434 /* Ex text structure initialization. */ 435 if (F_ISSET(sp, SC_EX)) 436 TAILQ_INIT(tiq); 437 break; 438 case 'g': 439 sp->g_suffix = !sp->g_suffix; 440 break; 441 case 'l': 442 lflag = 1; 443 break; 444 case 'p': 445 pflag = 1; 446 break; 447 case 'r': 448 if (LF_ISSET(SUB_FIRST)) { 449 msgq(sp, M_ERR, 450 "155|Regular expression specified; r flag meaningless"); 451 return (1); 452 } 453 if (!F_ISSET(sp, SC_RE_SEARCH)) { 454 ex_emsg(sp, NULL, EXM_NOPREVRE); 455 return (1); 456 } 457 rflag = 1; 458 re = &sp->re_c; 459 break; 460 default: 461 goto usage; 462 } 463 464 if (*s != '\0' || (!rflag && LF_ISSET(SUB_MUSTSETR))) { 465 usage: ex_emsg(sp, cmdp->cmd->usage, EXM_USAGE); 466 return (1); 467 } 468 469 noargs: if (F_ISSET(sp, SC_VI) && sp->c_suffix && (lflag || nflag || pflag)) { 470 msgq(sp, M_ERR, 471 "156|The #, l and p flags may not be combined with the c flag in vi mode"); 472 return (1); 473 } 474 475 /* 476 * bp: if interactive, line cache 477 * blen: if interactive, line cache length 478 * lb: build buffer pointer. 479 * lbclen: current length of built buffer. 480 * lblen; length of build buffer. 481 */ 482 bp = lb = NULL; 483 blen = lbclen = lblen = 0; 484 485 /* For each line... */ 486 lno = cmdp->addr1.lno == 0 ? 1 : cmdp->addr1.lno; 487 for (matched = quit = 0, 488 elno = cmdp->addr2.lno; !quit && lno <= elno; ++lno) { 489 490 /* Someone's unhappy, time to stop. */ 491 if (INTERRUPTED(sp)) 492 break; 493 494 /* Get the line. */ 495 if (db_get(sp, lno, DBG_FATAL, &s, &llen)) 496 goto err; 497 498 /* 499 * Make a local copy if doing confirmation -- when calling 500 * the confirm routine we're likely to lose the cached copy. 501 */ 502 if (sp->c_suffix) { 503 if (bp == NULL) { 504 GET_SPACE_RETW(sp, bp, blen, llen); 505 } else 506 ADD_SPACE_RETW(sp, bp, blen, llen); 507 MEMCPY(bp, s, llen); 508 s = bp; 509 } 510 511 /* Start searching from the beginning. */ 512 offset = 0; 513 len = llen; 514 515 /* Reset the build buffer offset. */ 516 lbclen = 0; 517 518 /* Reset empty match flag. */ 519 empty_ok = 1; 520 521 /* 522 * We don't want to have to do a setline if the line didn't 523 * change -- keep track of whether or not this line changed. 524 * If doing confirmations, don't want to keep setting the 525 * line if change is refused -- keep track of substitutions. 526 */ 527 didsub = linechanged = 0; 528 529 /* New line, do an EOL match. */ 530 do_eol_match = 1; 531 532 /* It's not nul terminated, but we pretend it is. */ 533 eflags = REG_STARTEND; 534 535 /* 536 * The search area is from s + offset to the EOL. 537 * 538 * Generally, match[0].rm_so is the offset of the start 539 * of the match from the start of the search, and offset 540 * is the offset of the start of the last search. 541 */ 542 nextmatch: match[0].rm_so = 0; 543 match[0].rm_eo = len; 544 545 /* Get the next match. */ 546 eval = regexec(re, s + offset, 10, match, eflags); 547 548 /* 549 * There wasn't a match or if there was an error, deal with 550 * it. If there was a previous match in this line, resolve 551 * the changes into the database. Otherwise, just move on. 552 */ 553 if (eval == REG_NOMATCH) 554 goto endmatch; 555 if (eval != 0) { 556 re_error(sp, eval, re); 557 goto err; 558 } 559 matched = 1; 560 561 /* Only the first search can match an anchored expression. */ 562 eflags |= REG_NOTBOL; 563 564 /* 565 * !!! 566 * It's possible to match 0-length strings -- for example, the 567 * command s;a*;X;, when matched against the string "aabb" will 568 * result in "XbXbX", i.e. the matches are "aa", the space 569 * between the b's and the space between the b's and the end of 570 * the string. There is a similar space between the beginning 571 * of the string and the a's. The rule that we use (because vi 572 * historically used it) is that any 0-length match, occurring 573 * immediately after a match, is ignored. Otherwise, the above 574 * example would have resulted in "XXbXbX". Another example is 575 * incorrectly using " *" to replace groups of spaces with one 576 * space. 577 * 578 * The way we do this is that if we just had a successful match, 579 * the starting offset does not skip characters, and the match 580 * is empty, ignore the match and move forward. If there's no 581 * more characters in the string, we were attempting to match 582 * after the last character, so quit. 583 */ 584 if (!empty_ok && match[0].rm_so == 0 && match[0].rm_eo == 0) { 585 empty_ok = 1; 586 if (len == 0) 587 goto endmatch; 588 BUILD(sp, s + offset, 1); 589 ++offset; 590 --len; 591 goto nextmatch; 592 } 593 594 /* Confirm change. */ 595 if (sp->c_suffix) { 596 /* 597 * Set the cursor position for confirmation. Note, 598 * if we matched on a '$', the cursor may be past 599 * the end of line. 600 */ 601 from.lno = to.lno = lno; 602 from.cno = match[0].rm_so + offset; 603 to.cno = match[0].rm_eo + offset; 604 /* 605 * Both ex and vi have to correct for a change before 606 * the first character in the line. 607 */ 608 if (llen == 0) 609 from.cno = to.cno = 0; 610 if (F_ISSET(sp, SC_VI)) { 611 /* 612 * Only vi has to correct for a change after 613 * the last character in the line. 614 * 615 * XXX 616 * It would be nice to change the vi code so 617 * that we could display a cursor past EOL. 618 */ 619 if (to.cno >= llen) 620 to.cno = llen - 1; 621 if (from.cno >= llen) 622 from.cno = llen - 1; 623 624 sp->lno = from.lno; 625 sp->cno = from.cno; 626 if (vs_refresh(sp, 1)) 627 goto err; 628 629 vs_update(sp, msg_cat(sp, 630 "169|Confirm change? [n]", NULL), NULL); 631 632 if (v_event_get(sp, &ev, 0, 0)) 633 goto err; 634 switch (ev.e_event) { 635 case E_CHARACTER: 636 break; 637 case E_EOF: 638 case E_ERR: 639 case E_INTERRUPT: 640 goto lquit; 641 default: 642 v_event_err(sp, &ev); 643 goto lquit; 644 } 645 } else { 646 if (ex_print(sp, cmdp, &from, &to, 0) || 647 ex_scprint(sp, &from, &to)) 648 goto lquit; 649 if (ex_txt(sp, tiq, 0, TXT_CR)) 650 goto err; 651 ev.e_c = TAILQ_FIRST(tiq)->lb[0]; 652 } 653 654 switch (ev.e_c) { 655 case CH_YES: 656 break; 657 default: 658 case CH_NO: 659 didsub = 0; 660 BUILD(sp, s +offset, match[0].rm_eo); 661 goto skip; 662 case CH_QUIT: 663 /* Set the quit/interrupted flags. */ 664 lquit: quit = 1; 665 F_SET(sp->gp, G_INTERRUPTED); 666 667 /* 668 * Resolve any changes, then return to (and 669 * exit from) the main loop. 670 */ 671 goto endmatch; 672 } 673 } 674 675 /* 676 * Set the cursor to the last position changed, converting 677 * from 1-based to 0-based. 678 */ 679 sp->lno = lno; 680 sp->cno = match[0].rm_so; 681 682 /* Copy the bytes before the match into the build buffer. */ 683 BUILD(sp, s + offset, match[0].rm_so); 684 685 /* Substitute the matching bytes. */ 686 didsub = 1; 687 if (re_sub(sp, s + offset, &lb, &lbclen, &lblen, match)) 688 goto err; 689 690 /* Set the change flag so we know this line was modified. */ 691 linechanged = 1; 692 693 /* Move past the matched bytes. */ 694 skip: offset += match[0].rm_eo; 695 len -= match[0].rm_eo; 696 697 /* A match cannot be followed by an empty pattern. */ 698 empty_ok = 0; 699 700 /* 701 * If doing a global change with confirmation, we have to 702 * update the screen. The basic idea is to store the line 703 * so the screen update routines can find it, and restart. 704 */ 705 if (didsub && sp->c_suffix && sp->g_suffix) { 706 /* 707 * The new search offset will be the end of the 708 * modified line. 709 */ 710 saved_offset = lbclen; 711 712 /* Copy the rest of the line. */ 713 if (len) 714 BUILD(sp, s + offset, len); 715 716 /* Set the new offset. */ 717 offset = saved_offset; 718 719 /* Store inserted lines, adjusting the build buffer. */ 720 last = 0; 721 if (sp->newl_cnt) { 722 for (cnt = 0; 723 cnt < sp->newl_cnt; ++cnt, ++lno, ++elno) { 724 if (db_insert(sp, lno, 725 lb + last, sp->newl[cnt] - last)) 726 goto err; 727 last = sp->newl[cnt] + 1; 728 ++sp->rptlines[L_ADDED]; 729 } 730 lbclen -= last; 731 offset -= last; 732 sp->newl_cnt = 0; 733 } 734 735 /* Store and retrieve the line. */ 736 if (db_set(sp, lno, lb + last, lbclen)) 737 goto err; 738 if (db_get(sp, lno, DBG_FATAL, &s, &llen)) 739 goto err; 740 ADD_SPACE_RETW(sp, bp, blen, llen); 741 MEMCPY(bp, s, llen); 742 s = bp; 743 len = llen - offset; 744 745 /* Restart the build. */ 746 lbclen = 0; 747 BUILD(sp, s, offset); 748 749 /* 750 * If we haven't already done the after-the-string 751 * match, do one. Set REG_NOTEOL so the '$' pattern 752 * only matches once. 753 */ 754 if (!do_eol_match) 755 goto endmatch; 756 if (offset == len) { 757 do_eol_match = 0; 758 eflags |= REG_NOTEOL; 759 } 760 goto nextmatch; 761 } 762 763 /* 764 * If it's a global: 765 * 766 * If at the end of the string, do a test for the after 767 * the string match. Set REG_NOTEOL so the '$' pattern 768 * only matches once. 769 */ 770 if (sp->g_suffix && do_eol_match) { 771 if (len == 0) { 772 do_eol_match = 0; 773 eflags |= REG_NOTEOL; 774 } 775 goto nextmatch; 776 } 777 778 endmatch: if (!linechanged) 779 continue; 780 781 /* Copy any remaining bytes into the build buffer. */ 782 if (len) 783 BUILD(sp, s + offset, len); 784 785 /* Store inserted lines, adjusting the build buffer. */ 786 last = 0; 787 if (sp->newl_cnt) { 788 for (cnt = 0; 789 cnt < sp->newl_cnt; ++cnt, ++lno, ++elno) { 790 if (db_insert(sp, 791 lno, lb + last, sp->newl[cnt] - last)) 792 goto err; 793 last = sp->newl[cnt] + 1; 794 ++sp->rptlines[L_ADDED]; 795 } 796 lbclen -= last; 797 sp->newl_cnt = 0; 798 } 799 800 /* Store the changed line. */ 801 if (db_set(sp, lno, lb + last, lbclen)) 802 goto err; 803 804 /* Update changed line counter. */ 805 if (sp->rptlchange != lno) { 806 sp->rptlchange = lno; 807 ++sp->rptlines[L_CHANGED]; 808 } 809 810 /* 811 * !!! 812 * Display as necessary. Historic practice is to only 813 * display the last line of a line split into multiple 814 * lines. 815 */ 816 if (lflag || nflag || pflag) { 817 from.lno = to.lno = lno; 818 from.cno = to.cno = 0; 819 if (lflag) 820 (void)ex_print(sp, cmdp, &from, &to, E_C_LIST); 821 if (nflag) 822 (void)ex_print(sp, cmdp, &from, &to, E_C_HASH); 823 if (pflag) 824 (void)ex_print(sp, cmdp, &from, &to, E_C_PRINT); 825 } 826 } 827 828 /* 829 * !!! 830 * Historically, vi attempted to leave the cursor at the same place if 831 * the substitution was done at the current cursor position. Otherwise 832 * it moved it to the first non-blank of the last line changed. There 833 * were some problems: for example, :s/$/foo/ with the cursor on the 834 * last character of the line left the cursor on the last character, or 835 * the & command with multiple occurrences of the matching string in the 836 * line usually left the cursor in a fairly random position. 837 * 838 * We try to do the same thing, with the exception that if the user is 839 * doing substitution with confirmation, we move to the last line about 840 * which the user was consulted, as opposed to the last line that they 841 * actually changed. This prevents a screen flash if the user doesn't 842 * change many of the possible lines. 843 */ 844 if (!sp->c_suffix && (sp->lno != slno || sp->cno != scno)) { 845 sp->cno = 0; 846 (void)nonblank(sp, sp->lno, &sp->cno); 847 } 848 849 /* 850 * If not in a global command, and nothing matched, say so. 851 * Else, if none of the lines displayed, put something up. 852 */ 853 rval = 0; 854 if (!matched) { 855 if (!F_ISSET(sp, SC_EX_GLOBAL)) { 856 msgq(sp, M_ERR, "157|No match found"); 857 goto err; 858 } 859 } else if (!lflag && !nflag && !pflag) 860 F_SET(cmdp, E_AUTOPRINT); 861 862 if (0) { 863 err: rval = 1; 864 } 865 866 if (bp != NULL) 867 FREE_SPACEW(sp, bp, blen); 868 free(lb); 869 return (rval); 870 } 871 872 /* 873 * re_compile -- 874 * Compile the RE. 875 * 876 * PUBLIC: int re_compile(SCR *, 877 * PUBLIC: CHAR_T *, size_t, CHAR_T **, size_t *, regex_t *, u_int); 878 */ 879 int 880 re_compile(SCR *sp, CHAR_T *ptrn, size_t plen, CHAR_T **ptrnp, size_t *lenp, regex_t *rep, u_int flags) 881 { 882 size_t len; 883 int reflags, replaced, rval; 884 CHAR_T *p; 885 886 /* Set RE flags. */ 887 reflags = 0; 888 if (!LF_ISSET(RE_C_CSCOPE | RE_C_TAG)) { 889 if (O_ISSET(sp, O_EXTENDED)) 890 reflags |= REG_EXTENDED; 891 if (O_ISSET(sp, O_IGNORECASE)) 892 reflags |= REG_ICASE; 893 if (O_ISSET(sp, O_ICLOWER)) { 894 for (p = ptrn, len = plen; len > 0; ++p, --len) 895 if (ISUPPER(*p)) 896 break; 897 if (len == 0) 898 reflags |= REG_ICASE; 899 } 900 } 901 902 /* If we're replacing a saved value, clear the old one. */ 903 if (LF_ISSET(RE_C_SEARCH) && F_ISSET(sp, SC_RE_SEARCH)) { 904 regfree(&sp->re_c); 905 F_CLR(sp, SC_RE_SEARCH); 906 } 907 if (LF_ISSET(RE_C_SUBST) && F_ISSET(sp, SC_RE_SUBST)) { 908 regfree(&sp->subre_c); 909 F_CLR(sp, SC_RE_SUBST); 910 } 911 912 /* 913 * If we're saving the string, it's a pattern we haven't seen before, 914 * so convert the vi-style RE's to POSIX 1003.2 RE's. Save a copy for 915 * later recompilation. Free any previously saved value. 916 */ 917 if (ptrnp != NULL) { 918 replaced = 0; 919 if (LF_ISSET(RE_C_CSCOPE)) { 920 if (re_cscope_conv(sp, &ptrn, &plen, &replaced)) 921 return (1); 922 /* 923 * XXX 924 * Currently, the match-any-<blank> expression used in 925 * re_cscope_conv() requires extended RE's. This may 926 * not be right or safe. 927 */ 928 reflags |= REG_EXTENDED; 929 } else if (LF_ISSET(RE_C_TAG)) { 930 if (re_tag_conv(sp, &ptrn, &plen, &replaced)) 931 return (1); 932 } else 933 if (re_conv(sp, &ptrn, &plen, &replaced)) 934 return (1); 935 936 /* Discard previous pattern. */ 937 free(*ptrnp); 938 *ptrnp = NULL; 939 940 if (lenp != NULL) 941 *lenp = plen; 942 943 /* 944 * Copy the string into allocated memory. 945 * 946 * XXX 947 * Regcomp isn't 8-bit clean, so the pattern is nul-terminated 948 * for now. There's just no other solution. 949 */ 950 MALLOC(sp, *ptrnp, (plen + 1) * sizeof(CHAR_T)); 951 if (*ptrnp != NULL) { 952 MEMCPY(*ptrnp, ptrn, plen); 953 (*ptrnp)[plen] = '\0'; 954 } 955 956 /* Free up conversion-routine-allocated memory. */ 957 if (replaced) 958 FREE_SPACEW(sp, ptrn, 0); 959 960 if (*ptrnp == NULL) 961 return (1); 962 963 ptrn = *ptrnp; 964 } 965 966 /* 967 * XXX 968 * Regcomp isn't 8-bit clean, so we just lost if the pattern 969 * contained a nul. Bummer! 970 */ 971 if ((rval = regcomp(rep, ptrn, /* plen, */ reflags)) != 0) { 972 if (!LF_ISSET(RE_C_SILENT)) 973 re_error(sp, rval, rep); 974 return (1); 975 } 976 977 if (LF_ISSET(RE_C_SEARCH)) 978 F_SET(sp, SC_RE_SEARCH); 979 if (LF_ISSET(RE_C_SUBST)) 980 F_SET(sp, SC_RE_SUBST); 981 982 return (0); 983 } 984 985 /* 986 * re_conv -- 987 * Convert vi's regular expressions into something that the 988 * the POSIX 1003.2 RE functions can handle. 989 * 990 * There are three conversions we make to make vi's RE's (specifically 991 * the global, search, and substitute patterns) work with POSIX RE's. 992 * 993 * 1: If O_MAGIC is not set, strip backslashes from the magic character 994 * set (.[*~) that have them, and add them to the ones that don't. 995 * 2: If O_MAGIC is not set, the string "\~" is replaced with the text 996 * from the last substitute command's replacement string. If O_MAGIC 997 * is set, it's the string "~". 998 * 3: The pattern \<ptrn\> does "word" searches, convert it to use the 999 * new RE escapes. 1000 * 1001 * !!!/XXX 1002 * This doesn't exactly match the historic behavior of vi because we do 1003 * the ~ substitution before calling the RE engine, so magic characters 1004 * in the replacement string will be expanded by the RE engine, and they 1005 * weren't historically. It's a bug. 1006 */ 1007 static int 1008 re_conv(SCR *sp, CHAR_T **ptrnp, size_t *plenp, int *replacedp) 1009 { 1010 size_t blen, len, needlen; 1011 int magic; 1012 CHAR_T *bp, *p, *t; 1013 1014 /* 1015 * First pass through, we figure out how much space we'll need. 1016 * We do it in two passes, on the grounds that most of the time 1017 * the user is doing a search and won't have magic characters. 1018 * That way we can skip most of the memory allocation and copies. 1019 */ 1020 magic = 0; 1021 for (p = *ptrnp, len = *plenp, needlen = 0; len > 0; ++p, --len) 1022 switch (*p) { 1023 case '\\': 1024 if (len > 1) { 1025 --len; 1026 switch (*++p) { 1027 case '<': 1028 magic = 1; 1029 needlen += RE_WSTART_LEN + 1; 1030 break; 1031 case '>': 1032 magic = 1; 1033 needlen += RE_WSTOP_LEN + 1; 1034 break; 1035 case '~': 1036 if (!O_ISSET(sp, O_MAGIC)) { 1037 magic = 1; 1038 needlen += sp->repl_len; 1039 } 1040 break; 1041 case '.': 1042 case '[': 1043 case '*': 1044 if (!O_ISSET(sp, O_MAGIC)) { 1045 magic = 1; 1046 needlen += 1; 1047 } 1048 break; 1049 default: 1050 needlen += 2; 1051 } 1052 } else 1053 needlen += 1; 1054 break; 1055 case '~': 1056 if (O_ISSET(sp, O_MAGIC)) { 1057 magic = 1; 1058 needlen += sp->repl_len; 1059 } 1060 break; 1061 case '.': 1062 case '[': 1063 case '*': 1064 if (!O_ISSET(sp, O_MAGIC)) { 1065 magic = 1; 1066 needlen += 2; 1067 } 1068 break; 1069 default: 1070 needlen += 1; 1071 break; 1072 } 1073 1074 if (!magic) { 1075 *replacedp = 0; 1076 return (0); 1077 } 1078 1079 /* Get enough memory to hold the final pattern. */ 1080 *replacedp = 1; 1081 GET_SPACE_RETW(sp, bp, blen, needlen); 1082 1083 for (p = *ptrnp, len = *plenp, t = bp; len > 0; ++p, --len) 1084 switch (*p) { 1085 case '\\': 1086 if (len > 1) { 1087 --len; 1088 switch (*++p) { 1089 case '<': 1090 MEMCPY(t, 1091 RE_WSTART, RE_WSTART_LEN); 1092 t += RE_WSTART_LEN; 1093 break; 1094 case '>': 1095 MEMCPY(t, 1096 RE_WSTOP, RE_WSTOP_LEN); 1097 t += RE_WSTOP_LEN; 1098 break; 1099 case '~': 1100 if (O_ISSET(sp, O_MAGIC)) 1101 *t++ = '~'; 1102 else { 1103 MEMCPY(t, 1104 sp->repl, sp->repl_len); 1105 t += sp->repl_len; 1106 } 1107 break; 1108 case '.': 1109 case '[': 1110 case '*': 1111 if (O_ISSET(sp, O_MAGIC)) 1112 *t++ = '\\'; 1113 *t++ = *p; 1114 break; 1115 default: 1116 *t++ = '\\'; 1117 *t++ = *p; 1118 } 1119 } else 1120 *t++ = '\\'; 1121 break; 1122 case '~': 1123 if (O_ISSET(sp, O_MAGIC)) { 1124 MEMCPY(t, sp->repl, sp->repl_len); 1125 t += sp->repl_len; 1126 } else 1127 *t++ = '~'; 1128 break; 1129 case '.': 1130 case '[': 1131 case '*': 1132 if (!O_ISSET(sp, O_MAGIC)) 1133 *t++ = '\\'; 1134 *t++ = *p; 1135 break; 1136 default: 1137 *t++ = *p; 1138 break; 1139 } 1140 1141 *ptrnp = bp; 1142 *plenp = t - bp; 1143 return (0); 1144 } 1145 1146 /* 1147 * re_tag_conv -- 1148 * Convert a tags search path into something that the POSIX 1149 * 1003.2 RE functions can handle. 1150 */ 1151 static int 1152 re_tag_conv(SCR *sp, CHAR_T **ptrnp, size_t *plenp, int *replacedp) 1153 { 1154 size_t blen, len; 1155 int lastdollar; 1156 CHAR_T *bp, *p, *t; 1157 1158 len = *plenp; 1159 1160 /* Max memory usage is 2 times the length of the string. */ 1161 *replacedp = 1; 1162 GET_SPACE_RETW(sp, bp, blen, len * 2); 1163 1164 p = *ptrnp; 1165 t = bp; 1166 1167 /* If the last character is a '/' or '?', we just strip it. */ 1168 if (len > 0 && (p[len - 1] == '/' || p[len - 1] == '?')) 1169 --len; 1170 1171 /* If the next-to-last or last character is a '$', it's magic. */ 1172 if (len > 0 && p[len - 1] == '$') { 1173 --len; 1174 lastdollar = 1; 1175 } else 1176 lastdollar = 0; 1177 1178 /* If the first character is a '/' or '?', we just strip it. */ 1179 if (len > 0 && (p[0] == '/' || p[0] == '?')) { 1180 ++p; 1181 --len; 1182 } 1183 1184 /* If the first or second character is a '^', it's magic. */ 1185 if (p[0] == '^') { 1186 *t++ = *p++; 1187 --len; 1188 } 1189 1190 /* 1191 * Escape every other magic character we can find, meanwhile stripping 1192 * the backslashes ctags inserts when escaping the search delimiter 1193 * characters. 1194 */ 1195 for (; len > 0; --len) { 1196 if (p[0] == '\\' && (p[1] == '/' || p[1] == '?')) { 1197 ++p; 1198 --len; 1199 } else if (STRCHR(L("^.[]$*"), p[0])) 1200 *t++ = '\\'; 1201 *t++ = *p++; 1202 } 1203 if (lastdollar) 1204 *t++ = '$'; 1205 1206 *ptrnp = bp; 1207 *plenp = t - bp; 1208 return (0); 1209 } 1210 1211 /* 1212 * re_cscope_conv -- 1213 * Convert a cscope search path into something that the POSIX 1214 * 1003.2 RE functions can handle. 1215 */ 1216 static int 1217 re_cscope_conv(SCR *sp, CHAR_T **ptrnp, size_t *plenp, int *replacedp) 1218 { 1219 size_t blen, len, nspaces; 1220 CHAR_T *bp, *t; 1221 CHAR_T *p; 1222 CHAR_T *wp; 1223 size_t wlen; 1224 1225 /* 1226 * Each space in the source line printed by cscope represents an 1227 * arbitrary sequence of spaces, tabs, and comments. 1228 */ 1229 #define CSCOPE_RE_SPACE "([ \t]|/\\*([^*]|\\*/)*\\*/)*" 1230 #define CSCOPE_LEN sizeof(CSCOPE_RE_SPACE) - 1 1231 CHAR2INT(sp, CSCOPE_RE_SPACE, CSCOPE_LEN, wp, wlen); 1232 for (nspaces = 0, p = *ptrnp, len = *plenp; len > 0; ++p, --len) 1233 if (*p == ' ') 1234 ++nspaces; 1235 1236 /* 1237 * Allocate plenty of space: 1238 * the string, plus potential escaping characters; 1239 * nspaces + 2 copies of CSCOPE_RE_SPACE; 1240 * ^, $, nul terminator characters. 1241 */ 1242 *replacedp = 1; 1243 len = (p - *ptrnp) * 2 + (nspaces + 2) * sizeof(CSCOPE_RE_SPACE) + 3; 1244 GET_SPACE_RETW(sp, bp, blen, len); 1245 1246 p = *ptrnp; 1247 t = bp; 1248 1249 *t++ = '^'; 1250 MEMCPY(t, wp, wlen); 1251 t += wlen; 1252 1253 for (len = *plenp; len > 0; ++p, --len) 1254 if (*p == ' ') { 1255 MEMCPY(t, wp, wlen); 1256 t += wlen; 1257 } else { 1258 if (STRCHR(L("\\^.[]$*+?()|{}"), *p)) 1259 *t++ = '\\'; 1260 *t++ = *p; 1261 } 1262 1263 MEMCPY(t, wp, wlen); 1264 t += wlen; 1265 *t++ = '$'; 1266 1267 *ptrnp = bp; 1268 *plenp = t - bp; 1269 return (0); 1270 } 1271 1272 /* 1273 * re_error -- 1274 * Report a regular expression error. 1275 * 1276 * PUBLIC: void re_error(SCR *, int, regex_t *); 1277 */ 1278 void 1279 re_error(SCR *sp, int errcode, regex_t *preg) 1280 { 1281 size_t s; 1282 char *oe; 1283 1284 s = regerror(errcode, preg, "", 0); 1285 MALLOC(sp, oe, s); 1286 if (oe != NULL) { 1287 (void)regerror(errcode, preg, oe, s); 1288 msgq(sp, M_ERR, "RE error: %s", oe); 1289 free(oe); 1290 } 1291 } 1292 1293 /* 1294 * re_sub -- 1295 * Do the substitution for a regular expression. 1296 */ 1297 static int 1298 re_sub( 1299 SCR *sp, 1300 CHAR_T *ip, /* Input line. */ 1301 CHAR_T **lbp, 1302 size_t *lbclenp, 1303 size_t *lblenp, 1304 regmatch_t match[10]) 1305 { 1306 enum { C_NOTSET, C_LOWER, C_ONELOWER, C_ONEUPPER, C_UPPER } conv; 1307 size_t lbclen, lblen; /* Local copies. */ 1308 size_t mlen; /* Match length. */ 1309 size_t rpl; /* Remaining replacement length. */ 1310 CHAR_T *rp; /* Replacement pointer. */ 1311 int ch; 1312 int no; /* Match replacement offset. */ 1313 CHAR_T *p, *t; /* Buffer pointers. */ 1314 CHAR_T *lb; /* Local copies. */ 1315 1316 lb = *lbp; /* Get local copies. */ 1317 lbclen = *lbclenp; 1318 lblen = *lblenp; 1319 1320 /* 1321 * QUOTING NOTE: 1322 * 1323 * There are some special sequences that vi provides in the 1324 * replacement patterns. 1325 * & string the RE matched (\& if nomagic set) 1326 * \# n-th regular subexpression 1327 * \E end \U, \L conversion 1328 * \e end \U, \L conversion 1329 * \l convert the next character to lower-case 1330 * \L convert to lower-case, until \E, \e, or end of replacement 1331 * \u convert the next character to upper-case 1332 * \U convert to upper-case, until \E, \e, or end of replacement 1333 * 1334 * Otherwise, since this is the lowest level of replacement, discard 1335 * all escaping characters. This (hopefully) matches historic practice. 1336 */ 1337 #define OUTCH(ch, nltrans) do { \ 1338 ARG_CHAR_T __ch = (ch); \ 1339 e_key_t __value = KEY_VAL(sp, __ch); \ 1340 if (nltrans && (__value == K_CR || __value == K_NL)) { \ 1341 NEEDNEWLINE(sp); \ 1342 sp->newl[sp->newl_cnt++] = lbclen; \ 1343 } else if (conv != C_NOTSET) { \ 1344 switch (conv) { \ 1345 case C_ONELOWER: \ 1346 conv = C_NOTSET; \ 1347 /* FALLTHROUGH */ \ 1348 case C_LOWER: \ 1349 if (ISUPPER(__ch)) \ 1350 __ch = TOLOWER(__ch); \ 1351 break; \ 1352 case C_ONEUPPER: \ 1353 conv = C_NOTSET; \ 1354 /* FALLTHROUGH */ \ 1355 case C_UPPER: \ 1356 if (ISLOWER(__ch)) \ 1357 __ch = TOUPPER(__ch); \ 1358 break; \ 1359 default: \ 1360 abort(); \ 1361 } \ 1362 } \ 1363 NEEDSP(sp, 1, p); \ 1364 *p++ = __ch; \ 1365 ++lbclen; \ 1366 } while (0) 1367 conv = C_NOTSET; 1368 for (rp = sp->repl, rpl = sp->repl_len, p = lb + lbclen; rpl--;) { 1369 switch (ch = *rp++) { 1370 case '&': 1371 if (O_ISSET(sp, O_MAGIC)) { 1372 no = 0; 1373 goto subzero; 1374 } 1375 break; 1376 case '\\': 1377 if (rpl == 0) 1378 break; 1379 --rpl; 1380 switch (ch = *rp) { 1381 case '&': 1382 ++rp; 1383 if (!O_ISSET(sp, O_MAGIC)) { 1384 no = 0; 1385 goto subzero; 1386 } 1387 break; 1388 case '0': case '1': case '2': case '3': case '4': 1389 case '5': case '6': case '7': case '8': case '9': 1390 no = *rp++ - '0'; 1391 subzero: if (match[no].rm_so == -1 || 1392 match[no].rm_eo == -1) 1393 break; 1394 mlen = match[no].rm_eo - match[no].rm_so; 1395 for (t = ip + match[no].rm_so; mlen--; ++t) 1396 OUTCH(*t, 0); 1397 continue; 1398 case 'e': 1399 case 'E': 1400 ++rp; 1401 conv = C_NOTSET; 1402 continue; 1403 case 'l': 1404 ++rp; 1405 conv = C_ONELOWER; 1406 continue; 1407 case 'L': 1408 ++rp; 1409 conv = C_LOWER; 1410 continue; 1411 case 'u': 1412 ++rp; 1413 conv = C_ONEUPPER; 1414 continue; 1415 case 'U': 1416 ++rp; 1417 conv = C_UPPER; 1418 continue; 1419 case '\r': 1420 OUTCH(ch, 0); 1421 continue; 1422 default: 1423 ++rp; 1424 break; 1425 } 1426 } 1427 OUTCH(ch, 1); 1428 } 1429 1430 *lbp = lb; /* Update caller's information. */ 1431 *lbclenp = lbclen; 1432 *lblenp = lblen; 1433 return (0); 1434 } 1435