app.c revision 130562
1/* This is the Assembler Pre-Processor 2 Copyright 1987, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 3 1999, 2000, 2002, 2003 4 Free Software Foundation, Inc. 5 6 This file is part of GAS, the GNU Assembler. 7 8 GAS is free software; you can redistribute it and/or modify 9 it under the terms of the GNU General Public License as published by 10 the Free Software Foundation; either version 2, or (at your option) 11 any later version. 12 13 GAS is distributed in the hope that it will be useful, 14 but WITHOUT ANY WARRANTY; without even the implied warranty of 15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 GNU General Public License for more details. 17 18 You should have received a copy of the GNU General Public License 19 along with GAS; see the file COPYING. If not, write to the Free 20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA 21 02111-1307, USA. */ 22 23/* Modified by Allen Wirfs-Brock, Instantiations Inc 2/90. */ 24/* App, the assembler pre-processor. This pre-processor strips out excess 25 spaces, turns single-quoted characters into a decimal constant, and turns 26 # <number> <filename> <garbage> into a .line <number>\n.file <filename> 27 pair. This needs better error-handling. */ 28 29#include <stdio.h> 30#include "as.h" /* For BAD_CASE() only. */ 31 32#if (__STDC__ != 1) 33#ifndef const 34#define const /* empty */ 35#endif 36#endif 37 38#ifdef TC_M68K 39/* Whether we are scrubbing in m68k MRI mode. This is different from 40 flag_m68k_mri, because the two flags will be affected by the .mri 41 pseudo-op at different times. */ 42static int scrub_m68k_mri; 43 44/* The pseudo-op which switches in and out of MRI mode. See the 45 comment in do_scrub_chars. */ 46static const char mri_pseudo[] = ".mri 0"; 47#else 48#define scrub_m68k_mri 0 49#endif 50 51#if defined TC_ARM && defined OBJ_ELF 52/* The pseudo-op for which we need to special-case `@' characters. 53 See the comment in do_scrub_chars. */ 54static const char symver_pseudo[] = ".symver"; 55static const char * symver_state; 56#endif 57 58static char lex[256]; 59static const char symbol_chars[] = 60"$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789"; 61 62#define LEX_IS_SYMBOL_COMPONENT 1 63#define LEX_IS_WHITESPACE 2 64#define LEX_IS_LINE_SEPARATOR 3 65#define LEX_IS_COMMENT_START 4 66#define LEX_IS_LINE_COMMENT_START 5 67#define LEX_IS_TWOCHAR_COMMENT_1ST 6 68#define LEX_IS_STRINGQUOTE 8 69#define LEX_IS_COLON 9 70#define LEX_IS_NEWLINE 10 71#define LEX_IS_ONECHAR_QUOTE 11 72#ifdef TC_V850 73#define LEX_IS_DOUBLEDASH_1ST 12 74#endif 75#ifdef TC_M32R 76#define DOUBLEBAR_PARALLEL 77#endif 78#ifdef DOUBLEBAR_PARALLEL 79#define LEX_IS_DOUBLEBAR_1ST 13 80#endif 81#define LEX_IS_PARALLEL_SEPARATOR 14 82#define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT) 83#define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE) 84#define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR) 85#define IS_PARALLEL_SEPARATOR(c) (lex[c] == LEX_IS_PARALLEL_SEPARATOR) 86#define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START) 87#define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START) 88#define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE) 89 90static int process_escape (int); 91 92/* FIXME-soon: The entire lexer/parser thingy should be 93 built statically at compile time rather than dynamically 94 each and every time the assembler is run. xoxorich. */ 95 96void 97do_scrub_begin (int m68k_mri ATTRIBUTE_UNUSED) 98{ 99 const char *p; 100 int c; 101 102 lex[' '] = LEX_IS_WHITESPACE; 103 lex['\t'] = LEX_IS_WHITESPACE; 104 lex['\r'] = LEX_IS_WHITESPACE; 105 lex['\n'] = LEX_IS_NEWLINE; 106 lex[':'] = LEX_IS_COLON; 107 108#ifdef TC_M68K 109 scrub_m68k_mri = m68k_mri; 110 111 if (! m68k_mri) 112#endif 113 { 114 lex['"'] = LEX_IS_STRINGQUOTE; 115 116#if ! defined (TC_HPPA) && ! defined (TC_I370) 117 /* I370 uses single-quotes to delimit integer, float constants. */ 118 lex['\''] = LEX_IS_ONECHAR_QUOTE; 119#endif 120 121#ifdef SINGLE_QUOTE_STRINGS 122 lex['\''] = LEX_IS_STRINGQUOTE; 123#endif 124 } 125 126 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop 127 in state 5 of do_scrub_chars must be changed. */ 128 129 /* Note that these override the previous defaults, e.g. if ';' is a 130 comment char, then it isn't a line separator. */ 131 for (p = symbol_chars; *p; ++p) 132 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT; 133 134 for (c = 128; c < 256; ++c) 135 lex[c] = LEX_IS_SYMBOL_COMPONENT; 136 137#ifdef tc_symbol_chars 138 /* This macro permits the processor to specify all characters which 139 may appears in an operand. This will prevent the scrubber from 140 discarding meaningful whitespace in certain cases. The i386 141 backend uses this to support prefixes, which can confuse the 142 scrubber as to whether it is parsing operands or opcodes. */ 143 for (p = tc_symbol_chars; *p; ++p) 144 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT; 145#endif 146 147 /* The m68k backend wants to be able to change comment_chars. */ 148#ifndef tc_comment_chars 149#define tc_comment_chars comment_chars 150#endif 151 for (p = tc_comment_chars; *p; p++) 152 lex[(unsigned char) *p] = LEX_IS_COMMENT_START; 153 154 for (p = line_comment_chars; *p; p++) 155 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START; 156 157 for (p = line_separator_chars; *p; p++) 158 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR; 159 160#ifdef tc_parallel_separator_chars 161 /* This macro permits the processor to specify all characters which 162 separate parallel insns on the same line. */ 163 for (p = tc_parallel_separator_chars; *p; p++) 164 lex[(unsigned char) *p] = LEX_IS_PARALLEL_SEPARATOR; 165#endif 166 167 /* Only allow slash-star comments if slash is not in use. 168 FIXME: This isn't right. We should always permit them. */ 169 if (lex['/'] == 0) 170 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST; 171 172#ifdef TC_M68K 173 if (m68k_mri) 174 { 175 lex['\''] = LEX_IS_STRINGQUOTE; 176 lex[';'] = LEX_IS_COMMENT_START; 177 lex['*'] = LEX_IS_LINE_COMMENT_START; 178 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but 179 then it can't be used in an expression. */ 180 lex['!'] = LEX_IS_LINE_COMMENT_START; 181 } 182#endif 183 184#ifdef TC_V850 185 lex['-'] = LEX_IS_DOUBLEDASH_1ST; 186#endif 187#ifdef DOUBLEBAR_PARALLEL 188 lex['|'] = LEX_IS_DOUBLEBAR_1ST; 189#endif 190#ifdef TC_D30V 191 /* Must do this is we want VLIW instruction with "->" or "<-". */ 192 lex['-'] = LEX_IS_SYMBOL_COMPONENT; 193#endif 194} 195 196/* Saved state of the scrubber. */ 197static int state; 198static int old_state; 199static char *out_string; 200static char out_buf[20]; 201static int add_newlines; 202static char *saved_input; 203static int saved_input_len; 204static char input_buffer[32 * 1024]; 205static const char *mri_state; 206static char mri_last_ch; 207 208/* Data structure for saving the state of app across #include's. Note that 209 app is called asynchronously to the parsing of the .include's, so our 210 state at the time .include is interpreted is completely unrelated. 211 That's why we have to save it all. */ 212 213struct app_save 214{ 215 int state; 216 int old_state; 217 char * out_string; 218 char out_buf[sizeof (out_buf)]; 219 int add_newlines; 220 char * saved_input; 221 int saved_input_len; 222#ifdef TC_M68K 223 int scrub_m68k_mri; 224#endif 225 const char * mri_state; 226 char mri_last_ch; 227#if defined TC_ARM && defined OBJ_ELF 228 const char * symver_state; 229#endif 230}; 231 232char * 233app_push (void) 234{ 235 register struct app_save *saved; 236 237 saved = (struct app_save *) xmalloc (sizeof (*saved)); 238 saved->state = state; 239 saved->old_state = old_state; 240 saved->out_string = out_string; 241 memcpy (saved->out_buf, out_buf, sizeof (out_buf)); 242 saved->add_newlines = add_newlines; 243 if (saved_input == NULL) 244 saved->saved_input = NULL; 245 else 246 { 247 saved->saved_input = xmalloc (saved_input_len); 248 memcpy (saved->saved_input, saved_input, saved_input_len); 249 saved->saved_input_len = saved_input_len; 250 } 251#ifdef TC_M68K 252 saved->scrub_m68k_mri = scrub_m68k_mri; 253#endif 254 saved->mri_state = mri_state; 255 saved->mri_last_ch = mri_last_ch; 256#if defined TC_ARM && defined OBJ_ELF 257 saved->symver_state = symver_state; 258#endif 259 260 /* do_scrub_begin() is not useful, just wastes time. */ 261 262 state = 0; 263 saved_input = NULL; 264 265 return (char *) saved; 266} 267 268void 269app_pop (char *arg) 270{ 271 register struct app_save *saved = (struct app_save *) arg; 272 273 /* There is no do_scrub_end (). */ 274 state = saved->state; 275 old_state = saved->old_state; 276 out_string = saved->out_string; 277 memcpy (out_buf, saved->out_buf, sizeof (out_buf)); 278 add_newlines = saved->add_newlines; 279 if (saved->saved_input == NULL) 280 saved_input = NULL; 281 else 282 { 283 assert (saved->saved_input_len <= (int) (sizeof input_buffer)); 284 memcpy (input_buffer, saved->saved_input, saved->saved_input_len); 285 saved_input = input_buffer; 286 saved_input_len = saved->saved_input_len; 287 free (saved->saved_input); 288 } 289#ifdef TC_M68K 290 scrub_m68k_mri = saved->scrub_m68k_mri; 291#endif 292 mri_state = saved->mri_state; 293 mri_last_ch = saved->mri_last_ch; 294#if defined TC_ARM && defined OBJ_ELF 295 symver_state = saved->symver_state; 296#endif 297 298 free (arg); 299} 300 301/* @@ This assumes that \n &c are the same on host and target. This is not 302 necessarily true. */ 303 304static int 305process_escape (int ch) 306{ 307 switch (ch) 308 { 309 case 'b': 310 return '\b'; 311 case 'f': 312 return '\f'; 313 case 'n': 314 return '\n'; 315 case 'r': 316 return '\r'; 317 case 't': 318 return '\t'; 319 case '\'': 320 return '\''; 321 case '"': 322 return '\"'; 323 default: 324 return ch; 325 } 326} 327 328/* This function is called to process input characters. The GET 329 parameter is used to retrieve more input characters. GET should 330 set its parameter to point to a buffer, and return the length of 331 the buffer; it should return 0 at end of file. The scrubbed output 332 characters are put into the buffer starting at TOSTART; the TOSTART 333 buffer is TOLEN bytes in length. The function returns the number 334 of scrubbed characters put into TOSTART. This will be TOLEN unless 335 end of file was seen. This function is arranged as a state 336 machine, and saves its state so that it may return at any point. 337 This is the way the old code used to work. */ 338 339int 340do_scrub_chars (int (*get) (char *, int), char *tostart, int tolen) 341{ 342 char *to = tostart; 343 char *toend = tostart + tolen; 344 char *from; 345 char *fromend; 346 int fromlen; 347 register int ch, ch2 = 0; 348 349 /*State 0: beginning of normal line 350 1: After first whitespace on line (flush more white) 351 2: After first non-white (opcode) on line (keep 1white) 352 3: after second white on line (into operands) (flush white) 353 4: after putting out a .line, put out digits 354 5: parsing a string, then go to old-state 355 6: putting out \ escape in a "d string. 356 7: After putting out a .appfile, put out string. 357 8: After putting out a .appfile string, flush until newline. 358 9: After seeing symbol char in state 3 (keep 1white after symchar) 359 10: After seeing whitespace in state 9 (keep white before symchar) 360 11: After seeing a symbol character in state 0 (eg a label definition) 361 -1: output string in out_string and go to the state in old_state 362 -2: flush text until a '*' '/' is seen, then go to state old_state 363#ifdef TC_V850 364 12: After seeing a dash, looking for a second dash as a start 365 of comment. 366#endif 367#ifdef DOUBLEBAR_PARALLEL 368 13: After seeing a vertical bar, looking for a second 369 vertical bar as a parallel expression separator. 370#endif 371#ifdef TC_IA64 372 14: After seeing a `(' at state 0, looking for a `)' as 373 predicate. 374 15: After seeing a `(' at state 1, looking for a `)' as 375 predicate. 376#endif 377 */ 378 379 /* I added states 9 and 10 because the MIPS ECOFF assembler uses 380 constructs like ``.loc 1 20''. This was turning into ``.loc 381 120''. States 9 and 10 ensure that a space is never dropped in 382 between characters which could appear in an identifier. Ian 383 Taylor, ian@cygnus.com. 384 385 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works 386 correctly on the PA (and any other target where colons are optional). 387 Jeff Law, law@cs.utah.edu. 388 389 I added state 13 so that something like "cmp r1, r2 || trap #1" does not 390 get squashed into "cmp r1,r2||trap#1", with the all important space 391 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */ 392 393 /* This macro gets the next input character. */ 394 395#define GET() \ 396 (from < fromend \ 397 ? * (unsigned char *) (from++) \ 398 : (saved_input = NULL, \ 399 fromlen = (*get) (input_buffer, sizeof input_buffer), \ 400 from = input_buffer, \ 401 fromend = from + fromlen, \ 402 (fromlen == 0 \ 403 ? EOF \ 404 : * (unsigned char *) (from++)))) 405 406 /* This macro pushes a character back on the input stream. */ 407 408#define UNGET(uch) (*--from = (uch)) 409 410 /* This macro puts a character into the output buffer. If this 411 character fills the output buffer, this macro jumps to the label 412 TOFULL. We use this rather ugly approach because we need to 413 handle two different termination conditions: EOF on the input 414 stream, and a full output buffer. It would be simpler if we 415 always read in the entire input stream before processing it, but 416 I don't want to make such a significant change to the assembler's 417 memory usage. */ 418 419#define PUT(pch) \ 420 do \ 421 { \ 422 *to++ = (pch); \ 423 if (to >= toend) \ 424 goto tofull; \ 425 } \ 426 while (0) 427 428 if (saved_input != NULL) 429 { 430 from = saved_input; 431 fromend = from + saved_input_len; 432 } 433 else 434 { 435 fromlen = (*get) (input_buffer, sizeof input_buffer); 436 if (fromlen == 0) 437 return 0; 438 from = input_buffer; 439 fromend = from + fromlen; 440 } 441 442 while (1) 443 { 444 /* The cases in this switch end with continue, in order to 445 branch back to the top of this while loop and generate the 446 next output character in the appropriate state. */ 447 switch (state) 448 { 449 case -1: 450 ch = *out_string++; 451 if (*out_string == '\0') 452 { 453 state = old_state; 454 old_state = 3; 455 } 456 PUT (ch); 457 continue; 458 459 case -2: 460 for (;;) 461 { 462 do 463 { 464 ch = GET (); 465 466 if (ch == EOF) 467 { 468 as_warn (_("end of file in comment")); 469 goto fromeof; 470 } 471 472 if (ch == '\n') 473 PUT ('\n'); 474 } 475 while (ch != '*'); 476 477 while ((ch = GET ()) == '*') 478 ; 479 480 if (ch == EOF) 481 { 482 as_warn (_("end of file in comment")); 483 goto fromeof; 484 } 485 486 if (ch == '/') 487 break; 488 489 UNGET (ch); 490 } 491 492 state = old_state; 493 UNGET (' '); 494 continue; 495 496 case 4: 497 ch = GET (); 498 if (ch == EOF) 499 goto fromeof; 500 else if (ch >= '0' && ch <= '9') 501 PUT (ch); 502 else 503 { 504 while (ch != EOF && IS_WHITESPACE (ch)) 505 ch = GET (); 506 if (ch == '"') 507 { 508 UNGET (ch); 509 if (scrub_m68k_mri) 510 out_string = "\n\tappfile "; 511 else 512 out_string = "\n\t.appfile "; 513 old_state = 7; 514 state = -1; 515 PUT (*out_string++); 516 } 517 else 518 { 519 while (ch != EOF && ch != '\n') 520 ch = GET (); 521 state = 0; 522 PUT (ch); 523 } 524 } 525 continue; 526 527 case 5: 528 /* We are going to copy everything up to a quote character, 529 with special handling for a backslash. We try to 530 optimize the copying in the simple case without using the 531 GET and PUT macros. */ 532 { 533 char *s; 534 int len; 535 536 for (s = from; s < fromend; s++) 537 { 538 ch = *s; 539 /* This condition must be changed if the type of any 540 other character can be LEX_IS_STRINGQUOTE. */ 541 if (ch == '\\' 542 || ch == '"' 543 || ch == '\'' 544 || ch == '\n') 545 break; 546 } 547 len = s - from; 548 if (len > toend - to) 549 len = toend - to; 550 if (len > 0) 551 { 552 memcpy (to, from, len); 553 to += len; 554 from += len; 555 } 556 } 557 558 ch = GET (); 559 if (ch == EOF) 560 { 561 as_warn (_("end of file in string; inserted '\"'")); 562 state = old_state; 563 UNGET ('\n'); 564 PUT ('"'); 565 } 566 else if (lex[ch] == LEX_IS_STRINGQUOTE) 567 { 568 state = old_state; 569 PUT (ch); 570 } 571#ifndef NO_STRING_ESCAPES 572 else if (ch == '\\') 573 { 574 state = 6; 575 PUT (ch); 576 } 577#endif 578 else if (scrub_m68k_mri && ch == '\n') 579 { 580 /* Just quietly terminate the string. This permits lines like 581 bne label loop if we haven't reach end yet. */ 582 state = old_state; 583 UNGET (ch); 584 PUT ('\''); 585 } 586 else 587 { 588 PUT (ch); 589 } 590 continue; 591 592 case 6: 593 state = 5; 594 ch = GET (); 595 switch (ch) 596 { 597 /* Handle strings broken across lines, by turning '\n' into 598 '\\' and 'n'. */ 599 case '\n': 600 UNGET ('n'); 601 add_newlines++; 602 PUT ('\\'); 603 continue; 604 605 case EOF: 606 as_warn (_("end of file in string; '\"' inserted")); 607 PUT ('"'); 608 continue; 609 610 case '"': 611 case '\\': 612 case 'b': 613 case 'f': 614 case 'n': 615 case 'r': 616 case 't': 617 case 'v': 618 case 'x': 619 case 'X': 620 case '0': 621 case '1': 622 case '2': 623 case '3': 624 case '4': 625 case '5': 626 case '6': 627 case '7': 628 break; 629 630 default: 631#ifdef ONLY_STANDARD_ESCAPES 632 as_warn (_("unknown escape '\\%c' in string; ignored"), ch); 633#endif 634 break; 635 } 636 PUT (ch); 637 continue; 638 639 case 7: 640 ch = GET (); 641 state = 5; 642 old_state = 8; 643 if (ch == EOF) 644 goto fromeof; 645 PUT (ch); 646 continue; 647 648 case 8: 649 do 650 ch = GET (); 651 while (ch != '\n' && ch != EOF); 652 if (ch == EOF) 653 goto fromeof; 654 state = 0; 655 PUT (ch); 656 continue; 657 658#ifdef DOUBLEBAR_PARALLEL 659 case 13: 660 ch = GET (); 661 if (ch != '|') 662 abort (); 663 664 /* Reset back to state 1 and pretend that we are parsing a 665 line from just after the first white space. */ 666 state = 1; 667 PUT ('|'); 668 continue; 669#endif 670 } 671 672 /* OK, we are somewhere in states 0 through 4 or 9 through 11. */ 673 674 /* flushchar: */ 675 ch = GET (); 676 677#ifdef TC_IA64 678 if (ch == '(' && (state == 0 || state == 1)) 679 { 680 state += 14; 681 PUT (ch); 682 continue; 683 } 684 else if (state == 14 || state == 15) 685 { 686 if (ch == ')') 687 { 688 state -= 14; 689 PUT (ch); 690 ch = GET (); 691 } 692 else 693 { 694 PUT (ch); 695 continue; 696 } 697 } 698#endif 699 700 recycle: 701 702#if defined TC_ARM && defined OBJ_ELF 703 /* We need to watch out for .symver directives. See the comment later 704 in this function. */ 705 if (symver_state == NULL) 706 { 707 if ((state == 0 || state == 1) && ch == symver_pseudo[0]) 708 symver_state = symver_pseudo + 1; 709 } 710 else 711 { 712 /* We advance to the next state if we find the right 713 character. */ 714 if (ch != '\0' && (*symver_state == ch)) 715 ++symver_state; 716 else if (*symver_state != '\0') 717 /* We did not get the expected character, or we didn't 718 get a valid terminating character after seeing the 719 entire pseudo-op, so we must go back to the beginning. */ 720 symver_state = NULL; 721 else 722 { 723 /* We've read the entire pseudo-op. If this is the end 724 of the line, go back to the beginning. */ 725 if (IS_NEWLINE (ch)) 726 symver_state = NULL; 727 } 728 } 729#endif /* TC_ARM && OBJ_ELF */ 730 731#ifdef TC_M68K 732 /* We want to have pseudo-ops which control whether we are in 733 MRI mode or not. Unfortunately, since m68k MRI mode affects 734 the scrubber, that means that we need a special purpose 735 recognizer here. */ 736 if (mri_state == NULL) 737 { 738 if ((state == 0 || state == 1) 739 && ch == mri_pseudo[0]) 740 mri_state = mri_pseudo + 1; 741 } 742 else 743 { 744 /* We advance to the next state if we find the right 745 character, or if we need a space character and we get any 746 whitespace character, or if we need a '0' and we get a 747 '1' (this is so that we only need one state to handle 748 ``.mri 0'' and ``.mri 1''). */ 749 if (ch != '\0' 750 && (*mri_state == ch 751 || (*mri_state == ' ' 752 && lex[ch] == LEX_IS_WHITESPACE) 753 || (*mri_state == '0' 754 && ch == '1'))) 755 { 756 mri_last_ch = ch; 757 ++mri_state; 758 } 759 else if (*mri_state != '\0' 760 || (lex[ch] != LEX_IS_WHITESPACE 761 && lex[ch] != LEX_IS_NEWLINE)) 762 { 763 /* We did not get the expected character, or we didn't 764 get a valid terminating character after seeing the 765 entire pseudo-op, so we must go back to the 766 beginning. */ 767 mri_state = NULL; 768 } 769 else 770 { 771 /* We've read the entire pseudo-op. mips_last_ch is 772 either '0' or '1' indicating whether to enter or 773 leave MRI mode. */ 774 do_scrub_begin (mri_last_ch == '1'); 775 mri_state = NULL; 776 777 /* We continue handling the character as usual. The 778 main gas reader must also handle the .mri pseudo-op 779 to control expression parsing and the like. */ 780 } 781 } 782#endif 783 784 if (ch == EOF) 785 { 786 if (state != 0) 787 { 788 as_warn (_("end of file not at end of a line; newline inserted")); 789 state = 0; 790 PUT ('\n'); 791 } 792 goto fromeof; 793 } 794 795 switch (lex[ch]) 796 { 797 case LEX_IS_WHITESPACE: 798 do 799 { 800 ch = GET (); 801 } 802 while (ch != EOF && IS_WHITESPACE (ch)); 803 if (ch == EOF) 804 goto fromeof; 805 806 if (state == 0) 807 { 808 /* Preserve a single whitespace character at the 809 beginning of a line. */ 810 state = 1; 811 UNGET (ch); 812 PUT (' '); 813 break; 814 } 815 816#ifdef KEEP_WHITE_AROUND_COLON 817 if (lex[ch] == LEX_IS_COLON) 818 { 819 /* Only keep this white if there's no white *after* the 820 colon. */ 821 ch2 = GET (); 822 UNGET (ch2); 823 if (!IS_WHITESPACE (ch2)) 824 { 825 state = 9; 826 UNGET (ch); 827 PUT (' '); 828 break; 829 } 830 } 831#endif 832 if (IS_COMMENT (ch) 833 || ch == '/' 834 || IS_LINE_SEPARATOR (ch) 835 || IS_PARALLEL_SEPARATOR (ch)) 836 { 837 if (scrub_m68k_mri) 838 { 839 /* In MRI mode, we keep these spaces. */ 840 UNGET (ch); 841 PUT (' '); 842 break; 843 } 844 goto recycle; 845 } 846 847 /* If we're in state 2 or 11, we've seen a non-white 848 character followed by whitespace. If the next character 849 is ':', this is whitespace after a label name which we 850 normally must ignore. In MRI mode, though, spaces are 851 not permitted between the label and the colon. */ 852 if ((state == 2 || state == 11) 853 && lex[ch] == LEX_IS_COLON 854 && ! scrub_m68k_mri) 855 { 856 state = 1; 857 PUT (ch); 858 break; 859 } 860 861 switch (state) 862 { 863 case 0: 864 state++; 865 goto recycle; /* Punted leading sp */ 866 case 1: 867 /* We can arrive here if we leave a leading whitespace 868 character at the beginning of a line. */ 869 goto recycle; 870 case 2: 871 state = 3; 872 if (to + 1 < toend) 873 { 874 /* Optimize common case by skipping UNGET/GET. */ 875 PUT (' '); /* Sp after opco */ 876 goto recycle; 877 } 878 UNGET (ch); 879 PUT (' '); 880 break; 881 case 3: 882 if (scrub_m68k_mri) 883 { 884 /* In MRI mode, we keep these spaces. */ 885 UNGET (ch); 886 PUT (' '); 887 break; 888 } 889 goto recycle; /* Sp in operands */ 890 case 9: 891 case 10: 892 if (scrub_m68k_mri) 893 { 894 /* In MRI mode, we keep these spaces. */ 895 state = 3; 896 UNGET (ch); 897 PUT (' '); 898 break; 899 } 900 state = 10; /* Sp after symbol char */ 901 goto recycle; 902 case 11: 903 if (LABELS_WITHOUT_COLONS || flag_m68k_mri) 904 state = 1; 905 else 906 { 907 /* We know that ch is not ':', since we tested that 908 case above. Therefore this is not a label, so it 909 must be the opcode, and we've just seen the 910 whitespace after it. */ 911 state = 3; 912 } 913 UNGET (ch); 914 PUT (' '); /* Sp after label definition. */ 915 break; 916 default: 917 BAD_CASE (state); 918 } 919 break; 920 921 case LEX_IS_TWOCHAR_COMMENT_1ST: 922 ch2 = GET (); 923 if (ch2 == '*') 924 { 925 for (;;) 926 { 927 do 928 { 929 ch2 = GET (); 930 if (ch2 != EOF && IS_NEWLINE (ch2)) 931 add_newlines++; 932 } 933 while (ch2 != EOF && ch2 != '*'); 934 935 while (ch2 == '*') 936 ch2 = GET (); 937 938 if (ch2 == EOF || ch2 == '/') 939 break; 940 941 /* This UNGET will ensure that we count newlines 942 correctly. */ 943 UNGET (ch2); 944 } 945 946 if (ch2 == EOF) 947 as_warn (_("end of file in multiline comment")); 948 949 ch = ' '; 950 goto recycle; 951 } 952#ifdef DOUBLESLASH_LINE_COMMENTS 953 else if (ch2 == '/') 954 { 955 do 956 { 957 ch = GET (); 958 } 959 while (ch != EOF && !IS_NEWLINE (ch)); 960 if (ch == EOF) 961 as_warn ("end of file in comment; newline inserted"); 962 state = 0; 963 PUT ('\n'); 964 break; 965 } 966#endif 967 else 968 { 969 if (ch2 != EOF) 970 UNGET (ch2); 971 if (state == 9 || state == 10) 972 state = 3; 973 PUT (ch); 974 } 975 break; 976 977 case LEX_IS_STRINGQUOTE: 978 if (state == 10) 979 { 980 /* Preserve the whitespace in foo "bar". */ 981 UNGET (ch); 982 state = 3; 983 PUT (' '); 984 985 /* PUT didn't jump out. We could just break, but we 986 know what will happen, so optimize a bit. */ 987 ch = GET (); 988 old_state = 3; 989 } 990 else if (state == 9) 991 old_state = 3; 992 else 993 old_state = state; 994 state = 5; 995 PUT (ch); 996 break; 997 998#ifndef IEEE_STYLE 999 case LEX_IS_ONECHAR_QUOTE: 1000 if (state == 10) 1001 { 1002 /* Preserve the whitespace in foo 'b'. */ 1003 UNGET (ch); 1004 state = 3; 1005 PUT (' '); 1006 break; 1007 } 1008 ch = GET (); 1009 if (ch == EOF) 1010 { 1011 as_warn (_("end of file after a one-character quote; \\0 inserted")); 1012 ch = 0; 1013 } 1014 if (ch == '\\') 1015 { 1016 ch = GET (); 1017 if (ch == EOF) 1018 { 1019 as_warn (_("end of file in escape character")); 1020 ch = '\\'; 1021 } 1022 else 1023 ch = process_escape (ch); 1024 } 1025 sprintf (out_buf, "%d", (int) (unsigned char) ch); 1026 1027 /* None of these 'x constants for us. We want 'x'. */ 1028 if ((ch = GET ()) != '\'') 1029 { 1030#ifdef REQUIRE_CHAR_CLOSE_QUOTE 1031 as_warn (_("missing close quote; (assumed)")); 1032#else 1033 if (ch != EOF) 1034 UNGET (ch); 1035#endif 1036 } 1037 if (strlen (out_buf) == 1) 1038 { 1039 PUT (out_buf[0]); 1040 break; 1041 } 1042 if (state == 9) 1043 old_state = 3; 1044 else 1045 old_state = state; 1046 state = -1; 1047 out_string = out_buf; 1048 PUT (*out_string++); 1049 break; 1050#endif 1051 1052 case LEX_IS_COLON: 1053#ifdef KEEP_WHITE_AROUND_COLON 1054 state = 9; 1055#else 1056 if (state == 9 || state == 10) 1057 state = 3; 1058 else if (state != 3) 1059 state = 1; 1060#endif 1061 PUT (ch); 1062 break; 1063 1064 case LEX_IS_NEWLINE: 1065 /* Roll out a bunch of newlines from inside comments, etc. */ 1066 if (add_newlines) 1067 { 1068 --add_newlines; 1069 UNGET (ch); 1070 } 1071 /* Fall through. */ 1072 1073 case LEX_IS_LINE_SEPARATOR: 1074 state = 0; 1075 PUT (ch); 1076 break; 1077 1078 case LEX_IS_PARALLEL_SEPARATOR: 1079 state = 1; 1080 PUT (ch); 1081 break; 1082 1083#ifdef TC_V850 1084 case LEX_IS_DOUBLEDASH_1ST: 1085 ch2 = GET (); 1086 if (ch2 != '-') 1087 { 1088 UNGET (ch2); 1089 goto de_fault; 1090 } 1091 /* Read and skip to end of line. */ 1092 do 1093 { 1094 ch = GET (); 1095 } 1096 while (ch != EOF && ch != '\n'); 1097 1098 if (ch == EOF) 1099 as_warn (_("end of file in comment; newline inserted")); 1100 1101 state = 0; 1102 PUT ('\n'); 1103 break; 1104#endif 1105#ifdef DOUBLEBAR_PARALLEL 1106 case LEX_IS_DOUBLEBAR_1ST: 1107 ch2 = GET (); 1108 UNGET (ch2); 1109 if (ch2 != '|') 1110 goto de_fault; 1111 1112 /* Handle '||' in two states as invoking PUT twice might 1113 result in the first one jumping out of this loop. We'd 1114 then lose track of the state and one '|' char. */ 1115 state = 13; 1116 PUT ('|'); 1117 break; 1118#endif 1119 case LEX_IS_LINE_COMMENT_START: 1120 /* FIXME-someday: The two character comment stuff was badly 1121 thought out. On i386, we want '/' as line comment start 1122 AND we want C style comments. hence this hack. The 1123 whole lexical process should be reworked. xoxorich. */ 1124 if (ch == '/') 1125 { 1126 ch2 = GET (); 1127 if (ch2 == '*') 1128 { 1129 old_state = 3; 1130 state = -2; 1131 break; 1132 } 1133 else 1134 { 1135 UNGET (ch2); 1136 } 1137 } 1138 1139 if (state == 0 || state == 1) /* Only comment at start of line. */ 1140 { 1141 int startch; 1142 1143 startch = ch; 1144 1145 do 1146 { 1147 ch = GET (); 1148 } 1149 while (ch != EOF && IS_WHITESPACE (ch)); 1150 1151 if (ch == EOF) 1152 { 1153 as_warn (_("end of file in comment; newline inserted")); 1154 PUT ('\n'); 1155 break; 1156 } 1157 1158 if (ch < '0' || ch > '9' || state != 0 || startch != '#') 1159 { 1160 /* Not a cpp line. */ 1161 while (ch != EOF && !IS_NEWLINE (ch)) 1162 ch = GET (); 1163 if (ch == EOF) 1164 as_warn (_("end of file in comment; newline inserted")); 1165 state = 0; 1166 PUT ('\n'); 1167 break; 1168 } 1169 /* Looks like `# 123 "filename"' from cpp. */ 1170 UNGET (ch); 1171 old_state = 4; 1172 state = -1; 1173 if (scrub_m68k_mri) 1174 out_string = "\tappline "; 1175 else 1176 out_string = "\t.appline "; 1177 PUT (*out_string++); 1178 break; 1179 } 1180 1181#ifdef TC_D10V 1182 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true. 1183 Trap is the only short insn that has a first operand that is 1184 neither register nor label. 1185 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 . 1186 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is 1187 already LEX_IS_LINE_COMMENT_START. However, it is the 1188 only character in line_comment_chars for d10v, hence we 1189 can recognize it as such. */ 1190 /* An alternative approach would be to reset the state to 1 when 1191 we see '||', '<'- or '->', but that seems to be overkill. */ 1192 if (state == 10) 1193 PUT (' '); 1194#endif 1195 /* We have a line comment character which is not at the 1196 start of a line. If this is also a normal comment 1197 character, fall through. Otherwise treat it as a default 1198 character. */ 1199 if (strchr (tc_comment_chars, ch) == NULL 1200 && (! scrub_m68k_mri 1201 || (ch != '!' && ch != '*'))) 1202 goto de_fault; 1203 if (scrub_m68k_mri 1204 && (ch == '!' || ch == '*' || ch == '#') 1205 && state != 1 1206 && state != 10) 1207 goto de_fault; 1208 /* Fall through. */ 1209 case LEX_IS_COMMENT_START: 1210#if defined TC_ARM && defined OBJ_ELF 1211 /* On the ARM, `@' is the comment character. 1212 Unfortunately this is also a special character in ELF .symver 1213 directives (and .type, though we deal with those another way). 1214 So we check if this line is such a directive, and treat 1215 the character as default if so. This is a hack. */ 1216 if ((symver_state != NULL) && (*symver_state == 0)) 1217 goto de_fault; 1218#endif 1219#ifdef WARN_COMMENTS 1220 if (!found_comment) 1221 as_where (&found_comment_file, &found_comment); 1222#endif 1223 do 1224 { 1225 ch = GET (); 1226 } 1227 while (ch != EOF && !IS_NEWLINE (ch)); 1228 if (ch == EOF) 1229 as_warn (_("end of file in comment; newline inserted")); 1230 state = 0; 1231 PUT ('\n'); 1232 break; 1233 1234 case LEX_IS_SYMBOL_COMPONENT: 1235 if (state == 10) 1236 { 1237 /* This is a symbol character following another symbol 1238 character, with whitespace in between. We skipped 1239 the whitespace earlier, so output it now. */ 1240 UNGET (ch); 1241 state = 3; 1242 PUT (' '); 1243 break; 1244 } 1245 1246 if (state == 3) 1247 state = 9; 1248 1249 /* This is a common case. Quickly copy CH and all the 1250 following symbol component or normal characters. */ 1251 if (to + 1 < toend 1252 && mri_state == NULL 1253#if defined TC_ARM && defined OBJ_ELF 1254 && symver_state == NULL 1255#endif 1256 ) 1257 { 1258 char *s; 1259 int len; 1260 1261 for (s = from; s < fromend; s++) 1262 { 1263 int type; 1264 1265 ch2 = *(unsigned char *) s; 1266 type = lex[ch2]; 1267 if (type != 0 1268 && type != LEX_IS_SYMBOL_COMPONENT) 1269 break; 1270 } 1271 1272 if (s > from) 1273 /* Handle the last character normally, for 1274 simplicity. */ 1275 --s; 1276 1277 len = s - from; 1278 1279 if (len > (toend - to) - 1) 1280 len = (toend - to) - 1; 1281 1282 if (len > 0) 1283 { 1284 PUT (ch); 1285 if (len > 8) 1286 { 1287 memcpy (to, from, len); 1288 to += len; 1289 from += len; 1290 } 1291 else 1292 { 1293 switch (len) 1294 { 1295 case 8: *to++ = *from++; 1296 case 7: *to++ = *from++; 1297 case 6: *to++ = *from++; 1298 case 5: *to++ = *from++; 1299 case 4: *to++ = *from++; 1300 case 3: *to++ = *from++; 1301 case 2: *to++ = *from++; 1302 case 1: *to++ = *from++; 1303 } 1304 } 1305 ch = GET (); 1306 } 1307 } 1308 1309 /* Fall through. */ 1310 default: 1311 de_fault: 1312 /* Some relatively `normal' character. */ 1313 if (state == 0) 1314 { 1315 state = 11; /* Now seeing label definition. */ 1316 } 1317 else if (state == 1) 1318 { 1319 state = 2; /* Ditto. */ 1320 } 1321 else if (state == 9) 1322 { 1323 if (!IS_SYMBOL_COMPONENT (ch)) 1324 state = 3; 1325 } 1326 else if (state == 10) 1327 { 1328 if (ch == '\\') 1329 { 1330 /* Special handling for backslash: a backslash may 1331 be the beginning of a formal parameter (of a 1332 macro) following another symbol character, with 1333 whitespace in between. If that is the case, we 1334 output a space before the parameter. Strictly 1335 speaking, correct handling depends upon what the 1336 macro parameter expands into; if the parameter 1337 expands into something which does not start with 1338 an operand character, then we don't want to keep 1339 the space. We don't have enough information to 1340 make the right choice, so here we are making the 1341 choice which is more likely to be correct. */ 1342 PUT (' '); 1343 } 1344 1345 state = 3; 1346 } 1347 PUT (ch); 1348 break; 1349 } 1350 } 1351 1352 /*NOTREACHED*/ 1353 1354 fromeof: 1355 /* We have reached the end of the input. */ 1356 return to - tostart; 1357 1358 tofull: 1359 /* The output buffer is full. Save any input we have not yet 1360 processed. */ 1361 if (fromend > from) 1362 { 1363 saved_input = from; 1364 saved_input_len = fromend - from; 1365 } 1366 else 1367 saved_input = NULL; 1368 1369 return to - tostart; 1370} 1371 1372