daily update
[binutils.git] / gas / app.c
blobb9832d14b71f02493e5087b66f3a1538f9262be2
1 /* This is the Assembler Pre-Processor
2 Copyright 1987, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3 1999, 2000, 2001, 2002, 2003, 2006, 2007
4 Free Software Foundation, Inc.
6 This file is part of GAS, the GNU Assembler.
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 3, or (at your option)
11 any later version.
13 GAS is distributed in the hope that it will be useful, but WITHOUT
14 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
15 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
16 License for more details.
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, 51 Franklin Street - Fifth Floor, Boston, MA
21 02110-1301, USA. */
23 /* Modified by Allen Wirfs-Brock, Instantiations Inc 2/90. */
24 /* App, the assembler pre-processor. This pre-processor strips out
25 excess spaces, turns single-quoted characters into a decimal
26 constant, and turns the # in # <number> <filename> <garbage> into a
27 .linefile. This needs better error-handling. */
29 #include "as.h"
31 #if (__STDC__ != 1)
32 #ifndef const
33 #define const /* empty */
34 #endif
35 #endif
37 #ifdef H_TICK_HEX
38 int enable_h_tick_hex = 0;
39 #endif
41 #ifdef TC_M68K
42 /* Whether we are scrubbing in m68k MRI mode. This is different from
43 flag_m68k_mri, because the two flags will be affected by the .mri
44 pseudo-op at different times. */
45 static int scrub_m68k_mri;
47 /* The pseudo-op which switches in and out of MRI mode. See the
48 comment in do_scrub_chars. */
49 static const char mri_pseudo[] = ".mri 0";
50 #else
51 #define scrub_m68k_mri 0
52 #endif
54 #if defined TC_ARM && defined OBJ_ELF
55 /* The pseudo-op for which we need to special-case `@' characters.
56 See the comment in do_scrub_chars. */
57 static const char symver_pseudo[] = ".symver";
58 static const char * symver_state;
59 #endif
61 static char lex[256];
62 static const char symbol_chars[] =
63 "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
65 #define LEX_IS_SYMBOL_COMPONENT 1
66 #define LEX_IS_WHITESPACE 2
67 #define LEX_IS_LINE_SEPARATOR 3
68 #define LEX_IS_COMMENT_START 4
69 #define LEX_IS_LINE_COMMENT_START 5
70 #define LEX_IS_TWOCHAR_COMMENT_1ST 6
71 #define LEX_IS_STRINGQUOTE 8
72 #define LEX_IS_COLON 9
73 #define LEX_IS_NEWLINE 10
74 #define LEX_IS_ONECHAR_QUOTE 11
75 #ifdef TC_V850
76 #define LEX_IS_DOUBLEDASH_1ST 12
77 #endif
78 #ifdef TC_M32R
79 #define DOUBLEBAR_PARALLEL
80 #endif
81 #ifdef DOUBLEBAR_PARALLEL
82 #define LEX_IS_DOUBLEBAR_1ST 13
83 #endif
84 #define LEX_IS_PARALLEL_SEPARATOR 14
85 #ifdef H_TICK_HEX
86 #define LEX_IS_H 15
87 #endif
88 #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
89 #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
90 #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
91 #define IS_PARALLEL_SEPARATOR(c) (lex[c] == LEX_IS_PARALLEL_SEPARATOR)
92 #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
93 #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
94 #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
96 static int process_escape (int);
98 /* FIXME-soon: The entire lexer/parser thingy should be
99 built statically at compile time rather than dynamically
100 each and every time the assembler is run. xoxorich. */
102 void
103 do_scrub_begin (int m68k_mri ATTRIBUTE_UNUSED)
105 const char *p;
106 int c;
108 lex[' '] = LEX_IS_WHITESPACE;
109 lex['\t'] = LEX_IS_WHITESPACE;
110 lex['\r'] = LEX_IS_WHITESPACE;
111 lex['\n'] = LEX_IS_NEWLINE;
112 lex[':'] = LEX_IS_COLON;
114 #ifdef TC_M68K
115 scrub_m68k_mri = m68k_mri;
117 if (! m68k_mri)
118 #endif
120 lex['"'] = LEX_IS_STRINGQUOTE;
122 #if ! defined (TC_HPPA) && ! defined (TC_I370)
123 /* I370 uses single-quotes to delimit integer, float constants. */
124 lex['\''] = LEX_IS_ONECHAR_QUOTE;
125 #endif
127 #ifdef SINGLE_QUOTE_STRINGS
128 lex['\''] = LEX_IS_STRINGQUOTE;
129 #endif
132 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
133 in state 5 of do_scrub_chars must be changed. */
135 /* Note that these override the previous defaults, e.g. if ';' is a
136 comment char, then it isn't a line separator. */
137 for (p = symbol_chars; *p; ++p)
138 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
140 for (c = 128; c < 256; ++c)
141 lex[c] = LEX_IS_SYMBOL_COMPONENT;
143 #ifdef tc_symbol_chars
144 /* This macro permits the processor to specify all characters which
145 may appears in an operand. This will prevent the scrubber from
146 discarding meaningful whitespace in certain cases. The i386
147 backend uses this to support prefixes, which can confuse the
148 scrubber as to whether it is parsing operands or opcodes. */
149 for (p = tc_symbol_chars; *p; ++p)
150 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
151 #endif
153 /* The m68k backend wants to be able to change comment_chars. */
154 #ifndef tc_comment_chars
155 #define tc_comment_chars comment_chars
156 #endif
157 for (p = tc_comment_chars; *p; p++)
158 lex[(unsigned char) *p] = LEX_IS_COMMENT_START;
160 for (p = line_comment_chars; *p; p++)
161 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START;
163 for (p = line_separator_chars; *p; p++)
164 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
166 #ifdef tc_parallel_separator_chars
167 /* This macro permits the processor to specify all characters which
168 separate parallel insns on the same line. */
169 for (p = tc_parallel_separator_chars; *p; p++)
170 lex[(unsigned char) *p] = LEX_IS_PARALLEL_SEPARATOR;
171 #endif
173 /* Only allow slash-star comments if slash is not in use.
174 FIXME: This isn't right. We should always permit them. */
175 if (lex['/'] == 0)
176 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
178 #ifdef TC_M68K
179 if (m68k_mri)
181 lex['\''] = LEX_IS_STRINGQUOTE;
182 lex[';'] = LEX_IS_COMMENT_START;
183 lex['*'] = LEX_IS_LINE_COMMENT_START;
184 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
185 then it can't be used in an expression. */
186 lex['!'] = LEX_IS_LINE_COMMENT_START;
188 #endif
190 #ifdef TC_V850
191 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
192 #endif
193 #ifdef DOUBLEBAR_PARALLEL
194 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
195 #endif
196 #ifdef TC_D30V
197 /* Must do this is we want VLIW instruction with "->" or "<-". */
198 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
199 #endif
201 #ifdef H_TICK_HEX
202 if (enable_h_tick_hex)
204 lex['h'] = LEX_IS_H;
205 lex['H'] = LEX_IS_H;
207 #endif
210 /* Saved state of the scrubber. */
211 static int state;
212 static int old_state;
213 static char *out_string;
214 static char out_buf[20];
215 static int add_newlines;
216 static char *saved_input;
217 static int saved_input_len;
218 static char input_buffer[32 * 1024];
219 static const char *mri_state;
220 static char mri_last_ch;
222 /* Data structure for saving the state of app across #include's. Note that
223 app is called asynchronously to the parsing of the .include's, so our
224 state at the time .include is interpreted is completely unrelated.
225 That's why we have to save it all. */
227 struct app_save
229 int state;
230 int old_state;
231 char * out_string;
232 char out_buf[sizeof (out_buf)];
233 int add_newlines;
234 char * saved_input;
235 int saved_input_len;
236 #ifdef TC_M68K
237 int scrub_m68k_mri;
238 #endif
239 const char * mri_state;
240 char mri_last_ch;
241 #if defined TC_ARM && defined OBJ_ELF
242 const char * symver_state;
243 #endif
246 char *
247 app_push (void)
249 register struct app_save *saved;
251 saved = (struct app_save *) xmalloc (sizeof (*saved));
252 saved->state = state;
253 saved->old_state = old_state;
254 saved->out_string = out_string;
255 memcpy (saved->out_buf, out_buf, sizeof (out_buf));
256 saved->add_newlines = add_newlines;
257 if (saved_input == NULL)
258 saved->saved_input = NULL;
259 else
261 saved->saved_input = xmalloc (saved_input_len);
262 memcpy (saved->saved_input, saved_input, saved_input_len);
263 saved->saved_input_len = saved_input_len;
265 #ifdef TC_M68K
266 saved->scrub_m68k_mri = scrub_m68k_mri;
267 #endif
268 saved->mri_state = mri_state;
269 saved->mri_last_ch = mri_last_ch;
270 #if defined TC_ARM && defined OBJ_ELF
271 saved->symver_state = symver_state;
272 #endif
274 /* do_scrub_begin() is not useful, just wastes time. */
276 state = 0;
277 saved_input = NULL;
279 return (char *) saved;
282 void
283 app_pop (char *arg)
285 register struct app_save *saved = (struct app_save *) arg;
287 /* There is no do_scrub_end (). */
288 state = saved->state;
289 old_state = saved->old_state;
290 out_string = saved->out_string;
291 memcpy (out_buf, saved->out_buf, sizeof (out_buf));
292 add_newlines = saved->add_newlines;
293 if (saved->saved_input == NULL)
294 saved_input = NULL;
295 else
297 assert (saved->saved_input_len <= (int) (sizeof input_buffer));
298 memcpy (input_buffer, saved->saved_input, saved->saved_input_len);
299 saved_input = input_buffer;
300 saved_input_len = saved->saved_input_len;
301 free (saved->saved_input);
303 #ifdef TC_M68K
304 scrub_m68k_mri = saved->scrub_m68k_mri;
305 #endif
306 mri_state = saved->mri_state;
307 mri_last_ch = saved->mri_last_ch;
308 #if defined TC_ARM && defined OBJ_ELF
309 symver_state = saved->symver_state;
310 #endif
312 free (arg);
315 /* @@ This assumes that \n &c are the same on host and target. This is not
316 necessarily true. */
318 static int
319 process_escape (int ch)
321 switch (ch)
323 case 'b':
324 return '\b';
325 case 'f':
326 return '\f';
327 case 'n':
328 return '\n';
329 case 'r':
330 return '\r';
331 case 't':
332 return '\t';
333 case '\'':
334 return '\'';
335 case '"':
336 return '\"';
337 default:
338 return ch;
342 /* This function is called to process input characters. The GET
343 parameter is used to retrieve more input characters. GET should
344 set its parameter to point to a buffer, and return the length of
345 the buffer; it should return 0 at end of file. The scrubbed output
346 characters are put into the buffer starting at TOSTART; the TOSTART
347 buffer is TOLEN bytes in length. The function returns the number
348 of scrubbed characters put into TOSTART. This will be TOLEN unless
349 end of file was seen. This function is arranged as a state
350 machine, and saves its state so that it may return at any point.
351 This is the way the old code used to work. */
354 do_scrub_chars (int (*get) (char *, int), char *tostart, int tolen)
356 char *to = tostart;
357 char *toend = tostart + tolen;
358 char *from;
359 char *fromend;
360 int fromlen;
361 register int ch, ch2 = 0;
362 /* Character that started the string we're working on. */
363 static char quotechar;
365 /*State 0: beginning of normal line
366 1: After first whitespace on line (flush more white)
367 2: After first non-white (opcode) on line (keep 1white)
368 3: after second white on line (into operands) (flush white)
369 4: after putting out a .linefile, put out digits
370 5: parsing a string, then go to old-state
371 6: putting out \ escape in a "d string.
372 7: no longer used
373 8: no longer used
374 9: After seeing symbol char in state 3 (keep 1white after symchar)
375 10: After seeing whitespace in state 9 (keep white before symchar)
376 11: After seeing a symbol character in state 0 (eg a label definition)
377 -1: output string in out_string and go to the state in old_state
378 -2: flush text until a '*' '/' is seen, then go to state old_state
379 #ifdef TC_V850
380 12: After seeing a dash, looking for a second dash as a start
381 of comment.
382 #endif
383 #ifdef DOUBLEBAR_PARALLEL
384 13: After seeing a vertical bar, looking for a second
385 vertical bar as a parallel expression separator.
386 #endif
387 #ifdef TC_IA64
388 14: After seeing a `(' at state 0, looking for a `)' as
389 predicate.
390 15: After seeing a `(' at state 1, looking for a `)' as
391 predicate.
392 #endif
393 #ifdef TC_Z80
394 16: After seeing an 'a' or an 'A' at the start of a symbol
395 17: After seeing an 'f' or an 'F' in state 16
396 #endif
399 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
400 constructs like ``.loc 1 20''. This was turning into ``.loc
401 120''. States 9 and 10 ensure that a space is never dropped in
402 between characters which could appear in an identifier. Ian
403 Taylor, ian@cygnus.com.
405 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
406 correctly on the PA (and any other target where colons are optional).
407 Jeff Law, law@cs.utah.edu.
409 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
410 get squashed into "cmp r1,r2||trap#1", with the all important space
411 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
413 /* This macro gets the next input character. */
415 #define GET() \
416 (from < fromend \
417 ? * (unsigned char *) (from++) \
418 : (saved_input = NULL, \
419 fromlen = (*get) (input_buffer, sizeof input_buffer), \
420 from = input_buffer, \
421 fromend = from + fromlen, \
422 (fromlen == 0 \
423 ? EOF \
424 : * (unsigned char *) (from++))))
426 /* This macro pushes a character back on the input stream. */
428 #define UNGET(uch) (*--from = (uch))
430 /* This macro puts a character into the output buffer. If this
431 character fills the output buffer, this macro jumps to the label
432 TOFULL. We use this rather ugly approach because we need to
433 handle two different termination conditions: EOF on the input
434 stream, and a full output buffer. It would be simpler if we
435 always read in the entire input stream before processing it, but
436 I don't want to make such a significant change to the assembler's
437 memory usage. */
439 #define PUT(pch) \
440 do \
442 *to++ = (pch); \
443 if (to >= toend) \
444 goto tofull; \
446 while (0)
448 if (saved_input != NULL)
450 from = saved_input;
451 fromend = from + saved_input_len;
453 else
455 fromlen = (*get) (input_buffer, sizeof input_buffer);
456 if (fromlen == 0)
457 return 0;
458 from = input_buffer;
459 fromend = from + fromlen;
462 while (1)
464 /* The cases in this switch end with continue, in order to
465 branch back to the top of this while loop and generate the
466 next output character in the appropriate state. */
467 switch (state)
469 case -1:
470 ch = *out_string++;
471 if (*out_string == '\0')
473 state = old_state;
474 old_state = 3;
476 PUT (ch);
477 continue;
479 case -2:
480 for (;;)
484 ch = GET ();
486 if (ch == EOF)
488 as_warn (_("end of file in comment"));
489 goto fromeof;
492 if (ch == '\n')
493 PUT ('\n');
495 while (ch != '*');
497 while ((ch = GET ()) == '*')
500 if (ch == EOF)
502 as_warn (_("end of file in comment"));
503 goto fromeof;
506 if (ch == '/')
507 break;
509 UNGET (ch);
512 state = old_state;
513 UNGET (' ');
514 continue;
516 case 4:
517 ch = GET ();
518 if (ch == EOF)
519 goto fromeof;
520 else if (ch >= '0' && ch <= '9')
521 PUT (ch);
522 else
524 while (ch != EOF && IS_WHITESPACE (ch))
525 ch = GET ();
526 if (ch == '"')
528 quotechar = ch;
529 state = 5;
530 old_state = 3;
531 PUT (ch);
533 else
535 while (ch != EOF && ch != '\n')
536 ch = GET ();
537 state = 0;
538 PUT (ch);
541 continue;
543 case 5:
544 /* We are going to copy everything up to a quote character,
545 with special handling for a backslash. We try to
546 optimize the copying in the simple case without using the
547 GET and PUT macros. */
549 char *s;
550 int len;
552 for (s = from; s < fromend; s++)
554 ch = *s;
555 if (ch == '\\'
556 || ch == quotechar
557 || ch == '\n')
558 break;
560 len = s - from;
561 if (len > toend - to)
562 len = toend - to;
563 if (len > 0)
565 memcpy (to, from, len);
566 to += len;
567 from += len;
568 if (to >= toend)
569 goto tofull;
573 ch = GET ();
574 if (ch == EOF)
576 /* This buffer is here specifically so
577 that the UNGET below will work. */
578 static char one_char_buf[1];
580 as_warn (_("end of file in string; '%c' inserted"), quotechar);
581 state = old_state;
582 from = fromend = one_char_buf + 1;
583 fromlen = 1;
584 UNGET ('\n');
585 PUT (quotechar);
587 else if (ch == quotechar)
589 state = old_state;
590 PUT (ch);
592 #ifndef NO_STRING_ESCAPES
593 else if (ch == '\\')
595 state = 6;
596 PUT (ch);
598 #endif
599 else if (scrub_m68k_mri && ch == '\n')
601 /* Just quietly terminate the string. This permits lines like
602 bne label loop if we haven't reach end yet. */
603 state = old_state;
604 UNGET (ch);
605 PUT ('\'');
607 else
609 PUT (ch);
611 continue;
613 case 6:
614 state = 5;
615 ch = GET ();
616 switch (ch)
618 /* Handle strings broken across lines, by turning '\n' into
619 '\\' and 'n'. */
620 case '\n':
621 UNGET ('n');
622 add_newlines++;
623 PUT ('\\');
624 continue;
626 case EOF:
627 as_warn (_("end of file in string; '%c' inserted"), quotechar);
628 PUT (quotechar);
629 continue;
631 case '"':
632 case '\\':
633 case 'b':
634 case 'f':
635 case 'n':
636 case 'r':
637 case 't':
638 case 'v':
639 case 'x':
640 case 'X':
641 case '0':
642 case '1':
643 case '2':
644 case '3':
645 case '4':
646 case '5':
647 case '6':
648 case '7':
649 break;
651 default:
652 #ifdef ONLY_STANDARD_ESCAPES
653 as_warn (_("unknown escape '\\%c' in string; ignored"), ch);
654 #endif
655 break;
657 PUT (ch);
658 continue;
660 #ifdef DOUBLEBAR_PARALLEL
661 case 13:
662 ch = GET ();
663 if (ch != '|')
664 abort ();
666 /* Reset back to state 1 and pretend that we are parsing a
667 line from just after the first white space. */
668 state = 1;
669 PUT ('|');
670 continue;
671 #endif
672 #ifdef TC_Z80
673 case 16:
674 /* We have seen an 'a' at the start of a symbol, look for an 'f'. */
675 ch = GET ();
676 if (ch == 'f' || ch == 'F')
678 state = 17;
679 PUT (ch);
681 else
683 state = 9;
684 break;
686 case 17:
687 /* We have seen "af" at the start of a symbol,
688 a ' here is a part of that symbol. */
689 ch = GET ();
690 state = 9;
691 if (ch == '\'')
692 /* Change to avoid warning about unclosed string. */
693 PUT ('`');
694 else if (ch != EOF)
695 UNGET (ch);
696 break;
697 #endif
700 /* OK, we are somewhere in states 0 through 4 or 9 through 11. */
702 /* flushchar: */
703 ch = GET ();
705 #ifdef TC_IA64
706 if (ch == '(' && (state == 0 || state == 1))
708 state += 14;
709 PUT (ch);
710 continue;
712 else if (state == 14 || state == 15)
714 if (ch == ')')
716 state -= 14;
717 PUT (ch);
718 ch = GET ();
720 else
722 PUT (ch);
723 continue;
726 #endif
728 recycle:
730 #if defined TC_ARM && defined OBJ_ELF
731 /* We need to watch out for .symver directives. See the comment later
732 in this function. */
733 if (symver_state == NULL)
735 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
736 symver_state = symver_pseudo + 1;
738 else
740 /* We advance to the next state if we find the right
741 character. */
742 if (ch != '\0' && (*symver_state == ch))
743 ++symver_state;
744 else if (*symver_state != '\0')
745 /* We did not get the expected character, or we didn't
746 get a valid terminating character after seeing the
747 entire pseudo-op, so we must go back to the beginning. */
748 symver_state = NULL;
749 else
751 /* We've read the entire pseudo-op. If this is the end
752 of the line, go back to the beginning. */
753 if (IS_NEWLINE (ch))
754 symver_state = NULL;
757 #endif /* TC_ARM && OBJ_ELF */
759 #ifdef TC_M68K
760 /* We want to have pseudo-ops which control whether we are in
761 MRI mode or not. Unfortunately, since m68k MRI mode affects
762 the scrubber, that means that we need a special purpose
763 recognizer here. */
764 if (mri_state == NULL)
766 if ((state == 0 || state == 1)
767 && ch == mri_pseudo[0])
768 mri_state = mri_pseudo + 1;
770 else
772 /* We advance to the next state if we find the right
773 character, or if we need a space character and we get any
774 whitespace character, or if we need a '0' and we get a
775 '1' (this is so that we only need one state to handle
776 ``.mri 0'' and ``.mri 1''). */
777 if (ch != '\0'
778 && (*mri_state == ch
779 || (*mri_state == ' '
780 && lex[ch] == LEX_IS_WHITESPACE)
781 || (*mri_state == '0'
782 && ch == '1')))
784 mri_last_ch = ch;
785 ++mri_state;
787 else if (*mri_state != '\0'
788 || (lex[ch] != LEX_IS_WHITESPACE
789 && lex[ch] != LEX_IS_NEWLINE))
791 /* We did not get the expected character, or we didn't
792 get a valid terminating character after seeing the
793 entire pseudo-op, so we must go back to the
794 beginning. */
795 mri_state = NULL;
797 else
799 /* We've read the entire pseudo-op. mips_last_ch is
800 either '0' or '1' indicating whether to enter or
801 leave MRI mode. */
802 do_scrub_begin (mri_last_ch == '1');
803 mri_state = NULL;
805 /* We continue handling the character as usual. The
806 main gas reader must also handle the .mri pseudo-op
807 to control expression parsing and the like. */
810 #endif
812 if (ch == EOF)
814 if (state != 0)
816 as_warn (_("end of file not at end of a line; newline inserted"));
817 state = 0;
818 PUT ('\n');
820 goto fromeof;
823 switch (lex[ch])
825 case LEX_IS_WHITESPACE:
828 ch = GET ();
830 while (ch != EOF && IS_WHITESPACE (ch));
831 if (ch == EOF)
832 goto fromeof;
834 if (state == 0)
836 /* Preserve a single whitespace character at the
837 beginning of a line. */
838 state = 1;
839 UNGET (ch);
840 PUT (' ');
841 break;
844 #ifdef KEEP_WHITE_AROUND_COLON
845 if (lex[ch] == LEX_IS_COLON)
847 /* Only keep this white if there's no white *after* the
848 colon. */
849 ch2 = GET ();
850 if (ch2 != EOF)
851 UNGET (ch2);
852 if (!IS_WHITESPACE (ch2))
854 state = 9;
855 UNGET (ch);
856 PUT (' ');
857 break;
860 #endif
861 if (IS_COMMENT (ch)
862 || ch == '/'
863 || IS_LINE_SEPARATOR (ch)
864 || IS_PARALLEL_SEPARATOR (ch))
866 if (scrub_m68k_mri)
868 /* In MRI mode, we keep these spaces. */
869 UNGET (ch);
870 PUT (' ');
871 break;
873 goto recycle;
876 /* If we're in state 2 or 11, we've seen a non-white
877 character followed by whitespace. If the next character
878 is ':', this is whitespace after a label name which we
879 normally must ignore. In MRI mode, though, spaces are
880 not permitted between the label and the colon. */
881 if ((state == 2 || state == 11)
882 && lex[ch] == LEX_IS_COLON
883 && ! scrub_m68k_mri)
885 state = 1;
886 PUT (ch);
887 break;
890 switch (state)
892 case 1:
893 /* We can arrive here if we leave a leading whitespace
894 character at the beginning of a line. */
895 goto recycle;
896 case 2:
897 state = 3;
898 if (to + 1 < toend)
900 /* Optimize common case by skipping UNGET/GET. */
901 PUT (' '); /* Sp after opco */
902 goto recycle;
904 UNGET (ch);
905 PUT (' ');
906 break;
907 case 3:
908 if (scrub_m68k_mri)
910 /* In MRI mode, we keep these spaces. */
911 UNGET (ch);
912 PUT (' ');
913 break;
915 goto recycle; /* Sp in operands */
916 case 9:
917 case 10:
918 if (scrub_m68k_mri)
920 /* In MRI mode, we keep these spaces. */
921 state = 3;
922 UNGET (ch);
923 PUT (' ');
924 break;
926 state = 10; /* Sp after symbol char */
927 goto recycle;
928 case 11:
929 if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
930 state = 1;
931 else
933 /* We know that ch is not ':', since we tested that
934 case above. Therefore this is not a label, so it
935 must be the opcode, and we've just seen the
936 whitespace after it. */
937 state = 3;
939 UNGET (ch);
940 PUT (' '); /* Sp after label definition. */
941 break;
942 default:
943 BAD_CASE (state);
945 break;
947 case LEX_IS_TWOCHAR_COMMENT_1ST:
948 ch2 = GET ();
949 if (ch2 == '*')
951 for (;;)
955 ch2 = GET ();
956 if (ch2 != EOF && IS_NEWLINE (ch2))
957 add_newlines++;
959 while (ch2 != EOF && ch2 != '*');
961 while (ch2 == '*')
962 ch2 = GET ();
964 if (ch2 == EOF || ch2 == '/')
965 break;
967 /* This UNGET will ensure that we count newlines
968 correctly. */
969 UNGET (ch2);
972 if (ch2 == EOF)
973 as_warn (_("end of file in multiline comment"));
975 ch = ' ';
976 goto recycle;
978 #ifdef DOUBLESLASH_LINE_COMMENTS
979 else if (ch2 == '/')
983 ch = GET ();
985 while (ch != EOF && !IS_NEWLINE (ch));
986 if (ch == EOF)
987 as_warn ("end of file in comment; newline inserted");
988 state = 0;
989 PUT ('\n');
990 break;
992 #endif
993 else
995 if (ch2 != EOF)
996 UNGET (ch2);
997 if (state == 9 || state == 10)
998 state = 3;
999 PUT (ch);
1001 break;
1003 case LEX_IS_STRINGQUOTE:
1004 quotechar = ch;
1005 if (state == 10)
1007 /* Preserve the whitespace in foo "bar". */
1008 UNGET (ch);
1009 state = 3;
1010 PUT (' ');
1012 /* PUT didn't jump out. We could just break, but we
1013 know what will happen, so optimize a bit. */
1014 ch = GET ();
1015 old_state = 3;
1017 else if (state == 9)
1018 old_state = 3;
1019 else
1020 old_state = state;
1021 state = 5;
1022 PUT (ch);
1023 break;
1025 #ifndef IEEE_STYLE
1026 case LEX_IS_ONECHAR_QUOTE:
1027 #ifdef H_TICK_HEX
1028 if (state == 9 && enable_h_tick_hex)
1030 char c;
1032 c = GET ();
1033 as_warn ("'%c found after symbol", c);
1034 UNGET (c);
1036 #endif
1037 if (state == 10)
1039 /* Preserve the whitespace in foo 'b'. */
1040 UNGET (ch);
1041 state = 3;
1042 PUT (' ');
1043 break;
1045 ch = GET ();
1046 if (ch == EOF)
1048 as_warn (_("end of file after a one-character quote; \\0 inserted"));
1049 ch = 0;
1051 if (ch == '\\')
1053 ch = GET ();
1054 if (ch == EOF)
1056 as_warn (_("end of file in escape character"));
1057 ch = '\\';
1059 else
1060 ch = process_escape (ch);
1062 sprintf (out_buf, "%d", (int) (unsigned char) ch);
1064 /* None of these 'x constants for us. We want 'x'. */
1065 if ((ch = GET ()) != '\'')
1067 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
1068 as_warn (_("missing close quote; (assumed)"));
1069 #else
1070 if (ch != EOF)
1071 UNGET (ch);
1072 #endif
1074 if (strlen (out_buf) == 1)
1076 PUT (out_buf[0]);
1077 break;
1079 if (state == 9)
1080 old_state = 3;
1081 else
1082 old_state = state;
1083 state = -1;
1084 out_string = out_buf;
1085 PUT (*out_string++);
1086 break;
1087 #endif
1089 case LEX_IS_COLON:
1090 #ifdef KEEP_WHITE_AROUND_COLON
1091 state = 9;
1092 #else
1093 if (state == 9 || state == 10)
1094 state = 3;
1095 else if (state != 3)
1096 state = 1;
1097 #endif
1098 PUT (ch);
1099 break;
1101 case LEX_IS_NEWLINE:
1102 /* Roll out a bunch of newlines from inside comments, etc. */
1103 if (add_newlines)
1105 --add_newlines;
1106 UNGET (ch);
1108 /* Fall through. */
1110 case LEX_IS_LINE_SEPARATOR:
1111 state = 0;
1112 PUT (ch);
1113 break;
1115 case LEX_IS_PARALLEL_SEPARATOR:
1116 state = 1;
1117 PUT (ch);
1118 break;
1120 #ifdef TC_V850
1121 case LEX_IS_DOUBLEDASH_1ST:
1122 ch2 = GET ();
1123 if (ch2 != '-')
1125 if (ch2 != EOF)
1126 UNGET (ch2);
1127 goto de_fault;
1129 /* Read and skip to end of line. */
1132 ch = GET ();
1134 while (ch != EOF && ch != '\n');
1136 if (ch == EOF)
1137 as_warn (_("end of file in comment; newline inserted"));
1139 state = 0;
1140 PUT ('\n');
1141 break;
1142 #endif
1143 #ifdef DOUBLEBAR_PARALLEL
1144 case LEX_IS_DOUBLEBAR_1ST:
1145 ch2 = GET ();
1146 if (ch2 != EOF)
1147 UNGET (ch2);
1148 if (ch2 != '|')
1149 goto de_fault;
1151 /* Handle '||' in two states as invoking PUT twice might
1152 result in the first one jumping out of this loop. We'd
1153 then lose track of the state and one '|' char. */
1154 state = 13;
1155 PUT ('|');
1156 break;
1157 #endif
1158 case LEX_IS_LINE_COMMENT_START:
1159 /* FIXME-someday: The two character comment stuff was badly
1160 thought out. On i386, we want '/' as line comment start
1161 AND we want C style comments. hence this hack. The
1162 whole lexical process should be reworked. xoxorich. */
1163 if (ch == '/')
1165 ch2 = GET ();
1166 if (ch2 == '*')
1168 old_state = 3;
1169 state = -2;
1170 break;
1172 else
1174 UNGET (ch2);
1178 if (state == 0 || state == 1) /* Only comment at start of line. */
1180 int startch;
1182 startch = ch;
1186 ch = GET ();
1188 while (ch != EOF && IS_WHITESPACE (ch));
1190 if (ch == EOF)
1192 as_warn (_("end of file in comment; newline inserted"));
1193 PUT ('\n');
1194 break;
1197 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1199 /* Not a cpp line. */
1200 while (ch != EOF && !IS_NEWLINE (ch))
1201 ch = GET ();
1202 if (ch == EOF)
1203 as_warn (_("end of file in comment; newline inserted"));
1204 state = 0;
1205 PUT ('\n');
1206 break;
1208 /* Looks like `# 123 "filename"' from cpp. */
1209 UNGET (ch);
1210 old_state = 4;
1211 state = -1;
1212 if (scrub_m68k_mri)
1213 out_string = "\tlinefile ";
1214 else
1215 out_string = "\t.linefile ";
1216 PUT (*out_string++);
1217 break;
1220 #ifdef TC_D10V
1221 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1222 Trap is the only short insn that has a first operand that is
1223 neither register nor label.
1224 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1225 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is
1226 already LEX_IS_LINE_COMMENT_START. However, it is the
1227 only character in line_comment_chars for d10v, hence we
1228 can recognize it as such. */
1229 /* An alternative approach would be to reset the state to 1 when
1230 we see '||', '<'- or '->', but that seems to be overkill. */
1231 if (state == 10)
1232 PUT (' ');
1233 #endif
1234 /* We have a line comment character which is not at the
1235 start of a line. If this is also a normal comment
1236 character, fall through. Otherwise treat it as a default
1237 character. */
1238 if (strchr (tc_comment_chars, ch) == NULL
1239 && (! scrub_m68k_mri
1240 || (ch != '!' && ch != '*')))
1241 goto de_fault;
1242 if (scrub_m68k_mri
1243 && (ch == '!' || ch == '*' || ch == '#')
1244 && state != 1
1245 && state != 10)
1246 goto de_fault;
1247 /* Fall through. */
1248 case LEX_IS_COMMENT_START:
1249 #if defined TC_ARM && defined OBJ_ELF
1250 /* On the ARM, `@' is the comment character.
1251 Unfortunately this is also a special character in ELF .symver
1252 directives (and .type, though we deal with those another way).
1253 So we check if this line is such a directive, and treat
1254 the character as default if so. This is a hack. */
1255 if ((symver_state != NULL) && (*symver_state == 0))
1256 goto de_fault;
1257 #endif
1259 #ifdef TC_ARM
1260 /* For the ARM, care is needed not to damage occurrences of \@
1261 by stripping the @ onwards. Yuck. */
1262 if (to > tostart && *(to - 1) == '\\')
1263 /* Do not treat the @ as a start-of-comment. */
1264 goto de_fault;
1265 #endif
1267 #ifdef WARN_COMMENTS
1268 if (!found_comment)
1269 as_where (&found_comment_file, &found_comment);
1270 #endif
1273 ch = GET ();
1275 while (ch != EOF && !IS_NEWLINE (ch));
1276 if (ch == EOF)
1277 as_warn (_("end of file in comment; newline inserted"));
1278 state = 0;
1279 PUT ('\n');
1280 break;
1282 #ifdef H_TICK_HEX
1283 case LEX_IS_H:
1284 /* Look for strings like H'[0-9A-Fa-f] and if found, replace
1285 the H' with 0x to make them gas-style hex characters. */
1286 if (enable_h_tick_hex)
1288 char quot;
1290 quot = GET ();
1291 if (quot == '\'')
1293 UNGET ('x');
1294 ch = '0';
1296 else
1297 UNGET (quot);
1299 /* FALL THROUGH */
1300 #endif
1302 case LEX_IS_SYMBOL_COMPONENT:
1303 if (state == 10)
1305 /* This is a symbol character following another symbol
1306 character, with whitespace in between. We skipped
1307 the whitespace earlier, so output it now. */
1308 UNGET (ch);
1309 state = 3;
1310 PUT (' ');
1311 break;
1314 #ifdef TC_Z80
1315 /* "af'" is a symbol containing '\''. */
1316 if (state == 3 && (ch == 'a' || ch == 'A'))
1318 state = 16;
1319 PUT (ch);
1320 ch = GET ();
1321 if (ch == 'f' || ch == 'F')
1323 state = 17;
1324 PUT (ch);
1325 break;
1327 else
1329 state = 9;
1330 if (!IS_SYMBOL_COMPONENT (ch))
1332 if (ch != EOF)
1333 UNGET (ch);
1334 break;
1338 #endif
1339 if (state == 3)
1340 state = 9;
1342 /* This is a common case. Quickly copy CH and all the
1343 following symbol component or normal characters. */
1344 if (to + 1 < toend
1345 && mri_state == NULL
1346 #if defined TC_ARM && defined OBJ_ELF
1347 && symver_state == NULL
1348 #endif
1351 char *s;
1352 int len;
1354 for (s = from; s < fromend; s++)
1356 int type;
1358 ch2 = *(unsigned char *) s;
1359 type = lex[ch2];
1360 if (type != 0
1361 && type != LEX_IS_SYMBOL_COMPONENT)
1362 break;
1365 if (s > from)
1366 /* Handle the last character normally, for
1367 simplicity. */
1368 --s;
1370 len = s - from;
1372 if (len > (toend - to) - 1)
1373 len = (toend - to) - 1;
1375 if (len > 0)
1377 PUT (ch);
1378 memcpy (to, from, len);
1379 to += len;
1380 from += len;
1381 if (to >= toend)
1382 goto tofull;
1383 ch = GET ();
1387 /* Fall through. */
1388 default:
1389 de_fault:
1390 /* Some relatively `normal' character. */
1391 if (state == 0)
1393 state = 11; /* Now seeing label definition. */
1395 else if (state == 1)
1397 state = 2; /* Ditto. */
1399 else if (state == 9)
1401 if (!IS_SYMBOL_COMPONENT (ch))
1402 state = 3;
1404 else if (state == 10)
1406 if (ch == '\\')
1408 /* Special handling for backslash: a backslash may
1409 be the beginning of a formal parameter (of a
1410 macro) following another symbol character, with
1411 whitespace in between. If that is the case, we
1412 output a space before the parameter. Strictly
1413 speaking, correct handling depends upon what the
1414 macro parameter expands into; if the parameter
1415 expands into something which does not start with
1416 an operand character, then we don't want to keep
1417 the space. We don't have enough information to
1418 make the right choice, so here we are making the
1419 choice which is more likely to be correct. */
1420 if (to + 1 >= toend)
1422 /* If we're near the end of the buffer, save the
1423 character for the next time round. Otherwise
1424 we'll lose our state. */
1425 UNGET (ch);
1426 goto tofull;
1428 *to++ = ' ';
1431 state = 3;
1433 PUT (ch);
1434 break;
1438 /*NOTREACHED*/
1440 fromeof:
1441 /* We have reached the end of the input. */
1442 return to - tostart;
1444 tofull:
1445 /* The output buffer is full. Save any input we have not yet
1446 processed. */
1447 if (fromend > from)
1449 saved_input = from;
1450 saved_input_len = fromend - from;
1452 else
1453 saved_input = NULL;
1455 return to - tostart;