source: trunk/src/binutils/gas/app.c@ 10

Last change on this file since 10 was 10, checked in by bird, 22 years ago

Initial revision
  • Property cvs2svn:cvs-rev set to 1.1
  • Property svn:eol-style set to native
  • Property svn:executable set to *
File size: 32.2 KB
Line 
1/* This is the Assembler Pre-Processor
2 Copyright 1987, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3 1999, 2000
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#else
44#define scrub_m68k_mri 0
45#endif
46
47/* The pseudo-op which switches in and out of MRI mode. See the
48 comment in do_scrub_chars. */
49static const char mri_pseudo[] = ".mri 0";
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 IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
82#define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
83#define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
84#define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
85#define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
86#define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
87
88static int process_escape PARAMS ((int));
89
90/* FIXME-soon: The entire lexer/parser thingy should be
91 built statically at compile time rather than dynamically
92 each and every time the assembler is run. xoxorich. */
93
94void
95do_scrub_begin (m68k_mri)
96 int m68k_mri ATTRIBUTE_UNUSED;
97{
98 const char *p;
99 int c;
100
101 lex[' '] = LEX_IS_WHITESPACE;
102 lex['\t'] = LEX_IS_WHITESPACE;
103 lex['\r'] = LEX_IS_WHITESPACE;
104 lex['\n'] = LEX_IS_NEWLINE;
105 lex[':'] = LEX_IS_COLON;
106
107#ifdef TC_M68K
108 scrub_m68k_mri = m68k_mri;
109
110 if (! m68k_mri)
111#endif
112 {
113 lex['"'] = LEX_IS_STRINGQUOTE;
114
115#if ! defined (TC_HPPA) && ! defined (TC_I370)
116 /* I370 uses single-quotes to delimit integer, float constants */
117 lex['\''] = LEX_IS_ONECHAR_QUOTE;
118#endif
119
120#ifdef SINGLE_QUOTE_STRINGS
121 lex['\''] = LEX_IS_STRINGQUOTE;
122#endif
123 }
124
125 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
126 in state 5 of do_scrub_chars must be changed. */
127
128 /* Note that these override the previous defaults, e.g. if ';' is a
129 comment char, then it isn't a line separator. */
130 for (p = symbol_chars; *p; ++p)
131 {
132 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
133 } /* declare symbol characters */
134
135 for (c = 128; c < 256; ++c)
136 lex[c] = LEX_IS_SYMBOL_COMPONENT;
137
138#ifdef tc_symbol_chars
139 /* This macro permits the processor to specify all characters which
140 may appears in an operand. This will prevent the scrubber from
141 discarding meaningful whitespace in certain cases. The i386
142 backend uses this to support prefixes, which can confuse the
143 scrubber as to whether it is parsing operands or opcodes. */
144 for (p = tc_symbol_chars; *p; ++p)
145 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
146#endif
147
148 /* The m68k backend wants to be able to change comment_chars. */
149#ifndef tc_comment_chars
150#define tc_comment_chars comment_chars
151#endif
152 for (p = tc_comment_chars; *p; p++)
153 {
154 lex[(unsigned char) *p] = LEX_IS_COMMENT_START;
155 } /* declare comment chars */
156
157 for (p = line_comment_chars; *p; p++)
158 {
159 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START;
160 } /* declare line comment chars */
161
162 for (p = line_separator_chars; *p; p++)
163 {
164 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
165 } /* declare line separators */
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 {
171 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
172 }
173
174#ifdef TC_M68K
175 if (m68k_mri)
176 {
177 lex['\''] = LEX_IS_STRINGQUOTE;
178 lex[';'] = LEX_IS_COMMENT_START;
179 lex['*'] = LEX_IS_LINE_COMMENT_START;
180 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
181 then it can't be used in an expression. */
182 lex['!'] = LEX_IS_LINE_COMMENT_START;
183 }
184#endif
185
186#ifdef TC_V850
187 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
188#endif
189#ifdef DOUBLEBAR_PARALLEL
190 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
191#endif
192#ifdef TC_D30V
193 /* must do this is we want VLIW instruction with "->" or "<-" */
194 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
195#endif
196} /* do_scrub_begin() */
197
198/* Saved state of the scrubber */
199static int state;
200static int old_state;
201static char *out_string;
202static char out_buf[20];
203static int add_newlines;
204static char *saved_input;
205static int saved_input_len;
206static char input_buffer[32 * 1024];
207static const char *mri_state;
208static char mri_last_ch;
209
210/* Data structure for saving the state of app across #include's. Note that
211 app is called asynchronously to the parsing of the .include's, so our
212 state at the time .include is interpreted is completely unrelated.
213 That's why we have to save it all. */
214
215struct app_save {
216 int state;
217 int old_state;
218 char * out_string;
219 char out_buf[sizeof (out_buf)];
220 int add_newlines;
221 char * saved_input;
222 int saved_input_len;
223#ifdef TC_M68K
224 int scrub_m68k_mri;
225#endif
226 const char * mri_state;
227 char mri_last_ch;
228#if defined TC_ARM && defined OBJ_ELF
229 const char * symver_state;
230#endif
231};
232
233char *
234app_push ()
235{
236 register struct app_save *saved;
237
238 saved = (struct app_save *) xmalloc (sizeof (*saved));
239 saved->state = state;
240 saved->old_state = old_state;
241 saved->out_string = out_string;
242 memcpy (saved->out_buf, out_buf, sizeof (out_buf));
243 saved->add_newlines = add_newlines;
244 if (saved_input == NULL)
245 saved->saved_input = NULL;
246 else
247 {
248 saved->saved_input = xmalloc (saved_input_len);
249 memcpy (saved->saved_input, saved_input, saved_input_len);
250 saved->saved_input_len = saved_input_len;
251 }
252#ifdef TC_M68K
253 saved->scrub_m68k_mri = scrub_m68k_mri;
254#endif
255 saved->mri_state = mri_state;
256 saved->mri_last_ch = mri_last_ch;
257#if defined TC_ARM && defined OBJ_ELF
258 saved->symver_state = symver_state;
259#endif
260
261 /* do_scrub_begin() is not useful, just wastes time. */
262
263 state = 0;
264 saved_input = NULL;
265
266 return (char *) saved;
267}
268
269void
270app_pop (arg)
271 char *arg;
272{
273 register struct app_save *saved = (struct app_save *) arg;
274
275 /* There is no do_scrub_end (). */
276 state = saved->state;
277 old_state = saved->old_state;
278 out_string = saved->out_string;
279 memcpy (out_buf, saved->out_buf, sizeof (out_buf));
280 add_newlines = saved->add_newlines;
281 if (saved->saved_input == NULL)
282 saved_input = NULL;
283 else
284 {
285 assert (saved->saved_input_len <= (int) (sizeof input_buffer));
286 memcpy (input_buffer, saved->saved_input, saved->saved_input_len);
287 saved_input = input_buffer;
288 saved_input_len = saved->saved_input_len;
289 free (saved->saved_input);
290 }
291#ifdef TC_M68K
292 scrub_m68k_mri = saved->scrub_m68k_mri;
293#endif
294 mri_state = saved->mri_state;
295 mri_last_ch = saved->mri_last_ch;
296#if defined TC_ARM && defined OBJ_ELF
297 symver_state = saved->symver_state;
298#endif
299
300 free (arg);
301} /* app_pop() */
302
303/* @@ This assumes that \n &c are the same on host and target. This is not
304 necessarily true. */
305static int
306process_escape (ch)
307 int ch;
308{
309 switch (ch)
310 {
311 case 'b':
312 return '\b';
313 case 'f':
314 return '\f';
315 case 'n':
316 return '\n';
317 case 'r':
318 return '\r';
319 case 't':
320 return '\t';
321 case '\'':
322 return '\'';
323 case '"':
324 return '\"';
325 default:
326 return ch;
327 }
328}
329
330/* This function is called to process input characters. The GET
331 parameter is used to retrieve more input characters. GET should
332 set its parameter to point to a buffer, and return the length of
333 the buffer; it should return 0 at end of file. The scrubbed output
334 characters are put into the buffer starting at TOSTART; the TOSTART
335 buffer is TOLEN bytes in length. The function returns the number
336 of scrubbed characters put into TOSTART. This will be TOLEN unless
337 end of file was seen. This function is arranged as a state
338 machine, and saves its state so that it may return at any point.
339 This is the way the old code used to work. */
340
341int
342do_scrub_chars (get, tostart, tolen)
343 int (*get) PARAMS ((char *, int));
344 char *tostart;
345 int tolen;
346{
347 char *to = tostart;
348 char *toend = tostart + tolen;
349 char *from;
350 char *fromend;
351 int fromlen;
352 register int ch, ch2 = 0;
353
354 /*State 0: beginning of normal line
355 1: After first whitespace on line (flush more white)
356 2: After first non-white (opcode) on line (keep 1white)
357 3: after second white on line (into operands) (flush white)
358 4: after putting out a .line, put out digits
359 5: parsing a string, then go to old-state
360 6: putting out \ escape in a "d string.
361 7: After putting out a .appfile, put out string.
362 8: After putting out a .appfile string, flush until newline.
363 9: After seeing symbol char in state 3 (keep 1white after symchar)
364 10: After seeing whitespace in state 9 (keep white before symchar)
365 11: After seeing a symbol character in state 0 (eg a label definition)
366 -1: output string in out_string and go to the state in old_state
367 -2: flush text until a '*' '/' is seen, then go to state old_state
368#ifdef TC_V850
369 12: After seeing a dash, looking for a second dash as a start of comment.
370#endif
371#ifdef DOUBLEBAR_PARALLEL
372 13: After seeing a vertical bar, looking for a second vertical bar as a parallel expression seperator.
373#endif
374 */
375
376 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
377 constructs like ``.loc 1 20''. This was turning into ``.loc
378 120''. States 9 and 10 ensure that a space is never dropped in
379 between characters which could appear in a identifier. Ian
380 Taylor, ian@cygnus.com.
381
382 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
383 correctly on the PA (and any other target where colons are optional).
384 Jeff Law, law@cs.utah.edu.
385
386 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
387 get squashed into "cmp r1,r2||trap#1", with the all important space
388 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
389
390 /* This macro gets the next input character. */
391
392#define GET() \
393 (from < fromend \
394 ? * (unsigned char *) (from++) \
395 : (saved_input = NULL, \
396 fromlen = (*get) (input_buffer, sizeof input_buffer), \
397 from = input_buffer, \
398 fromend = from + fromlen, \
399 (fromlen == 0 \
400 ? EOF \
401 : * (unsigned char *) (from++))))
402
403 /* This macro pushes a character back on the input stream. */
404
405#define UNGET(uch) (*--from = (uch))
406
407 /* This macro puts a character into the output buffer. If this
408 character fills the output buffer, this macro jumps to the label
409 TOFULL. We use this rather ugly approach because we need to
410 handle two different termination conditions: EOF on the input
411 stream, and a full output buffer. It would be simpler if we
412 always read in the entire input stream before processing it, but
413 I don't want to make such a significant change to the assembler's
414 memory usage. */
415
416#define PUT(pch) \
417 do \
418 { \
419 *to++ = (pch); \
420 if (to >= toend) \
421 goto tofull; \
422 } \
423 while (0)
424
425 if (saved_input != NULL)
426 {
427 from = saved_input;
428 fromend = from + saved_input_len;
429 }
430 else
431 {
432 fromlen = (*get) (input_buffer, sizeof input_buffer);
433 if (fromlen == 0)
434 return 0;
435 from = input_buffer;
436 fromend = from + fromlen;
437 }
438
439 while (1)
440 {
441 /* The cases in this switch end with continue, in order to
442 branch back to the top of this while loop and generate the
443 next output character in the appropriate state. */
444 switch (state)
445 {
446 case -1:
447 ch = *out_string++;
448 if (*out_string == '\0')
449 {
450 state = old_state;
451 old_state = 3;
452 }
453 PUT (ch);
454 continue;
455
456 case -2:
457 for (;;)
458 {
459 do
460 {
461 ch = GET ();
462
463 if (ch == EOF)
464 {
465 as_warn (_("end of file in comment"));
466 goto fromeof;
467 }
468
469 if (ch == '\n')
470 PUT ('\n');
471 }
472 while (ch != '*');
473
474 while ((ch = GET ()) == '*')
475 ;
476
477 if (ch == EOF)
478 {
479 as_warn (_("end of file in comment"));
480 goto fromeof;
481 }
482
483 if (ch == '/')
484 break;
485
486 UNGET (ch);
487 }
488
489 state = old_state;
490 UNGET (' ');
491 continue;
492
493 case 4:
494 ch = GET ();
495 if (ch == EOF)
496 goto fromeof;
497 else if (ch >= '0' && ch <= '9')
498 PUT (ch);
499 else
500 {
501 while (ch != EOF && IS_WHITESPACE (ch))
502 ch = GET ();
503 if (ch == '"')
504 {
505 UNGET (ch);
506 if (scrub_m68k_mri)
507 out_string = "\n\tappfile ";
508 else
509 out_string = "\n\t.appfile ";
510 old_state = 7;
511 state = -1;
512 PUT (*out_string++);
513 }
514 else
515 {
516 while (ch != EOF && ch != '\n')
517 ch = GET ();
518 state = 0;
519 PUT (ch);
520 }
521 }
522 continue;
523
524 case 5:
525 /* We are going to copy everything up to a quote character,
526 with special handling for a backslash. We try to
527 optimize the copying in the simple case without using the
528 GET and PUT macros. */
529 {
530 char *s;
531 int len;
532
533 for (s = from; s < fromend; s++)
534 {
535 ch = *s;
536 /* This condition must be changed if the type of any
537 other character can be LEX_IS_STRINGQUOTE. */
538 if (ch == '\\'
539 || ch == '"'
540 || ch == '\''
541 || ch == '\n')
542 break;
543 }
544 len = s - from;
545 if (len > toend - to)
546 len = toend - to;
547 if (len > 0)
548 {
549 memcpy (to, from, len);
550 to += len;
551 from += len;
552 }
553 }
554
555 ch = GET ();
556 if (ch == EOF)
557 {
558 as_warn (_("end of file in string: inserted '\"'"));
559 state = old_state;
560 UNGET ('\n');
561 PUT ('"');
562 }
563 else if (lex[ch] == LEX_IS_STRINGQUOTE)
564 {
565 state = old_state;
566 PUT (ch);
567 }
568#ifndef NO_STRING_ESCAPES
569 else if (ch == '\\')
570 {
571 state = 6;
572 PUT (ch);
573 }
574#endif
575 else if (scrub_m68k_mri && ch == '\n')
576 {
577 /* Just quietly terminate the string. This permits lines like
578 bne label loop if we haven't reach end yet
579 */
580 state = old_state;
581 UNGET (ch);
582 PUT ('\'');
583 }
584 else
585 {
586 PUT (ch);
587 }
588 continue;
589
590 case 6:
591 state = 5;
592 ch = GET ();
593 switch (ch)
594 {
595 /* Handle strings broken across lines, by turning '\n' into
596 '\\' and 'n'. */
597 case '\n':
598 UNGET ('n');
599 add_newlines++;
600 PUT ('\\');
601 continue;
602
603 case '"':
604 case '\\':
605 case 'b':
606 case 'f':
607 case 'n':
608 case 'r':
609 case 't':
610 case 'v':
611 case 'x':
612 case 'X':
613 case '0':
614 case '1':
615 case '2':
616 case '3':
617 case '4':
618 case '5':
619 case '6':
620 case '7':
621 break;
622#if defined(IGNORE_NONSTANDARD_ESCAPES) | defined(ONLY_STANDARD_ESCAPES)
623 default:
624 as_warn (_("Unknown escape '\\%c' in string: Ignored"), ch);
625 break;
626#else /* ONLY_STANDARD_ESCAPES */
627 default:
628 /* Accept \x as x for any x */
629 break;
630#endif /* ONLY_STANDARD_ESCAPES */
631
632 case EOF:
633 as_warn (_("End of file in string: '\"' inserted"));
634 PUT ('"');
635 continue;
636 }
637 PUT (ch);
638 continue;
639
640 case 7:
641 ch = GET ();
642 state = 5;
643 old_state = 8;
644 if (ch == EOF)
645 goto fromeof;
646 PUT (ch);
647 continue;
648
649 case 8:
650 do
651 ch = GET ();
652 while (ch != '\n' && ch != EOF);
653 if (ch == EOF)
654 goto fromeof;
655 state = 0;
656 PUT (ch);
657 continue;
658 }
659
660 /* OK, we are somewhere in states 0 through 4 or 9 through 11 */
661
662 /* flushchar: */
663 ch = GET ();
664
665 recycle:
666
667#if defined TC_ARM && defined OBJ_ELF
668 /* We need to watch out for .symver directives. See the comment later
669 in this function. */
670 if (symver_state == NULL)
671 {
672 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
673 symver_state = symver_pseudo + 1;
674 }
675 else
676 {
677 /* We advance to the next state if we find the right
678 character. */
679 if (ch != '\0' && (*symver_state == ch))
680 ++symver_state;
681 else if (*symver_state != '\0')
682 /* We did not get the expected character, or we didn't
683 get a valid terminating character after seeing the
684 entire pseudo-op, so we must go back to the beginning. */
685 symver_state = NULL;
686 else
687 {
688 /* We've read the entire pseudo-op. If this is the end
689 of the line, go back to the beginning. */
690 if (IS_NEWLINE (ch))
691 symver_state = NULL;
692 }
693 }
694#endif /* TC_ARM && OBJ_ELF */
695
696#ifdef TC_M68K
697 /* We want to have pseudo-ops which control whether we are in
698 MRI mode or not. Unfortunately, since m68k MRI mode affects
699 the scrubber, that means that we need a special purpose
700 recognizer here. */
701 if (mri_state == NULL)
702 {
703 if ((state == 0 || state == 1)
704 && ch == mri_pseudo[0])
705 mri_state = mri_pseudo + 1;
706 }
707 else
708 {
709 /* We advance to the next state if we find the right
710 character, or if we need a space character and we get any
711 whitespace character, or if we need a '0' and we get a
712 '1' (this is so that we only need one state to handle
713 ``.mri 0'' and ``.mri 1''). */
714 if (ch != '\0'
715 && (*mri_state == ch
716 || (*mri_state == ' '
717 && lex[ch] == LEX_IS_WHITESPACE)
718 || (*mri_state == '0'
719 && ch == '1')))
720 {
721 mri_last_ch = ch;
722 ++mri_state;
723 }
724 else if (*mri_state != '\0'
725 || (lex[ch] != LEX_IS_WHITESPACE
726 && lex[ch] != LEX_IS_NEWLINE))
727 {
728 /* We did not get the expected character, or we didn't
729 get a valid terminating character after seeing the
730 entire pseudo-op, so we must go back to the
731 beginning. */
732 mri_state = NULL;
733 }
734 else
735 {
736 /* We've read the entire pseudo-op. mips_last_ch is
737 either '0' or '1' indicating whether to enter or
738 leave MRI mode. */
739 do_scrub_begin (mri_last_ch == '1');
740 mri_state = NULL;
741
742 /* We continue handling the character as usual. The
743 main gas reader must also handle the .mri pseudo-op
744 to control expression parsing and the like. */
745 }
746 }
747#endif
748
749 if (ch == EOF)
750 {
751 if (state != 0)
752 {
753 as_warn (_("end of file not at end of a line; newline inserted"));
754 state = 0;
755 PUT ('\n');
756 }
757 goto fromeof;
758 }
759
760 switch (lex[ch])
761 {
762 case LEX_IS_WHITESPACE:
763 do
764 {
765 ch = GET ();
766 }
767 while (ch != EOF && IS_WHITESPACE (ch));
768 if (ch == EOF)
769 goto fromeof;
770
771 if (state == 0)
772 {
773 /* Preserve a single whitespace character at the
774 beginning of a line. */
775 state = 1;
776 UNGET (ch);
777 PUT (' ');
778 break;
779 }
780
781#ifdef KEEP_WHITE_AROUND_COLON
782 if (lex[ch] == LEX_IS_COLON)
783 {
784 /* Only keep this white if there's no white *after* the
785 colon. */
786 ch2 = GET ();
787 UNGET (ch2);
788 if (!IS_WHITESPACE (ch2))
789 {
790 state = 9;
791 UNGET (ch);
792 PUT (' ');
793 break;
794 }
795 }
796#endif
797 if (IS_COMMENT (ch)
798 || ch == '/'
799 || IS_LINE_SEPARATOR (ch))
800 {
801 if (scrub_m68k_mri)
802 {
803 /* In MRI mode, we keep these spaces. */
804 UNGET (ch);
805 PUT (' ');
806 break;
807 }
808 goto recycle;
809 }
810
811 /* If we're in state 2 or 11, we've seen a non-white
812 character followed by whitespace. If the next character
813 is ':', this is whitespace after a label name which we
814 normally must ignore. In MRI mode, though, spaces are
815 not permitted between the label and the colon. */
816 if ((state == 2 || state == 11)
817 && lex[ch] == LEX_IS_COLON
818 && ! scrub_m68k_mri)
819 {
820 state = 1;
821 PUT (ch);
822 break;
823 }
824
825 switch (state)
826 {
827 case 0:
828 state++;
829 goto recycle; /* Punted leading sp */
830 case 1:
831 /* We can arrive here if we leave a leading whitespace
832 character at the beginning of a line. */
833 goto recycle;
834 case 2:
835 state = 3;
836 if (to + 1 < toend)
837 {
838 /* Optimize common case by skipping UNGET/GET. */
839 PUT (' '); /* Sp after opco */
840 goto recycle;
841 }
842 UNGET (ch);
843 PUT (' ');
844 break;
845 case 3:
846 if (scrub_m68k_mri)
847 {
848 /* In MRI mode, we keep these spaces. */
849 UNGET (ch);
850 PUT (' ');
851 break;
852 }
853 goto recycle; /* Sp in operands */
854 case 9:
855 case 10:
856 if (scrub_m68k_mri)
857 {
858 /* In MRI mode, we keep these spaces. */
859 state = 3;
860 UNGET (ch);
861 PUT (' ');
862 break;
863 }
864 state = 10; /* Sp after symbol char */
865 goto recycle;
866 case 11:
867 if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
868 state = 1;
869 else
870 {
871 /* We know that ch is not ':', since we tested that
872 case above. Therefore this is not a label, so it
873 must be the opcode, and we've just seen the
874 whitespace after it. */
875 state = 3;
876 }
877 UNGET (ch);
878 PUT (' '); /* Sp after label definition. */
879 break;
880 default:
881 BAD_CASE (state);
882 }
883 break;
884
885 case LEX_IS_TWOCHAR_COMMENT_1ST:
886 ch2 = GET ();
887 if (ch2 == '*')
888 {
889 for (;;)
890 {
891 do
892 {
893 ch2 = GET ();
894 if (ch2 != EOF && IS_NEWLINE (ch2))
895 add_newlines++;
896 }
897 while (ch2 != EOF && ch2 != '*');
898
899 while (ch2 == '*')
900 ch2 = GET ();
901
902 if (ch2 == EOF || ch2 == '/')
903 break;
904
905 /* This UNGET will ensure that we count newlines
906 correctly. */
907 UNGET (ch2);
908 }
909
910 if (ch2 == EOF)
911 as_warn (_("end of file in multiline comment"));
912
913 ch = ' ';
914 goto recycle;
915 }
916#ifdef DOUBLESLASH_LINE_COMMENTS
917 else if (ch2 == '/')
918 {
919 do
920 {
921 ch = GET ();
922 }
923 while (ch != EOF && !IS_NEWLINE (ch));
924 if (ch == EOF)
925 as_warn ("end of file in comment; newline inserted");
926 state = 0;
927 PUT ('\n');
928 break;
929 }
930#endif
931 else
932 {
933 if (ch2 != EOF)
934 UNGET (ch2);
935 if (state == 9 || state == 10)
936 state = 3;
937 PUT (ch);
938 }
939 break;
940
941 case LEX_IS_STRINGQUOTE:
942 if (state == 10)
943 {
944 /* Preserve the whitespace in foo "bar" */
945 UNGET (ch);
946 state = 3;
947 PUT (' ');
948
949 /* PUT didn't jump out. We could just break, but we
950 know what will happen, so optimize a bit. */
951 ch = GET ();
952 old_state = 3;
953 }
954 else if (state == 9)
955 old_state = 3;
956 else
957 old_state = state;
958 state = 5;
959 PUT (ch);
960 break;
961
962#ifndef IEEE_STYLE
963 case LEX_IS_ONECHAR_QUOTE:
964 if (state == 10)
965 {
966 /* Preserve the whitespace in foo 'b' */
967 UNGET (ch);
968 state = 3;
969 PUT (' ');
970 break;
971 }
972 ch = GET ();
973 if (ch == EOF)
974 {
975 as_warn (_("end of file after a one-character quote; \\0 inserted"));
976 ch = 0;
977 }
978 if (ch == '\\')
979 {
980 ch = GET ();
981 if (ch == EOF)
982 {
983 as_warn (_("end of file in escape character"));
984 ch = '\\';
985 }
986 else
987 ch = process_escape (ch);
988 }
989 sprintf (out_buf, "%d", (int) (unsigned char) ch);
990
991 /* None of these 'x constants for us. We want 'x'. */
992 if ((ch = GET ()) != '\'')
993 {
994#ifdef REQUIRE_CHAR_CLOSE_QUOTE
995 as_warn (_("Missing close quote: (assumed)"));
996#else
997 if (ch != EOF)
998 UNGET (ch);
999#endif
1000 }
1001 if (strlen (out_buf) == 1)
1002 {
1003 PUT (out_buf[0]);
1004 break;
1005 }
1006 if (state == 9)
1007 old_state = 3;
1008 else
1009 old_state = state;
1010 state = -1;
1011 out_string = out_buf;
1012 PUT (*out_string++);
1013 break;
1014#endif
1015
1016 case LEX_IS_COLON:
1017#ifdef KEEP_WHITE_AROUND_COLON
1018 state = 9;
1019#else
1020 if (state == 9 || state == 10)
1021 state = 3;
1022 else if (state != 3)
1023 state = 1;
1024#endif
1025 PUT (ch);
1026 break;
1027
1028 case LEX_IS_NEWLINE:
1029 /* Roll out a bunch of newlines from inside comments, etc. */
1030 if (add_newlines)
1031 {
1032 --add_newlines;
1033 UNGET (ch);
1034 }
1035 /* Fall through. */
1036
1037 case LEX_IS_LINE_SEPARATOR:
1038 state = 0;
1039 PUT (ch);
1040 break;
1041
1042#ifdef TC_V850
1043 case LEX_IS_DOUBLEDASH_1ST:
1044 ch2 = GET ();
1045 if (ch2 != '-')
1046 {
1047 UNGET (ch2);
1048 goto de_fault;
1049 }
1050 /* Read and skip to end of line. */
1051 do
1052 {
1053 ch = GET ();
1054 }
1055 while (ch != EOF && ch != '\n');
1056 if (ch == EOF)
1057 {
1058 as_warn (_("end of file in comment; newline inserted"));
1059 }
1060 state = 0;
1061 PUT ('\n');
1062 break;
1063#endif
1064#ifdef DOUBLEBAR_PARALLEL
1065 case LEX_IS_DOUBLEBAR_1ST:
1066 ch2 = GET ();
1067 if (ch2 != '|')
1068 {
1069 UNGET (ch2);
1070 goto de_fault;
1071 }
1072 /* Reset back to state 1 and pretend that we are parsing a line from
1073 just after the first white space. */
1074 state = 1;
1075 PUT ('|');
1076 PUT ('|');
1077 break;
1078#endif
1079 case LEX_IS_LINE_COMMENT_START:
1080 /* FIXME-someday: The two character comment stuff was badly
1081 thought out. On i386, we want '/' as line comment start
1082 AND we want C style comments. hence this hack. The
1083 whole lexical process should be reworked. xoxorich. */
1084 if (ch == '/')
1085 {
1086 ch2 = GET ();
1087 if (ch2 == '*')
1088 {
1089 old_state = 3;
1090 state = -2;
1091 break;
1092 }
1093 else
1094 {
1095 UNGET (ch2);
1096 }
1097 } /* bad hack */
1098
1099 if (state == 0 || state == 1) /* Only comment at start of line. */
1100 {
1101 int startch;
1102
1103 startch = ch;
1104
1105 do
1106 {
1107 ch = GET ();
1108 }
1109 while (ch != EOF && IS_WHITESPACE (ch));
1110 if (ch == EOF)
1111 {
1112 as_warn (_("end of file in comment; newline inserted"));
1113 PUT ('\n');
1114 break;
1115 }
1116 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1117 {
1118 /* Not a cpp line. */
1119 while (ch != EOF && !IS_NEWLINE (ch))
1120 ch = GET ();
1121 if (ch == EOF)
1122 as_warn (_("EOF in Comment: Newline inserted"));
1123 state = 0;
1124 PUT ('\n');
1125 break;
1126 }
1127 /* Looks like `# 123 "filename"' from cpp. */
1128 UNGET (ch);
1129 old_state = 4;
1130 state = -1;
1131 if (scrub_m68k_mri)
1132 out_string = "\tappline ";
1133 else
1134 out_string = "\t.appline ";
1135 PUT (*out_string++);
1136 break;
1137 }
1138
1139#ifdef TC_D10V
1140 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1141 Trap is the only short insn that has a first operand that is
1142 neither register nor label.
1143 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1144 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is
1145 already LEX_IS_LINE_COMMENT_START. However, it is the
1146 only character in line_comment_chars for d10v, hence we
1147 can recognize it as such. */
1148 /* An alternative approach would be to reset the state to 1 when
1149 we see '||', '<'- or '->', but that seems to be overkill. */
1150 if (state == 10)
1151 PUT (' ');
1152#endif
1153 /* We have a line comment character which is not at the
1154 start of a line. If this is also a normal comment
1155 character, fall through. Otherwise treat it as a default
1156 character. */
1157 if (strchr (tc_comment_chars, ch) == NULL
1158 && (! scrub_m68k_mri
1159 || (ch != '!' && ch != '*')))
1160 goto de_fault;
1161 if (scrub_m68k_mri
1162 && (ch == '!' || ch == '*' || ch == '#')
1163 && state != 1
1164 && state != 10)
1165 goto de_fault;
1166 /* Fall through. */
1167 case LEX_IS_COMMENT_START:
1168#if defined TC_ARM && defined OBJ_ELF
1169 /* On the ARM, `@' is the comment character.
1170 Unfortunately this is also a special character in ELF .symver
1171 directives (and .type, though we deal with those another way).
1172 So we check if this line is such a directive, and treat
1173 the character as default if so. This is a hack. */
1174 if ((symver_state != NULL) && (*symver_state == 0))
1175 goto de_fault;
1176#endif
1177#ifdef WARN_COMMENTS
1178 if (!found_comment)
1179 as_where (&found_comment_file, &found_comment);
1180#endif
1181 do
1182 {
1183 ch = GET ();
1184 }
1185 while (ch != EOF && !IS_NEWLINE (ch));
1186 if (ch == EOF)
1187 as_warn (_("end of file in comment; newline inserted"));
1188 state = 0;
1189 PUT ('\n');
1190 break;
1191
1192 case LEX_IS_SYMBOL_COMPONENT:
1193 if (state == 10)
1194 {
1195 /* This is a symbol character following another symbol
1196 character, with whitespace in between. We skipped
1197 the whitespace earlier, so output it now. */
1198 UNGET (ch);
1199 state = 3;
1200 PUT (' ');
1201 break;
1202 }
1203
1204 if (state == 3)
1205 state = 9;
1206
1207 /* This is a common case. Quickly copy CH and all the
1208 following symbol component or normal characters. */
1209 if (to + 1 < toend
1210 && mri_state == NULL
1211#if defined TC_ARM && defined OBJ_ELF
1212 && symver_state == NULL
1213#endif
1214 )
1215 {
1216 char *s;
1217 int len;
1218
1219 for (s = from; s < fromend; s++)
1220 {
1221 int type;
1222
1223 ch2 = *(unsigned char *) s;
1224 type = lex[ch2];
1225 if (type != 0
1226 && type != LEX_IS_SYMBOL_COMPONENT)
1227 break;
1228 }
1229 if (s > from)
1230 {
1231 /* Handle the last character normally, for
1232 simplicity. */
1233 --s;
1234 }
1235 len = s - from;
1236 if (len > (toend - to) - 1)
1237 len = (toend - to) - 1;
1238 if (len > 0)
1239 {
1240 PUT (ch);
1241 if (len > 8)
1242 {
1243 memcpy (to, from, len);
1244 to += len;
1245 from += len;
1246 }
1247 else
1248 {
1249 switch (len)
1250 {
1251 case 8: *to++ = *from++;
1252 case 7: *to++ = *from++;
1253 case 6: *to++ = *from++;
1254 case 5: *to++ = *from++;
1255 case 4: *to++ = *from++;
1256 case 3: *to++ = *from++;
1257 case 2: *to++ = *from++;
1258 case 1: *to++ = *from++;
1259 }
1260 }
1261 ch = GET ();
1262 }
1263 }
1264
1265 /* Fall through. */
1266 default:
1267 de_fault:
1268 /* Some relatively `normal' character. */
1269 if (state == 0)
1270 {
1271 state = 11; /* Now seeing label definition */
1272 }
1273 else if (state == 1)
1274 {
1275 state = 2; /* Ditto */
1276 }
1277 else if (state == 9)
1278 {
1279 if (lex[ch] != LEX_IS_SYMBOL_COMPONENT)
1280 state = 3;
1281 }
1282 else if (state == 10)
1283 {
1284 if (ch == '\\')
1285 {
1286 /* Special handling for backslash: a backslash may
1287 be the beginning of a formal parameter (of a
1288 macro) following another symbol character, with
1289 whitespace in between. If that is the case, we
1290 output a space before the parameter. Strictly
1291 speaking, correct handling depends upon what the
1292 macro parameter expands into; if the parameter
1293 expands into something which does not start with
1294 an operand character, then we don't want to keep
1295 the space. We don't have enough information to
1296 make the right choice, so here we are making the
1297 choice which is more likely to be correct. */
1298 PUT (' ');
1299 }
1300
1301 state = 3;
1302 }
1303 PUT (ch);
1304 break;
1305 }
1306 }
1307
1308 /*NOTREACHED*/
1309
1310 fromeof:
1311 /* We have reached the end of the input. */
1312 return to - tostart;
1313
1314 tofull:
1315 /* The output buffer is full. Save any input we have not yet
1316 processed. */
1317 if (fromend > from)
1318 {
1319 saved_input = from;
1320 saved_input_len = fromend - from;
1321 }
1322 else
1323 saved_input = NULL;
1324
1325 return to - tostart;
1326}
1327
1328/* end of app.c */
Note: See TracBrowser for help on using the repository browser.