Commit | Line | Data |
---|---|---|
252b5132 RH |
1 | /* This is the Assembler Pre-Processor |
2 | Copyright (C) 1987, 90, 91, 92, 93, 94, 95, 96, 97, 98, 1999 | |
3 | Free Software Foundation, Inc. | |
4 | ||
5 | This file is part of GAS, the GNU Assembler. | |
6 | ||
7 | GAS is free software; you can redistribute it and/or modify | |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
11 | ||
12 | GAS is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with GAS; see the file COPYING. If not, write to the Free | |
19 | Software Foundation, 59 Temple Place - Suite 330, Boston, MA | |
20 | 02111-1307, USA. */ | |
21 | ||
22 | /* Modified by Allen Wirfs-Brock, Instantiations Inc 2/90 */ | |
23 | /* App, the assembler pre-processor. This pre-processor strips out excess | |
24 | spaces, turns single-quoted characters into a decimal constant, and turns | |
25 | # <number> <filename> <garbage> into a .line <number>\n.file <filename> | |
26 | pair. This needs better error-handling. */ | |
27 | ||
28 | #include <stdio.h> | |
29 | #include "as.h" /* For BAD_CASE() only */ | |
30 | ||
31 | #if (__STDC__ != 1) | |
32 | #ifndef const | |
33 | #define const /* empty */ | |
34 | #endif | |
35 | #endif | |
36 | ||
37 | /* Whether we are scrubbing in m68k MRI mode. This is different from | |
38 | flag_m68k_mri, because the two flags will be affected by the .mri | |
39 | pseudo-op at different times. */ | |
40 | static int scrub_m68k_mri; | |
41 | ||
42 | /* The pseudo-op which switches in and out of MRI mode. See the | |
43 | comment in do_scrub_chars. */ | |
44 | static const char mri_pseudo[] = ".mri 0"; | |
45 | ||
46 | #if defined TC_ARM && defined OBJ_ELF | |
47 | /* The pseudo-op for which we need to special-case `@' characters. | |
48 | See the comment in do_scrub_chars. */ | |
49 | static const char symver_pseudo[] = ".symver"; | |
50 | static const char * symver_state; | |
51 | #endif | |
52 | ||
53 | static char lex[256]; | |
54 | static const char symbol_chars[] = | |
55 | "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789"; | |
56 | ||
57 | #define LEX_IS_SYMBOL_COMPONENT 1 | |
58 | #define LEX_IS_WHITESPACE 2 | |
59 | #define LEX_IS_LINE_SEPARATOR 3 | |
60 | #define LEX_IS_COMMENT_START 4 | |
61 | #define LEX_IS_LINE_COMMENT_START 5 | |
62 | #define LEX_IS_TWOCHAR_COMMENT_1ST 6 | |
63 | #define LEX_IS_STRINGQUOTE 8 | |
64 | #define LEX_IS_COLON 9 | |
65 | #define LEX_IS_NEWLINE 10 | |
66 | #define LEX_IS_ONECHAR_QUOTE 11 | |
67 | #ifdef TC_V850 | |
68 | #define LEX_IS_DOUBLEDASH_1ST 12 | |
69 | #endif | |
70 | #ifdef TC_M32R | |
71 | #define LEX_IS_DOUBLEBAR_1ST 13 | |
72 | #endif | |
73 | #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT) | |
74 | #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE) | |
75 | #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR) | |
76 | #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START) | |
77 | #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START) | |
78 | #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE) | |
79 | ||
80 | static int process_escape PARAMS ((int)); | |
81 | ||
82 | /* FIXME-soon: The entire lexer/parser thingy should be | |
83 | built statically at compile time rather than dynamically | |
84 | each and every time the assembler is run. xoxorich. */ | |
85 | ||
86 | void | |
87 | do_scrub_begin (m68k_mri) | |
88 | int m68k_mri; | |
89 | { | |
90 | const char *p; | |
91 | int c; | |
92 | ||
93 | scrub_m68k_mri = m68k_mri; | |
94 | ||
95 | lex[' '] = LEX_IS_WHITESPACE; | |
96 | lex['\t'] = LEX_IS_WHITESPACE; | |
97 | lex['\r'] = LEX_IS_WHITESPACE; | |
98 | lex['\n'] = LEX_IS_NEWLINE; | |
99 | lex[';'] = LEX_IS_LINE_SEPARATOR; | |
100 | lex[':'] = LEX_IS_COLON; | |
101 | ||
102 | if (! m68k_mri) | |
103 | { | |
104 | lex['"'] = LEX_IS_STRINGQUOTE; | |
105 | ||
106 | #ifndef TC_HPPA | |
107 | lex['\''] = LEX_IS_ONECHAR_QUOTE; | |
108 | #endif | |
109 | ||
110 | #ifdef SINGLE_QUOTE_STRINGS | |
111 | lex['\''] = LEX_IS_STRINGQUOTE; | |
112 | #endif | |
113 | } | |
114 | ||
115 | /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop | |
116 | in state 5 of do_scrub_chars must be changed. */ | |
117 | ||
118 | /* Note that these override the previous defaults, e.g. if ';' is a | |
119 | comment char, then it isn't a line separator. */ | |
120 | for (p = symbol_chars; *p; ++p) | |
121 | { | |
122 | lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT; | |
123 | } /* declare symbol characters */ | |
124 | ||
125 | for (c = 128; c < 256; ++c) | |
126 | lex[c] = LEX_IS_SYMBOL_COMPONENT; | |
127 | ||
128 | #ifdef tc_symbol_chars | |
129 | /* This macro permits the processor to specify all characters which | |
130 | may appears in an operand. This will prevent the scrubber from | |
131 | discarding meaningful whitespace in certain cases. The i386 | |
132 | backend uses this to support prefixes, which can confuse the | |
133 | scrubber as to whether it is parsing operands or opcodes. */ | |
134 | for (p = tc_symbol_chars; *p; ++p) | |
135 | lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT; | |
136 | #endif | |
137 | ||
138 | /* The m68k backend wants to be able to change comment_chars. */ | |
139 | #ifndef tc_comment_chars | |
140 | #define tc_comment_chars comment_chars | |
141 | #endif | |
142 | for (p = tc_comment_chars; *p; p++) | |
143 | { | |
144 | lex[(unsigned char) *p] = LEX_IS_COMMENT_START; | |
145 | } /* declare comment chars */ | |
146 | ||
147 | for (p = line_comment_chars; *p; p++) | |
148 | { | |
149 | lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START; | |
150 | } /* declare line comment chars */ | |
151 | ||
152 | for (p = line_separator_chars; *p; p++) | |
153 | { | |
154 | lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR; | |
155 | } /* declare line separators */ | |
156 | ||
157 | /* Only allow slash-star comments if slash is not in use. | |
158 | FIXME: This isn't right. We should always permit them. */ | |
159 | if (lex['/'] == 0) | |
160 | { | |
161 | lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST; | |
162 | } | |
163 | ||
164 | if (m68k_mri) | |
165 | { | |
166 | lex['\''] = LEX_IS_STRINGQUOTE; | |
167 | lex[';'] = LEX_IS_COMMENT_START; | |
168 | lex['*'] = LEX_IS_LINE_COMMENT_START; | |
169 | /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but | |
170 | then it can't be used in an expression. */ | |
171 | lex['!'] = LEX_IS_LINE_COMMENT_START; | |
172 | } | |
173 | ||
174 | #ifdef TC_V850 | |
175 | lex['-'] = LEX_IS_DOUBLEDASH_1ST; | |
176 | #endif | |
177 | #ifdef TC_M32R | |
178 | lex['|'] = LEX_IS_DOUBLEBAR_1ST; | |
179 | #endif | |
180 | #ifdef TC_D30V | |
181 | /* must do this is we want VLIW instruction with "->" or "<-" */ | |
182 | lex['-'] = LEX_IS_SYMBOL_COMPONENT; | |
183 | #endif | |
184 | } /* do_scrub_begin() */ | |
185 | ||
186 | /* Saved state of the scrubber */ | |
187 | static int state; | |
188 | static int old_state; | |
189 | static char *out_string; | |
190 | static char out_buf[20]; | |
191 | static int add_newlines; | |
192 | static char *saved_input; | |
193 | static int saved_input_len; | |
194 | static const char *mri_state; | |
195 | static char mri_last_ch; | |
196 | ||
197 | /* Data structure for saving the state of app across #include's. Note that | |
198 | app is called asynchronously to the parsing of the .include's, so our | |
199 | state at the time .include is interpreted is completely unrelated. | |
200 | That's why we have to save it all. */ | |
201 | ||
202 | struct app_save | |
203 | { | |
204 | int state; | |
205 | int old_state; | |
206 | char * out_string; | |
207 | char out_buf[sizeof (out_buf)]; | |
208 | int add_newlines; | |
209 | char * saved_input; | |
210 | int saved_input_len; | |
211 | int scrub_m68k_mri; | |
212 | const char * mri_state; | |
213 | char mri_last_ch; | |
214 | #if defined TC_ARM && defined OBJ_ELF | |
215 | const char * symver_state; | |
216 | #endif | |
217 | }; | |
218 | ||
219 | char * | |
220 | app_push () | |
221 | { | |
222 | register struct app_save *saved; | |
223 | ||
224 | saved = (struct app_save *) xmalloc (sizeof (*saved)); | |
225 | saved->state = state; | |
226 | saved->old_state = old_state; | |
227 | saved->out_string = out_string; | |
228 | memcpy (saved->out_buf, out_buf, sizeof (out_buf)); | |
229 | saved->add_newlines = add_newlines; | |
230 | saved->saved_input = saved_input; | |
231 | saved->saved_input_len = saved_input_len; | |
232 | saved->scrub_m68k_mri = scrub_m68k_mri; | |
233 | saved->mri_state = mri_state; | |
234 | saved->mri_last_ch = mri_last_ch; | |
235 | #if defined TC_ARM && defined OBJ_ELF | |
236 | saved->symver_state = symver_state; | |
237 | #endif | |
238 | ||
239 | /* do_scrub_begin() is not useful, just wastes time. */ | |
240 | ||
241 | state = 0; | |
242 | saved_input = NULL; | |
243 | ||
244 | return (char *) saved; | |
245 | } | |
246 | ||
247 | void | |
248 | app_pop (arg) | |
249 | char *arg; | |
250 | { | |
251 | register struct app_save *saved = (struct app_save *) arg; | |
252 | ||
253 | /* There is no do_scrub_end (). */ | |
254 | state = saved->state; | |
255 | old_state = saved->old_state; | |
256 | out_string = saved->out_string; | |
257 | memcpy (out_buf, saved->out_buf, sizeof (out_buf)); | |
258 | add_newlines = saved->add_newlines; | |
259 | saved_input = saved->saved_input; | |
260 | saved_input_len = saved->saved_input_len; | |
261 | scrub_m68k_mri = saved->scrub_m68k_mri; | |
262 | mri_state = saved->mri_state; | |
263 | mri_last_ch = saved->mri_last_ch; | |
264 | #if defined TC_ARM && defined OBJ_ELF | |
265 | symver_state = saved->symver_state; | |
266 | #endif | |
267 | ||
268 | free (arg); | |
269 | } /* app_pop() */ | |
270 | ||
271 | /* @@ This assumes that \n &c are the same on host and target. This is not | |
272 | necessarily true. */ | |
273 | static int | |
274 | process_escape (ch) | |
275 | int ch; | |
276 | { | |
277 | switch (ch) | |
278 | { | |
279 | case 'b': | |
280 | return '\b'; | |
281 | case 'f': | |
282 | return '\f'; | |
283 | case 'n': | |
284 | return '\n'; | |
285 | case 'r': | |
286 | return '\r'; | |
287 | case 't': | |
288 | return '\t'; | |
289 | case '\'': | |
290 | return '\''; | |
291 | case '"': | |
292 | return '\"'; | |
293 | default: | |
294 | return ch; | |
295 | } | |
296 | } | |
297 | ||
298 | /* This function is called to process input characters. The GET | |
299 | parameter is used to retrieve more input characters. GET should | |
300 | set its parameter to point to a buffer, and return the length of | |
301 | the buffer; it should return 0 at end of file. The scrubbed output | |
302 | characters are put into the buffer starting at TOSTART; the TOSTART | |
303 | buffer is TOLEN bytes in length. The function returns the number | |
304 | of scrubbed characters put into TOSTART. This will be TOLEN unless | |
305 | end of file was seen. This function is arranged as a state | |
306 | machine, and saves its state so that it may return at any point. | |
307 | This is the way the old code used to work. */ | |
308 | ||
309 | int | |
310 | do_scrub_chars (get, tostart, tolen) | |
311 | int (*get) PARAMS ((char **)); | |
312 | char *tostart; | |
313 | int tolen; | |
314 | { | |
315 | char *to = tostart; | |
316 | char *toend = tostart + tolen; | |
317 | char *from; | |
318 | char *fromend; | |
319 | int fromlen; | |
320 | register int ch, ch2 = 0; | |
321 | ||
322 | /*State 0: beginning of normal line | |
323 | 1: After first whitespace on line (flush more white) | |
324 | 2: After first non-white (opcode) on line (keep 1white) | |
325 | 3: after second white on line (into operands) (flush white) | |
326 | 4: after putting out a .line, put out digits | |
327 | 5: parsing a string, then go to old-state | |
328 | 6: putting out \ escape in a "d string. | |
329 | 7: After putting out a .appfile, put out string. | |
330 | 8: After putting out a .appfile string, flush until newline. | |
331 | 9: After seeing symbol char in state 3 (keep 1white after symchar) | |
332 | 10: After seeing whitespace in state 9 (keep white before symchar) | |
333 | 11: After seeing a symbol character in state 0 (eg a label definition) | |
334 | -1: output string in out_string and go to the state in old_state | |
335 | -2: flush text until a '*' '/' is seen, then go to state old_state | |
336 | #ifdef TC_V850 | |
337 | 12: After seeing a dash, looking for a second dash as a start of comment. | |
338 | #endif | |
339 | #ifdef TC_M32R | |
340 | 13: After seeing a vertical bar, looking for a second vertical bar as a parallel expression seperator. | |
341 | #endif | |
342 | */ | |
343 | ||
344 | /* I added states 9 and 10 because the MIPS ECOFF assembler uses | |
345 | constructs like ``.loc 1 20''. This was turning into ``.loc | |
346 | 120''. States 9 and 10 ensure that a space is never dropped in | |
347 | between characters which could appear in a identifier. Ian | |
348 | Taylor, ian@cygnus.com. | |
349 | ||
350 | I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works | |
351 | correctly on the PA (and any other target where colons are optional). | |
352 | Jeff Law, law@cs.utah.edu. | |
353 | ||
354 | I added state 13 so that something like "cmp r1, r2 || trap #1" does not | |
355 | get squashed into "cmp r1,r2||trap#1", with the all important space | |
356 | between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */ | |
357 | ||
358 | /* This macro gets the next input character. */ | |
359 | ||
360 | #define GET() \ | |
361 | (from < fromend \ | |
362 | ? * (unsigned char *) (from++) \ | |
363 | : ((saved_input != NULL \ | |
364 | ? (free (saved_input), \ | |
365 | saved_input = NULL, \ | |
366 | 0) \ | |
367 | : 0), \ | |
368 | fromlen = (*get) (&from), \ | |
369 | fromend = from + fromlen, \ | |
370 | (fromlen == 0 \ | |
371 | ? EOF \ | |
372 | : * (unsigned char *) (from++)))) | |
373 | ||
374 | /* This macro pushes a character back on the input stream. */ | |
375 | ||
376 | #define UNGET(uch) (*--from = (uch)) | |
377 | ||
378 | /* This macro puts a character into the output buffer. If this | |
379 | character fills the output buffer, this macro jumps to the label | |
380 | TOFULL. We use this rather ugly approach because we need to | |
381 | handle two different termination conditions: EOF on the input | |
382 | stream, and a full output buffer. It would be simpler if we | |
383 | always read in the entire input stream before processing it, but | |
384 | I don't want to make such a significant change to the assembler's | |
385 | memory usage. */ | |
386 | ||
387 | #define PUT(pch) \ | |
388 | do \ | |
389 | { \ | |
390 | *to++ = (pch); \ | |
391 | if (to >= toend) \ | |
392 | goto tofull; \ | |
393 | } \ | |
394 | while (0) | |
395 | ||
396 | if (saved_input != NULL) | |
397 | { | |
398 | from = saved_input; | |
399 | fromend = from + saved_input_len; | |
400 | } | |
401 | else | |
402 | { | |
403 | fromlen = (*get) (&from); | |
404 | if (fromlen == 0) | |
405 | return 0; | |
406 | fromend = from + fromlen; | |
407 | } | |
408 | ||
409 | while (1) | |
410 | { | |
411 | /* The cases in this switch end with continue, in order to | |
412 | branch back to the top of this while loop and generate the | |
413 | next output character in the appropriate state. */ | |
414 | switch (state) | |
415 | { | |
416 | case -1: | |
417 | ch = *out_string++; | |
418 | if (*out_string == '\0') | |
419 | { | |
420 | state = old_state; | |
421 | old_state = 3; | |
422 | } | |
423 | PUT (ch); | |
424 | continue; | |
425 | ||
426 | case -2: | |
427 | for (;;) | |
428 | { | |
429 | do | |
430 | { | |
431 | ch = GET (); | |
432 | ||
433 | if (ch == EOF) | |
434 | { | |
435 | as_warn (_("end of file in comment")); | |
436 | goto fromeof; | |
437 | } | |
438 | ||
439 | if (ch == '\n') | |
440 | PUT ('\n'); | |
441 | } | |
442 | while (ch != '*'); | |
443 | ||
444 | while ((ch = GET ()) == '*') | |
445 | ; | |
446 | ||
447 | if (ch == EOF) | |
448 | { | |
449 | as_warn (_("end of file in comment")); | |
450 | goto fromeof; | |
451 | } | |
452 | ||
453 | if (ch == '/') | |
454 | break; | |
455 | ||
456 | UNGET (ch); | |
457 | } | |
458 | ||
459 | state = old_state; | |
460 | UNGET (' '); | |
461 | continue; | |
462 | ||
463 | case 4: | |
464 | ch = GET (); | |
465 | if (ch == EOF) | |
466 | goto fromeof; | |
467 | else if (ch >= '0' && ch <= '9') | |
468 | PUT (ch); | |
469 | else | |
470 | { | |
471 | while (ch != EOF && IS_WHITESPACE (ch)) | |
472 | ch = GET (); | |
473 | if (ch == '"') | |
474 | { | |
475 | UNGET (ch); | |
476 | if (scrub_m68k_mri) | |
477 | out_string = "\n\tappfile "; | |
478 | else | |
479 | out_string = "\n\t.appfile "; | |
480 | old_state = 7; | |
481 | state = -1; | |
482 | PUT (*out_string++); | |
483 | } | |
484 | else | |
485 | { | |
486 | while (ch != EOF && ch != '\n') | |
487 | ch = GET (); | |
488 | state = 0; | |
489 | PUT (ch); | |
490 | } | |
491 | } | |
492 | continue; | |
493 | ||
494 | case 5: | |
495 | /* We are going to copy everything up to a quote character, | |
496 | with special handling for a backslash. We try to | |
497 | optimize the copying in the simple case without using the | |
498 | GET and PUT macros. */ | |
499 | { | |
500 | char *s; | |
501 | int len; | |
502 | ||
503 | for (s = from; s < fromend; s++) | |
504 | { | |
505 | ch = *s; | |
506 | /* This condition must be changed if the type of any | |
507 | other character can be LEX_IS_STRINGQUOTE. */ | |
508 | if (ch == '\\' | |
509 | || ch == '"' | |
510 | || ch == '\'' | |
511 | || ch == '\n') | |
512 | break; | |
513 | } | |
514 | len = s - from; | |
515 | if (len > toend - to) | |
516 | len = toend - to; | |
517 | if (len > 0) | |
518 | { | |
519 | memcpy (to, from, len); | |
520 | to += len; | |
521 | from += len; | |
522 | } | |
523 | } | |
524 | ||
525 | ch = GET (); | |
526 | if (ch == EOF) | |
527 | { | |
528 | as_warn (_("end of file in string: inserted '\"'")); | |
529 | state = old_state; | |
530 | UNGET ('\n'); | |
531 | PUT ('"'); | |
532 | } | |
533 | else if (lex[ch] == LEX_IS_STRINGQUOTE) | |
534 | { | |
535 | state = old_state; | |
536 | PUT (ch); | |
537 | } | |
538 | #ifndef NO_STRING_ESCAPES | |
539 | else if (ch == '\\') | |
540 | { | |
541 | state = 6; | |
542 | PUT (ch); | |
543 | } | |
544 | #endif | |
545 | else if (scrub_m68k_mri && ch == '\n') | |
546 | { | |
547 | /* Just quietly terminate the string. This permits lines like | |
548 | bne label loop if we haven't reach end yet | |
549 | */ | |
550 | state = old_state; | |
551 | UNGET (ch); | |
552 | PUT ('\''); | |
553 | } | |
554 | else | |
555 | { | |
556 | PUT (ch); | |
557 | } | |
558 | continue; | |
559 | ||
560 | case 6: | |
561 | state = 5; | |
562 | ch = GET (); | |
563 | switch (ch) | |
564 | { | |
565 | /* Handle strings broken across lines, by turning '\n' into | |
566 | '\\' and 'n'. */ | |
567 | case '\n': | |
568 | UNGET ('n'); | |
569 | add_newlines++; | |
570 | PUT ('\\'); | |
571 | continue; | |
572 | ||
573 | case '"': | |
574 | case '\\': | |
575 | case 'b': | |
576 | case 'f': | |
577 | case 'n': | |
578 | case 'r': | |
579 | case 't': | |
580 | case 'v': | |
581 | case 'x': | |
582 | case 'X': | |
583 | case '0': | |
584 | case '1': | |
585 | case '2': | |
586 | case '3': | |
587 | case '4': | |
588 | case '5': | |
589 | case '6': | |
590 | case '7': | |
591 | break; | |
592 | #if defined(IGNORE_NONSTANDARD_ESCAPES) | defined(ONLY_STANDARD_ESCAPES) | |
593 | default: | |
594 | as_warn (_("Unknown escape '\\%c' in string: Ignored"), ch); | |
595 | break; | |
596 | #else /* ONLY_STANDARD_ESCAPES */ | |
597 | default: | |
598 | /* Accept \x as x for any x */ | |
599 | break; | |
600 | #endif /* ONLY_STANDARD_ESCAPES */ | |
601 | ||
602 | case EOF: | |
603 | as_warn (_("End of file in string: '\"' inserted")); | |
604 | PUT ('"'); | |
605 | continue; | |
606 | } | |
607 | PUT (ch); | |
608 | continue; | |
609 | ||
610 | case 7: | |
611 | ch = GET (); | |
612 | state = 5; | |
613 | old_state = 8; | |
614 | if (ch == EOF) | |
615 | goto fromeof; | |
616 | PUT (ch); | |
617 | continue; | |
618 | ||
619 | case 8: | |
620 | do | |
621 | ch = GET (); | |
622 | while (ch != '\n' && ch != EOF); | |
623 | if (ch == EOF) | |
624 | goto fromeof; | |
625 | state = 0; | |
626 | PUT (ch); | |
627 | continue; | |
628 | } | |
629 | ||
630 | /* OK, we are somewhere in states 0 through 4 or 9 through 11 */ | |
631 | ||
632 | /* flushchar: */ | |
633 | ch = GET (); | |
634 | ||
635 | recycle: | |
636 | ||
637 | #if defined TC_ARM && defined OBJ_ELF | |
638 | /* We need to watch out for .symver directives. See the comment later | |
639 | in this function. */ | |
640 | if (symver_state == NULL) | |
641 | { | |
642 | if ((state == 0 || state == 1) && ch == symver_pseudo[0]) | |
643 | symver_state = symver_pseudo + 1; | |
644 | } | |
645 | else | |
646 | { | |
647 | /* We advance to the next state if we find the right | |
648 | character. */ | |
649 | if (ch != '\0' && (*symver_state == ch)) | |
650 | ++symver_state; | |
651 | else if (*symver_state != '\0') | |
652 | /* We did not get the expected character, or we didn't | |
653 | get a valid terminating character after seeing the | |
654 | entire pseudo-op, so we must go back to the beginning. */ | |
655 | symver_state = NULL; | |
656 | else | |
657 | { | |
658 | /* We've read the entire pseudo-op. If this is the end | |
659 | of the line, go back to the beginning. */ | |
660 | if (IS_NEWLINE (ch)) | |
661 | symver_state = NULL; | |
662 | } | |
663 | } | |
664 | #endif /* TC_ARM && OBJ_ELF */ | |
665 | ||
666 | #ifdef TC_M68K | |
667 | /* We want to have pseudo-ops which control whether we are in | |
668 | MRI mode or not. Unfortunately, since m68k MRI mode affects | |
669 | the scrubber, that means that we need a special purpose | |
670 | recognizer here. */ | |
671 | if (mri_state == NULL) | |
672 | { | |
673 | if ((state == 0 || state == 1) | |
674 | && ch == mri_pseudo[0]) | |
675 | mri_state = mri_pseudo + 1; | |
676 | } | |
677 | else | |
678 | { | |
679 | /* We advance to the next state if we find the right | |
680 | character, or if we need a space character and we get any | |
681 | whitespace character, or if we need a '0' and we get a | |
682 | '1' (this is so that we only need one state to handle | |
683 | ``.mri 0'' and ``.mri 1''). */ | |
684 | if (ch != '\0' | |
685 | && (*mri_state == ch | |
686 | || (*mri_state == ' ' | |
687 | && lex[ch] == LEX_IS_WHITESPACE) | |
688 | || (*mri_state == '0' | |
689 | && ch == '1'))) | |
690 | { | |
691 | mri_last_ch = ch; | |
692 | ++mri_state; | |
693 | } | |
694 | else if (*mri_state != '\0' | |
695 | || (lex[ch] != LEX_IS_WHITESPACE | |
696 | && lex[ch] != LEX_IS_NEWLINE)) | |
697 | { | |
698 | /* We did not get the expected character, or we didn't | |
699 | get a valid terminating character after seeing the | |
700 | entire pseudo-op, so we must go back to the | |
701 | beginning. */ | |
702 | mri_state = NULL; | |
703 | } | |
704 | else | |
705 | { | |
706 | /* We've read the entire pseudo-op. mips_last_ch is | |
707 | either '0' or '1' indicating whether to enter or | |
708 | leave MRI mode. */ | |
709 | do_scrub_begin (mri_last_ch == '1'); | |
710 | mri_state = NULL; | |
711 | ||
712 | /* We continue handling the character as usual. The | |
713 | main gas reader must also handle the .mri pseudo-op | |
714 | to control expression parsing and the like. */ | |
715 | } | |
716 | } | |
717 | #endif | |
718 | ||
719 | if (ch == EOF) | |
720 | { | |
721 | if (state != 0) | |
722 | { | |
723 | as_warn (_("end of file not at end of a line; newline inserted")); | |
724 | state = 0; | |
725 | PUT ('\n'); | |
726 | } | |
727 | goto fromeof; | |
728 | } | |
729 | ||
730 | switch (lex[ch]) | |
731 | { | |
732 | case LEX_IS_WHITESPACE: | |
733 | do | |
734 | { | |
735 | ch = GET (); | |
736 | } | |
737 | while (ch != EOF && IS_WHITESPACE (ch)); | |
738 | if (ch == EOF) | |
739 | goto fromeof; | |
740 | ||
741 | if (state == 0) | |
742 | { | |
743 | /* Preserve a single whitespace character at the | |
744 | beginning of a line. */ | |
745 | state = 1; | |
746 | UNGET (ch); | |
747 | PUT (' '); | |
748 | break; | |
749 | } | |
750 | ||
751 | if (IS_COMMENT (ch) | |
752 | || ch == '/' | |
753 | || IS_LINE_SEPARATOR (ch)) | |
754 | { | |
755 | if (scrub_m68k_mri) | |
756 | { | |
757 | /* In MRI mode, we keep these spaces. */ | |
758 | UNGET (ch); | |
759 | PUT (' '); | |
760 | break; | |
761 | } | |
762 | goto recycle; | |
763 | } | |
764 | ||
765 | /* If we're in state 2 or 11, we've seen a non-white | |
766 | character followed by whitespace. If the next character | |
767 | is ':', this is whitespace after a label name which we | |
768 | normally must ignore. In MRI mode, though, spaces are | |
769 | not permitted between the label and the colon. */ | |
770 | if ((state == 2 || state == 11) | |
771 | && lex[ch] == LEX_IS_COLON | |
772 | && ! scrub_m68k_mri) | |
773 | { | |
774 | state = 1; | |
775 | PUT (ch); | |
776 | break; | |
777 | } | |
778 | ||
779 | switch (state) | |
780 | { | |
781 | case 0: | |
782 | state++; | |
783 | goto recycle; /* Punted leading sp */ | |
784 | case 1: | |
785 | /* We can arrive here if we leave a leading whitespace | |
786 | character at the beginning of a line. */ | |
787 | goto recycle; | |
788 | case 2: | |
789 | state = 3; | |
790 | if (to + 1 < toend) | |
791 | { | |
792 | /* Optimize common case by skipping UNGET/GET. */ | |
793 | PUT (' '); /* Sp after opco */ | |
794 | goto recycle; | |
795 | } | |
796 | UNGET (ch); | |
797 | PUT (' '); | |
798 | break; | |
799 | case 3: | |
800 | if (scrub_m68k_mri) | |
801 | { | |
802 | /* In MRI mode, we keep these spaces. */ | |
803 | UNGET (ch); | |
804 | PUT (' '); | |
805 | break; | |
806 | } | |
807 | goto recycle; /* Sp in operands */ | |
808 | case 9: | |
809 | case 10: | |
810 | if (scrub_m68k_mri) | |
811 | { | |
812 | /* In MRI mode, we keep these spaces. */ | |
813 | state = 3; | |
814 | UNGET (ch); | |
815 | PUT (' '); | |
816 | break; | |
817 | } | |
818 | state = 10; /* Sp after symbol char */ | |
819 | goto recycle; | |
820 | case 11: | |
821 | if (flag_m68k_mri | |
822 | #ifdef LABELS_WITHOUT_COLONS | |
823 | || 1 | |
824 | #endif | |
825 | ) | |
826 | state = 1; | |
827 | else | |
828 | { | |
829 | /* We know that ch is not ':', since we tested that | |
830 | case above. Therefore this is not a label, so it | |
831 | must be the opcode, and we've just seen the | |
832 | whitespace after it. */ | |
833 | state = 3; | |
834 | } | |
835 | UNGET (ch); | |
836 | PUT (' '); /* Sp after label definition. */ | |
837 | break; | |
838 | default: | |
839 | BAD_CASE (state); | |
840 | } | |
841 | break; | |
842 | ||
843 | case LEX_IS_TWOCHAR_COMMENT_1ST: | |
844 | ch2 = GET (); | |
845 | if (ch2 == '*') | |
846 | { | |
847 | for (;;) | |
848 | { | |
849 | do | |
850 | { | |
851 | ch2 = GET (); | |
852 | if (ch2 != EOF && IS_NEWLINE (ch2)) | |
853 | add_newlines++; | |
854 | } | |
855 | while (ch2 != EOF && ch2 != '*'); | |
856 | ||
857 | while (ch2 == '*') | |
858 | ch2 = GET (); | |
859 | ||
860 | if (ch2 == EOF || ch2 == '/') | |
861 | break; | |
862 | ||
863 | /* This UNGET will ensure that we count newlines | |
864 | correctly. */ | |
865 | UNGET (ch2); | |
866 | } | |
867 | ||
868 | if (ch2 == EOF) | |
869 | as_warn (_("end of file in multiline comment")); | |
870 | ||
871 | ch = ' '; | |
872 | goto recycle; | |
873 | } | |
874 | else | |
875 | { | |
876 | if (ch2 != EOF) | |
877 | UNGET (ch2); | |
878 | if (state == 9 || state == 10) | |
879 | state = 3; | |
880 | PUT (ch); | |
881 | } | |
882 | break; | |
883 | ||
884 | case LEX_IS_STRINGQUOTE: | |
885 | if (state == 10) | |
886 | { | |
887 | /* Preserve the whitespace in foo "bar" */ | |
888 | UNGET (ch); | |
889 | state = 3; | |
890 | PUT (' '); | |
891 | ||
892 | /* PUT didn't jump out. We could just break, but we | |
893 | know what will happen, so optimize a bit. */ | |
894 | ch = GET (); | |
895 | old_state = 3; | |
896 | } | |
897 | else if (state == 9) | |
898 | old_state = 3; | |
899 | else | |
900 | old_state = state; | |
901 | state = 5; | |
902 | PUT (ch); | |
903 | break; | |
904 | ||
905 | #ifndef IEEE_STYLE | |
906 | case LEX_IS_ONECHAR_QUOTE: | |
907 | if (state == 10) | |
908 | { | |
909 | /* Preserve the whitespace in foo 'b' */ | |
910 | UNGET (ch); | |
911 | state = 3; | |
912 | PUT (' '); | |
913 | break; | |
914 | } | |
915 | ch = GET (); | |
916 | if (ch == EOF) | |
917 | { | |
918 | as_warn (_("end of file after a one-character quote; \\0 inserted")); | |
919 | ch = 0; | |
920 | } | |
921 | if (ch == '\\') | |
922 | { | |
923 | ch = GET (); | |
924 | if (ch == EOF) | |
925 | { | |
926 | as_warn (_("end of file in escape character")); | |
927 | ch = '\\'; | |
928 | } | |
929 | else | |
930 | ch = process_escape (ch); | |
931 | } | |
932 | sprintf (out_buf, "%d", (int) (unsigned char) ch); | |
933 | ||
934 | /* None of these 'x constants for us. We want 'x'. */ | |
935 | if ((ch = GET ()) != '\'') | |
936 | { | |
937 | #ifdef REQUIRE_CHAR_CLOSE_QUOTE | |
938 | as_warn (_("Missing close quote: (assumed)")); | |
939 | #else | |
940 | if (ch != EOF) | |
941 | UNGET (ch); | |
942 | #endif | |
943 | } | |
944 | if (strlen (out_buf) == 1) | |
945 | { | |
946 | PUT (out_buf[0]); | |
947 | break; | |
948 | } | |
949 | if (state == 9) | |
950 | old_state = 3; | |
951 | else | |
952 | old_state = state; | |
953 | state = -1; | |
954 | out_string = out_buf; | |
955 | PUT (*out_string++); | |
956 | break; | |
957 | #endif | |
958 | ||
959 | case LEX_IS_COLON: | |
960 | if (state == 9 || state == 10) | |
961 | state = 3; | |
962 | else if (state != 3) | |
963 | state = 1; | |
964 | PUT (ch); | |
965 | break; | |
966 | ||
967 | case LEX_IS_NEWLINE: | |
968 | /* Roll out a bunch of newlines from inside comments, etc. */ | |
969 | if (add_newlines) | |
970 | { | |
971 | --add_newlines; | |
972 | UNGET (ch); | |
973 | } | |
974 | /* fall thru into... */ | |
975 | ||
976 | case LEX_IS_LINE_SEPARATOR: | |
977 | state = 0; | |
978 | PUT (ch); | |
979 | break; | |
980 | ||
981 | #ifdef TC_V850 | |
982 | case LEX_IS_DOUBLEDASH_1ST: | |
983 | ch2 = GET(); | |
984 | if (ch2 != '-') | |
985 | { | |
986 | UNGET (ch2); | |
987 | goto de_fault; | |
988 | } | |
989 | /* read and skip to end of line */ | |
990 | do | |
991 | { | |
992 | ch = GET (); | |
993 | } | |
994 | while (ch != EOF && ch != '\n'); | |
995 | if (ch == EOF) | |
996 | { | |
997 | as_warn (_("end of file in comment; newline inserted")); | |
998 | } | |
999 | state = 0; | |
1000 | PUT ('\n'); | |
1001 | break; | |
1002 | #endif | |
1003 | #ifdef TC_M32R | |
1004 | case LEX_IS_DOUBLEBAR_1ST: | |
1005 | ch2 = GET(); | |
1006 | if (ch2 != '|') | |
1007 | { | |
1008 | UNGET (ch2); | |
1009 | goto de_fault; | |
1010 | } | |
1011 | /* Reset back to state 1 and pretend that we are parsing a line from | |
1012 | just after the first white space. */ | |
1013 | state = 1; | |
1014 | PUT ('|'); | |
1015 | PUT ('|'); | |
1016 | break; | |
1017 | #endif | |
1018 | case LEX_IS_LINE_COMMENT_START: | |
1019 | /* FIXME-someday: The two character comment stuff was badly | |
1020 | thought out. On i386, we want '/' as line comment start | |
1021 | AND we want C style comments. hence this hack. The | |
1022 | whole lexical process should be reworked. xoxorich. */ | |
1023 | if (ch == '/') | |
1024 | { | |
1025 | ch2 = GET (); | |
1026 | if (ch2 == '*') | |
1027 | { | |
1028 | old_state = 3; | |
1029 | state = -2; | |
1030 | break; | |
1031 | } | |
1032 | else | |
1033 | { | |
1034 | UNGET (ch2); | |
1035 | } | |
1036 | } /* bad hack */ | |
1037 | ||
1038 | if (state == 0 || state == 1) /* Only comment at start of line. */ | |
1039 | { | |
1040 | int startch; | |
1041 | ||
1042 | startch = ch; | |
1043 | ||
1044 | do | |
1045 | { | |
1046 | ch = GET (); | |
1047 | } | |
1048 | while (ch != EOF && IS_WHITESPACE (ch)); | |
1049 | if (ch == EOF) | |
1050 | { | |
1051 | as_warn (_("end of file in comment; newline inserted")); | |
1052 | PUT ('\n'); | |
1053 | break; | |
1054 | } | |
1055 | if (ch < '0' || ch > '9' || state != 0 || startch != '#') | |
1056 | { | |
1057 | /* Not a cpp line. */ | |
1058 | while (ch != EOF && !IS_NEWLINE (ch)) | |
1059 | ch = GET (); | |
1060 | if (ch == EOF) | |
1061 | as_warn (_("EOF in Comment: Newline inserted")); | |
1062 | state = 0; | |
1063 | PUT ('\n'); | |
1064 | break; | |
1065 | } | |
1066 | /* Loks like `# 123 "filename"' from cpp. */ | |
1067 | UNGET (ch); | |
1068 | old_state = 4; | |
1069 | state = -1; | |
1070 | if (scrub_m68k_mri) | |
1071 | out_string = "\tappline "; | |
1072 | else | |
1073 | out_string = "\t.appline "; | |
1074 | PUT (*out_string++); | |
1075 | break; | |
1076 | } | |
1077 | ||
1078 | #ifdef TC_D10V | |
1079 | /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true. | |
1080 | Trap is the only short insn that has a first operand that is | |
1081 | neither register nor label. | |
1082 | We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 . | |
1083 | We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is already | |
1084 | LEX_IS_LINE_COMMENT_START. However, it is the only character in | |
1085 | line_comment_chars for d10v, hence we can recognize it as such. */ | |
1086 | /* An alternative approach would be to reset the state to 1 when | |
1087 | we see '||', '<'- or '->', but that seems to be overkill. */ | |
1088 | if (state == 10) PUT (' '); | |
1089 | #endif | |
1090 | /* We have a line comment character which is not at the | |
1091 | start of a line. If this is also a normal comment | |
1092 | character, fall through. Otherwise treat it as a default | |
1093 | character. */ | |
1094 | if (strchr (tc_comment_chars, ch) == NULL | |
1095 | && (! scrub_m68k_mri | |
1096 | || (ch != '!' && ch != '*'))) | |
1097 | goto de_fault; | |
1098 | if (scrub_m68k_mri | |
1099 | && (ch == '!' || ch == '*' || ch == '#') | |
1100 | && state != 1 | |
1101 | && state != 10) | |
1102 | goto de_fault; | |
1103 | /* Fall through. */ | |
1104 | case LEX_IS_COMMENT_START: | |
1105 | #if defined TC_ARM && defined OBJ_ELF | |
1106 | /* On the ARM, `@' is the comment character. | |
1107 | Unfortunately this is also a special character in ELF .symver | |
1108 | directives (and .type, though we deal with those another way). So | |
1109 | we check if this line is such a directive, and treat the character | |
1110 | as default if so. This is a hack. */ | |
1111 | if ((symver_state != NULL) && (*symver_state == 0)) | |
1112 | goto de_fault; | |
1113 | #endif | |
1114 | do | |
1115 | { | |
1116 | ch = GET (); | |
1117 | } | |
1118 | while (ch != EOF && !IS_NEWLINE (ch)); | |
1119 | if (ch == EOF) | |
1120 | as_warn (_("end of file in comment; newline inserted")); | |
1121 | state = 0; | |
1122 | PUT ('\n'); | |
1123 | break; | |
1124 | ||
1125 | case LEX_IS_SYMBOL_COMPONENT: | |
1126 | if (state == 10) | |
1127 | { | |
1128 | /* This is a symbol character following another symbol | |
1129 | character, with whitespace in between. We skipped | |
1130 | the whitespace earlier, so output it now. */ | |
1131 | UNGET (ch); | |
1132 | state = 3; | |
1133 | PUT (' '); | |
1134 | break; | |
1135 | } | |
1136 | ||
1137 | if (state == 3) | |
1138 | state = 9; | |
1139 | ||
1140 | /* This is a common case. Quickly copy CH and all the | |
1141 | following symbol component or normal characters. */ | |
1142 | if (to + 1 < toend | |
1143 | && mri_state == NULL | |
1144 | #if defined TC_ARM && defined OBJ_ELF | |
1145 | && symver_state == NULL | |
1146 | #endif | |
1147 | ) | |
1148 | { | |
1149 | char *s; | |
1150 | int len; | |
1151 | ||
1152 | for (s = from; s < fromend; s++) | |
1153 | { | |
1154 | int type; | |
1155 | ||
1156 | ch2 = * (unsigned char *) s; | |
1157 | type = lex[ch2]; | |
1158 | if (type != 0 | |
1159 | && type != LEX_IS_SYMBOL_COMPONENT) | |
1160 | break; | |
1161 | } | |
1162 | if (s > from) | |
1163 | { | |
1164 | /* Handle the last character normally, for | |
1165 | simplicity. */ | |
1166 | --s; | |
1167 | } | |
1168 | len = s - from; | |
1169 | if (len > (toend - to) - 1) | |
1170 | len = (toend - to) - 1; | |
1171 | if (len > 0) | |
1172 | { | |
1173 | PUT (ch); | |
1174 | if (len > 8) | |
1175 | { | |
1176 | memcpy (to, from, len); | |
1177 | to += len; | |
1178 | from += len; | |
1179 | } | |
1180 | else | |
1181 | { | |
1182 | switch (len) | |
1183 | { | |
1184 | case 8: *to++ = *from++; | |
1185 | case 7: *to++ = *from++; | |
1186 | case 6: *to++ = *from++; | |
1187 | case 5: *to++ = *from++; | |
1188 | case 4: *to++ = *from++; | |
1189 | case 3: *to++ = *from++; | |
1190 | case 2: *to++ = *from++; | |
1191 | case 1: *to++ = *from++; | |
1192 | } | |
1193 | } | |
1194 | ch = GET (); | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | /* Fall through. */ | |
1199 | default: | |
1200 | de_fault: | |
1201 | /* Some relatively `normal' character. */ | |
1202 | if (state == 0) | |
1203 | { | |
1204 | state = 11; /* Now seeing label definition */ | |
1205 | } | |
1206 | else if (state == 1) | |
1207 | { | |
1208 | state = 2; /* Ditto */ | |
1209 | } | |
1210 | else if (state == 9) | |
1211 | { | |
1212 | if (lex[ch] != LEX_IS_SYMBOL_COMPONENT) | |
1213 | state = 3; | |
1214 | } | |
1215 | else if (state == 10) | |
1216 | { | |
1217 | state = 3; | |
1218 | } | |
1219 | PUT (ch); | |
1220 | break; | |
1221 | } | |
1222 | } | |
1223 | ||
1224 | /*NOTREACHED*/ | |
1225 | ||
1226 | fromeof: | |
1227 | /* We have reached the end of the input. */ | |
1228 | return to - tostart; | |
1229 | ||
1230 | tofull: | |
1231 | /* The output buffer is full. Save any input we have not yet | |
1232 | processed. */ | |
1233 | if (fromend > from) | |
1234 | { | |
1235 | char *save; | |
1236 | ||
1237 | save = (char *) xmalloc (fromend - from); | |
1238 | memcpy (save, from, fromend - from); | |
1239 | if (saved_input != NULL) | |
1240 | free (saved_input); | |
1241 | saved_input = save; | |
1242 | saved_input_len = fromend - from; | |
1243 | } | |
1244 | else | |
1245 | { | |
1246 | if (saved_input != NULL) | |
1247 | { | |
1248 | free (saved_input); | |
1249 | saved_input = NULL; | |
1250 | } | |
1251 | } | |
1252 | return to - tostart; | |
1253 | } | |
1254 | ||
1255 | /* end of app.c */ |