1 /* This is the Assembler Pre-Processor
2 Copyright (C) 1987-2014 Free Software Foundation, Inc.
4 This file is part of GAS, the GNU Assembler.
6 GAS is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3, or (at your option)
11 GAS is distributed in the hope that it will be useful, but WITHOUT
12 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
13 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
14 License for more details.
16 You should have received a copy of the GNU General Public License
17 along with GAS; see the file COPYING. If not, write to the Free
18 Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
21 /* Modified by Allen Wirfs-Brock, Instantiations Inc 2/90. */
22 /* App, the assembler pre-processor. This pre-processor strips out
23 excess spaces, turns single-quoted characters into a decimal
24 constant, and turns the # in # <number> <filename> <garbage> into a
25 .linefile. This needs better error-handling. */
31 #define const /* empty */
36 int enable_h_tick_hex = 0;
40 /* Whether we are scrubbing in m68k MRI mode. This is different from
41 flag_m68k_mri, because the two flags will be affected by the .mri
42 pseudo-op at different times. */
43 static int scrub_m68k_mri;
45 /* The pseudo-op which switches in and out of MRI mode. See the
46 comment in do_scrub_chars. */
47 static const char mri_pseudo[] = ".mri 0";
49 #define scrub_m68k_mri 0
52 #if defined TC_ARM && defined OBJ_ELF
53 /* The pseudo-op for which we need to special-case `@' characters.
54 See the comment in do_scrub_chars. */
55 static const char symver_pseudo[] = ".symver";
56 static const char * symver_state;
60 static const char symbol_chars[] =
61 "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
63 #define LEX_IS_SYMBOL_COMPONENT 1
64 #define LEX_IS_WHITESPACE 2
65 #define LEX_IS_LINE_SEPARATOR 3
66 #define LEX_IS_COMMENT_START 4
67 #define LEX_IS_LINE_COMMENT_START 5
68 #define LEX_IS_TWOCHAR_COMMENT_1ST 6
69 #define LEX_IS_STRINGQUOTE 8
70 #define LEX_IS_COLON 9
71 #define LEX_IS_NEWLINE 10
72 #define LEX_IS_ONECHAR_QUOTE 11
74 #define LEX_IS_DOUBLEDASH_1ST 12
77 #define DOUBLEBAR_PARALLEL
79 #ifdef DOUBLEBAR_PARALLEL
80 #define LEX_IS_DOUBLEBAR_1ST 13
82 #define LEX_IS_PARALLEL_SEPARATOR 14
86 #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
87 #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
88 #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
89 #define IS_PARALLEL_SEPARATOR(c) (lex[c] == LEX_IS_PARALLEL_SEPARATOR)
90 #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
91 #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
92 #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
94 static int process_escape (int);
96 /* FIXME-soon: The entire lexer/parser thingy should be
97 built statically at compile time rather than dynamically
98 each and every time the assembler is run. xoxorich. */
101 do_scrub_begin (int m68k_mri ATTRIBUTE_UNUSED)
106 lex[' '] = LEX_IS_WHITESPACE;
107 lex['\t'] = LEX_IS_WHITESPACE;
108 lex['\r'] = LEX_IS_WHITESPACE;
109 lex['\n'] = LEX_IS_NEWLINE;
110 lex[':'] = LEX_IS_COLON;
113 scrub_m68k_mri = m68k_mri;
118 lex['"'] = LEX_IS_STRINGQUOTE;
120 #if ! defined (TC_HPPA) && ! defined (TC_I370)
121 /* I370 uses single-quotes to delimit integer, float constants. */
122 lex['\''] = LEX_IS_ONECHAR_QUOTE;
125 #ifdef SINGLE_QUOTE_STRINGS
126 lex['\''] = LEX_IS_STRINGQUOTE;
130 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
131 in state 5 of do_scrub_chars must be changed. */
133 /* Note that these override the previous defaults, e.g. if ';' is a
134 comment char, then it isn't a line separator. */
135 for (p = symbol_chars; *p; ++p)
136 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
138 for (c = 128; c < 256; ++c)
139 lex[c] = LEX_IS_SYMBOL_COMPONENT;
141 #ifdef tc_symbol_chars
142 /* This macro permits the processor to specify all characters which
143 may appears in an operand. This will prevent the scrubber from
144 discarding meaningful whitespace in certain cases. The i386
145 backend uses this to support prefixes, which can confuse the
146 scrubber as to whether it is parsing operands or opcodes. */
147 for (p = tc_symbol_chars; *p; ++p)
148 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
151 /* The m68k backend wants to be able to change comment_chars. */
152 #ifndef tc_comment_chars
153 #define tc_comment_chars comment_chars
155 for (p = tc_comment_chars; *p; p++)
156 lex[(unsigned char) *p] = LEX_IS_COMMENT_START;
158 for (p = line_comment_chars; *p; p++)
159 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START;
161 for (p = line_separator_chars; *p; p++)
162 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
164 #ifdef tc_parallel_separator_chars
165 /* This macro permits the processor to specify all characters which
166 separate parallel insns on the same line. */
167 for (p = tc_parallel_separator_chars; *p; p++)
168 lex[(unsigned char) *p] = LEX_IS_PARALLEL_SEPARATOR;
171 /* Only allow slash-star comments if slash is not in use.
172 FIXME: This isn't right. We should always permit them. */
174 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
179 lex['\''] = LEX_IS_STRINGQUOTE;
180 lex[';'] = LEX_IS_COMMENT_START;
181 lex['*'] = LEX_IS_LINE_COMMENT_START;
182 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
183 then it can't be used in an expression. */
184 lex['!'] = LEX_IS_LINE_COMMENT_START;
189 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
191 #ifdef DOUBLEBAR_PARALLEL
192 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
195 /* Must do this is we want VLIW instruction with "->" or "<-". */
196 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
200 if (enable_h_tick_hex)
208 /* Saved state of the scrubber. */
210 static int old_state;
211 static char *out_string;
212 static char out_buf[20];
213 static int add_newlines;
214 static char *saved_input;
215 static size_t saved_input_len;
216 static char input_buffer[32 * 1024];
217 static const char *mri_state;
218 static char mri_last_ch;
220 /* Data structure for saving the state of app across #include's. Note that
221 app is called asynchronously to the parsing of the .include's, so our
222 state at the time .include is interpreted is completely unrelated.
223 That's why we have to save it all. */
230 char out_buf[sizeof (out_buf)];
233 size_t saved_input_len;
237 const char * mri_state;
239 #if defined TC_ARM && defined OBJ_ELF
240 const char * symver_state;
247 register struct app_save *saved;
249 saved = (struct app_save *) xmalloc (sizeof (*saved));
250 saved->state = state;
251 saved->old_state = old_state;
252 saved->out_string = out_string;
253 memcpy (saved->out_buf, out_buf, sizeof (out_buf));
254 saved->add_newlines = add_newlines;
255 if (saved_input == NULL)
256 saved->saved_input = NULL;
259 saved->saved_input = (char *) xmalloc (saved_input_len);
260 memcpy (saved->saved_input, saved_input, saved_input_len);
261 saved->saved_input_len = saved_input_len;
264 saved->scrub_m68k_mri = scrub_m68k_mri;
266 saved->mri_state = mri_state;
267 saved->mri_last_ch = mri_last_ch;
268 #if defined TC_ARM && defined OBJ_ELF
269 saved->symver_state = symver_state;
272 /* do_scrub_begin() is not useful, just wastes time. */
278 return (char *) saved;
284 register struct app_save *saved = (struct app_save *) arg;
286 /* There is no do_scrub_end (). */
287 state = saved->state;
288 old_state = saved->old_state;
289 out_string = saved->out_string;
290 memcpy (out_buf, saved->out_buf, sizeof (out_buf));
291 add_newlines = saved->add_newlines;
292 if (saved->saved_input == NULL)
296 gas_assert (saved->saved_input_len <= sizeof (input_buffer));
297 memcpy (input_buffer, saved->saved_input, saved->saved_input_len);
298 saved_input = input_buffer;
299 saved_input_len = saved->saved_input_len;
300 free (saved->saved_input);
303 scrub_m68k_mri = saved->scrub_m68k_mri;
305 mri_state = saved->mri_state;
306 mri_last_ch = saved->mri_last_ch;
307 #if defined TC_ARM && defined OBJ_ELF
308 symver_state = saved->symver_state;
314 /* @@ This assumes that \n &c are the same on host and target. This is not
318 process_escape (int ch)
341 /* This function is called to process input characters. The GET
342 parameter is used to retrieve more input characters. GET should
343 set its parameter to point to a buffer, and return the length of
344 the buffer; it should return 0 at end of file. The scrubbed output
345 characters are put into the buffer starting at TOSTART; the TOSTART
346 buffer is TOLEN bytes in length. The function returns the number
347 of scrubbed characters put into TOSTART. This will be TOLEN unless
348 end of file was seen. This function is arranged as a state
349 machine, and saves its state so that it may return at any point.
350 This is the way the old code used to work. */
353 do_scrub_chars (size_t (*get) (char *, size_t), char *tostart, size_t tolen)
356 char *toend = tostart + tolen;
360 register int ch, ch2 = 0;
361 /* Character that started the string we're working on. */
362 static char quotechar;
364 /*State 0: beginning of normal line
365 1: After first whitespace on line (flush more white)
366 2: After first non-white (opcode) on line (keep 1white)
367 3: after second white on line (into operands) (flush white)
368 4: after putting out a .linefile, put out digits
369 5: parsing a string, then go to old-state
370 6: putting out \ escape in a "d string.
373 9: After seeing symbol char in state 3 (keep 1white after symchar)
374 10: After seeing whitespace in state 9 (keep white before symchar)
375 11: After seeing a symbol character in state 0 (eg a label definition)
376 -1: output string in out_string and go to the state in old_state
377 -2: flush text until a '*' '/' is seen, then go to state old_state
379 12: After seeing a dash, looking for a second dash as a start
382 #ifdef DOUBLEBAR_PARALLEL
383 13: After seeing a vertical bar, looking for a second
384 vertical bar as a parallel expression separator.
386 #ifdef TC_PREDICATE_START_CHAR
387 14: After seeing a predicate start character at state 0, looking
388 for a predicate end character as predicate.
389 15: After seeing a predicate start character at state 1, looking
390 for a predicate end character as predicate.
393 16: After seeing an 'a' or an 'A' at the start of a symbol
394 17: After seeing an 'f' or an 'F' in state 16
398 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
399 constructs like ``.loc 1 20''. This was turning into ``.loc
400 120''. States 9 and 10 ensure that a space is never dropped in
401 between characters which could appear in an identifier. Ian
402 Taylor, ian@cygnus.com.
404 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
405 correctly on the PA (and any other target where colons are optional).
406 Jeff Law, law@cs.utah.edu.
408 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
409 get squashed into "cmp r1,r2||trap#1", with the all important space
410 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
412 /* This macro gets the next input character. */
416 ? * (unsigned char *) (from++) \
417 : (saved_input = NULL, \
418 fromlen = (*get) (input_buffer, sizeof input_buffer), \
419 from = input_buffer, \
420 fromend = from + fromlen, \
423 : * (unsigned char *) (from++))))
425 /* This macro pushes a character back on the input stream. */
427 #define UNGET(uch) (*--from = (uch))
429 /* This macro puts a character into the output buffer. If this
430 character fills the output buffer, this macro jumps to the label
431 TOFULL. We use this rather ugly approach because we need to
432 handle two different termination conditions: EOF on the input
433 stream, and a full output buffer. It would be simpler if we
434 always read in the entire input stream before processing it, but
435 I don't want to make such a significant change to the assembler's
447 if (saved_input != NULL)
450 fromend = from + saved_input_len;
454 fromlen = (*get) (input_buffer, sizeof input_buffer);
458 fromend = from + fromlen;
463 /* The cases in this switch end with continue, in order to
464 branch back to the top of this while loop and generate the
465 next output character in the appropriate state. */
470 if (*out_string == '\0')
487 as_warn (_("end of file in comment"));
496 while ((ch = GET ()) == '*')
501 as_warn (_("end of file in comment"));
519 else if (ch >= '0' && ch <= '9')
523 while (ch != EOF && IS_WHITESPACE (ch))
534 while (ch != EOF && ch != '\n')
543 /* We are going to copy everything up to a quote character,
544 with special handling for a backslash. We try to
545 optimize the copying in the simple case without using the
546 GET and PUT macros. */
551 for (s = from; s < fromend; s++)
560 if (len > toend - to)
564 memcpy (to, from, len);
575 /* This buffer is here specifically so
576 that the UNGET below will work. */
577 static char one_char_buf[1];
579 as_warn (_("end of file in string; '%c' inserted"), quotechar);
581 from = fromend = one_char_buf + 1;
586 else if (ch == quotechar)
591 #ifndef NO_STRING_ESCAPES
598 else if (scrub_m68k_mri && ch == '\n')
600 /* Just quietly terminate the string. This permits lines like
601 bne label loop if we haven't reach end yet. */
617 /* Handle strings broken across lines, by turning '\n' into
626 as_warn (_("end of file in string; '%c' inserted"), quotechar);
651 #ifdef ONLY_STANDARD_ESCAPES
652 as_warn (_("unknown escape '\\%c' in string; ignored"), ch);
659 #ifdef DOUBLEBAR_PARALLEL
665 /* Reset back to state 1 and pretend that we are parsing a
666 line from just after the first white space. */
670 /* "||^" is used for SPMASKed instructions. */
683 /* We have seen an 'a' at the start of a symbol, look for an 'f'. */
685 if (ch == 'f' || ch == 'F')
696 /* We have seen "af" at the start of a symbol,
697 a ' here is a part of that symbol. */
701 /* Change to avoid warning about unclosed string. */
709 /* OK, we are somewhere in states 0 through 4 or 9 through 11. */
714 #ifdef TC_PREDICATE_START_CHAR
715 if (ch == TC_PREDICATE_START_CHAR && (state == 0 || state == 1))
721 else if (state == 14 || state == 15)
723 if (ch == TC_PREDICATE_END_CHAR)
739 #if defined TC_ARM && defined OBJ_ELF
740 /* We need to watch out for .symver directives. See the comment later
742 if (symver_state == NULL)
744 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
745 symver_state = symver_pseudo + 1;
749 /* We advance to the next state if we find the right
751 if (ch != '\0' && (*symver_state == ch))
753 else if (*symver_state != '\0')
754 /* We did not get the expected character, or we didn't
755 get a valid terminating character after seeing the
756 entire pseudo-op, so we must go back to the beginning. */
760 /* We've read the entire pseudo-op. If this is the end
761 of the line, go back to the beginning. */
766 #endif /* TC_ARM && OBJ_ELF */
769 /* We want to have pseudo-ops which control whether we are in
770 MRI mode or not. Unfortunately, since m68k MRI mode affects
771 the scrubber, that means that we need a special purpose
773 if (mri_state == NULL)
775 if ((state == 0 || state == 1)
776 && ch == mri_pseudo[0])
777 mri_state = mri_pseudo + 1;
781 /* We advance to the next state if we find the right
782 character, or if we need a space character and we get any
783 whitespace character, or if we need a '0' and we get a
784 '1' (this is so that we only need one state to handle
785 ``.mri 0'' and ``.mri 1''). */
788 || (*mri_state == ' '
789 && lex[ch] == LEX_IS_WHITESPACE)
790 || (*mri_state == '0'
796 else if (*mri_state != '\0'
797 || (lex[ch] != LEX_IS_WHITESPACE
798 && lex[ch] != LEX_IS_NEWLINE))
800 /* We did not get the expected character, or we didn't
801 get a valid terminating character after seeing the
802 entire pseudo-op, so we must go back to the
808 /* We've read the entire pseudo-op. mips_last_ch is
809 either '0' or '1' indicating whether to enter or
811 do_scrub_begin (mri_last_ch == '1');
814 /* We continue handling the character as usual. The
815 main gas reader must also handle the .mri pseudo-op
816 to control expression parsing and the like. */
825 as_warn (_("end of file not at end of a line; newline inserted"));
834 case LEX_IS_WHITESPACE:
839 while (ch != EOF && IS_WHITESPACE (ch));
845 /* Preserve a single whitespace character at the
846 beginning of a line. */
853 #ifdef KEEP_WHITE_AROUND_COLON
854 if (lex[ch] == LEX_IS_COLON)
856 /* Only keep this white if there's no white *after* the
861 if (!IS_WHITESPACE (ch2))
872 || IS_LINE_SEPARATOR (ch)
873 || IS_PARALLEL_SEPARATOR (ch))
877 /* In MRI mode, we keep these spaces. */
885 /* If we're in state 2 or 11, we've seen a non-white
886 character followed by whitespace. If the next character
887 is ':', this is whitespace after a label name which we
888 normally must ignore. In MRI mode, though, spaces are
889 not permitted between the label and the colon. */
890 if ((state == 2 || state == 11)
891 && lex[ch] == LEX_IS_COLON
902 /* We can arrive here if we leave a leading whitespace
903 character at the beginning of a line. */
909 /* Optimize common case by skipping UNGET/GET. */
910 PUT (' '); /* Sp after opco */
917 #ifndef TC_KEEP_OPERAND_SPACES
918 /* For TI C6X, we keep these spaces as they may separate
919 functional unit specifiers from operands. */
923 /* In MRI mode, we keep these spaces. */
928 goto recycle; /* Sp in operands */
931 #ifndef TC_KEEP_OPERAND_SPACES
935 /* In MRI mode, we keep these spaces. */
941 state = 10; /* Sp after symbol char */
944 if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
948 /* We know that ch is not ':', since we tested that
949 case above. Therefore this is not a label, so it
950 must be the opcode, and we've just seen the
951 whitespace after it. */
955 PUT (' '); /* Sp after label definition. */
962 case LEX_IS_TWOCHAR_COMMENT_1ST:
971 if (ch2 != EOF && IS_NEWLINE (ch2))
974 while (ch2 != EOF && ch2 != '*');
979 if (ch2 == EOF || ch2 == '/')
982 /* This UNGET will ensure that we count newlines
988 as_warn (_("end of file in multiline comment"));
993 #ifdef DOUBLESLASH_LINE_COMMENTS
1000 while (ch != EOF && !IS_NEWLINE (ch));
1002 as_warn ("end of file in comment; newline inserted");
1012 if (state == 9 || state == 10)
1018 case LEX_IS_STRINGQUOTE:
1022 /* Preserve the whitespace in foo "bar". */
1027 /* PUT didn't jump out. We could just break, but we
1028 know what will happen, so optimize a bit. */
1032 else if (state == 9)
1041 case LEX_IS_ONECHAR_QUOTE:
1043 if (state == 9 && enable_h_tick_hex)
1048 as_warn ("'%c found after symbol", c);
1054 /* Preserve the whitespace in foo 'b'. */
1063 as_warn (_("end of file after a one-character quote; \\0 inserted"));
1071 as_warn (_("end of file in escape character"));
1075 ch = process_escape (ch);
1077 sprintf (out_buf, "%d", (int) (unsigned char) ch);
1079 /* None of these 'x constants for us. We want 'x'. */
1080 if ((ch = GET ()) != '\'')
1082 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
1083 as_warn (_("missing close quote; (assumed)"));
1089 if (strlen (out_buf) == 1)
1099 out_string = out_buf;
1100 PUT (*out_string++);
1105 #ifdef KEEP_WHITE_AROUND_COLON
1108 if (state == 9 || state == 10)
1110 else if (state != 3)
1116 case LEX_IS_NEWLINE:
1117 /* Roll out a bunch of newlines from inside comments, etc. */
1125 case LEX_IS_LINE_SEPARATOR:
1130 case LEX_IS_PARALLEL_SEPARATOR:
1136 case LEX_IS_DOUBLEDASH_1ST:
1144 /* Read and skip to end of line. */
1149 while (ch != EOF && ch != '\n');
1152 as_warn (_("end of file in comment; newline inserted"));
1158 #ifdef DOUBLEBAR_PARALLEL
1159 case LEX_IS_DOUBLEBAR_1ST:
1166 /* Handle '||' in two states as invoking PUT twice might
1167 result in the first one jumping out of this loop. We'd
1168 then lose track of the state and one '|' char. */
1173 case LEX_IS_LINE_COMMENT_START:
1174 /* FIXME-someday: The two character comment stuff was badly
1175 thought out. On i386, we want '/' as line comment start
1176 AND we want C style comments. hence this hack. The
1177 whole lexical process should be reworked. xoxorich. */
1193 if (state == 0 || state == 1) /* Only comment at start of line. */
1203 while (ch != EOF && IS_WHITESPACE (ch));
1207 as_warn (_("end of file in comment; newline inserted"));
1212 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1214 /* Not a cpp line. */
1215 while (ch != EOF && !IS_NEWLINE (ch))
1219 as_warn (_("end of file in comment; newline inserted"));
1222 else /* IS_NEWLINE (ch) */
1224 /* To process non-zero add_newlines. */
1230 /* Looks like `# 123 "filename"' from cpp. */
1235 out_string = "\tlinefile ";
1237 out_string = "\t.linefile ";
1238 PUT (*out_string++);
1243 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1244 Trap is the only short insn that has a first operand that is
1245 neither register nor label.
1246 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1247 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is
1248 already LEX_IS_LINE_COMMENT_START. However, it is the
1249 only character in line_comment_chars for d10v, hence we
1250 can recognize it as such. */
1251 /* An alternative approach would be to reset the state to 1 when
1252 we see '||', '<'- or '->', but that seems to be overkill. */
1256 /* We have a line comment character which is not at the
1257 start of a line. If this is also a normal comment
1258 character, fall through. Otherwise treat it as a default
1260 if (strchr (tc_comment_chars, ch) == NULL
1261 && (! scrub_m68k_mri
1262 || (ch != '!' && ch != '*')))
1265 && (ch == '!' || ch == '*' || ch == '#')
1270 case LEX_IS_COMMENT_START:
1271 #if defined TC_ARM && defined OBJ_ELF
1272 /* On the ARM, `@' is the comment character.
1273 Unfortunately this is also a special character in ELF .symver
1274 directives (and .type, though we deal with those another way).
1275 So we check if this line is such a directive, and treat
1276 the character as default if so. This is a hack. */
1277 if ((symver_state != NULL) && (*symver_state == 0))
1282 /* For the ARM, care is needed not to damage occurrences of \@
1283 by stripping the @ onwards. Yuck. */
1284 if (to > tostart && *(to - 1) == '\\')
1285 /* Do not treat the @ as a start-of-comment. */
1289 #ifdef WARN_COMMENTS
1291 as_where (&found_comment_file, &found_comment);
1297 while (ch != EOF && !IS_NEWLINE (ch));
1299 as_warn (_("end of file in comment; newline inserted"));
1306 /* Look for strings like H'[0-9A-Fa-f] and if found, replace
1307 the H' with 0x to make them gas-style hex characters. */
1308 if (enable_h_tick_hex)
1324 case LEX_IS_SYMBOL_COMPONENT:
1327 /* This is a symbol character following another symbol
1328 character, with whitespace in between. We skipped
1329 the whitespace earlier, so output it now. */
1337 /* "af'" is a symbol containing '\''. */
1338 if (state == 3 && (ch == 'a' || ch == 'A'))
1343 if (ch == 'f' || ch == 'F')
1352 if (ch == EOF || !IS_SYMBOL_COMPONENT (ch))
1364 /* This is a common case. Quickly copy CH and all the
1365 following symbol component or normal characters. */
1367 && mri_state == NULL
1368 #if defined TC_ARM && defined OBJ_ELF
1369 && symver_state == NULL
1376 for (s = from; s < fromend; s++)
1380 ch2 = *(unsigned char *) s;
1383 && type != LEX_IS_SYMBOL_COMPONENT)
1388 /* Handle the last character normally, for
1394 if (len > (toend - to) - 1)
1395 len = (toend - to) - 1;
1400 memcpy (to, from, len);
1412 /* Some relatively `normal' character. */
1415 state = 11; /* Now seeing label definition. */
1417 else if (state == 1)
1419 state = 2; /* Ditto. */
1421 else if (state == 9)
1423 if (!IS_SYMBOL_COMPONENT (ch))
1426 else if (state == 10)
1430 /* Special handling for backslash: a backslash may
1431 be the beginning of a formal parameter (of a
1432 macro) following another symbol character, with
1433 whitespace in between. If that is the case, we
1434 output a space before the parameter. Strictly
1435 speaking, correct handling depends upon what the
1436 macro parameter expands into; if the parameter
1437 expands into something which does not start with
1438 an operand character, then we don't want to keep
1439 the space. We don't have enough information to
1440 make the right choice, so here we are making the
1441 choice which is more likely to be correct. */
1442 if (to + 1 >= toend)
1444 /* If we're near the end of the buffer, save the
1445 character for the next time round. Otherwise
1446 we'll lose our state. */
1463 /* We have reached the end of the input. */
1464 return to - tostart;
1467 /* The output buffer is full. Save any input we have not yet
1472 saved_input_len = fromend - from;
1477 return to - tostart;