1 /* This is the Assembler Pre-Processor
2 Copyright (C) 1987-2019 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;
59 static char last_char;
63 static const char symbol_chars[] =
64 "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
66 #define LEX_IS_SYMBOL_COMPONENT 1
67 #define LEX_IS_WHITESPACE 2
68 #define LEX_IS_LINE_SEPARATOR 3
69 #define LEX_IS_COMMENT_START 4
70 #define LEX_IS_LINE_COMMENT_START 5
71 #define LEX_IS_TWOCHAR_COMMENT_1ST 6
72 #define LEX_IS_STRINGQUOTE 8
73 #define LEX_IS_COLON 9
74 #define LEX_IS_NEWLINE 10
75 #define LEX_IS_ONECHAR_QUOTE 11
77 #define LEX_IS_DOUBLEDASH_1ST 12
80 #define DOUBLEBAR_PARALLEL
82 #ifdef DOUBLEBAR_PARALLEL
83 #define LEX_IS_DOUBLEBAR_1ST 13
85 #define LEX_IS_PARALLEL_SEPARATOR 14
89 #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
90 #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
91 #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
92 #define IS_PARALLEL_SEPARATOR(c) (lex[c] == LEX_IS_PARALLEL_SEPARATOR)
93 #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
94 #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
95 #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
97 static int process_escape (int);
99 /* FIXME-soon: The entire lexer/parser thingy should be
100 built statically at compile time rather than dynamically
101 each and every time the assembler is run. xoxorich. */
104 do_scrub_begin (int m68k_mri ATTRIBUTE_UNUSED)
109 lex[' '] = LEX_IS_WHITESPACE;
110 lex['\t'] = LEX_IS_WHITESPACE;
111 lex['\r'] = LEX_IS_WHITESPACE;
112 lex['\n'] = LEX_IS_NEWLINE;
113 lex[':'] = LEX_IS_COLON;
116 scrub_m68k_mri = m68k_mri;
121 lex['"'] = LEX_IS_STRINGQUOTE;
123 #if ! defined (TC_HPPA)
124 lex['\''] = LEX_IS_ONECHAR_QUOTE;
127 #ifdef SINGLE_QUOTE_STRINGS
128 lex['\''] = LEX_IS_STRINGQUOTE;
132 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
133 in state 5 of do_scrub_chars must be changed. */
135 /* Note that these override the previous defaults, e.g. if ';' is a
136 comment char, then it isn't a line separator. */
137 for (p = symbol_chars; *p; ++p)
138 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
140 for (c = 128; c < 256; ++c)
141 lex[c] = LEX_IS_SYMBOL_COMPONENT;
143 #ifdef tc_symbol_chars
144 /* This macro permits the processor to specify all characters which
145 may appears in an operand. This will prevent the scrubber from
146 discarding meaningful whitespace in certain cases. The i386
147 backend uses this to support prefixes, which can confuse the
148 scrubber as to whether it is parsing operands or opcodes. */
149 for (p = tc_symbol_chars; *p; ++p)
150 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
153 /* The m68k backend wants to be able to change comment_chars. */
154 #ifndef tc_comment_chars
155 #define tc_comment_chars comment_chars
157 for (p = tc_comment_chars; *p; p++)
158 lex[(unsigned char) *p] = LEX_IS_COMMENT_START;
160 for (p = line_comment_chars; *p; p++)
161 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START;
163 #ifndef tc_line_separator_chars
164 #define tc_line_separator_chars line_separator_chars
166 for (p = tc_line_separator_chars; *p; p++)
167 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
169 #ifdef tc_parallel_separator_chars
170 /* This macro permits the processor to specify all characters which
171 separate parallel insns on the same line. */
172 for (p = tc_parallel_separator_chars; *p; p++)
173 lex[(unsigned char) *p] = LEX_IS_PARALLEL_SEPARATOR;
176 /* Only allow slash-star comments if slash is not in use.
177 FIXME: This isn't right. We should always permit them. */
179 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
184 lex['\''] = LEX_IS_STRINGQUOTE;
185 lex[';'] = LEX_IS_COMMENT_START;
186 lex['*'] = LEX_IS_LINE_COMMENT_START;
187 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
188 then it can't be used in an expression. */
189 lex['!'] = LEX_IS_LINE_COMMENT_START;
194 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
196 #ifdef DOUBLEBAR_PARALLEL
197 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
200 /* Must do this is we want VLIW instruction with "->" or "<-". */
201 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
205 if (enable_h_tick_hex)
213 /* Saved state of the scrubber. */
215 static int old_state;
216 static const char *out_string;
217 static char out_buf[20];
218 static int add_newlines;
219 static char *saved_input;
220 static size_t saved_input_len;
221 static char input_buffer[32 * 1024];
222 static const char *mri_state;
223 static char mri_last_ch;
225 /* Data structure for saving the state of app across #include's. Note that
226 app is called asynchronously to the parsing of the .include's, so our
227 state at the time .include is interpreted is completely unrelated.
228 That's why we have to save it all. */
234 const char * out_string;
235 char out_buf[sizeof (out_buf)];
238 size_t saved_input_len;
242 const char * mri_state;
244 #if defined TC_ARM && defined OBJ_ELF
245 const char * symver_state;
255 struct app_save *saved;
257 saved = XNEW (struct app_save);
258 saved->state = state;
259 saved->old_state = old_state;
260 saved->out_string = out_string;
261 memcpy (saved->out_buf, out_buf, sizeof (out_buf));
262 saved->add_newlines = add_newlines;
263 if (saved_input == NULL)
264 saved->saved_input = NULL;
267 saved->saved_input = XNEWVEC (char, saved_input_len);
268 memcpy (saved->saved_input, saved_input, saved_input_len);
269 saved->saved_input_len = saved_input_len;
272 saved->scrub_m68k_mri = scrub_m68k_mri;
274 saved->mri_state = mri_state;
275 saved->mri_last_ch = mri_last_ch;
276 #if defined TC_ARM && defined OBJ_ELF
277 saved->symver_state = symver_state;
280 saved->last_char = last_char;
283 /* do_scrub_begin() is not useful, just wastes time. */
289 return (char *) saved;
295 struct app_save *saved = (struct app_save *) arg;
297 /* There is no do_scrub_end (). */
298 state = saved->state;
299 old_state = saved->old_state;
300 out_string = saved->out_string;
301 memcpy (out_buf, saved->out_buf, sizeof (out_buf));
302 add_newlines = saved->add_newlines;
303 if (saved->saved_input == NULL)
307 gas_assert (saved->saved_input_len <= sizeof (input_buffer));
308 memcpy (input_buffer, saved->saved_input, saved->saved_input_len);
309 saved_input = input_buffer;
310 saved_input_len = saved->saved_input_len;
311 free (saved->saved_input);
314 scrub_m68k_mri = saved->scrub_m68k_mri;
316 mri_state = saved->mri_state;
317 mri_last_ch = saved->mri_last_ch;
318 #if defined TC_ARM && defined OBJ_ELF
319 symver_state = saved->symver_state;
322 last_char = saved->last_char;
328 /* @@ This assumes that \n &c are the same on host and target. This is not
332 process_escape (int ch)
355 /* This function is called to process input characters. The GET
356 parameter is used to retrieve more input characters. GET should
357 set its parameter to point to a buffer, and return the length of
358 the buffer; it should return 0 at end of file. The scrubbed output
359 characters are put into the buffer starting at TOSTART; the TOSTART
360 buffer is TOLEN bytes in length. The function returns the number
361 of scrubbed characters put into TOSTART. This will be TOLEN unless
362 end of file was seen. This function is arranged as a state
363 machine, and saves its state so that it may return at any point.
364 This is the way the old code used to work. */
367 do_scrub_chars (size_t (*get) (char *, size_t), char *tostart, size_t tolen)
370 char *toend = tostart + tolen;
375 /* Character that started the string we're working on. */
376 static char quotechar;
378 /*State 0: beginning of normal line
379 1: After first whitespace on line (flush more white)
380 2: After first non-white (opcode) on line (keep 1white)
381 3: after second white on line (into operands) (flush white)
382 4: after putting out a .linefile, put out digits
383 5: parsing a string, then go to old-state
384 6: putting out \ escape in a "d string.
387 9: After seeing symbol char in state 3 (keep 1white after symchar)
388 10: After seeing whitespace in state 9 (keep white before symchar)
389 11: After seeing a symbol character in state 0 (eg a label definition)
390 -1: output string in out_string and go to the state in old_state
391 -2: flush text until a '*' '/' is seen, then go to state old_state
393 12: After seeing a dash, looking for a second dash as a start
396 #ifdef DOUBLEBAR_PARALLEL
397 13: After seeing a vertical bar, looking for a second
398 vertical bar as a parallel expression separator.
400 #ifdef TC_PREDICATE_START_CHAR
401 14: After seeing a predicate start character at state 0, looking
402 for a predicate end character as predicate.
403 15: After seeing a predicate start character at state 1, looking
404 for a predicate end character as predicate.
407 16: After seeing an 'a' or an 'A' at the start of a symbol
408 17: After seeing an 'f' or an 'F' in state 16
412 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
413 constructs like ``.loc 1 20''. This was turning into ``.loc
414 120''. States 9 and 10 ensure that a space is never dropped in
415 between characters which could appear in an identifier. Ian
416 Taylor, ian@cygnus.com.
418 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
419 correctly on the PA (and any other target where colons are optional).
420 Jeff Law, law@cs.utah.edu.
422 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
423 get squashed into "cmp r1,r2||trap#1", with the all important space
424 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
426 /* This macro gets the next input character. */
430 ? * (unsigned char *) (from++) \
431 : (saved_input = NULL, \
432 fromlen = (*get) (input_buffer, sizeof input_buffer), \
433 from = input_buffer, \
434 fromend = from + fromlen, \
437 : * (unsigned char *) (from++))))
439 /* This macro pushes a character back on the input stream. */
441 #define UNGET(uch) (*--from = (uch))
443 /* This macro puts a character into the output buffer. If this
444 character fills the output buffer, this macro jumps to the label
445 TOFULL. We use this rather ugly approach because we need to
446 handle two different termination conditions: EOF on the input
447 stream, and a full output buffer. It would be simpler if we
448 always read in the entire input stream before processing it, but
449 I don't want to make such a significant change to the assembler's
461 if (saved_input != NULL)
464 fromend = from + saved_input_len;
468 fromlen = (*get) (input_buffer, sizeof input_buffer);
472 fromend = from + fromlen;
477 /* The cases in this switch end with continue, in order to
478 branch back to the top of this while loop and generate the
479 next output character in the appropriate state. */
484 if (*out_string == '\0')
501 as_warn (_("end of file in comment"));
510 while ((ch = GET ()) == '*')
515 as_warn (_("end of file in comment"));
533 else if (ch >= '0' && ch <= '9')
537 while (ch != EOF && IS_WHITESPACE (ch))
548 while (ch != EOF && ch != '\n')
557 /* We are going to copy everything up to a quote character,
558 with special handling for a backslash. We try to
559 optimize the copying in the simple case without using the
560 GET and PUT macros. */
565 for (s = from; s < fromend; s++)
574 if (len > toend - to)
578 memcpy (to, from, len);
589 /* This buffer is here specifically so
590 that the UNGET below will work. */
591 static char one_char_buf[1];
593 as_warn (_("end of file in string; '%c' inserted"), quotechar);
595 from = fromend = one_char_buf + 1;
600 else if (ch == quotechar)
605 #ifndef NO_STRING_ESCAPES
612 else if (scrub_m68k_mri && ch == '\n')
614 /* Just quietly terminate the string. This permits lines like
615 bne label loop if we haven't reach end yet. */
631 /* Handle strings broken across lines, by turning '\n' into
640 as_warn (_("end of file in string; '%c' inserted"), quotechar);
665 #ifdef ONLY_STANDARD_ESCAPES
666 as_warn (_("unknown escape '\\%c' in string; ignored"), ch);
673 #ifdef DOUBLEBAR_PARALLEL
679 /* Reset back to state 1 and pretend that we are parsing a
680 line from just after the first white space. */
684 /* "||^" is used for SPMASKed instructions. */
697 /* We have seen an 'a' at the start of a symbol, look for an 'f'. */
699 if (ch == 'f' || ch == 'F')
711 /* We have seen "af" at the start of a symbol,
712 a ' here is a part of that symbol. */
716 /* Change to avoid warning about unclosed string. */
724 /* OK, we are somewhere in states 0 through 4 or 9 through 11. */
729 #ifdef TC_PREDICATE_START_CHAR
730 if (ch == TC_PREDICATE_START_CHAR && (state == 0 || state == 1))
736 else if (state == 14 || state == 15)
738 if (ch == TC_PREDICATE_END_CHAR)
754 #if defined TC_ARM && defined OBJ_ELF
755 /* We need to watch out for .symver directives. See the comment later
757 if (symver_state == NULL)
759 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
760 symver_state = symver_pseudo + 1;
764 /* We advance to the next state if we find the right
766 if (ch != '\0' && (*symver_state == ch))
768 else if (*symver_state != '\0')
769 /* We did not get the expected character, or we didn't
770 get a valid terminating character after seeing the
771 entire pseudo-op, so we must go back to the beginning. */
775 /* We've read the entire pseudo-op. If this is the end
776 of the line, go back to the beginning. */
781 #endif /* TC_ARM && OBJ_ELF */
784 /* We want to have pseudo-ops which control whether we are in
785 MRI mode or not. Unfortunately, since m68k MRI mode affects
786 the scrubber, that means that we need a special purpose
788 if (mri_state == NULL)
790 if ((state == 0 || state == 1)
791 && ch == mri_pseudo[0])
792 mri_state = mri_pseudo + 1;
796 /* We advance to the next state if we find the right
797 character, or if we need a space character and we get any
798 whitespace character, or if we need a '0' and we get a
799 '1' (this is so that we only need one state to handle
800 ``.mri 0'' and ``.mri 1''). */
803 || (*mri_state == ' '
804 && lex[ch] == LEX_IS_WHITESPACE)
805 || (*mri_state == '0'
811 else if (*mri_state != '\0'
812 || (lex[ch] != LEX_IS_WHITESPACE
813 && lex[ch] != LEX_IS_NEWLINE))
815 /* We did not get the expected character, or we didn't
816 get a valid terminating character after seeing the
817 entire pseudo-op, so we must go back to the
823 /* We've read the entire pseudo-op. mips_last_ch is
824 either '0' or '1' indicating whether to enter or
826 do_scrub_begin (mri_last_ch == '1');
829 /* We continue handling the character as usual. The
830 main gas reader must also handle the .mri pseudo-op
831 to control expression parsing and the like. */
840 as_warn (_("end of file not at end of a line; newline inserted"));
849 case LEX_IS_WHITESPACE:
854 while (ch != EOF && IS_WHITESPACE (ch));
860 /* Preserve a single whitespace character at the
861 beginning of a line. */
868 #ifdef KEEP_WHITE_AROUND_COLON
869 if (lex[ch] == LEX_IS_COLON)
871 /* Only keep this white if there's no white *after* the
876 if (!IS_WHITESPACE (ch2))
887 || IS_LINE_SEPARATOR (ch)
888 || IS_PARALLEL_SEPARATOR (ch))
892 /* In MRI mode, we keep these spaces. */
900 /* If we're in state 2 or 11, we've seen a non-white
901 character followed by whitespace. If the next character
902 is ':', this is whitespace after a label name which we
903 normally must ignore. In MRI mode, though, spaces are
904 not permitted between the label and the colon. */
905 if ((state == 2 || state == 11)
906 && lex[ch] == LEX_IS_COLON
917 /* We can arrive here if we leave a leading whitespace
918 character at the beginning of a line. */
924 /* Optimize common case by skipping UNGET/GET. */
925 PUT (' '); /* Sp after opco */
932 #ifndef TC_KEEP_OPERAND_SPACES
933 /* For TI C6X, we keep these spaces as they may separate
934 functional unit specifiers from operands. */
938 /* In MRI mode, we keep these spaces. */
943 goto recycle; /* Sp in operands */
946 #ifndef TC_KEEP_OPERAND_SPACES
950 /* In MRI mode, we keep these spaces. */
956 state = 10; /* Sp after symbol char */
959 if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
963 /* We know that ch is not ':', since we tested that
964 case above. Therefore this is not a label, so it
965 must be the opcode, and we've just seen the
966 whitespace after it. */
970 PUT (' '); /* Sp after label definition. */
977 case LEX_IS_TWOCHAR_COMMENT_1ST:
986 if (ch2 != EOF && IS_NEWLINE (ch2))
989 while (ch2 != EOF && ch2 != '*');
994 if (ch2 == EOF || ch2 == '/')
997 /* This UNGET will ensure that we count newlines
1003 as_warn (_("end of file in multiline comment"));
1008 #ifdef DOUBLESLASH_LINE_COMMENTS
1009 else if (ch2 == '/')
1015 while (ch != EOF && !IS_NEWLINE (ch));
1017 as_warn ("end of file in comment; newline inserted");
1027 if (state == 9 || state == 10)
1033 case LEX_IS_STRINGQUOTE:
1037 /* Preserve the whitespace in foo "bar". */
1042 /* PUT didn't jump out. We could just break, but we
1043 know what will happen, so optimize a bit. */
1047 else if (state == 9)
1055 case LEX_IS_ONECHAR_QUOTE:
1057 if (state == 9 && enable_h_tick_hex)
1062 as_warn ("'%c found after symbol", c);
1068 /* Preserve the whitespace in foo 'b'. */
1077 as_warn (_("end of file after a one-character quote; \\0 inserted"));
1085 as_warn (_("end of file in escape character"));
1089 ch = process_escape (ch);
1091 sprintf (out_buf, "%d", (int) (unsigned char) ch);
1093 /* None of these 'x constants for us. We want 'x'. */
1094 if ((ch = GET ()) != '\'')
1096 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
1097 as_warn (_("missing close quote; (assumed)"));
1103 if (strlen (out_buf) == 1)
1113 out_string = out_buf;
1114 PUT (*out_string++);
1118 #ifdef KEEP_WHITE_AROUND_COLON
1121 if (state == 9 || state == 10)
1123 else if (state != 3)
1129 case LEX_IS_NEWLINE:
1130 /* Roll out a bunch of newlines from inside comments, etc. */
1138 case LEX_IS_LINE_SEPARATOR:
1143 case LEX_IS_PARALLEL_SEPARATOR:
1149 case LEX_IS_DOUBLEDASH_1ST:
1157 /* Read and skip to end of line. */
1162 while (ch != EOF && ch != '\n');
1165 as_warn (_("end of file in comment; newline inserted"));
1171 #ifdef DOUBLEBAR_PARALLEL
1172 case LEX_IS_DOUBLEBAR_1ST:
1179 /* Handle '||' in two states as invoking PUT twice might
1180 result in the first one jumping out of this loop. We'd
1181 then lose track of the state and one '|' char. */
1186 case LEX_IS_LINE_COMMENT_START:
1187 /* FIXME-someday: The two character comment stuff was badly
1188 thought out. On i386, we want '/' as line comment start
1189 AND we want C style comments. hence this hack. The
1190 whole lexical process should be reworked. xoxorich. */
1200 else if (ch2 != EOF)
1206 if (state == 0 || state == 1) /* Only comment at start of line. */
1216 while (ch != EOF && IS_WHITESPACE (ch));
1220 as_warn (_("end of file in comment; newline inserted"));
1225 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1227 /* Not a cpp line. */
1228 while (ch != EOF && !IS_NEWLINE (ch))
1232 as_warn (_("end of file in comment; newline inserted"));
1235 else /* IS_NEWLINE (ch) */
1237 /* To process non-zero add_newlines. */
1243 /* Looks like `# 123 "filename"' from cpp. */
1248 out_string = "\tlinefile ";
1250 out_string = "\t.linefile ";
1251 PUT (*out_string++);
1256 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1257 Trap is the only short insn that has a first operand that is
1258 neither register nor label.
1259 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1260 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is
1261 already LEX_IS_LINE_COMMENT_START. However, it is the
1262 only character in line_comment_chars for d10v, hence we
1263 can recognize it as such. */
1264 /* An alternative approach would be to reset the state to 1 when
1265 we see '||', '<'- or '->', but that seems to be overkill. */
1269 /* We have a line comment character which is not at the
1270 start of a line. If this is also a normal comment
1271 character, fall through. Otherwise treat it as a default
1273 if (strchr (tc_comment_chars, ch) == NULL
1274 && (! scrub_m68k_mri
1275 || (ch != '!' && ch != '*')))
1278 && (ch == '!' || ch == '*' || ch == '#')
1283 case LEX_IS_COMMENT_START:
1284 #if defined TC_ARM && defined OBJ_ELF
1285 /* On the ARM, `@' is the comment character.
1286 Unfortunately this is also a special character in ELF .symver
1287 directives (and .type, though we deal with those another way).
1288 So we check if this line is such a directive, and treat
1289 the character as default if so. This is a hack. */
1290 if ((symver_state != NULL) && (*symver_state == 0))
1295 /* For the ARM, care is needed not to damage occurrences of \@
1296 by stripping the @ onwards. Yuck. */
1297 if ((to > tostart ? to[-1] : last_char) == '\\')
1298 /* Do not treat the @ as a start-of-comment. */
1302 #ifdef WARN_COMMENTS
1304 found_comment_file = as_where (&found_comment);
1310 while (ch != EOF && !IS_NEWLINE (ch));
1312 as_warn (_("end of file in comment; newline inserted"));
1319 /* Look for strings like H'[0-9A-Fa-f] and if found, replace
1320 the H' with 0x to make them gas-style hex characters. */
1321 if (enable_h_tick_hex)
1337 case LEX_IS_SYMBOL_COMPONENT:
1340 /* This is a symbol character following another symbol
1341 character, with whitespace in between. We skipped
1342 the whitespace earlier, so output it now. */
1350 /* "af'" is a symbol containing '\''. */
1351 if (state == 3 && (ch == 'a' || ch == 'A'))
1356 if (ch == 'f' || ch == 'F')
1365 if (ch == EOF || !IS_SYMBOL_COMPONENT (ch))
1377 /* This is a common case. Quickly copy CH and all the
1378 following symbol component or normal characters. */
1380 && mri_state == NULL
1381 #if defined TC_ARM && defined OBJ_ELF
1382 && symver_state == NULL
1389 for (s = from; s < fromend; s++)
1393 ch2 = *(unsigned char *) s;
1396 && type != LEX_IS_SYMBOL_COMPONENT)
1401 /* Handle the last character normally, for
1407 if (len > (toend - to) - 1)
1408 len = (toend - to) - 1;
1413 memcpy (to, from, len);
1425 /* Some relatively `normal' character. */
1428 state = 11; /* Now seeing label definition. */
1430 else if (state == 1)
1432 state = 2; /* Ditto. */
1434 else if (state == 9)
1436 if (!IS_SYMBOL_COMPONENT (ch))
1439 else if (state == 10)
1443 /* Special handling for backslash: a backslash may
1444 be the beginning of a formal parameter (of a
1445 macro) following another symbol character, with
1446 whitespace in between. If that is the case, we
1447 output a space before the parameter. Strictly
1448 speaking, correct handling depends upon what the
1449 macro parameter expands into; if the parameter
1450 expands into something which does not start with
1451 an operand character, then we don't want to keep
1452 the space. We don't have enough information to
1453 make the right choice, so here we are making the
1454 choice which is more likely to be correct. */
1455 if (to + 1 >= toend)
1457 /* If we're near the end of the buffer, save the
1458 character for the next time round. Otherwise
1459 we'll lose our state. */
1476 /* We have reached the end of the input. */
1481 return to - tostart;
1484 /* The output buffer is full. Save any input we have not yet
1489 saved_input_len = fromend - from;
1498 return to - tostart;