1 /* This is the Assembler Pre-Processor
2 Copyright (C) 1987-2017 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 #ifndef tc_line_separator_chars
162 #define tc_line_separator_chars line_separator_chars
164 for (p = tc_line_separator_chars; *p; p++)
165 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
167 #ifdef tc_parallel_separator_chars
168 /* This macro permits the processor to specify all characters which
169 separate parallel insns on the same line. */
170 for (p = tc_parallel_separator_chars; *p; p++)
171 lex[(unsigned char) *p] = LEX_IS_PARALLEL_SEPARATOR;
174 /* Only allow slash-star comments if slash is not in use.
175 FIXME: This isn't right. We should always permit them. */
177 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
182 lex['\''] = LEX_IS_STRINGQUOTE;
183 lex[';'] = LEX_IS_COMMENT_START;
184 lex['*'] = LEX_IS_LINE_COMMENT_START;
185 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
186 then it can't be used in an expression. */
187 lex['!'] = LEX_IS_LINE_COMMENT_START;
192 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
194 #ifdef DOUBLEBAR_PARALLEL
195 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
198 /* Must do this is we want VLIW instruction with "->" or "<-". */
199 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
203 if (enable_h_tick_hex)
211 /* Saved state of the scrubber. */
213 static int old_state;
214 static const char *out_string;
215 static char out_buf[20];
216 static int add_newlines;
217 static char *saved_input;
218 static size_t saved_input_len;
219 static char input_buffer[32 * 1024];
220 static const char *mri_state;
221 static char mri_last_ch;
223 /* Data structure for saving the state of app across #include's. Note that
224 app is called asynchronously to the parsing of the .include's, so our
225 state at the time .include is interpreted is completely unrelated.
226 That's why we have to save it all. */
232 const char * out_string;
233 char out_buf[sizeof (out_buf)];
236 size_t saved_input_len;
240 const char * mri_state;
242 #if defined TC_ARM && defined OBJ_ELF
243 const char * symver_state;
250 struct app_save *saved;
252 saved = XNEW (struct app_save);
253 saved->state = state;
254 saved->old_state = old_state;
255 saved->out_string = out_string;
256 memcpy (saved->out_buf, out_buf, sizeof (out_buf));
257 saved->add_newlines = add_newlines;
258 if (saved_input == NULL)
259 saved->saved_input = NULL;
262 saved->saved_input = XNEWVEC (char, saved_input_len);
263 memcpy (saved->saved_input, saved_input, saved_input_len);
264 saved->saved_input_len = saved_input_len;
267 saved->scrub_m68k_mri = scrub_m68k_mri;
269 saved->mri_state = mri_state;
270 saved->mri_last_ch = mri_last_ch;
271 #if defined TC_ARM && defined OBJ_ELF
272 saved->symver_state = symver_state;
275 /* do_scrub_begin() is not useful, just wastes time. */
281 return (char *) saved;
287 struct app_save *saved = (struct app_save *) arg;
289 /* There is no do_scrub_end (). */
290 state = saved->state;
291 old_state = saved->old_state;
292 out_string = saved->out_string;
293 memcpy (out_buf, saved->out_buf, sizeof (out_buf));
294 add_newlines = saved->add_newlines;
295 if (saved->saved_input == NULL)
299 gas_assert (saved->saved_input_len <= sizeof (input_buffer));
300 memcpy (input_buffer, saved->saved_input, saved->saved_input_len);
301 saved_input = input_buffer;
302 saved_input_len = saved->saved_input_len;
303 free (saved->saved_input);
306 scrub_m68k_mri = saved->scrub_m68k_mri;
308 mri_state = saved->mri_state;
309 mri_last_ch = saved->mri_last_ch;
310 #if defined TC_ARM && defined OBJ_ELF
311 symver_state = saved->symver_state;
317 /* @@ This assumes that \n &c are the same on host and target. This is not
321 process_escape (int ch)
344 /* This function is called to process input characters. The GET
345 parameter is used to retrieve more input characters. GET should
346 set its parameter to point to a buffer, and return the length of
347 the buffer; it should return 0 at end of file. The scrubbed output
348 characters are put into the buffer starting at TOSTART; the TOSTART
349 buffer is TOLEN bytes in length. The function returns the number
350 of scrubbed characters put into TOSTART. This will be TOLEN unless
351 end of file was seen. This function is arranged as a state
352 machine, and saves its state so that it may return at any point.
353 This is the way the old code used to work. */
356 do_scrub_chars (size_t (*get) (char *, size_t), char *tostart, size_t tolen)
359 char *toend = tostart + tolen;
364 /* Character that started the string we're working on. */
365 static char quotechar;
367 /*State 0: beginning of normal line
368 1: After first whitespace on line (flush more white)
369 2: After first non-white (opcode) on line (keep 1white)
370 3: after second white on line (into operands) (flush white)
371 4: after putting out a .linefile, put out digits
372 5: parsing a string, then go to old-state
373 6: putting out \ escape in a "d string.
376 9: After seeing symbol char in state 3 (keep 1white after symchar)
377 10: After seeing whitespace in state 9 (keep white before symchar)
378 11: After seeing a symbol character in state 0 (eg a label definition)
379 -1: output string in out_string and go to the state in old_state
380 -2: flush text until a '*' '/' is seen, then go to state old_state
382 12: After seeing a dash, looking for a second dash as a start
385 #ifdef DOUBLEBAR_PARALLEL
386 13: After seeing a vertical bar, looking for a second
387 vertical bar as a parallel expression separator.
389 #ifdef TC_PREDICATE_START_CHAR
390 14: After seeing a predicate start character at state 0, looking
391 for a predicate end character as predicate.
392 15: After seeing a predicate start character at state 1, looking
393 for a predicate end character as predicate.
396 16: After seeing an 'a' or an 'A' at the start of a symbol
397 17: After seeing an 'f' or an 'F' in state 16
401 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
402 constructs like ``.loc 1 20''. This was turning into ``.loc
403 120''. States 9 and 10 ensure that a space is never dropped in
404 between characters which could appear in an identifier. Ian
405 Taylor, ian@cygnus.com.
407 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
408 correctly on the PA (and any other target where colons are optional).
409 Jeff Law, law@cs.utah.edu.
411 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
412 get squashed into "cmp r1,r2||trap#1", with the all important space
413 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
415 /* This macro gets the next input character. */
419 ? * (unsigned char *) (from++) \
420 : (saved_input = NULL, \
421 fromlen = (*get) (input_buffer, sizeof input_buffer), \
422 from = input_buffer, \
423 fromend = from + fromlen, \
426 : * (unsigned char *) (from++))))
428 /* This macro pushes a character back on the input stream. */
430 #define UNGET(uch) (*--from = (uch))
432 /* This macro puts a character into the output buffer. If this
433 character fills the output buffer, this macro jumps to the label
434 TOFULL. We use this rather ugly approach because we need to
435 handle two different termination conditions: EOF on the input
436 stream, and a full output buffer. It would be simpler if we
437 always read in the entire input stream before processing it, but
438 I don't want to make such a significant change to the assembler's
450 if (saved_input != NULL)
453 fromend = from + saved_input_len;
457 fromlen = (*get) (input_buffer, sizeof input_buffer);
461 fromend = from + fromlen;
466 /* The cases in this switch end with continue, in order to
467 branch back to the top of this while loop and generate the
468 next output character in the appropriate state. */
473 if (*out_string == '\0')
490 as_warn (_("end of file in comment"));
499 while ((ch = GET ()) == '*')
504 as_warn (_("end of file in comment"));
522 else if (ch >= '0' && ch <= '9')
526 while (ch != EOF && IS_WHITESPACE (ch))
537 while (ch != EOF && ch != '\n')
546 /* We are going to copy everything up to a quote character,
547 with special handling for a backslash. We try to
548 optimize the copying in the simple case without using the
549 GET and PUT macros. */
554 for (s = from; s < fromend; s++)
563 if (len > toend - to)
567 memcpy (to, from, len);
578 /* This buffer is here specifically so
579 that the UNGET below will work. */
580 static char one_char_buf[1];
582 as_warn (_("end of file in string; '%c' inserted"), quotechar);
584 from = fromend = one_char_buf + 1;
589 else if (ch == quotechar)
594 #ifndef NO_STRING_ESCAPES
601 else if (scrub_m68k_mri && ch == '\n')
603 /* Just quietly terminate the string. This permits lines like
604 bne label loop if we haven't reach end yet. */
620 /* Handle strings broken across lines, by turning '\n' into
629 as_warn (_("end of file in string; '%c' inserted"), quotechar);
654 #ifdef ONLY_STANDARD_ESCAPES
655 as_warn (_("unknown escape '\\%c' in string; ignored"), ch);
662 #ifdef DOUBLEBAR_PARALLEL
668 /* Reset back to state 1 and pretend that we are parsing a
669 line from just after the first white space. */
673 /* "||^" is used for SPMASKed instructions. */
686 /* We have seen an 'a' at the start of a symbol, look for an 'f'. */
688 if (ch == 'f' || ch == 'F')
700 /* We have seen "af" at the start of a symbol,
701 a ' here is a part of that symbol. */
705 /* Change to avoid warning about unclosed string. */
713 /* OK, we are somewhere in states 0 through 4 or 9 through 11. */
718 #ifdef TC_PREDICATE_START_CHAR
719 if (ch == TC_PREDICATE_START_CHAR && (state == 0 || state == 1))
725 else if (state == 14 || state == 15)
727 if (ch == TC_PREDICATE_END_CHAR)
743 #if defined TC_ARM && defined OBJ_ELF
744 /* We need to watch out for .symver directives. See the comment later
746 if (symver_state == NULL)
748 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
749 symver_state = symver_pseudo + 1;
753 /* We advance to the next state if we find the right
755 if (ch != '\0' && (*symver_state == ch))
757 else if (*symver_state != '\0')
758 /* We did not get the expected character, or we didn't
759 get a valid terminating character after seeing the
760 entire pseudo-op, so we must go back to the beginning. */
764 /* We've read the entire pseudo-op. If this is the end
765 of the line, go back to the beginning. */
770 #endif /* TC_ARM && OBJ_ELF */
773 /* We want to have pseudo-ops which control whether we are in
774 MRI mode or not. Unfortunately, since m68k MRI mode affects
775 the scrubber, that means that we need a special purpose
777 if (mri_state == NULL)
779 if ((state == 0 || state == 1)
780 && ch == mri_pseudo[0])
781 mri_state = mri_pseudo + 1;
785 /* We advance to the next state if we find the right
786 character, or if we need a space character and we get any
787 whitespace character, or if we need a '0' and we get a
788 '1' (this is so that we only need one state to handle
789 ``.mri 0'' and ``.mri 1''). */
792 || (*mri_state == ' '
793 && lex[ch] == LEX_IS_WHITESPACE)
794 || (*mri_state == '0'
800 else if (*mri_state != '\0'
801 || (lex[ch] != LEX_IS_WHITESPACE
802 && lex[ch] != LEX_IS_NEWLINE))
804 /* We did not get the expected character, or we didn't
805 get a valid terminating character after seeing the
806 entire pseudo-op, so we must go back to the
812 /* We've read the entire pseudo-op. mips_last_ch is
813 either '0' or '1' indicating whether to enter or
815 do_scrub_begin (mri_last_ch == '1');
818 /* We continue handling the character as usual. The
819 main gas reader must also handle the .mri pseudo-op
820 to control expression parsing and the like. */
829 as_warn (_("end of file not at end of a line; newline inserted"));
838 case LEX_IS_WHITESPACE:
843 while (ch != EOF && IS_WHITESPACE (ch));
849 /* Preserve a single whitespace character at the
850 beginning of a line. */
857 #ifdef KEEP_WHITE_AROUND_COLON
858 if (lex[ch] == LEX_IS_COLON)
860 /* Only keep this white if there's no white *after* the
865 if (!IS_WHITESPACE (ch2))
876 || IS_LINE_SEPARATOR (ch)
877 || IS_PARALLEL_SEPARATOR (ch))
881 /* In MRI mode, we keep these spaces. */
889 /* If we're in state 2 or 11, we've seen a non-white
890 character followed by whitespace. If the next character
891 is ':', this is whitespace after a label name which we
892 normally must ignore. In MRI mode, though, spaces are
893 not permitted between the label and the colon. */
894 if ((state == 2 || state == 11)
895 && lex[ch] == LEX_IS_COLON
906 /* We can arrive here if we leave a leading whitespace
907 character at the beginning of a line. */
913 /* Optimize common case by skipping UNGET/GET. */
914 PUT (' '); /* Sp after opco */
921 #ifndef TC_KEEP_OPERAND_SPACES
922 /* For TI C6X, we keep these spaces as they may separate
923 functional unit specifiers from operands. */
927 /* In MRI mode, we keep these spaces. */
932 goto recycle; /* Sp in operands */
935 #ifndef TC_KEEP_OPERAND_SPACES
939 /* In MRI mode, we keep these spaces. */
945 state = 10; /* Sp after symbol char */
948 if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
952 /* We know that ch is not ':', since we tested that
953 case above. Therefore this is not a label, so it
954 must be the opcode, and we've just seen the
955 whitespace after it. */
959 PUT (' '); /* Sp after label definition. */
966 case LEX_IS_TWOCHAR_COMMENT_1ST:
975 if (ch2 != EOF && IS_NEWLINE (ch2))
978 while (ch2 != EOF && ch2 != '*');
983 if (ch2 == EOF || ch2 == '/')
986 /* This UNGET will ensure that we count newlines
992 as_warn (_("end of file in multiline comment"));
997 #ifdef DOUBLESLASH_LINE_COMMENTS
1004 while (ch != EOF && !IS_NEWLINE (ch));
1006 as_warn ("end of file in comment; newline inserted");
1016 if (state == 9 || state == 10)
1022 case LEX_IS_STRINGQUOTE:
1026 /* Preserve the whitespace in foo "bar". */
1031 /* PUT didn't jump out. We could just break, but we
1032 know what will happen, so optimize a bit. */
1036 else if (state == 9)
1045 case LEX_IS_ONECHAR_QUOTE:
1047 if (state == 9 && enable_h_tick_hex)
1052 as_warn ("'%c found after symbol", c);
1058 /* Preserve the whitespace in foo 'b'. */
1067 as_warn (_("end of file after a one-character quote; \\0 inserted"));
1075 as_warn (_("end of file in escape character"));
1079 ch = process_escape (ch);
1081 sprintf (out_buf, "%d", (int) (unsigned char) ch);
1083 /* None of these 'x constants for us. We want 'x'. */
1084 if ((ch = GET ()) != '\'')
1086 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
1087 as_warn (_("missing close quote; (assumed)"));
1093 if (strlen (out_buf) == 1)
1103 out_string = out_buf;
1104 PUT (*out_string++);
1109 #ifdef KEEP_WHITE_AROUND_COLON
1112 if (state == 9 || state == 10)
1114 else if (state != 3)
1120 case LEX_IS_NEWLINE:
1121 /* Roll out a bunch of newlines from inside comments, etc. */
1129 case LEX_IS_LINE_SEPARATOR:
1134 case LEX_IS_PARALLEL_SEPARATOR:
1140 case LEX_IS_DOUBLEDASH_1ST:
1148 /* Read and skip to end of line. */
1153 while (ch != EOF && ch != '\n');
1156 as_warn (_("end of file in comment; newline inserted"));
1162 #ifdef DOUBLEBAR_PARALLEL
1163 case LEX_IS_DOUBLEBAR_1ST:
1170 /* Handle '||' in two states as invoking PUT twice might
1171 result in the first one jumping out of this loop. We'd
1172 then lose track of the state and one '|' char. */
1177 case LEX_IS_LINE_COMMENT_START:
1178 /* FIXME-someday: The two character comment stuff was badly
1179 thought out. On i386, we want '/' as line comment start
1180 AND we want C style comments. hence this hack. The
1181 whole lexical process should be reworked. xoxorich. */
1191 else if (ch2 != EOF)
1197 if (state == 0 || state == 1) /* Only comment at start of line. */
1207 while (ch != EOF && IS_WHITESPACE (ch));
1211 as_warn (_("end of file in comment; newline inserted"));
1216 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1218 /* Not a cpp line. */
1219 while (ch != EOF && !IS_NEWLINE (ch))
1223 as_warn (_("end of file in comment; newline inserted"));
1226 else /* IS_NEWLINE (ch) */
1228 /* To process non-zero add_newlines. */
1234 /* Looks like `# 123 "filename"' from cpp. */
1239 out_string = "\tlinefile ";
1241 out_string = "\t.linefile ";
1242 PUT (*out_string++);
1247 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1248 Trap is the only short insn that has a first operand that is
1249 neither register nor label.
1250 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1251 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is
1252 already LEX_IS_LINE_COMMENT_START. However, it is the
1253 only character in line_comment_chars for d10v, hence we
1254 can recognize it as such. */
1255 /* An alternative approach would be to reset the state to 1 when
1256 we see '||', '<'- or '->', but that seems to be overkill. */
1260 /* We have a line comment character which is not at the
1261 start of a line. If this is also a normal comment
1262 character, fall through. Otherwise treat it as a default
1264 if (strchr (tc_comment_chars, ch) == NULL
1265 && (! scrub_m68k_mri
1266 || (ch != '!' && ch != '*')))
1269 && (ch == '!' || ch == '*' || ch == '#')
1274 case LEX_IS_COMMENT_START:
1275 #if defined TC_ARM && defined OBJ_ELF
1276 /* On the ARM, `@' is the comment character.
1277 Unfortunately this is also a special character in ELF .symver
1278 directives (and .type, though we deal with those another way).
1279 So we check if this line is such a directive, and treat
1280 the character as default if so. This is a hack. */
1281 if ((symver_state != NULL) && (*symver_state == 0))
1286 /* For the ARM, care is needed not to damage occurrences of \@
1287 by stripping the @ onwards. Yuck. */
1288 if (to > tostart && *(to - 1) == '\\')
1289 /* Do not treat the @ as a start-of-comment. */
1293 #ifdef WARN_COMMENTS
1295 found_comment_file = as_where (&found_comment);
1301 while (ch != EOF && !IS_NEWLINE (ch));
1303 as_warn (_("end of file in comment; newline inserted"));
1310 /* Look for strings like H'[0-9A-Fa-f] and if found, replace
1311 the H' with 0x to make them gas-style hex characters. */
1312 if (enable_h_tick_hex)
1328 case LEX_IS_SYMBOL_COMPONENT:
1331 /* This is a symbol character following another symbol
1332 character, with whitespace in between. We skipped
1333 the whitespace earlier, so output it now. */
1341 /* "af'" is a symbol containing '\''. */
1342 if (state == 3 && (ch == 'a' || ch == 'A'))
1347 if (ch == 'f' || ch == 'F')
1356 if (ch == EOF || !IS_SYMBOL_COMPONENT (ch))
1368 /* This is a common case. Quickly copy CH and all the
1369 following symbol component or normal characters. */
1371 && mri_state == NULL
1372 #if defined TC_ARM && defined OBJ_ELF
1373 && symver_state == NULL
1380 for (s = from; s < fromend; s++)
1384 ch2 = *(unsigned char *) s;
1387 && type != LEX_IS_SYMBOL_COMPONENT)
1392 /* Handle the last character normally, for
1398 if (len > (toend - to) - 1)
1399 len = (toend - to) - 1;
1404 memcpy (to, from, len);
1416 /* Some relatively `normal' character. */
1419 state = 11; /* Now seeing label definition. */
1421 else if (state == 1)
1423 state = 2; /* Ditto. */
1425 else if (state == 9)
1427 if (!IS_SYMBOL_COMPONENT (ch))
1430 else if (state == 10)
1434 /* Special handling for backslash: a backslash may
1435 be the beginning of a formal parameter (of a
1436 macro) following another symbol character, with
1437 whitespace in between. If that is the case, we
1438 output a space before the parameter. Strictly
1439 speaking, correct handling depends upon what the
1440 macro parameter expands into; if the parameter
1441 expands into something which does not start with
1442 an operand character, then we don't want to keep
1443 the space. We don't have enough information to
1444 make the right choice, so here we are making the
1445 choice which is more likely to be correct. */
1446 if (to + 1 >= toend)
1448 /* If we're near the end of the buffer, save the
1449 character for the next time round. Otherwise
1450 we'll lose our state. */
1467 /* We have reached the end of the input. */
1468 return to - tostart;
1471 /* The output buffer is full. Save any input we have not yet
1476 saved_input_len = fromend - from;
1481 return to - tostart;