1 /* This is the Assembler Pre-Processor
2 Copyright (C) 1987-2021 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
;
62 static const char symbol_chars
[] =
63 "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
65 #define LEX_IS_SYMBOL_COMPONENT 1
66 #define LEX_IS_WHITESPACE 2
67 #define LEX_IS_LINE_SEPARATOR 3
68 #define LEX_IS_COMMENT_START 4
69 #define LEX_IS_LINE_COMMENT_START 5
70 #define LEX_IS_TWOCHAR_COMMENT_1ST 6
71 #define LEX_IS_STRINGQUOTE 8
72 #define LEX_IS_COLON 9
73 #define LEX_IS_NEWLINE 10
74 #define LEX_IS_ONECHAR_QUOTE 11
76 #define LEX_IS_DOUBLEDASH_1ST 12
79 #define DOUBLEBAR_PARALLEL
81 #ifdef DOUBLEBAR_PARALLEL
82 #define LEX_IS_DOUBLEBAR_1ST 13
84 #define LEX_IS_PARALLEL_SEPARATOR 14
88 #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
89 #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
90 #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
91 #define IS_PARALLEL_SEPARATOR(c) (lex[c] == LEX_IS_PARALLEL_SEPARATOR)
92 #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
93 #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
94 #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
96 static int process_escape (int);
98 /* FIXME-soon: The entire lexer/parser thingy should be
99 built statically at compile time rather than dynamically
100 each and every time the assembler is run. xoxorich. */
103 do_scrub_begin (int m68k_mri ATTRIBUTE_UNUSED
)
108 lex
[' '] = LEX_IS_WHITESPACE
;
109 lex
['\t'] = LEX_IS_WHITESPACE
;
110 lex
['\r'] = LEX_IS_WHITESPACE
;
111 lex
['\n'] = LEX_IS_NEWLINE
;
112 lex
[':'] = LEX_IS_COLON
;
115 scrub_m68k_mri
= m68k_mri
;
120 lex
['"'] = LEX_IS_STRINGQUOTE
;
122 #if ! defined (TC_HPPA)
123 lex
['\''] = LEX_IS_ONECHAR_QUOTE
;
126 #ifdef SINGLE_QUOTE_STRINGS
127 lex
['\''] = LEX_IS_STRINGQUOTE
;
131 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
132 in state 5 of do_scrub_chars must be changed. */
134 /* Note that these override the previous defaults, e.g. if ';' is a
135 comment char, then it isn't a line separator. */
136 for (p
= symbol_chars
; *p
; ++p
)
137 lex
[(unsigned char) *p
] = LEX_IS_SYMBOL_COMPONENT
;
139 for (c
= 128; c
< 256; ++c
)
140 lex
[c
] = LEX_IS_SYMBOL_COMPONENT
;
142 #ifdef tc_symbol_chars
143 /* This macro permits the processor to specify all characters which
144 may appears in an operand. This will prevent the scrubber from
145 discarding meaningful whitespace in certain cases. The i386
146 backend uses this to support prefixes, which can confuse the
147 scrubber as to whether it is parsing operands or opcodes. */
148 for (p
= tc_symbol_chars
; *p
; ++p
)
149 lex
[(unsigned char) *p
] = LEX_IS_SYMBOL_COMPONENT
;
152 /* The m68k backend wants to be able to change comment_chars. */
153 #ifndef tc_comment_chars
154 #define tc_comment_chars comment_chars
156 for (p
= tc_comment_chars
; *p
; p
++)
157 lex
[(unsigned char) *p
] = LEX_IS_COMMENT_START
;
159 for (p
= line_comment_chars
; *p
; p
++)
160 lex
[(unsigned char) *p
] = LEX_IS_LINE_COMMENT_START
;
162 #ifndef tc_line_separator_chars
163 #define tc_line_separator_chars line_separator_chars
165 for (p
= tc_line_separator_chars
; *p
; p
++)
166 lex
[(unsigned char) *p
] = LEX_IS_LINE_SEPARATOR
;
168 #ifdef tc_parallel_separator_chars
169 /* This macro permits the processor to specify all characters which
170 separate parallel insns on the same line. */
171 for (p
= tc_parallel_separator_chars
; *p
; p
++)
172 lex
[(unsigned char) *p
] = LEX_IS_PARALLEL_SEPARATOR
;
175 /* Only allow slash-star comments if slash is not in use.
176 FIXME: This isn't right. We should always permit them. */
178 lex
['/'] = LEX_IS_TWOCHAR_COMMENT_1ST
;
183 lex
['\''] = LEX_IS_STRINGQUOTE
;
184 lex
[';'] = LEX_IS_COMMENT_START
;
185 lex
['*'] = LEX_IS_LINE_COMMENT_START
;
186 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
187 then it can't be used in an expression. */
188 lex
['!'] = LEX_IS_LINE_COMMENT_START
;
193 lex
['-'] = LEX_IS_DOUBLEDASH_1ST
;
195 #ifdef DOUBLEBAR_PARALLEL
196 lex
['|'] = LEX_IS_DOUBLEBAR_1ST
;
199 /* Must do this is we want VLIW instruction with "->" or "<-". */
200 lex
['-'] = LEX_IS_SYMBOL_COMPONENT
;
204 if (enable_h_tick_hex
)
212 /* Saved state of the scrubber. */
214 static int old_state
;
215 static const char *out_string
;
216 static char out_buf
[20];
217 static int add_newlines
;
218 static char *saved_input
;
219 static size_t saved_input_len
;
220 static char input_buffer
[32 * 1024];
221 static const char *mri_state
;
222 static char mri_last_ch
;
224 /* Data structure for saving the state of app across #include's. Note that
225 app is called asynchronously to the parsing of the .include's, so our
226 state at the time .include is interpreted is completely unrelated.
227 That's why we have to save it all. */
233 const char * out_string
;
234 char out_buf
[sizeof (out_buf
)];
237 size_t saved_input_len
;
241 const char * mri_state
;
243 #if defined TC_ARM && defined OBJ_ELF
244 const char * symver_state
;
252 struct app_save
*saved
;
254 saved
= XNEW (struct app_save
);
255 saved
->state
= state
;
256 saved
->old_state
= old_state
;
257 saved
->out_string
= out_string
;
258 memcpy (saved
->out_buf
, out_buf
, sizeof (out_buf
));
259 saved
->add_newlines
= add_newlines
;
260 if (saved_input
== NULL
)
261 saved
->saved_input
= NULL
;
264 saved
->saved_input
= XNEWVEC (char, saved_input_len
);
265 memcpy (saved
->saved_input
, saved_input
, saved_input_len
);
266 saved
->saved_input_len
= saved_input_len
;
269 saved
->scrub_m68k_mri
= scrub_m68k_mri
;
271 saved
->mri_state
= mri_state
;
272 saved
->mri_last_ch
= mri_last_ch
;
273 #if defined TC_ARM && defined OBJ_ELF
274 saved
->symver_state
= symver_state
;
276 saved
->last_char
= last_char
;
278 /* do_scrub_begin() is not useful, just wastes time. */
284 return (char *) saved
;
290 struct app_save
*saved
= (struct app_save
*) arg
;
292 /* There is no do_scrub_end (). */
293 state
= saved
->state
;
294 old_state
= saved
->old_state
;
295 out_string
= saved
->out_string
;
296 memcpy (out_buf
, saved
->out_buf
, sizeof (out_buf
));
297 add_newlines
= saved
->add_newlines
;
298 if (saved
->saved_input
== NULL
)
302 gas_assert (saved
->saved_input_len
<= sizeof (input_buffer
));
303 memcpy (input_buffer
, saved
->saved_input
, saved
->saved_input_len
);
304 saved_input
= input_buffer
;
305 saved_input_len
= saved
->saved_input_len
;
306 free (saved
->saved_input
);
309 scrub_m68k_mri
= saved
->scrub_m68k_mri
;
311 mri_state
= saved
->mri_state
;
312 mri_last_ch
= saved
->mri_last_ch
;
313 #if defined TC_ARM && defined OBJ_ELF
314 symver_state
= saved
->symver_state
;
316 last_char
= saved
->last_char
;
321 /* @@ This assumes that \n &c are the same on host and target. This is not
325 process_escape (int ch
)
348 /* This function is called to process input characters. The GET
349 parameter is used to retrieve more input characters. GET should
350 set its parameter to point to a buffer, and return the length of
351 the buffer; it should return 0 at end of file. The scrubbed output
352 characters are put into the buffer starting at TOSTART; the TOSTART
353 buffer is TOLEN bytes in length. The function returns the number
354 of scrubbed characters put into TOSTART. This will be TOLEN unless
355 end of file was seen. This function is arranged as a state
356 machine, and saves its state so that it may return at any point.
357 This is the way the old code used to work. */
360 do_scrub_chars (size_t (*get
) (char *, size_t), char *tostart
, size_t tolen
)
363 char *toend
= tostart
+ tolen
;
368 /* Character that started the string we're working on. */
369 static char quotechar
;
371 /*State 0: beginning of normal line
372 1: After first whitespace on line (flush more white)
373 2: After first non-white (opcode) on line (keep 1white)
374 3: after second white on line (into operands) (flush white)
375 4: after putting out a .linefile, put out digits
376 5: parsing a string, then go to old-state
377 6: putting out \ escape in a "d string.
380 9: After seeing symbol char in state 3 (keep 1white after symchar)
381 10: After seeing whitespace in state 9 (keep white before symchar)
382 11: After seeing a symbol character in state 0 (eg a label definition)
383 -1: output string in out_string and go to the state in old_state
384 -2: flush text until a '*' '/' is seen, then go to state old_state
386 12: After seeing a dash, looking for a second dash as a start
389 #ifdef DOUBLEBAR_PARALLEL
390 13: After seeing a vertical bar, looking for a second
391 vertical bar as a parallel expression separator.
393 #ifdef TC_PREDICATE_START_CHAR
394 14: After seeing a predicate start character at state 0, looking
395 for a predicate end character as predicate.
396 15: After seeing a predicate start character at state 1, looking
397 for a predicate end character as predicate.
400 16: After seeing an 'a' or an 'A' at the start of a symbol
401 17: After seeing an 'f' or an 'F' in state 16
405 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
406 constructs like ``.loc 1 20''. This was turning into ``.loc
407 120''. States 9 and 10 ensure that a space is never dropped in
408 between characters which could appear in an identifier. Ian
409 Taylor, ian@cygnus.com.
411 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
412 correctly on the PA (and any other target where colons are optional).
413 Jeff Law, law@cs.utah.edu.
415 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
416 get squashed into "cmp r1,r2||trap#1", with the all important space
417 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
419 /* This macro gets the next input character. */
423 ? * (unsigned char *) (from++) \
424 : (saved_input = NULL, \
425 fromlen = (*get) (input_buffer, sizeof input_buffer), \
426 from = input_buffer, \
427 fromend = from + fromlen, \
430 : * (unsigned char *) (from++))))
432 /* This macro pushes a character back on the input stream. */
434 #define UNGET(uch) (*--from = (uch))
436 /* This macro puts a character into the output buffer. If this
437 character fills the output buffer, this macro jumps to the label
438 TOFULL. We use this rather ugly approach because we need to
439 handle two different termination conditions: EOF on the input
440 stream, and a full output buffer. It would be simpler if we
441 always read in the entire input stream before processing it, but
442 I don't want to make such a significant change to the assembler's
454 if (saved_input
!= NULL
)
457 fromend
= from
+ saved_input_len
;
461 fromlen
= (*get
) (input_buffer
, sizeof input_buffer
);
465 fromend
= from
+ fromlen
;
470 /* The cases in this switch end with continue, in order to
471 branch back to the top of this while loop and generate the
472 next output character in the appropriate state. */
477 if (*out_string
== '\0')
494 as_warn (_("end of file in comment"));
503 while ((ch
= GET ()) == '*')
508 as_warn (_("end of file in comment"));
526 else if (ch
>= '0' && ch
<= '9')
530 while (ch
!= EOF
&& IS_WHITESPACE (ch
))
541 while (ch
!= EOF
&& ch
!= '\n')
550 /* We are going to copy everything up to a quote character,
551 with special handling for a backslash. We try to
552 optimize the copying in the simple case without using the
553 GET and PUT macros. */
558 for (s
= from
; s
< fromend
; s
++)
567 if (len
> toend
- to
)
571 memcpy (to
, from
, len
);
582 /* This buffer is here specifically so
583 that the UNGET below will work. */
584 static char one_char_buf
[1];
586 as_warn (_("end of file in string; '%c' inserted"), quotechar
);
588 from
= fromend
= one_char_buf
+ 1;
593 else if (ch
== quotechar
)
598 else if (TC_STRING_ESCAPES
&& ch
== '\\')
603 else if (scrub_m68k_mri
&& ch
== '\n')
605 /* Just quietly terminate the string. This permits lines like
606 bne label loop if we haven't reach end yet. */
622 /* Handle strings broken across lines, by turning '\n' into
631 as_warn (_("end of file in string; '%c' inserted"), quotechar
);
656 #ifdef ONLY_STANDARD_ESCAPES
657 as_warn (_("unknown escape '\\%c' in string; ignored"), ch
);
664 #ifdef DOUBLEBAR_PARALLEL
670 /* Reset back to state 1 and pretend that we are parsing a
671 line from just after the first white space. */
675 /* "||^" is used for SPMASKed instructions. */
688 /* We have seen an 'a' at the start of a symbol, look for an 'f'. */
690 if (ch
== 'f' || ch
== 'F')
702 /* We have seen "af" at the start of a symbol,
703 a ' here is a part of that symbol. */
707 /* Change to avoid warning about unclosed string. */
715 /* OK, we are somewhere in states 0 through 4 or 9 through 11. */
720 #ifdef TC_PREDICATE_START_CHAR
721 if (ch
== TC_PREDICATE_START_CHAR
&& (state
== 0 || state
== 1))
727 else if (state
== 14 || state
== 15)
729 if (ch
== TC_PREDICATE_END_CHAR
)
745 #if defined TC_ARM && defined OBJ_ELF
746 /* We need to watch out for .symver directives. See the comment later
748 if (symver_state
== NULL
)
750 if ((state
== 0 || state
== 1) && ch
== symver_pseudo
[0])
751 symver_state
= symver_pseudo
+ 1;
755 /* We advance to the next state if we find the right
757 if (ch
!= '\0' && (*symver_state
== ch
))
759 else if (*symver_state
!= '\0')
760 /* We did not get the expected character, or we didn't
761 get a valid terminating character after seeing the
762 entire pseudo-op, so we must go back to the beginning. */
766 /* We've read the entire pseudo-op. If this is the end
767 of the line, go back to the beginning. */
772 #endif /* TC_ARM && OBJ_ELF */
775 /* We want to have pseudo-ops which control whether we are in
776 MRI mode or not. Unfortunately, since m68k MRI mode affects
777 the scrubber, that means that we need a special purpose
779 if (mri_state
== NULL
)
781 if ((state
== 0 || state
== 1)
782 && ch
== mri_pseudo
[0])
783 mri_state
= mri_pseudo
+ 1;
787 /* We advance to the next state if we find the right
788 character, or if we need a space character and we get any
789 whitespace character, or if we need a '0' and we get a
790 '1' (this is so that we only need one state to handle
791 ``.mri 0'' and ``.mri 1''). */
794 || (*mri_state
== ' '
795 && lex
[ch
] == LEX_IS_WHITESPACE
)
796 || (*mri_state
== '0'
802 else if (*mri_state
!= '\0'
803 || (lex
[ch
] != LEX_IS_WHITESPACE
804 && lex
[ch
] != LEX_IS_NEWLINE
))
806 /* We did not get the expected character, or we didn't
807 get a valid terminating character after seeing the
808 entire pseudo-op, so we must go back to the
814 /* We've read the entire pseudo-op. mips_last_ch is
815 either '0' or '1' indicating whether to enter or
817 do_scrub_begin (mri_last_ch
== '1');
820 /* We continue handling the character as usual. The
821 main gas reader must also handle the .mri pseudo-op
822 to control expression parsing and the like. */
831 as_warn (_("end of file not at end of a line; newline inserted"));
840 case LEX_IS_WHITESPACE
:
845 while (ch
!= EOF
&& IS_WHITESPACE (ch
));
851 /* Preserve a single whitespace character at the
852 beginning of a line. */
859 #ifdef KEEP_WHITE_AROUND_COLON
860 if (lex
[ch
] == LEX_IS_COLON
)
862 /* Only keep this white if there's no white *after* the
867 if (!IS_WHITESPACE (ch2
))
878 || IS_LINE_SEPARATOR (ch
)
879 || IS_PARALLEL_SEPARATOR (ch
))
883 /* In MRI mode, we keep these spaces. */
891 /* If we're in state 2 or 11, we've seen a non-white
892 character followed by whitespace. If the next character
893 is ':', this is whitespace after a label name which we
894 normally must ignore. In MRI mode, though, spaces are
895 not permitted between the label and the colon. */
896 if ((state
== 2 || state
== 11)
897 && lex
[ch
] == LEX_IS_COLON
908 /* We can arrive here if we leave a leading whitespace
909 character at the beginning of a line. */
915 /* Optimize common case by skipping UNGET/GET. */
916 PUT (' '); /* Sp after opco */
923 #ifndef TC_KEEP_OPERAND_SPACES
924 /* For TI C6X, we keep these spaces as they may separate
925 functional unit specifiers from operands. */
929 /* In MRI mode, we keep these spaces. */
934 goto recycle
; /* Sp in operands */
937 #ifndef TC_KEEP_OPERAND_SPACES
941 /* In MRI mode, we keep these spaces. */
947 state
= 10; /* Sp after symbol char */
950 if (LABELS_WITHOUT_COLONS
|| flag_m68k_mri
)
954 /* We know that ch is not ':', since we tested that
955 case above. Therefore this is not a label, so it
956 must be the opcode, and we've just seen the
957 whitespace after it. */
961 PUT (' '); /* Sp after label definition. */
968 case LEX_IS_TWOCHAR_COMMENT_1ST
:
977 if (ch2
!= EOF
&& IS_NEWLINE (ch2
))
980 while (ch2
!= EOF
&& ch2
!= '*');
985 if (ch2
== EOF
|| ch2
== '/')
988 /* This UNGET will ensure that we count newlines
994 as_warn (_("end of file in multiline comment"));
999 #ifdef DOUBLESLASH_LINE_COMMENTS
1000 else if (ch2
== '/')
1006 while (ch
!= EOF
&& !IS_NEWLINE (ch
));
1008 as_warn ("end of file in comment; newline inserted");
1018 if (state
== 9 || state
== 10)
1024 case LEX_IS_STRINGQUOTE
:
1028 /* Preserve the whitespace in foo "bar". */
1033 /* PUT didn't jump out. We could just break, but we
1034 know what will happen, so optimize a bit. */
1038 else if (state
== 9)
1046 case LEX_IS_ONECHAR_QUOTE
:
1048 if (state
== 9 && enable_h_tick_hex
)
1053 as_warn ("'%c found after symbol", c
);
1059 /* Preserve the whitespace in foo 'b'. */
1068 as_warn (_("end of file after a one-character quote; \\0 inserted"));
1076 as_warn (_("end of file in escape character"));
1080 ch
= process_escape (ch
);
1082 sprintf (out_buf
, "%d", (int) (unsigned char) ch
);
1084 /* None of these 'x constants for us. We want 'x'. */
1085 if ((ch
= GET ()) != '\'')
1087 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
1088 as_warn (_("missing close quote; (assumed)"));
1094 if (strlen (out_buf
) == 1)
1104 out_string
= out_buf
;
1105 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))
1285 /* Care is needed not to damage occurrences of \<comment-char>
1286 by stripping the <comment-char> onwards. Yuck. */
1287 if ((to
> tostart
? to
[-1] : last_char
) == '\\')
1288 /* Do not treat the <comment-char> as a start-of-comment. */
1291 #ifdef WARN_COMMENTS
1293 found_comment_file
= as_where (&found_comment
);
1299 while (ch
!= EOF
&& !IS_NEWLINE (ch
));
1301 as_warn (_("end of file in comment; newline inserted"));
1308 /* Look for strings like H'[0-9A-Fa-f] and if found, replace
1309 the H' with 0x to make them gas-style hex characters. */
1310 if (enable_h_tick_hex
)
1326 case LEX_IS_SYMBOL_COMPONENT
:
1329 /* This is a symbol character following another symbol
1330 character, with whitespace in between. We skipped
1331 the whitespace earlier, so output it now. */
1339 /* "af'" is a symbol containing '\''. */
1340 if (state
== 3 && (ch
== 'a' || ch
== 'A'))
1345 if (ch
== 'f' || ch
== 'F')
1354 if (ch
== EOF
|| !IS_SYMBOL_COMPONENT (ch
))
1366 /* This is a common case. Quickly copy CH and all the
1367 following symbol component or normal characters. */
1369 && mri_state
== NULL
1370 #if defined TC_ARM && defined OBJ_ELF
1371 && symver_state
== NULL
1378 for (s
= from
; s
< fromend
; s
++)
1382 ch2
= *(unsigned char *) s
;
1385 && type
!= LEX_IS_SYMBOL_COMPONENT
)
1390 /* Handle the last character normally, for
1396 if (len
> (toend
- to
) - 1)
1397 len
= (toend
- to
) - 1;
1402 memcpy (to
, from
, len
);
1414 /* Some relatively `normal' character. */
1417 state
= 11; /* Now seeing label definition. */
1419 else if (state
== 1)
1421 state
= 2; /* Ditto. */
1423 else if (state
== 9)
1425 if (!IS_SYMBOL_COMPONENT (ch
))
1428 else if (state
== 10)
1432 /* Special handling for backslash: a backslash may
1433 be the beginning of a formal parameter (of a
1434 macro) following another symbol character, with
1435 whitespace in between. If that is the case, we
1436 output a space before the parameter. Strictly
1437 speaking, correct handling depends upon what the
1438 macro parameter expands into; if the parameter
1439 expands into something which does not start with
1440 an operand character, then we don't want to keep
1441 the space. We don't have enough information to
1442 make the right choice, so here we are making the
1443 choice which is more likely to be correct. */
1444 if (to
+ 1 >= toend
)
1446 /* If we're near the end of the buffer, save the
1447 character for the next time round. Otherwise
1448 we'll lose our state. */
1465 /* 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
;
1483 return to
- tostart
;