1 /* This is the Assembler Pre-Processor
2 Copyright 1987, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
4 Free Software Foundation, Inc.
6 This file is part of GAS, the GNU Assembler.
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
23 /* Modified by Allen Wirfs-Brock, Instantiations Inc 2/90 */
24 /* App, the assembler pre-processor. This pre-processor strips out excess
25 spaces, turns single-quoted characters into a decimal constant, and turns
26 # <number> <filename> <garbage> into a .line <number>\n.file <filename>
27 pair. This needs better error-handling. */
30 #include "as.h" /* For BAD_CASE() only */
34 #define const /* empty */
39 /* Whether we are scrubbing in m68k MRI mode. This is different from
40 flag_m68k_mri, because the two flags will be affected by the .mri
41 pseudo-op at different times. */
42 static int scrub_m68k_mri;
44 /* The pseudo-op which switches in and out of MRI mode. See the
45 comment in do_scrub_chars. */
46 static const char mri_pseudo[] = ".mri 0";
48 #define scrub_m68k_mri 0
51 #if defined TC_ARM && defined OBJ_ELF
52 /* The pseudo-op for which we need to special-case `@' characters.
53 See the comment in do_scrub_chars. */
54 static const char symver_pseudo[] = ".symver";
55 static const char * symver_state;
59 static const char symbol_chars[] =
60 "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
62 #define LEX_IS_SYMBOL_COMPONENT 1
63 #define LEX_IS_WHITESPACE 2
64 #define LEX_IS_LINE_SEPARATOR 3
65 #define LEX_IS_COMMENT_START 4
66 #define LEX_IS_LINE_COMMENT_START 5
67 #define LEX_IS_TWOCHAR_COMMENT_1ST 6
68 #define LEX_IS_STRINGQUOTE 8
69 #define LEX_IS_COLON 9
70 #define LEX_IS_NEWLINE 10
71 #define LEX_IS_ONECHAR_QUOTE 11
73 #define LEX_IS_DOUBLEDASH_1ST 12
76 #define DOUBLEBAR_PARALLEL
78 #ifdef DOUBLEBAR_PARALLEL
79 #define LEX_IS_DOUBLEBAR_1ST 13
81 #define LEX_IS_PARALLEL_SEPARATOR 14
82 #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
83 #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
84 #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
85 #define IS_PARALLEL_SEPARATOR(c) (lex[c] == LEX_IS_PARALLEL_SEPARATOR)
86 #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
87 #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
88 #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
90 static int process_escape PARAMS ((int));
92 /* FIXME-soon: The entire lexer/parser thingy should be
93 built statically at compile time rather than dynamically
94 each and every time the assembler is run. xoxorich. */
97 do_scrub_begin (m68k_mri)
98 int m68k_mri ATTRIBUTE_UNUSED;
103 lex[' '] = LEX_IS_WHITESPACE;
104 lex['\t'] = LEX_IS_WHITESPACE;
105 lex['\r'] = LEX_IS_WHITESPACE;
106 lex['\n'] = LEX_IS_NEWLINE;
107 lex[':'] = LEX_IS_COLON;
110 scrub_m68k_mri = m68k_mri;
115 lex['"'] = LEX_IS_STRINGQUOTE;
117 #if ! defined (TC_HPPA) && ! defined (TC_I370)
118 /* I370 uses single-quotes to delimit integer, float constants */
119 lex['\''] = LEX_IS_ONECHAR_QUOTE;
122 #ifdef SINGLE_QUOTE_STRINGS
123 lex['\''] = LEX_IS_STRINGQUOTE;
127 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
128 in state 5 of do_scrub_chars must be changed. */
130 /* Note that these override the previous defaults, e.g. if ';' is a
131 comment char, then it isn't a line separator. */
132 for (p = symbol_chars; *p; ++p)
134 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
135 } /* declare symbol characters */
137 for (c = 128; c < 256; ++c)
138 lex[c] = LEX_IS_SYMBOL_COMPONENT;
140 #ifdef tc_symbol_chars
141 /* This macro permits the processor to specify all characters which
142 may appears in an operand. This will prevent the scrubber from
143 discarding meaningful whitespace in certain cases. The i386
144 backend uses this to support prefixes, which can confuse the
145 scrubber as to whether it is parsing operands or opcodes. */
146 for (p = tc_symbol_chars; *p; ++p)
147 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
150 /* The m68k backend wants to be able to change comment_chars. */
151 #ifndef tc_comment_chars
152 #define tc_comment_chars comment_chars
154 for (p = tc_comment_chars; *p; p++)
156 lex[(unsigned char) *p] = LEX_IS_COMMENT_START;
157 } /* declare comment chars */
159 for (p = line_comment_chars; *p; p++)
161 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START;
162 } /* declare line comment chars */
164 for (p = line_separator_chars; *p; p++)
166 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
167 } /* declare line separators */
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++)
174 lex[(unsigned char) *p] = LEX_IS_PARALLEL_SEPARATOR;
175 } /* declare parallel separators */
178 /* Only allow slash-star comments if slash is not in use.
179 FIXME: This isn't right. We should always permit them. */
182 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
188 lex['\''] = LEX_IS_STRINGQUOTE;
189 lex[';'] = LEX_IS_COMMENT_START;
190 lex['*'] = LEX_IS_LINE_COMMENT_START;
191 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
192 then it can't be used in an expression. */
193 lex['!'] = LEX_IS_LINE_COMMENT_START;
198 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
200 #ifdef DOUBLEBAR_PARALLEL
201 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
204 /* must do this is we want VLIW instruction with "->" or "<-" */
205 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
207 } /* do_scrub_begin() */
209 /* Saved state of the scrubber */
211 static int old_state;
212 static char *out_string;
213 static char out_buf[20];
214 static int add_newlines;
215 static char *saved_input;
216 static int saved_input_len;
217 static char input_buffer[32 * 1024];
218 static const char *mri_state;
219 static char mri_last_ch;
221 /* Data structure for saving the state of app across #include's. Note that
222 app is called asynchronously to the parsing of the .include's, so our
223 state at the time .include is interpreted is completely unrelated.
224 That's why we have to save it all. */
230 char out_buf[sizeof (out_buf)];
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 = 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. */
277 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 assert (saved->saved_input_len <= (int) (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
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 (get, tostart, tolen)
354 int (*get) PARAMS ((char *, int));
359 char *toend = tostart + tolen;
363 register int ch, ch2 = 0;
365 /*State 0: beginning of normal line
366 1: After first whitespace on line (flush more white)
367 2: After first non-white (opcode) on line (keep 1white)
368 3: after second white on line (into operands) (flush white)
369 4: after putting out a .line, put out digits
370 5: parsing a string, then go to old-state
371 6: putting out \ escape in a "d string.
372 7: After putting out a .appfile, put out string.
373 8: After putting out a .appfile string, flush until newline.
374 9: After seeing symbol char in state 3 (keep 1white after symchar)
375 10: After seeing whitespace in state 9 (keep white before symchar)
376 11: After seeing a symbol character in state 0 (eg a label definition)
377 -1: output string in out_string and go to the state in old_state
378 -2: flush text until a '*' '/' is seen, then go to state old_state
380 12: After seeing a dash, looking for a second dash as a start of comment.
382 #ifdef DOUBLEBAR_PARALLEL
383 13: After seeing a vertical bar, looking for a second vertical bar as a parallel expression seperator.
387 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
388 constructs like ``.loc 1 20''. This was turning into ``.loc
389 120''. States 9 and 10 ensure that a space is never dropped in
390 between characters which could appear in an identifier. Ian
391 Taylor, ian@cygnus.com.
393 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
394 correctly on the PA (and any other target where colons are optional).
395 Jeff Law, law@cs.utah.edu.
397 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
398 get squashed into "cmp r1,r2||trap#1", with the all important space
399 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
401 /* This macro gets the next input character. */
405 ? * (unsigned char *) (from++) \
406 : (saved_input = NULL, \
407 fromlen = (*get) (input_buffer, sizeof input_buffer), \
408 from = input_buffer, \
409 fromend = from + fromlen, \
412 : * (unsigned char *) (from++))))
414 /* This macro pushes a character back on the input stream. */
416 #define UNGET(uch) (*--from = (uch))
418 /* This macro puts a character into the output buffer. If this
419 character fills the output buffer, this macro jumps to the label
420 TOFULL. We use this rather ugly approach because we need to
421 handle two different termination conditions: EOF on the input
422 stream, and a full output buffer. It would be simpler if we
423 always read in the entire input stream before processing it, but
424 I don't want to make such a significant change to the assembler's
436 if (saved_input != NULL)
439 fromend = from + saved_input_len;
443 fromlen = (*get) (input_buffer, sizeof input_buffer);
447 fromend = from + fromlen;
452 /* The cases in this switch end with continue, in order to
453 branch back to the top of this while loop and generate the
454 next output character in the appropriate state. */
459 if (*out_string == '\0')
476 as_warn (_("end of file in comment"));
485 while ((ch = GET ()) == '*')
490 as_warn (_("end of file in comment"));
508 else if (ch >= '0' && ch <= '9')
512 while (ch != EOF && IS_WHITESPACE (ch))
518 out_string = "\n\tappfile ";
520 out_string = "\n\t.appfile ";
527 while (ch != EOF && ch != '\n')
536 /* We are going to copy everything up to a quote character,
537 with special handling for a backslash. We try to
538 optimize the copying in the simple case without using the
539 GET and PUT macros. */
544 for (s = from; s < fromend; s++)
547 /* This condition must be changed if the type of any
548 other character can be LEX_IS_STRINGQUOTE. */
556 if (len > toend - to)
560 memcpy (to, from, len);
569 as_warn (_("end of file in string; inserted '\"'"));
574 else if (lex[ch] == LEX_IS_STRINGQUOTE)
579 #ifndef NO_STRING_ESCAPES
586 else if (scrub_m68k_mri && ch == '\n')
588 /* Just quietly terminate the string. This permits lines like
589 bne label loop if we haven't reach end yet
606 /* Handle strings broken across lines, by turning '\n' into
615 as_warn (_("end of file in string; '\"' inserted"));
640 #ifdef ONLY_STANDARD_ESCAPES
641 as_warn (_("unknown escape '\\%c' in string; ignored"), ch);
660 while (ch != '\n' && ch != EOF);
668 /* OK, we are somewhere in states 0 through 4 or 9 through 11 */
675 #if defined TC_ARM && defined OBJ_ELF
676 /* We need to watch out for .symver directives. See the comment later
678 if (symver_state == NULL)
680 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
681 symver_state = symver_pseudo + 1;
685 /* We advance to the next state if we find the right
687 if (ch != '\0' && (*symver_state == ch))
689 else if (*symver_state != '\0')
690 /* We did not get the expected character, or we didn't
691 get a valid terminating character after seeing the
692 entire pseudo-op, so we must go back to the beginning. */
696 /* We've read the entire pseudo-op. If this is the end
697 of the line, go back to the beginning. */
702 #endif /* TC_ARM && OBJ_ELF */
705 /* We want to have pseudo-ops which control whether we are in
706 MRI mode or not. Unfortunately, since m68k MRI mode affects
707 the scrubber, that means that we need a special purpose
709 if (mri_state == NULL)
711 if ((state == 0 || state == 1)
712 && ch == mri_pseudo[0])
713 mri_state = mri_pseudo + 1;
717 /* We advance to the next state if we find the right
718 character, or if we need a space character and we get any
719 whitespace character, or if we need a '0' and we get a
720 '1' (this is so that we only need one state to handle
721 ``.mri 0'' and ``.mri 1''). */
724 || (*mri_state == ' '
725 && lex[ch] == LEX_IS_WHITESPACE)
726 || (*mri_state == '0'
732 else if (*mri_state != '\0'
733 || (lex[ch] != LEX_IS_WHITESPACE
734 && lex[ch] != LEX_IS_NEWLINE))
736 /* We did not get the expected character, or we didn't
737 get a valid terminating character after seeing the
738 entire pseudo-op, so we must go back to the
744 /* We've read the entire pseudo-op. mips_last_ch is
745 either '0' or '1' indicating whether to enter or
747 do_scrub_begin (mri_last_ch == '1');
750 /* We continue handling the character as usual. The
751 main gas reader must also handle the .mri pseudo-op
752 to control expression parsing and the like. */
761 as_warn (_("end of file not at end of a line; newline inserted"));
770 case LEX_IS_WHITESPACE:
775 while (ch != EOF && IS_WHITESPACE (ch));
781 /* Preserve a single whitespace character at the
782 beginning of a line. */
789 #ifdef KEEP_WHITE_AROUND_COLON
790 if (lex[ch] == LEX_IS_COLON)
792 /* Only keep this white if there's no white *after* the
796 if (!IS_WHITESPACE (ch2))
807 || IS_LINE_SEPARATOR (ch)
808 || IS_PARALLEL_SEPARATOR (ch))
812 /* In MRI mode, we keep these spaces. */
820 /* If we're in state 2 or 11, we've seen a non-white
821 character followed by whitespace. If the next character
822 is ':', this is whitespace after a label name which we
823 normally must ignore. In MRI mode, though, spaces are
824 not permitted between the label and the colon. */
825 if ((state == 2 || state == 11)
826 && lex[ch] == LEX_IS_COLON
838 goto recycle; /* Punted leading sp */
840 /* We can arrive here if we leave a leading whitespace
841 character at the beginning of a line. */
847 /* Optimize common case by skipping UNGET/GET. */
848 PUT (' '); /* Sp after opco */
857 /* In MRI mode, we keep these spaces. */
862 goto recycle; /* Sp in operands */
867 /* In MRI mode, we keep these spaces. */
873 state = 10; /* Sp after symbol char */
876 if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
880 /* We know that ch is not ':', since we tested that
881 case above. Therefore this is not a label, so it
882 must be the opcode, and we've just seen the
883 whitespace after it. */
887 PUT (' '); /* Sp after label definition. */
894 case LEX_IS_TWOCHAR_COMMENT_1ST:
903 if (ch2 != EOF && IS_NEWLINE (ch2))
906 while (ch2 != EOF && ch2 != '*');
911 if (ch2 == EOF || ch2 == '/')
914 /* This UNGET will ensure that we count newlines
920 as_warn (_("end of file in multiline comment"));
925 #ifdef DOUBLESLASH_LINE_COMMENTS
932 while (ch != EOF && !IS_NEWLINE (ch));
934 as_warn ("end of file in comment; newline inserted");
944 if (state == 9 || state == 10)
950 case LEX_IS_STRINGQUOTE:
953 /* Preserve the whitespace in foo "bar" */
958 /* PUT didn't jump out. We could just break, but we
959 know what will happen, so optimize a bit. */
972 case LEX_IS_ONECHAR_QUOTE:
975 /* Preserve the whitespace in foo 'b' */
984 as_warn (_("end of file after a one-character quote; \\0 inserted"));
992 as_warn (_("end of file in escape character"));
996 ch = process_escape (ch);
998 sprintf (out_buf, "%d", (int) (unsigned char) ch);
1000 /* None of these 'x constants for us. We want 'x'. */
1001 if ((ch = GET ()) != '\'')
1003 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
1004 as_warn (_("missing close quote; (assumed)"));
1010 if (strlen (out_buf) == 1)
1020 out_string = out_buf;
1021 PUT (*out_string++);
1026 #ifdef KEEP_WHITE_AROUND_COLON
1029 if (state == 9 || state == 10)
1031 else if (state != 3)
1037 case LEX_IS_NEWLINE:
1038 /* Roll out a bunch of newlines from inside comments, etc. */
1046 case LEX_IS_LINE_SEPARATOR:
1051 case LEX_IS_PARALLEL_SEPARATOR:
1057 case LEX_IS_DOUBLEDASH_1ST:
1064 /* Read and skip to end of line. */
1069 while (ch != EOF && ch != '\n');
1072 as_warn (_("end of file in comment; newline inserted"));
1078 #ifdef DOUBLEBAR_PARALLEL
1079 case LEX_IS_DOUBLEBAR_1ST:
1086 /* Reset back to state 1 and pretend that we are parsing a line from
1087 just after the first white space. */
1093 case LEX_IS_LINE_COMMENT_START:
1094 /* FIXME-someday: The two character comment stuff was badly
1095 thought out. On i386, we want '/' as line comment start
1096 AND we want C style comments. hence this hack. The
1097 whole lexical process should be reworked. xoxorich. */
1113 if (state == 0 || state == 1) /* Only comment at start of line. */
1123 while (ch != EOF && IS_WHITESPACE (ch));
1126 as_warn (_("end of file in comment; newline inserted"));
1130 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1132 /* Not a cpp line. */
1133 while (ch != EOF && !IS_NEWLINE (ch))
1136 as_warn (_("end of file in comment; newline inserted"));
1141 /* Looks like `# 123 "filename"' from cpp. */
1146 out_string = "\tappline ";
1148 out_string = "\t.appline ";
1149 PUT (*out_string++);
1154 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1155 Trap is the only short insn that has a first operand that is
1156 neither register nor label.
1157 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1158 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is
1159 already LEX_IS_LINE_COMMENT_START. However, it is the
1160 only character in line_comment_chars for d10v, hence we
1161 can recognize it as such. */
1162 /* An alternative approach would be to reset the state to 1 when
1163 we see '||', '<'- or '->', but that seems to be overkill. */
1167 /* We have a line comment character which is not at the
1168 start of a line. If this is also a normal comment
1169 character, fall through. Otherwise treat it as a default
1171 if (strchr (tc_comment_chars, ch) == NULL
1172 && (! scrub_m68k_mri
1173 || (ch != '!' && ch != '*')))
1176 && (ch == '!' || ch == '*' || ch == '#')
1181 case LEX_IS_COMMENT_START:
1182 #if defined TC_ARM && defined OBJ_ELF
1183 /* On the ARM, `@' is the comment character.
1184 Unfortunately this is also a special character in ELF .symver
1185 directives (and .type, though we deal with those another way).
1186 So we check if this line is such a directive, and treat
1187 the character as default if so. This is a hack. */
1188 if ((symver_state != NULL) && (*symver_state == 0))
1191 #ifdef WARN_COMMENTS
1193 as_where (&found_comment_file, &found_comment);
1199 while (ch != EOF && !IS_NEWLINE (ch));
1201 as_warn (_("end of file in comment; newline inserted"));
1206 case LEX_IS_SYMBOL_COMPONENT:
1209 /* This is a symbol character following another symbol
1210 character, with whitespace in between. We skipped
1211 the whitespace earlier, so output it now. */
1221 /* This is a common case. Quickly copy CH and all the
1222 following symbol component or normal characters. */
1224 && mri_state == NULL
1225 #if defined TC_ARM && defined OBJ_ELF
1226 && symver_state == NULL
1233 for (s = from; s < fromend; s++)
1237 ch2 = *(unsigned char *) s;
1240 && type != LEX_IS_SYMBOL_COMPONENT)
1245 /* Handle the last character normally, for
1250 if (len > (toend - to) - 1)
1251 len = (toend - to) - 1;
1257 memcpy (to, from, len);
1265 case 8: *to++ = *from++;
1266 case 7: *to++ = *from++;
1267 case 6: *to++ = *from++;
1268 case 5: *to++ = *from++;
1269 case 4: *to++ = *from++;
1270 case 3: *to++ = *from++;
1271 case 2: *to++ = *from++;
1272 case 1: *to++ = *from++;
1282 /* Some relatively `normal' character. */
1285 state = 11; /* Now seeing label definition */
1287 else if (state == 1)
1289 state = 2; /* Ditto */
1291 else if (state == 9)
1293 if (lex[ch] != LEX_IS_SYMBOL_COMPONENT)
1296 else if (state == 10)
1300 /* Special handling for backslash: a backslash may
1301 be the beginning of a formal parameter (of a
1302 macro) following another symbol character, with
1303 whitespace in between. If that is the case, we
1304 output a space before the parameter. Strictly
1305 speaking, correct handling depends upon what the
1306 macro parameter expands into; if the parameter
1307 expands into something which does not start with
1308 an operand character, then we don't want to keep
1309 the space. We don't have enough information to
1310 make the right choice, so here we are making the
1311 choice which is more likely to be correct. */
1325 /* We have reached the end of the input. */
1326 return to - tostart;
1329 /* The output buffer is full. Save any input we have not yet
1334 saved_input_len = fromend - from;
1339 return to - tostart;