1 /* This is the Assembler Pre-Processor
2 Copyright (C) 1987, 90, 91, 92, 93, 94, 95, 96, 97, 98, 1999
3 Free Software Foundation, Inc.
5 This file is part of GAS, the GNU Assembler.
7 GAS is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 GAS is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GAS; see the file COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
22 /* Modified by Allen Wirfs-Brock, Instantiations Inc 2/90 */
23 /* App, the assembler pre-processor. This pre-processor strips out excess
24 spaces, turns single-quoted characters into a decimal constant, and turns
25 # <number> <filename> <garbage> into a .line <number>\n.file <filename>
26 pair. This needs better error-handling. */
29 #include "as.h" /* For BAD_CASE() only */
33 #define const /* empty */
37 /* Whether we are scrubbing in m68k MRI mode. This is different from
38 flag_m68k_mri, because the two flags will be affected by the .mri
39 pseudo-op at different times. */
40 static int scrub_m68k_mri;
42 /* The pseudo-op which switches in and out of MRI mode. See the
43 comment in do_scrub_chars. */
44 static const char mri_pseudo[] = ".mri 0";
46 #if defined TC_ARM && defined OBJ_ELF
47 /* The pseudo-op for which we need to special-case `@' characters.
48 See the comment in do_scrub_chars. */
49 static const char symver_pseudo[] = ".symver";
50 static const char * symver_state;
54 static const char symbol_chars[] =
55 "$._ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
57 #define LEX_IS_SYMBOL_COMPONENT 1
58 #define LEX_IS_WHITESPACE 2
59 #define LEX_IS_LINE_SEPARATOR 3
60 #define LEX_IS_COMMENT_START 4
61 #define LEX_IS_LINE_COMMENT_START 5
62 #define LEX_IS_TWOCHAR_COMMENT_1ST 6
63 #define LEX_IS_STRINGQUOTE 8
64 #define LEX_IS_COLON 9
65 #define LEX_IS_NEWLINE 10
66 #define LEX_IS_ONECHAR_QUOTE 11
68 #define LEX_IS_DOUBLEDASH_1ST 12
71 #define DOUBLEBAR_PARALLEL
73 #ifdef DOUBLEBAR_PARALLEL
74 #define LEX_IS_DOUBLEBAR_1ST 13
76 #define IS_SYMBOL_COMPONENT(c) (lex[c] == LEX_IS_SYMBOL_COMPONENT)
77 #define IS_WHITESPACE(c) (lex[c] == LEX_IS_WHITESPACE)
78 #define IS_LINE_SEPARATOR(c) (lex[c] == LEX_IS_LINE_SEPARATOR)
79 #define IS_COMMENT(c) (lex[c] == LEX_IS_COMMENT_START)
80 #define IS_LINE_COMMENT(c) (lex[c] == LEX_IS_LINE_COMMENT_START)
81 #define IS_NEWLINE(c) (lex[c] == LEX_IS_NEWLINE)
83 static int process_escape PARAMS ((int));
85 /* FIXME-soon: The entire lexer/parser thingy should be
86 built statically at compile time rather than dynamically
87 each and every time the assembler is run. xoxorich. */
90 do_scrub_begin (m68k_mri)
96 scrub_m68k_mri = m68k_mri;
98 lex[' '] = LEX_IS_WHITESPACE;
99 lex['\t'] = LEX_IS_WHITESPACE;
100 lex['\r'] = LEX_IS_WHITESPACE;
101 lex['\n'] = LEX_IS_NEWLINE;
102 lex[';'] = LEX_IS_LINE_SEPARATOR;
103 lex[':'] = LEX_IS_COLON;
107 lex['"'] = LEX_IS_STRINGQUOTE;
109 #if ! defined (TC_HPPA) && ! defined (TC_I370)
110 /* I370 uses single-quotes to delimit integer, float constants */
111 lex['\''] = LEX_IS_ONECHAR_QUOTE;
114 #ifdef SINGLE_QUOTE_STRINGS
115 lex['\''] = LEX_IS_STRINGQUOTE;
119 /* Note: if any other character can be LEX_IS_STRINGQUOTE, the loop
120 in state 5 of do_scrub_chars must be changed. */
122 /* Note that these override the previous defaults, e.g. if ';' is a
123 comment char, then it isn't a line separator. */
124 for (p = symbol_chars; *p; ++p)
126 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
127 } /* declare symbol characters */
129 for (c = 128; c < 256; ++c)
130 lex[c] = LEX_IS_SYMBOL_COMPONENT;
132 #ifdef tc_symbol_chars
133 /* This macro permits the processor to specify all characters which
134 may appears in an operand. This will prevent the scrubber from
135 discarding meaningful whitespace in certain cases. The i386
136 backend uses this to support prefixes, which can confuse the
137 scrubber as to whether it is parsing operands or opcodes. */
138 for (p = tc_symbol_chars; *p; ++p)
139 lex[(unsigned char) *p] = LEX_IS_SYMBOL_COMPONENT;
142 /* The m68k backend wants to be able to change comment_chars. */
143 #ifndef tc_comment_chars
144 #define tc_comment_chars comment_chars
146 for (p = tc_comment_chars; *p; p++)
148 lex[(unsigned char) *p] = LEX_IS_COMMENT_START;
149 } /* declare comment chars */
151 for (p = line_comment_chars; *p; p++)
153 lex[(unsigned char) *p] = LEX_IS_LINE_COMMENT_START;
154 } /* declare line comment chars */
156 for (p = line_separator_chars; *p; p++)
158 lex[(unsigned char) *p] = LEX_IS_LINE_SEPARATOR;
159 } /* declare line separators */
161 /* Only allow slash-star comments if slash is not in use.
162 FIXME: This isn't right. We should always permit them. */
165 lex['/'] = LEX_IS_TWOCHAR_COMMENT_1ST;
170 lex['\''] = LEX_IS_STRINGQUOTE;
171 lex[';'] = LEX_IS_COMMENT_START;
172 lex['*'] = LEX_IS_LINE_COMMENT_START;
173 /* The MRI documentation says '!' is LEX_IS_COMMENT_START, but
174 then it can't be used in an expression. */
175 lex['!'] = LEX_IS_LINE_COMMENT_START;
179 lex['-'] = LEX_IS_DOUBLEDASH_1ST;
181 #ifdef DOUBLEBAR_PARALLEL
182 lex['|'] = LEX_IS_DOUBLEBAR_1ST;
185 /* must do this is we want VLIW instruction with "->" or "<-" */
186 lex['-'] = LEX_IS_SYMBOL_COMPONENT;
188 } /* do_scrub_begin() */
190 /* Saved state of the scrubber */
192 static int old_state;
193 static char *out_string;
194 static char out_buf[20];
195 static int add_newlines;
196 static char *saved_input;
197 static int saved_input_len;
198 static char input_buffer[32 * 1024];
199 static const char *mri_state;
200 static char mri_last_ch;
202 /* Data structure for saving the state of app across #include's. Note that
203 app is called asynchronously to the parsing of the .include's, so our
204 state at the time .include is interpreted is completely unrelated.
205 That's why we have to save it all. */
212 char out_buf[sizeof (out_buf)];
217 const char * mri_state;
219 #if defined TC_ARM && defined OBJ_ELF
220 const char * symver_state;
227 register struct app_save *saved;
229 saved = (struct app_save *) xmalloc (sizeof (*saved));
230 saved->state = state;
231 saved->old_state = old_state;
232 saved->out_string = out_string;
233 memcpy (saved->out_buf, out_buf, sizeof (out_buf));
234 saved->add_newlines = add_newlines;
235 if (saved_input == NULL)
236 saved->saved_input = NULL;
239 saved->saved_input = xmalloc (saved_input_len);
240 memcpy (saved->saved_input, saved_input, saved_input_len);
241 saved->saved_input_len = saved_input_len;
243 saved->scrub_m68k_mri = scrub_m68k_mri;
244 saved->mri_state = mri_state;
245 saved->mri_last_ch = mri_last_ch;
246 #if defined TC_ARM && defined OBJ_ELF
247 saved->symver_state = symver_state;
250 /* do_scrub_begin() is not useful, just wastes time. */
255 return (char *) saved;
262 register struct app_save *saved = (struct app_save *) arg;
264 /* There is no do_scrub_end (). */
265 state = saved->state;
266 old_state = saved->old_state;
267 out_string = saved->out_string;
268 memcpy (out_buf, saved->out_buf, sizeof (out_buf));
269 add_newlines = saved->add_newlines;
270 if (saved->saved_input == NULL)
274 assert (saved->saved_input_len <= (int) (sizeof input_buffer));
275 memcpy (input_buffer, saved->saved_input, saved->saved_input_len);
276 saved_input = input_buffer;
277 saved_input_len = saved->saved_input_len;
278 free (saved->saved_input);
280 scrub_m68k_mri = saved->scrub_m68k_mri;
281 mri_state = saved->mri_state;
282 mri_last_ch = saved->mri_last_ch;
283 #if defined TC_ARM && defined OBJ_ELF
284 symver_state = saved->symver_state;
290 /* @@ This assumes that \n &c are the same on host and target. This is not
317 /* This function is called to process input characters. The GET
318 parameter is used to retrieve more input characters. GET should
319 set its parameter to point to a buffer, and return the length of
320 the buffer; it should return 0 at end of file. The scrubbed output
321 characters are put into the buffer starting at TOSTART; the TOSTART
322 buffer is TOLEN bytes in length. The function returns the number
323 of scrubbed characters put into TOSTART. This will be TOLEN unless
324 end of file was seen. This function is arranged as a state
325 machine, and saves its state so that it may return at any point.
326 This is the way the old code used to work. */
329 do_scrub_chars (get, tostart, tolen)
330 int (*get) PARAMS ((char *, int));
335 char *toend = tostart + tolen;
339 register int ch, ch2 = 0;
341 /*State 0: beginning of normal line
342 1: After first whitespace on line (flush more white)
343 2: After first non-white (opcode) on line (keep 1white)
344 3: after second white on line (into operands) (flush white)
345 4: after putting out a .line, put out digits
346 5: parsing a string, then go to old-state
347 6: putting out \ escape in a "d string.
348 7: After putting out a .appfile, put out string.
349 8: After putting out a .appfile string, flush until newline.
350 9: After seeing symbol char in state 3 (keep 1white after symchar)
351 10: After seeing whitespace in state 9 (keep white before symchar)
352 11: After seeing a symbol character in state 0 (eg a label definition)
353 -1: output string in out_string and go to the state in old_state
354 -2: flush text until a '*' '/' is seen, then go to state old_state
356 12: After seeing a dash, looking for a second dash as a start of comment.
358 #ifdef DOUBLEBAR_PARALLEL
359 13: After seeing a vertical bar, looking for a second vertical bar as a parallel expression seperator.
363 /* I added states 9 and 10 because the MIPS ECOFF assembler uses
364 constructs like ``.loc 1 20''. This was turning into ``.loc
365 120''. States 9 and 10 ensure that a space is never dropped in
366 between characters which could appear in a identifier. Ian
367 Taylor, ian@cygnus.com.
369 I added state 11 so that something like "Lfoo add %r25,%r26,%r27" works
370 correctly on the PA (and any other target where colons are optional).
371 Jeff Law, law@cs.utah.edu.
373 I added state 13 so that something like "cmp r1, r2 || trap #1" does not
374 get squashed into "cmp r1,r2||trap#1", with the all important space
375 between the 'trap' and the '#1' being eliminated. nickc@cygnus.com */
377 /* This macro gets the next input character. */
381 ? * (unsigned char *) (from++) \
382 : (saved_input = NULL, \
383 fromlen = (*get) (input_buffer, sizeof input_buffer), \
384 from = input_buffer, \
385 fromend = from + fromlen, \
388 : * (unsigned char *) (from++))))
390 /* This macro pushes a character back on the input stream. */
392 #define UNGET(uch) (*--from = (uch))
394 /* This macro puts a character into the output buffer. If this
395 character fills the output buffer, this macro jumps to the label
396 TOFULL. We use this rather ugly approach because we need to
397 handle two different termination conditions: EOF on the input
398 stream, and a full output buffer. It would be simpler if we
399 always read in the entire input stream before processing it, but
400 I don't want to make such a significant change to the assembler's
412 if (saved_input != NULL)
415 fromend = from + saved_input_len;
419 fromlen = (*get) (input_buffer, sizeof input_buffer);
423 fromend = from + fromlen;
428 /* The cases in this switch end with continue, in order to
429 branch back to the top of this while loop and generate the
430 next output character in the appropriate state. */
435 if (*out_string == '\0')
452 as_warn (_("end of file in comment"));
461 while ((ch = GET ()) == '*')
466 as_warn (_("end of file in comment"));
484 else if (ch >= '0' && ch <= '9')
488 while (ch != EOF && IS_WHITESPACE (ch))
494 out_string = "\n\tappfile ";
496 out_string = "\n\t.appfile ";
503 while (ch != EOF && ch != '\n')
512 /* We are going to copy everything up to a quote character,
513 with special handling for a backslash. We try to
514 optimize the copying in the simple case without using the
515 GET and PUT macros. */
520 for (s = from; s < fromend; s++)
523 /* This condition must be changed if the type of any
524 other character can be LEX_IS_STRINGQUOTE. */
532 if (len > toend - to)
536 memcpy (to, from, len);
545 as_warn (_("end of file in string: inserted '\"'"));
550 else if (lex[ch] == LEX_IS_STRINGQUOTE)
555 #ifndef NO_STRING_ESCAPES
562 else if (scrub_m68k_mri && ch == '\n')
564 /* Just quietly terminate the string. This permits lines like
565 bne label loop if we haven't reach end yet
582 /* Handle strings broken across lines, by turning '\n' into
609 #if defined(IGNORE_NONSTANDARD_ESCAPES) | defined(ONLY_STANDARD_ESCAPES)
611 as_warn (_("Unknown escape '\\%c' in string: Ignored"), ch);
613 #else /* ONLY_STANDARD_ESCAPES */
615 /* Accept \x as x for any x */
617 #endif /* ONLY_STANDARD_ESCAPES */
620 as_warn (_("End of file in string: '\"' inserted"));
639 while (ch != '\n' && ch != EOF);
647 /* OK, we are somewhere in states 0 through 4 or 9 through 11 */
654 #if defined TC_ARM && defined OBJ_ELF
655 /* We need to watch out for .symver directives. See the comment later
657 if (symver_state == NULL)
659 if ((state == 0 || state == 1) && ch == symver_pseudo[0])
660 symver_state = symver_pseudo + 1;
664 /* We advance to the next state if we find the right
666 if (ch != '\0' && (*symver_state == ch))
668 else if (*symver_state != '\0')
669 /* We did not get the expected character, or we didn't
670 get a valid terminating character after seeing the
671 entire pseudo-op, so we must go back to the beginning. */
675 /* We've read the entire pseudo-op. If this is the end
676 of the line, go back to the beginning. */
681 #endif /* TC_ARM && OBJ_ELF */
684 /* We want to have pseudo-ops which control whether we are in
685 MRI mode or not. Unfortunately, since m68k MRI mode affects
686 the scrubber, that means that we need a special purpose
688 if (mri_state == NULL)
690 if ((state == 0 || state == 1)
691 && ch == mri_pseudo[0])
692 mri_state = mri_pseudo + 1;
696 /* We advance to the next state if we find the right
697 character, or if we need a space character and we get any
698 whitespace character, or if we need a '0' and we get a
699 '1' (this is so that we only need one state to handle
700 ``.mri 0'' and ``.mri 1''). */
703 || (*mri_state == ' '
704 && lex[ch] == LEX_IS_WHITESPACE)
705 || (*mri_state == '0'
711 else if (*mri_state != '\0'
712 || (lex[ch] != LEX_IS_WHITESPACE
713 && lex[ch] != LEX_IS_NEWLINE))
715 /* We did not get the expected character, or we didn't
716 get a valid terminating character after seeing the
717 entire pseudo-op, so we must go back to the
723 /* We've read the entire pseudo-op. mips_last_ch is
724 either '0' or '1' indicating whether to enter or
726 do_scrub_begin (mri_last_ch == '1');
729 /* We continue handling the character as usual. The
730 main gas reader must also handle the .mri pseudo-op
731 to control expression parsing and the like. */
740 as_warn (_("end of file not at end of a line; newline inserted"));
749 case LEX_IS_WHITESPACE:
754 while (ch != EOF && IS_WHITESPACE (ch));
760 /* Preserve a single whitespace character at the
761 beginning of a line. */
768 #ifdef KEEP_WHITE_AROUND_COLON
769 if (lex[ch] == LEX_IS_COLON)
771 /* only keep this white if there's no white *after* the colon */
774 if (!IS_WHITESPACE (ch2))
785 || IS_LINE_SEPARATOR (ch))
789 /* In MRI mode, we keep these spaces. */
797 /* If we're in state 2 or 11, we've seen a non-white
798 character followed by whitespace. If the next character
799 is ':', this is whitespace after a label name which we
800 normally must ignore. In MRI mode, though, spaces are
801 not permitted between the label and the colon. */
802 if ((state == 2 || state == 11)
803 && lex[ch] == LEX_IS_COLON
815 goto recycle; /* Punted leading sp */
817 /* We can arrive here if we leave a leading whitespace
818 character at the beginning of a line. */
824 /* Optimize common case by skipping UNGET/GET. */
825 PUT (' '); /* Sp after opco */
834 /* In MRI mode, we keep these spaces. */
839 goto recycle; /* Sp in operands */
844 /* In MRI mode, we keep these spaces. */
850 state = 10; /* Sp after symbol char */
854 #ifdef LABELS_WITHOUT_COLONS
861 /* We know that ch is not ':', since we tested that
862 case above. Therefore this is not a label, so it
863 must be the opcode, and we've just seen the
864 whitespace after it. */
868 PUT (' '); /* Sp after label definition. */
875 case LEX_IS_TWOCHAR_COMMENT_1ST:
884 if (ch2 != EOF && IS_NEWLINE (ch2))
887 while (ch2 != EOF && ch2 != '*');
892 if (ch2 == EOF || ch2 == '/')
895 /* This UNGET will ensure that we count newlines
901 as_warn (_("end of file in multiline comment"));
910 if (state == 9 || state == 10)
916 case LEX_IS_STRINGQUOTE:
919 /* Preserve the whitespace in foo "bar" */
924 /* PUT didn't jump out. We could just break, but we
925 know what will happen, so optimize a bit. */
938 case LEX_IS_ONECHAR_QUOTE:
941 /* Preserve the whitespace in foo 'b' */
950 as_warn (_("end of file after a one-character quote; \\0 inserted"));
958 as_warn (_("end of file in escape character"));
962 ch = process_escape (ch);
964 sprintf (out_buf, "%d", (int) (unsigned char) ch);
966 /* None of these 'x constants for us. We want 'x'. */
967 if ((ch = GET ()) != '\'')
969 #ifdef REQUIRE_CHAR_CLOSE_QUOTE
970 as_warn (_("Missing close quote: (assumed)"));
976 if (strlen (out_buf) == 1)
986 out_string = out_buf;
992 #ifdef KEEP_WHITE_AROUND_COLON
995 if (state == 9 || state == 10)
1003 case LEX_IS_NEWLINE:
1004 /* Roll out a bunch of newlines from inside comments, etc. */
1010 /* fall thru into... */
1012 case LEX_IS_LINE_SEPARATOR:
1018 case LEX_IS_DOUBLEDASH_1ST:
1025 /* read and skip to end of line */
1030 while (ch != EOF && ch != '\n');
1033 as_warn (_("end of file in comment; newline inserted"));
1039 #ifdef DOUBLEBAR_PARALLEL
1040 case LEX_IS_DOUBLEBAR_1ST:
1047 /* Reset back to state 1 and pretend that we are parsing a line from
1048 just after the first white space. */
1054 case LEX_IS_LINE_COMMENT_START:
1055 /* FIXME-someday: The two character comment stuff was badly
1056 thought out. On i386, we want '/' as line comment start
1057 AND we want C style comments. hence this hack. The
1058 whole lexical process should be reworked. xoxorich. */
1074 if (state == 0 || state == 1) /* Only comment at start of line. */
1084 while (ch != EOF && IS_WHITESPACE (ch));
1087 as_warn (_("end of file in comment; newline inserted"));
1091 if (ch < '0' || ch > '9' || state != 0 || startch != '#')
1093 /* Not a cpp line. */
1094 while (ch != EOF && !IS_NEWLINE (ch))
1097 as_warn (_("EOF in Comment: Newline inserted"));
1102 /* Loks like `# 123 "filename"' from cpp. */
1107 out_string = "\tappline ";
1109 out_string = "\t.appline ";
1110 PUT (*out_string++);
1115 /* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
1116 Trap is the only short insn that has a first operand that is
1117 neither register nor label.
1118 We must prevent exef0f ||trap #1 to degenerate to exef0f ||trap#1 .
1119 We can't make '#' LEX_IS_SYMBOL_COMPONENT because it is already
1120 LEX_IS_LINE_COMMENT_START. However, it is the only character in
1121 line_comment_chars for d10v, hence we can recognize it as such. */
1122 /* An alternative approach would be to reset the state to 1 when
1123 we see '||', '<'- or '->', but that seems to be overkill. */
1124 if (state == 10) PUT (' ');
1126 /* We have a line comment character which is not at the
1127 start of a line. If this is also a normal comment
1128 character, fall through. Otherwise treat it as a default
1130 if (strchr (tc_comment_chars, ch) == NULL
1131 && (! scrub_m68k_mri
1132 || (ch != '!' && ch != '*')))
1135 && (ch == '!' || ch == '*' || ch == '#')
1140 case LEX_IS_COMMENT_START:
1141 #if defined TC_ARM && defined OBJ_ELF
1142 /* On the ARM, `@' is the comment character.
1143 Unfortunately this is also a special character in ELF .symver
1144 directives (and .type, though we deal with those another way). So
1145 we check if this line is such a directive, and treat the character
1146 as default if so. This is a hack. */
1147 if ((symver_state != NULL) && (*symver_state == 0))
1154 while (ch != EOF && !IS_NEWLINE (ch));
1156 as_warn (_("end of file in comment; newline inserted"));
1161 case LEX_IS_SYMBOL_COMPONENT:
1164 /* This is a symbol character following another symbol
1165 character, with whitespace in between. We skipped
1166 the whitespace earlier, so output it now. */
1176 /* This is a common case. Quickly copy CH and all the
1177 following symbol component or normal characters. */
1179 && mri_state == NULL
1180 #if defined TC_ARM && defined OBJ_ELF
1181 && symver_state == NULL
1188 for (s = from; s < fromend; s++)
1192 ch2 = * (unsigned char *) s;
1195 && type != LEX_IS_SYMBOL_COMPONENT)
1200 /* Handle the last character normally, for
1205 if (len > (toend - to) - 1)
1206 len = (toend - to) - 1;
1212 memcpy (to, from, len);
1220 case 8: *to++ = *from++;
1221 case 7: *to++ = *from++;
1222 case 6: *to++ = *from++;
1223 case 5: *to++ = *from++;
1224 case 4: *to++ = *from++;
1225 case 3: *to++ = *from++;
1226 case 2: *to++ = *from++;
1227 case 1: *to++ = *from++;
1237 /* Some relatively `normal' character. */
1240 state = 11; /* Now seeing label definition */
1242 else if (state == 1)
1244 state = 2; /* Ditto */
1246 else if (state == 9)
1248 if (lex[ch] != LEX_IS_SYMBOL_COMPONENT)
1251 else if (state == 10)
1255 /* Special handling for backslash: a backslash may
1256 be the beginning of a formal parameter (of a
1257 macro) following another symbol character, with
1258 whitespace in between. If that is the case, we
1259 output a space before the parameter. Strictly
1260 speaking, correct handling depends upon what the
1261 macro parameter expands into; if the parameter
1262 expands into something which does not start with
1263 an operand character, then we don't want to keep
1264 the space. We don't have enough information to
1265 make the right choice, so here we are making the
1266 choice which is more likely to be correct. */
1280 /* We have reached the end of the input. */
1281 return to - tostart;
1284 /* The output buffer is full. Save any input we have not yet
1289 saved_input_len = fromend - from;
1294 return to - tostart;