cfgrtl.c (commit_one_edge_insertion): Fix warning.
[official-gcc.git] / gcc / tradcpp.c
blobdbc4334bcbe5b4ceb3cea002c3f125bd888605ab
1 /* C Compatible Compiler Preprocessor (CCCP)
2 Copyright (C) 1986, 1987, 1989, 2000, 2001 Free Software Foundation, Inc.
3 Written by Paul Rubin, June 1986
4 Adapted to ANSI C, Richard Stallman, Jan 1987
5 Dusted off, polished, and adapted for use as traditional
6 preprocessor only, Zack Weinberg, Jul 2000
8 This program is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by the
10 Free Software Foundation; either version 2, or (at your option) any
11 later version.
13 This program 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 this program; if not, write to the Free Software
20 Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
22 #include "config.h"
23 #include "system.h"
24 #include "version.h"
25 #include "cppdefault.h"
26 #include "tradcpp.h"
27 #include "mkdeps.h"
28 #include "intl.h"
30 typedef unsigned char U_CHAR;
32 /* Name under which this program was invoked. */
34 static const char *progname;
36 /* Current maximum length of directory names in the search path
37 for include files. (Altered as we get more of them.) */
39 size_t max_include_len;
41 /* Nonzero means copy comments into the output file. */
43 int put_out_comments = 0;
45 /* mkdeps.h opaque structure that encapsulates dependency information. */
46 struct deps *deps;
48 /* Nonzero means print the names of included files rather than
49 the preprocessed output. 1 means just the #include "...",
50 2 means #include <...> as well. */
52 int print_deps = 0;
54 /* Nonzero means print dummy targets for each header file. */
56 int print_deps_phony_targets = 0;
58 /* If true, fopen (deps_file, "a") else fopen (deps_file, "w"). */
60 int deps_append = 0;
62 /* File name which deps are being written to. This is 0 if deps are
63 being written to stdout. */
65 const char *deps_file = 0;
67 /* Nonzero if missing .h files in -M output are assumed to be
68 generated files and not errors. */
70 int deps_missing_files = 0;
72 /* Nonzero means don't output line number information. */
74 int no_line_commands;
76 /* Nonzero means inhibit output of the preprocessed text
77 and instead output the definitions of all user-defined macros
78 in a form suitable for use as input to cccp. */
80 int dump_macros;
82 /* Nonzero means don't print warning messages. -w. */
84 int inhibit_warnings = 0;
86 /* Non-0 means don't output the preprocessed program. */
87 int inhibit_output = 0;
89 /* Nonzero means chars are signed. */
90 #if DEFAULT_SIGNED_CHAR
91 int flag_signed_char = 1;
92 #else
93 int flag_signed_char = 0;
94 #endif
96 /* Nonzero means warn if slash-star appears in a comment. */
98 int warn_comments;
100 /* Nonzero causes output not to be done,
101 but directives such as #define that have side effects
102 are still obeyed. */
104 int no_output;
106 /* I/O buffer structure.
107 The `fname' field is nonzero for source files and #include files
108 and for the dummy text used for -D and -U.
109 It is zero for rescanning results of macro expansion
110 and for expanding macro arguments. */
111 #define INPUT_STACK_MAX 200
112 struct file_name_list;
113 struct file_buf {
114 const char *fname;
115 int lineno;
116 int length;
117 U_CHAR *buf;
118 U_CHAR *bufp;
119 /* Macro that this level is the expansion of.
120 Included so that we can reenable the macro
121 at the end of this level. */
122 struct hashnode *macro;
123 /* Value of if_stack at start of this file.
124 Used to prohibit unmatched #endif (etc) in an include file. */
125 struct if_stack *if_stack;
126 /* Object to be freed at end of input at this level. */
127 U_CHAR *free_ptr;
128 /* Position to start scanning for #include_next in this file. */
129 struct file_name_list *next_header_dir;
130 } instack[INPUT_STACK_MAX];
132 typedef struct file_buf FILE_BUF;
134 /* Current nesting level of input sources.
135 `instack[indepth]' is the level currently being read. */
136 int indepth = -1;
137 #define CHECK_DEPTH(code) \
138 if (indepth >= (INPUT_STACK_MAX - 1)) \
140 error_with_line (line_for_error (instack[indepth].lineno), \
141 "macro or #include recursion too deep"); \
142 code; \
145 /* Current depth in #include directives that use <...>. */
146 int system_include_depth = 0;
148 /* The output buffer. Its LENGTH field is the amount of room allocated
149 for the buffer, not the number of chars actually present. To get
150 that, subtract outbuf.buf from outbuf.bufp. */
152 #define OUTBUF_SIZE 10 /* initial size of output buffer */
153 FILE_BUF outbuf;
155 /* Grow output buffer OBUF points at
156 so it can hold at least NEEDED more chars. */
158 #define check_expand(OBUF, NEEDED) do { \
159 if ((OBUF)->length - ((OBUF)->bufp - (OBUF)->buf) <= (NEEDED)) \
160 grow_outbuf ((OBUF), (NEEDED)); \
161 } while (0)
163 struct file_name_list
165 struct file_name_list *next;
166 const char *fname;
169 struct file_name_list *include = 0; /* First dir to search */
170 /* First dir to search for <file> */
171 struct file_name_list *first_bracket_include = 0;
172 struct file_name_list *last_include = 0; /* Last in chain */
174 /* List of included files that contained #once. */
175 struct file_name_list *dont_repeat_files = 0;
177 /* List of other included files. */
178 struct file_name_list *all_include_files = 0;
180 /* Structure allocated for every #define. For a simple replacement
181 such as
182 #define foo bar ,
183 nargs = -1, the `pattern' list is null, and the expansion is just
184 the replacement text. Nargs = 0 means a functionlike macro with no args,
185 e.g.,
186 #define getchar() getc (stdin) .
187 When there are args, the expansion is the replacement text with the
188 args squashed out, and the reflist is a list describing how to
189 build the output from the input: e.g., "3 chars, then the 1st arg,
190 then 9 chars, then the 3rd arg, then 0 chars, then the 2nd arg".
191 The chars here come from the expansion. Whatever is left of the
192 expansion after the last arg-occurrence is copied after that arg.
193 Note that the reflist can be arbitrarily long---
194 its length depends on the number of times the arguments appear in
195 the replacement text, not how many args there are. Example:
196 #define f(x) x+x+x+x+x+x+x would have replacement text "++++++" and
197 pattern list
198 { (0, 1), (1, 1), (1, 1), ..., (1, 1), NULL }
199 where (x, y) means (nchars, argno). */
201 typedef struct definition DEFINITION;
202 struct definition {
203 int nargs;
204 int length; /* length of expansion string */
205 U_CHAR *expansion;
206 struct reflist {
207 struct reflist *next;
208 char stringify; /* nonzero if this arg was preceded by a
209 # operator. */
210 char raw_before; /* Nonzero if a ## operator before arg. */
211 char raw_after; /* Nonzero if a ## operator after arg. */
212 int nchars; /* Number of literal chars to copy before
213 this arg occurrence. */
214 int argno; /* Number of arg to substitute (origin-0) */
215 } *pattern;
216 /* Names of macro args, concatenated in reverse order
217 with comma-space between them.
218 The only use of this is that we warn on redefinition
219 if this differs between the old and new definitions. */
220 const U_CHAR *argnames;
223 /* Chained list of answers to an assertion. */
224 struct answer
226 struct answer *next;
227 const unsigned char *answer;
228 size_t len;
231 /* different kinds of things that can appear in the value field
232 of a hash node. Actually, this may be useless now. */
233 union hashval {
234 const char *cpval;
235 DEFINITION *defn;
236 struct answer *answers;
239 /* The structure of a node in the hash table. The hash table
240 has entries for all tokens defined by #define commands (type T_MACRO),
241 plus some special tokens like __LINE__ (these each have their own
242 type, and the appropriate code is run when that type of node is seen.
243 It does not contain control words like "#define", which are recognized
244 by a separate piece of code. */
246 /* different flavors of hash nodes --- also used in keyword table */
247 enum node_type {
248 T_DEFINE = 1, /* `#define' */
249 T_INCLUDE, /* `#include' */
250 T_INCLUDE_NEXT,/* `#include_next' */
251 T_IFDEF, /* `#ifdef' */
252 T_IFNDEF, /* `#ifndef' */
253 T_IF, /* `#if' */
254 T_ELSE, /* `#else' */
255 T_ELIF, /* `#elif' */
256 T_UNDEF, /* `#undef' */
257 T_LINE, /* `#line' */
258 T_ENDIF, /* `#endif' */
259 T_ERROR, /* `#error' */
260 T_WARNING, /* `#warning' */
261 T_ASSERT, /* `#assert' */
262 T_UNASSERT, /* `#unassert' */
263 T_SPECLINE, /* special symbol `__LINE__' */
264 T_DATE, /* `__DATE__' */
265 T_FILE, /* `__FILE__' */
266 T_BASE_FILE, /* `__BASE_FILE__' */
267 T_INCLUDE_LEVEL, /* `__INCLUDE_LEVEL__' */
268 T_VERSION, /* `__VERSION__' */
269 T_TIME, /* `__TIME__' */
270 T_CONST, /* Constant value, used by `__STDC__' */
271 T_MACRO, /* macro defined by `#define' */
272 T_SPEC_DEFINED, /* special `defined' macro for use in #if statements */
273 T_UNUSED /* Used for something not defined. */
276 struct hashnode {
277 struct hashnode *next; /* double links for easy deletion */
278 struct hashnode *prev;
279 struct hashnode **bucket_hdr; /* also, a back pointer to this node's hash
280 chain is kept, in case the node is the head
281 of the chain and gets deleted. */
282 enum node_type type; /* type of special token */
283 int length; /* length of token, for quick comparison */
284 U_CHAR *name; /* the actual name */
285 union hashval value; /* pointer to expansion, or whatever */
288 typedef struct hashnode HASHNODE;
290 static HASHNODE *parse_assertion PARAMS ((const unsigned char *,
291 const unsigned char *,
292 struct answer **, int));
293 static struct answer **find_answer PARAMS ((HASHNODE *,
294 const struct answer *));
295 static int parse_answer PARAMS ((const unsigned char *, const unsigned char *,
296 struct answer **, int));
297 static unsigned char *canonicalize_text PARAMS ((const unsigned char *,
298 const unsigned char *,
299 const unsigned char **));
301 /* Some definitions for the hash table. The hash function MUST be
302 computed as shown in hashf () below. That is because the rescan
303 loop computes the hash value `on the fly' for most tokens,
304 in order to avoid the overhead of a lot of procedure calls to
305 the hashf () function. Hashf () only exists for the sake of
306 politeness, for use when speed isn't so important. */
308 #define HASHSIZE 1403
309 HASHNODE *hashtab[HASHSIZE];
310 #define HASHSTEP(old, c) ((old << 2) + c)
311 #define MAKE_POS(v) (v & 0x7fffffff) /* make number positive */
313 /* `struct directive' defines one #-directive, including how to handle it. */
315 struct directive {
316 const int length; /* Length of name */
317 void (*const func) PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
318 /* Function to handle directive */
319 const char *const name; /* Name of directive */
320 const enum node_type type; /* Code which describes which directive. */
323 /* Last arg to output_line_command. */
324 enum file_change_code {same_file, enter_file, leave_file};
326 /* This structure represents one parsed argument in a macro call.
327 `raw' points to the argument text as written (`raw_length' is its length).
328 `expanded' points to the argument's macro-expansion
329 (its length is `expand_length').
330 `stringified_length' is the length the argument would have
331 if stringified.
332 `free1' and `free2', if nonzero, point to blocks to be freed
333 when the macro argument data is no longer needed. */
335 struct argdata {
336 U_CHAR *raw, *expanded;
337 int raw_length, expand_length;
338 int stringified_length;
339 U_CHAR *free1, *free2;
340 char newlines;
341 char comments;
344 /* The arglist structure is built by do_define to tell
345 collect_definition where the argument names begin. That
346 is, for a define like "#define f(x,y,z) foo+x-bar*y", the arglist
347 would contain pointers to the strings x, y, and z.
348 Collect_definition would then build a DEFINITION node,
349 with reflist nodes pointing to the places x, y, and z had
350 appeared. So the arglist is just convenience data passed
351 between these two routines. It is not kept around after
352 the current #define has been processed and entered into the
353 hash table. */
355 struct arglist {
356 struct arglist *next;
357 U_CHAR *name;
358 int length;
359 int argno;
362 /* Function prototypes. */
364 static void do_define PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
365 static void do_error PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
366 static void do_warning PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
367 static void do_line PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
368 static void do_include PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
369 static void do_include_next PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
370 static void do_undef PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
371 static void do_if PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
372 static void do_ifdef PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
373 static void do_ifndef PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
374 static void do_else PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
375 static void do_elif PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
376 static void do_endif PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
377 static void do_assert PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
378 static void do_unassert PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
379 static void do_xifdef PARAMS ((U_CHAR *, U_CHAR *, enum node_type));
381 static struct hashnode *install PARAMS ((const U_CHAR *, int, enum node_type, int));
382 static int hashf PARAMS ((const U_CHAR *, int, int));
383 static int compare_defs PARAMS ((DEFINITION *, DEFINITION *));
384 static int comp_def_part PARAMS ((int, const U_CHAR *, int,
385 const U_CHAR *, int, int));
386 static void delete_macro PARAMS ((HASHNODE *));
388 /* First arg to v_message. */
389 enum msgtype { MT_WARNING = 0, MT_ERROR, MT_FATAL };
390 static void v_message PARAMS ((enum msgtype mtype, int line,
391 const char *msgid, va_list ap))
392 ATTRIBUTE_PRINTF (3, 0);
394 static int line_for_error PARAMS ((int));
396 /* We know perfectly well which file this is, so we don't need to
397 use __FILE__. */
398 #undef abort
399 #if (GCC_VERSION >= 2007)
400 #define abort() fancy_abort(__LINE__, __FUNCTION__)
401 #else
402 #define abort() fancy_abort(__LINE__, 0);
403 #endif
405 static void macroexpand PARAMS ((HASHNODE *, FILE_BUF *));
406 static void special_symbol PARAMS ((HASHNODE *, FILE_BUF *));
407 static void dump_all_macros PARAMS ((void));
408 static void dump_defn_1 PARAMS ((const U_CHAR *, int, int));
409 static void dump_arg_n PARAMS ((DEFINITION *, int));
410 static void conditional_skip PARAMS ((FILE_BUF *, int, enum node_type));
411 static void skip_if_group PARAMS ((FILE_BUF *, int));
412 static void output_line_command PARAMS ((FILE_BUF *, FILE_BUF *,
413 int, enum file_change_code));
415 static int eval_if_expression PARAMS ((const U_CHAR *, int));
417 static void output_deps PARAMS ((void));
418 static void initialize_builtins PARAMS ((void));
419 static void run_directive PARAMS ((const char *, size_t,
420 enum node_type));
421 static void make_definition PARAMS ((const char *));
422 static void make_undef PARAMS ((const char *));
423 static void make_assertion PARAMS ((const char *));
425 static void grow_outbuf PARAMS ((FILE_BUF *, int));
426 static int handle_directive PARAMS ((FILE_BUF *, FILE_BUF *));
427 static void process_include PARAMS ((struct file_name_list *,
428 const U_CHAR *, int, int, FILE_BUF *));
429 static void fixup_newlines PARAMS ((FILE_BUF *));
430 static void finclude PARAMS ((int, const char *,
431 struct file_name_list *, FILE_BUF *));
432 static void init_dependency_output PARAMS ((void));
433 static void rescan PARAMS ((FILE_BUF *, int));
434 static void newline_fix PARAMS ((U_CHAR *));
435 static void name_newline_fix PARAMS ((U_CHAR *));
436 static U_CHAR *macarg1 PARAMS ((U_CHAR *, const U_CHAR *, int *,
437 int *, int *));
438 static const char *macarg PARAMS ((struct argdata *));
439 static int discard_comments PARAMS ((U_CHAR *, int, int));
440 static int file_size_and_mode PARAMS ((int, int *, long *));
442 static U_CHAR *skip_to_end_of_comment PARAMS ((FILE_BUF *, int *));
443 static U_CHAR *skip_quoted_string PARAMS ((const U_CHAR *, const U_CHAR *,
444 int, int *, int *, int *));
446 int main PARAMS ((int, char **));
448 /* Convenience. Write U"string" to get an unsigned string constant. */
449 #define U (const unsigned char *)
451 /* Here is the actual list of #-directives, most-often-used first. */
453 static const struct directive directive_table[] = {
454 { 6, do_define, "define", T_DEFINE },
455 { 7, do_include, "include", T_INCLUDE },
456 { 5, do_endif, "endif", T_ENDIF },
457 { 5, do_ifdef, "ifdef", T_IFDEF },
458 { 2, do_if, "if", T_IF, },
459 { 4, do_else, "else", T_ELSE },
460 { 6, do_ifndef, "ifndef", T_IFNDEF },
461 { 5, do_undef, "undef", T_UNDEF },
462 { 4, do_line, "line", T_LINE },
463 { 4, do_elif, "elif", T_ELIF },
464 { 5, do_error, "error", T_ERROR },
465 { 7, do_warning, "warning", T_WARNING },
466 { 12, do_include_next, "include_next", T_INCLUDE_NEXT },
467 { 6, do_assert, "assert", T_ASSERT },
468 { 8, do_unassert,"unassert",T_UNASSERT},
469 { -1, 0, "", T_UNUSED},
472 #define SKIP_WHITE_SPACE(p) do { while (is_nvspace(*p)) p++; } while (0)
473 #define SKIP_ALL_WHITE_SPACE(p) do { while (is_space(*p)) p++; } while (0)
475 int errors = 0; /* Error counter for exit code */
477 static FILE_BUF expand_to_temp_buffer PARAMS ((const U_CHAR *, const U_CHAR *, int));
478 static DEFINITION *collect_expansion PARAMS ((U_CHAR *, U_CHAR *, int,
479 struct arglist *));
481 /* Stack of conditionals currently in progress
482 (including both successful and failing conditionals). */
484 struct if_stack {
485 struct if_stack *next; /* for chaining to the next stack frame */
486 const char *fname; /* copied from input when frame is made */
487 int lineno; /* similarly */
488 int if_succeeded; /* true if a leg of this if-group
489 has been passed through rescan */
490 enum node_type type; /* type of last directive seen in this group */
492 typedef struct if_stack IF_STACK_FRAME;
493 IF_STACK_FRAME *if_stack = NULL;
495 /* Nonzero means -I- has been seen,
496 so don't look for #include "foo" the source-file directory. */
497 int ignore_srcdir;
499 /* Pending directives. */
500 enum pending_dir_t {PD_NONE = 0, PD_DEFINE, PD_UNDEF, PD_ASSERTION, PD_FILE};
502 typedef struct pending_dir pending_dir;
503 struct pending_dir
505 const char *arg;
506 enum pending_dir_t type;
510 main (argc, argv)
511 int argc;
512 char **argv;
514 int st_mode;
515 long st_size;
516 const char *in_fname, *out_fname;
517 int f, i;
518 FILE_BUF *fp;
519 pending_dir *pend = (pending_dir *) xcalloc (argc, sizeof (pending_dir));
520 int no_standard_includes = 0;
522 hex_init ();
524 #ifdef RLIMIT_STACK
525 /* Get rid of any avoidable limit on stack size. */
527 struct rlimit rlim;
529 /* Set the stack limit huge so that alloca (particularly stringtab
530 * in dbxread.c) does not fail. */
531 getrlimit (RLIMIT_STACK, &rlim);
532 rlim.rlim_cur = rlim.rlim_max;
533 setrlimit (RLIMIT_STACK, &rlim);
535 #endif /* RLIMIT_STACK defined */
537 progname = argv[0];
539 in_fname = NULL;
540 out_fname = NULL;
542 no_line_commands = 0;
543 dump_macros = 0;
544 no_output = 0;
546 max_include_len = cpp_GCC_INCLUDE_DIR_len + 7; /* ??? */
548 gcc_init_libintl ();
550 /* It's simplest to just create this struct whether or not it will
551 be needed. */
552 deps = deps_init ();
554 /* Process switches and find input file name. */
556 for (i = 1; i < argc; i++) {
557 if (argv[i][0] != '-') {
558 if (out_fname != NULL)
559 fatal ("usage: %s [switches] input output", argv[0]);
560 else if (in_fname != NULL)
561 out_fname = argv[i];
562 else
563 in_fname = argv[i];
564 } else {
565 int c = argv[i][1];
567 switch (c) {
568 case 'E':
569 case '$':
570 break; /* Ignore for compatibility with ISO/extended cpp. */
572 case 'l':
573 if (!strcmp (argv[i], "-lang-c++")
574 || !strcmp (argv[i], "-lang-objc++"))
575 fatal ("-traditional is not supported in C++");
576 else if (!strcmp (argv[i], "-lang-c89"))
577 fatal ("-traditional and -ansi are mutually exclusive");
578 else if (!strcmp (argv[i], "-lang-objc"))
579 pend[i].type = PD_DEFINE, pend[i].arg = "__OBJC__";
580 else if (!strcmp (argv[i], "-lang-asm"))
581 pend[i].type = PD_DEFINE, pend[i].arg = "__ASSEMBLER__";
582 else if (!strcmp (argv[i], "-lang-fortran"))
583 pend[i].type = PD_DEFINE, pend[i].arg = "_LANGUAGE_FORTRAN";
584 /* All other possibilities ignored. */
585 break;
587 case 'i':
588 if (!strcmp (argv[i], "-include"))
590 if (i + 1 == argc)
591 fatal ("filename missing after -i option");
592 else
593 pend[i].type = PD_FILE, pend[i].arg = argv[i + 1], i++;
595 else if (!strcmp (argv[i], "-iprefix"))
596 i++; /* Ignore for compatibility */
597 else if (!strcmp (argv[i], "-isystem")
598 || !strcmp (argv[i], "-iwithprefix")
599 || !strcmp (argv[i], "-iwithprefixbefore")
600 || !strcmp (argv[i], "-idirafter"))
601 goto add_include; /* best we can do */
603 break;
605 case 'o':
606 if (out_fname != NULL)
607 fatal ("output filename specified twice");
608 if (i + 1 == argc)
609 fatal ("filename missing after -o option");
610 out_fname = argv[++i];
611 if (!strcmp (out_fname, "-"))
612 out_fname = "";
613 break;
615 case 'w':
616 inhibit_warnings = 1;
617 break;
619 case 'W':
620 if (!strcmp (argv[i], "-Wcomments"))
621 warn_comments = 1;
622 else if (!strcmp (argv[i], "-Wcomment"))
623 warn_comments = 1;
624 else if (!strcmp (argv[i], "-Wall")) {
625 warn_comments = 1;
627 break;
629 case 'f':
630 if (!strcmp (argv[i], "-fsigned-char"))
631 flag_signed_char = 1;
632 else if (!strcmp (argv[i], "-funsigned-char"))
633 flag_signed_char = 0;
634 break;
636 /* Ignore target-specific and optimization flags. */
637 case 'm':
638 case 'O':
639 break;
641 case 'M':
643 char *p = NULL;
645 /* -MD and -MMD for tradcpp are deprecated and undocumented
646 (use -M or -MM with -MF instead), and probably should be
647 removed with the next major GCC version. For the moment
648 we allow these for the benefit of Automake 1.4, which
649 uses these when dependency tracking is enabled. Automake
650 1.5 will fix this. */
651 if (!strncmp (argv[i], "-MD", 3)) {
652 p = argv[i] + 3;
653 print_deps = 2;
654 } else if (!strncmp (argv[i], "-MMD", 4)) {
655 p = argv[i] + 4;
656 print_deps = 1;
657 } else if (!strcmp (argv[i], "-M")) {
658 print_deps = 2;
659 } else if (!strcmp (argv[i], "-MM")) {
660 print_deps = 1;
661 } else if (!strcmp (argv[i], "-MG")) {
662 deps_missing_files = 1;
663 } else if (!strcmp (argv[i], "-MF")) {
664 p = argv[i] + 3;
665 } else if (!strcmp (argv[i], "-MP")) {
666 print_deps_phony_targets = 1;
667 } else if (!strcmp (argv[i], "-MQ") || !strcmp (argv[i], "-MT")) {
668 /* Add a target. -MQ quotes for Make. */
669 const char *tgt = argv[i] + 3;
670 int quoted = argv[i][2] == 'Q';
672 if (*tgt == '\0' && i + 1 == argc)
673 fatal ("target missing after %s option", argv[i]);
674 else
676 if (*tgt == '\0')
677 tgt = argv[++i];
679 deps_add_target (deps, tgt, quoted);
683 if (p) {
684 if (*p)
685 deps_file = p;
686 else if (i + 1 == argc)
687 fatal ("filename missing after %s option", argv[i]);
688 else
689 deps_file = argv[++i];
692 break;
694 case 'd':
695 dump_macros = 1;
696 no_output = 1;
697 break;
699 case 'v':
700 fprintf (stderr, "GNU traditional CPP version %s\n", version_string);
701 break;
703 case 'D':
704 case 'U':
705 case 'A':
707 char *p;
709 if (argv[i][2] != 0)
710 p = argv[i] + 2;
711 else if (i + 1 == argc)
712 fatal ("macro name missing after -%c option", c);
713 else
714 p = argv[++i];
716 if (c == 'D')
717 pend[i].type = PD_DEFINE;
718 else if (c == 'U')
719 pend[i].type = PD_UNDEF;
720 else
721 pend[i].type = PD_ASSERTION;
722 pend[i].arg = p;
724 break;
726 case 'C':
727 put_out_comments = 1;
728 break;
730 case 'p':
731 if (!strcmp (argv[i], "-pedantic"))
732 fatal ("-pedantic and -traditional are mutually exclusive");
733 break;
735 case 't':
736 if (!strcmp (argv[i], "-trigraphs"))
737 fatal ("-trigraphs and -traditional are mutually exclusive");
738 break;
740 case 'P':
741 no_line_commands = 1;
742 break;
744 case 'I': /* Add directory to path for includes. */
745 add_include:
747 struct file_name_list *dirtmp;
749 if (! ignore_srcdir && !strcmp (argv[i] + 2, "-"))
750 ignore_srcdir = 1;
751 else {
752 dirtmp = (struct file_name_list *)
753 xmalloc (sizeof (struct file_name_list));
754 dirtmp->next = 0; /* New one goes on the end */
755 if (include == 0)
756 include = dirtmp;
757 else
758 last_include->next = dirtmp;
759 last_include = dirtmp; /* Tail follows the last one */
760 if (argv[i][1] == 'I' && argv[i][2] != 0)
761 dirtmp->fname = argv[i] + 2;
762 else if (i + 1 == argc)
763 fatal ("directory name missing after -I option");
764 else
765 dirtmp->fname = argv[++i];
766 if (strlen (dirtmp->fname) > max_include_len)
767 max_include_len = strlen (dirtmp->fname);
768 if (ignore_srcdir && first_bracket_include == 0)
769 first_bracket_include = dirtmp;
772 break;
774 case 'n':
775 /* -nostdinc causes no default include directories.
776 You must specify all include-file directories with -I. */
777 no_standard_includes = 1;
778 break;
780 case 'q':
781 /* Accept -quiet silently. */
782 break;
784 case '\0': /* JF handle '-' as file name meaning stdin or stdout */
785 if (in_fname == NULL) {
786 in_fname = "";
787 break;
788 } else if (out_fname == NULL) {
789 out_fname = "";
790 break;
791 } /* else fall through into error */
793 default:
794 fatal ("invalid option `%s'", argv[i]);
799 init_dependency_output ();
801 /* After checking the environment variables, check if -M or -MM has
802 not been specified, but other -M options have. */
803 if (print_deps == 0
804 && (deps_missing_files || deps_file || print_deps_phony_targets))
805 fatal ("you must additionally specify either -M or -MM");
807 if (print_deps)
809 /* Set the default target (if there is none already), and
810 the dependency on the main file. */
811 deps_add_default_target (deps, in_fname);
813 deps_add_dep (deps, in_fname);
816 /* Install __LINE__, etc. Must follow option processing. */
817 initialize_builtins ();
819 /* Do defines specified with -D and undefines specified with -U. */
820 for (i = 1; i < argc; i++)
821 if (pend[i].type == PD_DEFINE)
822 make_definition (pend[i].arg);
823 else if (pend[i].type == PD_UNDEF)
824 make_undef (pend[i].arg);
825 else if (pend[i].type == PD_ASSERTION)
826 make_assertion (pend[i].arg);
828 /* Unless -fnostdinc,
829 tack on the standard include file dirs to the specified list */
830 if (!no_standard_includes) {
831 const struct default_include *di;
832 struct file_name_list *old_last_include = last_include;
833 struct file_name_list *dirtmp;
834 for (di = cpp_include_defaults; di->fname; di++) {
835 if (di->cplusplus)
836 continue;
837 dirtmp = (struct file_name_list *)
838 xmalloc (sizeof (struct file_name_list));
839 dirtmp->next = 0; /* New one goes on the end */
840 if (include == 0)
841 include = dirtmp;
842 else
843 last_include->next = dirtmp;
844 last_include = dirtmp; /* Tail follows the last one */
845 dirtmp->fname = di->fname;
846 if (strlen (dirtmp->fname) > max_include_len)
847 max_include_len = strlen (dirtmp->fname);
850 if (ignore_srcdir && first_bracket_include == 0)
851 first_bracket_include = old_last_include->next;
854 /* Initialize output buffer */
856 outbuf.buf = (U_CHAR *) xmalloc (OUTBUF_SIZE);
857 outbuf.bufp = outbuf.buf;
858 outbuf.length = OUTBUF_SIZE;
860 /* Scan the -i files before the main input.
861 Much like #including them, but with no_output set
862 so that only their macro definitions matter. */
864 no_output++;
865 indepth++;
866 for (i = 1; i < argc; i++)
867 if (pend[i].type == PD_FILE)
869 int fd = open (pend[i].arg, O_RDONLY, 0666);
870 if (fd < 0)
872 perror_with_name (pend[i].arg);
873 return FATAL_EXIT_CODE;
876 /* For -M, add this file to the dependencies. */
877 if (print_deps)
878 deps_add_dep (deps, pend[i].arg);
880 finclude (fd, pend[i].arg, 0, &outbuf);
882 indepth--;
883 no_output--;
885 /* Pending directives no longer needed. */
886 free ((PTR) pend);
888 /* Create an input stack level for the main input file
889 and copy the entire contents of the file into it. */
891 fp = &instack[++indepth];
893 /* JF check for stdin */
894 if (in_fname == NULL || *in_fname == 0) {
895 in_fname = "";
896 f = 0;
897 } else if ((f = open (in_fname, O_RDONLY, 0666)) < 0)
898 goto sys_error;
900 if (file_size_and_mode (f, &st_mode, &st_size))
901 goto sys_error;
902 fp->fname = in_fname;
903 fp->lineno = 1;
904 /* JF all this is mine about reading pipes and ttys */
905 if (!S_ISREG (st_mode)) {
906 /* Read input from a file that is not a normal disk file.
907 We cannot preallocate a buffer with the correct size,
908 so we must read in the file a piece at the time and make it bigger. */
909 int size;
910 int bsize;
911 int cnt;
912 U_CHAR *bufp;
914 bsize = 2000;
915 size = 0;
916 fp->buf = (U_CHAR *) xmalloc (bsize + 2);
917 bufp = fp->buf;
918 for (;;) {
919 cnt = read (f, bufp, bsize - size);
920 if (cnt < 0) goto sys_error; /* error! */
921 if (cnt == 0) break; /* End of file */
922 size += cnt;
923 bufp += cnt;
924 if (bsize == size) { /* Buffer is full! */
925 bsize *= 2;
926 fp->buf = (U_CHAR *) xrealloc (fp->buf, bsize + 2);
927 bufp = fp->buf + size; /* May have moved */
930 fp->length = size;
931 } else {
932 /* Read a file whose size we can determine in advance.
933 For the sake of VMS, st_size is just an upper bound. */
934 long i;
935 fp->length = 0;
936 fp->buf = (U_CHAR *) xmalloc (st_size + 2);
938 while (st_size > 0) {
939 i = read (f, fp->buf + fp->length, st_size);
940 if (i <= 0) {
941 if (i == 0) break;
942 goto sys_error;
944 fp->length += i;
945 st_size -= i;
948 fp->bufp = fp->buf;
949 fp->if_stack = if_stack;
950 fixup_newlines (fp);
952 /* Make sure data ends with a newline. And put a null after it. */
954 if (fp->length > 0 && fp->buf[fp->length-1] != '\n')
955 fp->buf[fp->length++] = '\n';
956 fp->buf[fp->length] = '\0';
958 /* Now that we know the input file is valid, open the output. */
960 if (!out_fname || !strcmp (out_fname, ""))
961 out_fname = "stdout";
962 else if (! freopen (out_fname, "w", stdout))
963 pfatal_with_name (out_fname);
965 output_line_command (fp, &outbuf, 0, same_file);
967 /* Scan the input, processing macros and directives. */
969 rescan (&outbuf, 0);
971 /* Now we have processed the entire input
972 Write whichever kind of output has been requested. */
975 if (dump_macros)
976 dump_all_macros ();
977 else if (! inhibit_output)
978 if (write (fileno (stdout), outbuf.buf, outbuf.bufp - outbuf.buf) < 0)
979 fatal ("I/O error on output");
981 /* Don't write the deps file if preprocessing has failed. */
982 if (print_deps && errors == 0)
983 output_deps ();
985 /* Destruct the deps object. */
986 deps_free (deps);
988 if (ferror (stdout))
989 fatal ("I/O error on output");
991 if (errors)
992 exit (FATAL_EXIT_CODE);
993 exit (SUCCESS_EXIT_CODE);
995 sys_error:
996 pfatal_with_name (in_fname);
999 /* Set up dependency-file output. */
1000 static void
1001 init_dependency_output ()
1003 char *spec, *s, *output_file;
1005 /* Either of two environment variables can specify output of deps.
1006 Its value is either "OUTPUT_FILE" or "OUTPUT_FILE DEPS_TARGET",
1007 where OUTPUT_FILE is the file to write deps info to
1008 and DEPS_TARGET is the target to mention in the deps. */
1010 if (print_deps == 0)
1012 spec = getenv ("DEPENDENCIES_OUTPUT");
1013 if (spec)
1014 print_deps = 1;
1015 else
1017 spec = getenv ("SUNPRO_DEPENDENCIES");
1018 if (spec)
1019 print_deps = 2;
1020 else
1021 return;
1024 /* Find the space before the DEPS_TARGET, if there is one. */
1025 s = strchr (spec, ' ');
1026 if (s)
1028 /* Let the caller perform MAKE quoting. */
1029 deps_add_target (deps, s + 1, 0);
1030 output_file = (char *) xmalloc (s - spec + 1);
1031 memcpy (output_file, spec, s - spec);
1032 output_file[s - spec] = 0;
1034 else
1035 output_file = spec;
1037 /* Command line overrides environment variables. */
1038 if (deps_file == 0)
1039 deps_file = output_file;
1040 deps_append = 1;
1043 /* If dependencies go to standard output, or -MG is used, we should
1044 suppress output. The user may be requesting other stuff to
1045 stdout, with -dM, -v etc. We let them shoot themselves in the
1046 foot. */
1047 if (deps_file == 0 || deps_missing_files)
1048 inhibit_output = 1;
1051 /* Use mkdeps.c to output dependency information. */
1052 static void
1053 output_deps ()
1055 /* Stream on which to print the dependency information. */
1056 FILE *deps_stream = 0;
1057 const char *const deps_mode = deps_append ? "a" : "w";
1059 if (deps_file == 0)
1060 deps_stream = stdout;
1061 else
1063 deps_stream = fopen (deps_file, deps_mode);
1064 if (deps_stream == 0)
1066 error_from_errno (deps_file);
1067 return;
1071 deps_write (deps, deps_stream, 72);
1073 if (print_deps_phony_targets)
1074 deps_phony_targets (deps, deps_stream);
1076 /* Don't close stdout. */
1077 if (deps_file)
1079 if (ferror (deps_stream) || fclose (deps_stream) != 0)
1080 fatal ("I/O error on output");
1084 /* Move all backslash-newline pairs out of embarrassing places.
1085 Exchange all such pairs following BP
1086 with any potentially-embarrasing characters that follow them.
1087 Potentially-embarrassing characters are / and *
1088 (because a backslash-newline inside a comment delimiter
1089 would cause it not to be recognized). */
1090 static void
1091 newline_fix (bp)
1092 U_CHAR *bp;
1094 U_CHAR *p = bp;
1095 int count = 0;
1097 /* First count the backslash-newline pairs here. */
1099 while (*p++ == '\\' && *p++ == '\n')
1100 count++;
1102 p = bp + count * 2;
1104 /* Exit if what follows the backslash-newlines is not embarrassing. */
1106 if (count == 0 || (*p != '/' && *p != '*'))
1107 return;
1109 /* Copy all potentially embarrassing characters
1110 that follow the backslash-newline pairs
1111 down to where the pairs originally started. */
1113 while (*p == '*' || *p == '/')
1114 *bp++ = *p++;
1116 /* Now write the same number of pairs after the embarrassing chars. */
1117 while (count-- > 0) {
1118 *bp++ = '\\';
1119 *bp++ = '\n';
1123 /* Like newline_fix but for use within a directive-name.
1124 Move any backslash-newlines up past any following symbol constituents. */
1125 static void
1126 name_newline_fix (bp)
1127 U_CHAR *bp;
1129 U_CHAR *p = bp;
1130 int count = 0;
1132 /* First count the backslash-newline pairs here. */
1134 while (*p++ == '\\' && *p++ == '\n')
1135 count++;
1137 p = bp + count * 2;
1139 /* What follows the backslash-newlines is not embarrassing. */
1141 if (count == 0 || !is_idchar (*p))
1142 return;
1144 /* Copy all potentially embarrassing characters
1145 that follow the backslash-newline pairs
1146 down to where the pairs originally started. */
1148 while (is_idchar (*p))
1149 *bp++ = *p++;
1151 /* Now write the same number of pairs after the embarrassing chars. */
1152 while (count-- > 0) {
1153 *bp++ = '\\';
1154 *bp++ = '\n';
1159 * The main loop of the program.
1161 * Read characters from the input stack, transferring them to the
1162 * output buffer OP.
1164 * Macros are expanded and push levels on the input stack.
1165 * At the end of such a level it is popped off and we keep reading.
1166 * At the end of any other kind of level, we return.
1167 * #-directives are handled, except within macros.
1169 * If OUTPUT_MARKS is nonzero, keep Newline markers found in the input
1170 * and insert them when appropriate. This is set while scanning macro
1171 * arguments before substitution. It is zero when scanning for final output.
1172 * There are three types of Newline markers:
1173 * * Newline - follows a macro name that was not expanded
1174 * because it appeared inside an expansion of the same macro.
1175 * This marker prevents future expansion of that identifier.
1176 * When the input is rescanned into the final output, these are deleted.
1177 * These are also deleted by ## concatenation.
1178 * * Newline Space (or Newline and any other whitespace character)
1179 * stands for a place that tokens must be separated or whitespace
1180 * is otherwise desirable, but where the ANSI standard specifies there
1181 * is no whitespace. This marker turns into a Space (or whichever other
1182 * whitespace char appears in the marker) in the final output,
1183 * but it turns into nothing in an argument that is stringified with #.
1184 * Such stringified arguments are the only place where the ANSI standard
1185 * specifies with precision that whitespace may not appear.
1187 * During this function, IP->bufp is kept cached in IBP for speed of access.
1188 * Likewise, OP->bufp is kept in OBP. Before calling a subroutine
1189 * IBP, IP and OBP must be copied back to memory. IP and IBP are
1190 * copied back with the RECACHE macro. OBP must be copied back from OP->bufp
1191 * explicitly, and before RECACHE, since RECACHE uses OBP.
1194 static void
1195 rescan (op, output_marks)
1196 FILE_BUF *op;
1197 int output_marks;
1199 /* Character being scanned in main loop. */
1200 U_CHAR c;
1202 /* Length of pending accumulated identifier. */
1203 int ident_length = 0;
1205 /* Hash code of pending accumulated identifier. */
1206 int hash = 0;
1208 /* Current input level (&instack[indepth]). */
1209 FILE_BUF *ip;
1211 /* Pointer for scanning input. */
1212 U_CHAR *ibp;
1214 /* Pointer to end of input. End of scan is controlled by LIMIT. */
1215 U_CHAR *limit;
1217 /* Pointer for storing output. */
1218 U_CHAR *obp;
1220 /* REDO_CHAR is nonzero if we are processing an identifier
1221 after backing up over the terminating character.
1222 Sometimes we process an identifier without backing up over
1223 the terminating character, if the terminating character
1224 is not special. Backing up is done so that the terminating character
1225 will be dispatched on again once the identifier is dealt with. */
1226 int redo_char = 0;
1228 /* 1 if within an identifier inside of which a concatenation
1229 marker (Newline -) has been seen. */
1230 int concatenated = 0;
1232 /* While scanning a comment or a string constant,
1233 this records the line it started on, for error messages. */
1234 int start_line;
1236 /* Record position of last `real' newline. */
1237 U_CHAR *beg_of_line;
1239 /* This has to be a global bacause of RECACHE. */
1240 U_CHAR *obufp_before_macroname = NULL;
1242 /* Pop the innermost input stack level, assuming it is a macro expansion. */
1244 #define POPMACRO \
1245 do { ip->macro->type = T_MACRO; \
1246 if (ip->free_ptr) free (ip->free_ptr); \
1247 --indepth; } while (0)
1249 /* Reload `rescan's local variables that describe the current
1250 level of the input stack. */
1252 #define RECACHE \
1253 do { ip = &instack[indepth]; \
1254 ibp = ip->bufp; \
1255 limit = ip->buf + ip->length; \
1256 op->bufp = obp; \
1257 check_expand (op, limit - ibp); \
1258 beg_of_line = 0; \
1259 obufp_before_macroname += op->bufp - obp; \
1260 obp = op->bufp; } while (0)
1262 if (no_output && instack[indepth].fname != 0)
1263 skip_if_group (&instack[indepth], 1);
1265 obp = op->bufp;
1266 RECACHE;
1267 beg_of_line = ibp;
1269 /* Our caller must always put a null after the end of
1270 the input at each input stack level. */
1271 if (*limit != 0)
1272 abort ();
1274 while (1) {
1275 c = *ibp++;
1276 *obp++ = c;
1278 switch (c) {
1279 case '\\':
1280 if (ibp >= limit)
1281 break;
1282 if (*ibp == '\n') {
1283 /* Always merge lines ending with backslash-newline,
1284 even in middle of identifier. */
1285 ++ibp;
1286 ++ip->lineno;
1287 --obp; /* remove backslash from obuf */
1288 break;
1290 /* Otherwise, backslash suppresses specialness of following char,
1291 so copy it here to prevent the switch from seeing it.
1292 But first get any pending identifier processed. */
1293 if (ident_length > 0)
1294 goto specialchar;
1295 *obp++ = *ibp++;
1296 break;
1298 case '#':
1299 /* If this is expanding a macro definition, don't recognize
1300 preprocessor directives. */
1301 if (ip->macro != 0)
1302 goto randomchar;
1303 if (ident_length)
1304 goto specialchar;
1306 /* # keyword: a # must be the first char on the line */
1307 if (beg_of_line == 0)
1308 goto randomchar;
1309 if (beg_of_line + 1 != ibp)
1310 goto randomchar;
1312 /* This # can start a directive. */
1314 --obp; /* Don't copy the '#' */
1316 ip->bufp = ibp;
1317 op->bufp = obp;
1318 if (! handle_directive (ip, op)) {
1319 #ifdef USE_C_ALLOCA
1320 alloca (0);
1321 #endif
1322 /* Not a known directive: treat it as ordinary text.
1323 IP, OP, IBP, etc. have not been changed. */
1324 if (no_output && instack[indepth].fname) {
1325 /* If not generating expanded output,
1326 what we do with ordinary text is skip it.
1327 Discard everything until next # directive. */
1328 skip_if_group (&instack[indepth], 1);
1329 RECACHE;
1330 beg_of_line = ibp;
1331 break;
1333 ++obp; /* Copy the '#' after all */
1334 goto randomchar;
1336 #ifdef USE_C_ALLOCA
1337 alloca (0);
1338 #endif
1339 /* A # directive has been successfully processed. */
1340 /* If not generating expanded output, ignore everything until
1341 next # directive. */
1342 if (no_output && instack[indepth].fname)
1343 skip_if_group (&instack[indepth], 1);
1344 obp = op->bufp;
1345 RECACHE;
1346 beg_of_line = ibp;
1347 break;
1349 case '\"': /* skip quoted string */
1350 case '\'':
1351 /* A single quoted string is treated like a double -- some
1352 programs (e.g., troff) are perverse this way */
1354 if (ident_length)
1355 goto specialchar;
1357 start_line = ip->lineno;
1359 /* Skip ahead to a matching quote. */
1361 while (1) {
1362 if (ibp >= limit) {
1363 if (ip->macro != 0) {
1364 /* try harder: this string crosses a macro expansion boundary */
1365 POPMACRO;
1366 RECACHE;
1367 continue;
1369 break;
1371 *obp++ = *ibp;
1372 switch (*ibp++) {
1373 case '\n':
1374 ++ip->lineno;
1375 ++op->lineno;
1376 /* Traditionally, end of line ends a string constant with no error.
1377 So exit the loop and record the new line. */
1378 beg_of_line = ibp;
1379 goto while2end;
1381 case '\\':
1382 if (ibp >= limit)
1383 break;
1384 if (*ibp == '\n') {
1385 /* Backslash newline is replaced by nothing at all,
1386 but keep the line counts correct. */
1387 --obp;
1388 ++ibp;
1389 ++ip->lineno;
1390 } else {
1391 /* ANSI stupidly requires that in \\ the second \
1392 is *not* prevented from combining with a newline. */
1393 while (*ibp == '\\' && ibp[1] == '\n') {
1394 ibp += 2;
1395 ++ip->lineno;
1397 *obp++ = *ibp++;
1399 break;
1401 case '\"':
1402 case '\'':
1403 if (ibp[-1] == c)
1404 goto while2end;
1405 break;
1408 while2end:
1409 break;
1411 case '/':
1412 if (*ibp == '\\' && ibp[1] == '\n')
1413 newline_fix (ibp);
1414 /* Don't look for comments inside a macro definition. */
1415 if (ip->macro != 0)
1416 goto randomchar;
1417 /* A comment constitutes white space, so it can terminate an identifier.
1418 Process the identifier, if any. */
1419 if (ident_length)
1420 goto specialchar;
1422 if (*ibp != '*')
1423 goto randomchar;
1425 /* We have a comment. Skip it, optionally copying it to output. */
1427 start_line = ip->lineno;
1429 ++ibp; /* Skip the star. */
1431 /* In K+R C, a comment is equivalent to nothing. Note that we
1432 already output the slash; we might not want it. */
1433 if (! put_out_comments)
1434 obp--;
1435 else
1436 *obp++ = '*';
1439 U_CHAR *before_bp = ibp;
1441 while (ibp < limit) {
1442 switch (*ibp++) {
1443 case '/':
1444 if (warn_comments && ibp < limit && *ibp == '*')
1445 warning("`/*' within comment");
1446 break;
1447 case '*':
1448 if (*ibp == '\\' && ibp[1] == '\n')
1449 newline_fix (ibp);
1450 if (ibp >= limit || *ibp == '/')
1451 goto comment_end;
1452 break;
1453 case '\n':
1454 ++ip->lineno;
1455 /* Copy the newline into the output buffer, in order to
1456 avoid the pain of a #line every time a multiline comment
1457 is seen. */
1458 if (!put_out_comments)
1459 *obp++ = '\n';
1460 ++op->lineno;
1463 comment_end:
1465 if (ibp >= limit)
1466 error_with_line (line_for_error (start_line),
1467 "unterminated comment");
1468 else {
1469 ibp++;
1470 if (put_out_comments) {
1471 memcpy (obp, before_bp, ibp - before_bp);
1472 obp += ibp - before_bp;
1476 break;
1478 case '0': case '1': case '2': case '3': case '4':
1479 case '5': case '6': case '7': case '8': case '9':
1480 /* If digit is not part of identifier, it starts a number,
1481 which means that following letters are not an identifier.
1482 "0x5" does not refer to an identifier "x5".
1483 So copy all alphanumerics that follow without accumulating
1484 as an identifier. Periods also, for sake of "3.e7". */
1486 if (ident_length == 0) {
1487 while (ibp < limit) {
1488 while (ibp < limit && ibp[0] == '\\' && ibp[1] == '\n') {
1489 ++ip->lineno;
1490 ibp += 2;
1492 c = *ibp++;
1493 if (! ISIDNUM (c) && c != '.') {
1494 --ibp;
1495 break;
1497 *obp++ = c;
1498 /* A sign can be part of a preprocessing number
1499 if it follows an e. */
1500 if (c == 'e' || c == 'E') {
1501 while (ibp < limit && ibp[0] == '\\' && ibp[1] == '\n') {
1502 ++ip->lineno;
1503 ibp += 2;
1505 if (ibp < limit && (*ibp == '+' || *ibp == '-')) {
1506 *obp++ = *ibp++;
1507 /* Traditional C does not let the token go past the sign. */
1508 break;
1512 break;
1514 /* fall through */
1516 case '_':
1517 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
1518 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
1519 case 'm': case 'n': case 'o': case 'p': case 'q': case 'r':
1520 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
1521 case 'y': case 'z':
1522 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
1523 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
1524 case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R':
1525 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
1526 case 'Y': case 'Z':
1527 ident_length++;
1528 /* Compute step of hash function, to avoid a proc call on every token */
1529 hash = HASHSTEP (hash, c);
1530 break;
1532 case '\n':
1533 /* If reprocessing a macro expansion, newline is a special marker. */
1534 if (ip->macro != 0) {
1535 /* Newline White is a "funny space" to separate tokens that are
1536 supposed to be separate but without space between.
1537 Here White means any horizontal whitespace character.
1538 Newline - marks a recursive macro use that is not
1539 supposed to be expandable. */
1541 if (*ibp == '-') {
1542 /* Newline - inhibits expansion of preceding token.
1543 If expanding a macro arg, we keep the newline -.
1544 In final output, it is deleted. */
1545 if (! concatenated) {
1546 ident_length = 0;
1547 hash = 0;
1549 ibp++;
1550 if (!output_marks) {
1551 obp--;
1552 } else {
1553 /* If expanding a macro arg, keep the newline -. */
1554 *obp++ = '-';
1556 } else if (is_space (*ibp)) {
1557 /* Newline Space does not prevent expansion of preceding token
1558 so expand the preceding token and then come back. */
1559 if (ident_length > 0)
1560 goto specialchar;
1562 /* If generating final output, newline space makes a space. */
1563 if (!output_marks) {
1564 obp[-1] = *ibp++;
1565 /* And Newline Newline makes a newline, so count it. */
1566 if (obp[-1] == '\n')
1567 op->lineno++;
1568 } else {
1569 /* If expanding a macro arg, keep the newline space.
1570 If the arg gets stringified, newline space makes nothing. */
1571 *obp++ = *ibp++;
1573 } else abort (); /* Newline followed by something random? */
1574 break;
1577 /* If there is a pending identifier, handle it and come back here. */
1578 if (ident_length > 0)
1579 goto specialchar;
1581 beg_of_line = ibp;
1583 /* Update the line counts and output a #line if necessary. */
1584 ++ip->lineno;
1585 ++op->lineno;
1586 if (ip->lineno != op->lineno) {
1587 op->bufp = obp;
1588 output_line_command (ip, op, 1, same_file);
1589 check_expand (op, ip->length - (ip->bufp - ip->buf));
1590 obp = op->bufp;
1592 break;
1594 /* Come here either after (1) a null character that is part of the input
1595 or (2) at the end of the input, because there is a null there. */
1596 case 0:
1597 if (ibp <= limit)
1598 /* Our input really contains a null character. */
1599 goto randomchar;
1601 /* At end of a macro-expansion level, pop it and read next level. */
1602 if (ip->macro != 0) {
1603 obp--;
1604 ibp--;
1605 /* If we have an identifier that ends here, process it now, so
1606 we get the right error for recursion. */
1607 if (ident_length && ! is_idchar (*instack[indepth - 1].bufp)) {
1608 redo_char = 1;
1609 goto randomchar;
1611 POPMACRO;
1612 RECACHE;
1613 break;
1616 /* If we don't have a pending identifier,
1617 return at end of input. */
1618 if (ident_length == 0) {
1619 obp--;
1620 ibp--;
1621 op->bufp = obp;
1622 ip->bufp = ibp;
1623 goto ending;
1626 /* If we do have a pending identifier, just consider this null
1627 a special character and arrange to dispatch on it again.
1628 The second time, IDENT_LENGTH will be zero so we will return. */
1630 /* Fall through */
1632 specialchar:
1634 /* Handle the case of a character such as /, ', " or null
1635 seen following an identifier. Back over it so that
1636 after the identifier is processed the special char
1637 will be dispatched on again. */
1639 ibp--;
1640 obp--;
1641 redo_char = 1;
1643 default:
1645 randomchar:
1647 if (ident_length > 0) {
1648 HASHNODE *hp;
1650 /* We have just seen an identifier end. If it's a macro, expand it.
1652 IDENT_LENGTH is the length of the identifier
1653 and HASH is its hash code.
1655 The identifier has already been copied to the output,
1656 so if it is a macro we must remove it.
1658 If REDO_CHAR is 0, the char that terminated the identifier
1659 has been skipped in the output and the input.
1660 OBP-IDENT_LENGTH-1 points to the identifier.
1661 If the identifier is a macro, we must back over the terminator.
1663 If REDO_CHAR is 1, the terminating char has already been
1664 backed over. OBP-IDENT_LENGTH points to the identifier. */
1666 for (hp = hashtab[MAKE_POS (hash) % HASHSIZE]; hp != NULL;
1667 hp = hp->next) {
1669 if (hp->length == ident_length) {
1670 /* obufp_before_macroname is used only in this block,
1671 but it has to be global because of RECACHE. */
1672 int op_lineno_before_macroname;
1673 int i = ident_length;
1674 U_CHAR *p = hp->name;
1675 U_CHAR *q = obp - i;
1677 if (! redo_char)
1678 q--;
1680 do { /* All this to avoid a strncmp () */
1681 if (*p++ != *q++)
1682 goto hashcollision;
1683 } while (--i);
1685 /* We found a use of a macro name.
1686 see if the context shows it is a macro call. */
1688 /* Back up over terminating character if not already done. */
1689 if (! redo_char) {
1690 ibp--;
1691 obp--;
1694 obufp_before_macroname = obp - ident_length;
1695 op_lineno_before_macroname = op->lineno;
1697 /* If macro wants an arglist, verify that a '(' follows.
1698 first skip all whitespace, copying it to the output
1699 after the macro name. Then, if there is no '(',
1700 decide this is not a macro call and leave things that way. */
1701 if (hp->type == T_MACRO && hp->value.defn->nargs >= 0)
1703 while (1) {
1704 /* Scan forward over whitespace, copying it to the output. */
1705 if (ibp == limit && ip->macro != 0) {
1706 POPMACRO;
1707 RECACHE;
1709 /* A comment: copy it unchanged or discard it. */
1710 else if (*ibp == '/' && ibp+1 != limit && ibp[1] == '*') {
1711 if (put_out_comments) {
1712 *obp++ = '/';
1713 *obp++ = '*';
1715 ibp += 2;
1716 while (ibp + 1 != limit
1717 && !(ibp[0] == '*' && ibp[1] == '/')) {
1718 /* We need not worry about newline-marks,
1719 since they are never found in comments. */
1720 if (*ibp == '\n') {
1721 /* Newline in a file. Count it. */
1722 ++ip->lineno;
1723 ++op->lineno;
1725 if (put_out_comments)
1726 *obp++ = *ibp++;
1727 else
1728 ibp++;
1730 ibp += 2;
1731 if (put_out_comments) {
1732 *obp++ = '*';
1733 *obp++ = '/';
1736 else if (is_space (*ibp)) {
1737 *obp++ = *ibp++;
1738 if (ibp[-1] == '\n') {
1739 if (ip->macro == 0) {
1740 /* Newline in a file. Count it. */
1741 ++ip->lineno;
1742 ++op->lineno;
1743 } else if (!output_marks) {
1744 /* A newline mark, and we don't want marks
1745 in the output. If it is newline-hyphen,
1746 discard it entirely. Otherwise, it is
1747 newline-whitechar, so keep the whitechar. */
1748 obp--;
1749 if (*ibp == '-')
1750 ibp++;
1751 else {
1752 if (*ibp == '\n')
1753 ++op->lineno;
1754 *obp++ = *ibp++;
1756 } else {
1757 /* A newline mark; copy both chars to the output. */
1758 *obp++ = *ibp++;
1762 else break;
1764 if (*ibp != '(')
1765 break;
1768 /* This is now known to be a macro call.
1769 Discard the macro name from the output,
1770 along with any following whitespace just copied. */
1771 obp = obufp_before_macroname;
1772 op->lineno = op_lineno_before_macroname;
1774 /* Expand the macro, reading arguments as needed,
1775 and push the expansion on the input stack. */
1776 ip->bufp = ibp;
1777 op->bufp = obp;
1778 macroexpand (hp, op);
1780 /* Reexamine input stack, since macroexpand has pushed
1781 a new level on it. */
1782 obp = op->bufp;
1783 RECACHE;
1784 break;
1786 hashcollision:
1788 } /* End hash-table-search loop */
1789 ident_length = hash = 0; /* Stop collecting identifier */
1790 redo_char = 0;
1791 concatenated = 0;
1792 } /* End if (ident_length > 0) */
1793 } /* End switch */
1794 } /* End per-char loop */
1796 /* Come here to return -- but first give an error message
1797 if there was an unterminated successful conditional. */
1798 ending:
1799 if (if_stack != ip->if_stack) {
1800 const char *str;
1801 switch (if_stack->type) {
1802 case T_IF:
1803 str = "if";
1804 break;
1805 case T_IFDEF:
1806 str = "ifdef";
1807 break;
1808 case T_IFNDEF:
1809 str = "ifndef";
1810 break;
1811 case T_ELSE:
1812 str = "else";
1813 break;
1814 case T_ELIF:
1815 str = "elif";
1816 break;
1817 default:
1818 abort ();
1820 error_with_line (line_for_error (if_stack->lineno),
1821 "unterminated #%s conditional", str);
1823 if_stack = ip->if_stack;
1827 * Rescan a string into a temporary buffer and return the result
1828 * as a FILE_BUF. Note this function returns a struct, not a pointer.
1830 * OUTPUT_MARKS nonzero means keep Newline markers found in the input
1831 * and insert such markers when appropriate. See `rescan' for details.
1832 * OUTPUT_MARKS is 1 for macroexpanding a macro argument separately
1833 * before substitution; it is 0 for other uses.
1835 static FILE_BUF
1836 expand_to_temp_buffer (buf, limit, output_marks)
1837 const U_CHAR *buf, *limit;
1838 int output_marks;
1840 FILE_BUF *ip;
1841 FILE_BUF obuf;
1842 int length = limit - buf;
1843 U_CHAR *buf1;
1844 int odepth = indepth;
1846 if (length < 0)
1847 abort ();
1849 /* Set up the input on the input stack. */
1851 buf1 = (U_CHAR *) alloca (length + 1);
1853 const U_CHAR *p1 = buf;
1854 U_CHAR *p2 = buf1;
1856 while (p1 != limit)
1857 *p2++ = *p1++;
1859 buf1[length] = 0;
1861 /* Set up to receive the output. */
1863 obuf.length = length * 2 + 100; /* Usually enough. Why be stingy? */
1864 obuf.bufp = obuf.buf = (U_CHAR *) xmalloc (obuf.length);
1865 obuf.fname = 0;
1866 obuf.macro = 0;
1867 obuf.free_ptr = 0;
1869 CHECK_DEPTH ({return obuf;});
1871 ++indepth;
1873 ip = &instack[indepth];
1874 ip->fname = 0;
1875 ip->macro = 0;
1876 ip->free_ptr = 0;
1877 ip->length = length;
1878 ip->buf = ip->bufp = buf1;
1879 ip->if_stack = if_stack;
1881 ip->lineno = obuf.lineno = 1;
1883 /* Scan the input, create the output. */
1885 rescan (&obuf, output_marks);
1887 /* Pop input stack to original state. */
1888 --indepth;
1890 if (indepth != odepth)
1891 abort ();
1893 /* Record the output. */
1894 obuf.length = obuf.bufp - obuf.buf;
1896 return obuf;
1900 * Process a # directive. Expects IP->bufp to point to the '#', as in
1901 * `#define foo bar'. Passes to the command handler
1902 * (do_define, do_include, etc.): the addresses of the 1st and
1903 * last chars of the command (starting immediately after the #
1904 * keyword), plus op and the keyword table pointer. If the command
1905 * contains comments it is copied into a temporary buffer sans comments
1906 * and the temporary buffer is passed to the command handler instead.
1907 * Likewise for backslash-newlines.
1909 * Returns nonzero if this was a known # directive.
1910 * Otherwise, returns zero, without advancing the input pointer.
1913 static int
1914 handle_directive (ip, op)
1915 FILE_BUF *ip, *op;
1917 U_CHAR *bp, *cp;
1918 const struct directive *kt;
1919 int ident_length;
1920 U_CHAR *resume_p;
1922 /* Nonzero means we must copy the entire command
1923 to get rid of comments or backslash-newlines. */
1924 int copy_command = 0;
1926 U_CHAR *ident, *after_ident;
1928 bp = ip->bufp;
1929 /* Skip whitespace and \-newline. */
1930 while (1) {
1931 if (is_nvspace (*bp))
1932 bp++;
1933 else if (*bp == '/' && (newline_fix (bp + 1), bp[1]) == '*') {
1934 ip->bufp = bp;
1935 skip_to_end_of_comment (ip, &ip->lineno);
1936 bp = ip->bufp;
1937 } else if (*bp == '\\' && bp[1] == '\n') {
1938 bp += 2; ip->lineno++;
1939 } else break;
1942 /* Now find end of directive name.
1943 If we encounter a backslash-newline, exchange it with any following
1944 symbol-constituents so that we end up with a contiguous name. */
1946 cp = bp;
1947 while (1) {
1948 if (is_idchar (*cp))
1949 cp++;
1950 else {
1951 if (*cp == '\\' && cp[1] == '\n')
1952 name_newline_fix (cp);
1953 if (is_idchar (*cp))
1954 cp++;
1955 else break;
1958 ident_length = cp - bp;
1959 ident = bp;
1960 after_ident = cp;
1962 /* A line of just `#' becomes blank. */
1964 if (ident_length == 0 && *after_ident == '\n') {
1965 ip->bufp = after_ident;
1966 return 1;
1970 * Decode the keyword and call the appropriate expansion
1971 * routine, after moving the input pointer up to the next line.
1973 for (kt = directive_table; kt->length > 0; kt++) {
1974 if (kt->length == ident_length
1975 && !strncmp (kt->name, (const char *)ident, ident_length)) {
1976 U_CHAR *buf;
1977 U_CHAR *limit = ip->buf + ip->length;
1978 int unterminated = 0;
1980 /* Nonzero means do not delete comments within the directive.
1981 #define needs this to detect traditional token paste. */
1982 int keep_comments = kt->type == T_DEFINE;
1984 /* Find the end of this command (first newline not backslashed
1985 and not in a string or comment).
1986 Set COPY_COMMAND if the command must be copied
1987 (it contains a backslash-newline or a comment). */
1989 buf = bp = after_ident;
1990 while (bp < limit) {
1991 U_CHAR c = *bp++;
1992 switch (c) {
1993 case '\\':
1994 if (bp < limit) {
1995 if (*bp == '\n') {
1996 ip->lineno++;
1997 copy_command = 1;
1999 bp++;
2001 break;
2003 case '\'':
2004 case '\"':
2005 bp = skip_quoted_string (bp - 1, limit, ip->lineno, &ip->lineno, &copy_command, &unterminated);
2006 if (unterminated) {
2007 /* Traditional preprocessing permits unterminated strings. */
2008 ip->bufp = bp;
2009 goto endloop1;
2011 break;
2013 /* <...> is special for #include. */
2014 case '<':
2015 if (kt->type != T_INCLUDE)
2016 break;
2017 while (*bp && *bp != '>') bp++;
2018 break;
2020 case '/':
2021 if (*bp == '\\' && bp[1] == '\n')
2022 newline_fix (bp);
2023 if (*bp == '*') {
2024 U_CHAR *obp = bp - 1;
2025 ip->bufp = bp + 1;
2026 skip_to_end_of_comment (ip, &ip->lineno);
2027 bp = ip->bufp;
2028 /* No need to copy the command because of a comment at the end;
2029 just don't include the comment in the directive. */
2030 if (bp == limit || *bp == '\n') {
2031 bp = obp;
2032 goto endloop1;
2034 /* Don't remove the comments if this is #define. */
2035 if (! keep_comments)
2036 copy_command++;
2038 break;
2040 case '\n':
2041 --bp; /* Point to the newline */
2042 ip->bufp = bp;
2043 goto endloop1;
2046 ip->bufp = bp;
2048 endloop1:
2049 resume_p = ip->bufp;
2050 /* BP is the end of the directive.
2051 RESUME_P is the next interesting data after the directive.
2052 A comment may come between. */
2054 if (copy_command) {
2055 U_CHAR *xp = buf;
2056 /* Need to copy entire command into temp buffer before dispatching */
2058 cp = (U_CHAR *) alloca (bp - buf + 5); /* room for cmd plus
2059 some slop */
2060 buf = cp;
2062 /* Copy to the new buffer, deleting comments
2063 and backslash-newlines (and whitespace surrounding the latter). */
2065 while (xp < bp) {
2066 U_CHAR c = *xp++;
2067 *cp++ = c;
2069 switch (c) {
2070 case '\n':
2071 break;
2073 /* <...> is special for #include. */
2074 case '<':
2075 if (kt->type != T_INCLUDE)
2076 break;
2077 while (xp < bp && c != '>') {
2078 c = *xp++;
2079 if (c == '\\' && xp < bp && *xp == '\n')
2080 xp++, ip->lineno++;
2081 else
2082 *cp++ = c;
2084 break;
2086 case '\\':
2087 if (*xp == '\n') {
2088 xp++;
2089 cp--;
2090 if (cp != buf && is_space (cp[-1])) {
2091 while (cp != buf && is_space(cp[-1])) cp--;
2092 cp++;
2093 SKIP_WHITE_SPACE (xp);
2094 } else if (is_nvspace (*xp)) {
2095 *cp++ = *xp++;
2096 SKIP_WHITE_SPACE (xp);
2098 } else {
2099 *cp++ = *xp++;
2101 break;
2103 case '\'':
2104 case '\"':
2106 const U_CHAR *bp1
2107 = skip_quoted_string (xp - 1, limit, ip->lineno, 0, 0, 0);
2108 while (xp != bp1)
2109 *cp++ = *xp++;
2111 break;
2113 case '/':
2114 if (*xp == '*') {
2115 ip->bufp = xp + 1;
2116 skip_to_end_of_comment (ip, 0);
2117 if (keep_comments)
2118 while (xp != ip->bufp)
2119 *cp++ = *xp++;
2120 /* Delete the slash. */
2121 else
2122 cp--;
2123 xp = ip->bufp;
2128 /* Null-terminate the copy. */
2130 *cp = 0;
2132 else
2133 cp = bp;
2135 ip->bufp = resume_p;
2137 /* Call the appropriate command handler. buf now points to
2138 either the appropriate place in the input buffer, or to
2139 the temp buffer if it was necessary to make one. cp
2140 points to the first char after the contents of the (possibly
2141 copied) command, in either case. */
2142 (*kt->func) (buf, cp, op);
2143 check_expand (op, ip->length - (ip->bufp - ip->buf));
2145 return 1;
2149 return 0;
2152 static const char *const
2153 monthnames[] = {"Jan", "Feb", "Mar", "Apr", "May", "Jun",
2154 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"};
2157 * expand things like __FILE__. Place the expansion into the output
2158 * buffer *without* rescanning.
2160 static void
2161 special_symbol (hp, op)
2162 HASHNODE *hp;
2163 FILE_BUF *op;
2165 const char *buf;
2166 time_t t;
2167 int i, len;
2168 int true_indepth;
2169 FILE_BUF *ip = NULL;
2170 static struct tm *timebuf = NULL;
2172 int paren = 0; /* For special `defined' keyword */
2174 for (i = indepth; i >= 0; i--)
2175 if (instack[i].fname != NULL) {
2176 ip = &instack[i];
2177 break;
2179 if (ip == NULL)
2180 fatal ("not in any file?!");
2182 switch (hp->type) {
2183 case T_FILE:
2184 case T_BASE_FILE:
2186 const char *string;
2187 if (hp->type == T_FILE)
2188 string = ip->fname;
2189 else
2190 string = instack[0].fname;
2192 if (string)
2194 char *tmp = (char *) alloca (3 + strlen (string));
2195 sprintf (tmp, "\"%s\"", string);
2196 buf = tmp;
2198 else
2199 buf = "";
2201 break;
2204 case T_INCLUDE_LEVEL:
2206 char *tmp = (char *) alloca (8); /* Eigth bytes ought to be more than enough */
2207 true_indepth = 0;
2208 for (i = indepth; i >= 0; i--)
2209 if (instack[i].fname != NULL)
2210 true_indepth++;
2212 sprintf (tmp, "%d", true_indepth - 1);
2213 buf = tmp;
2214 break;
2217 case T_VERSION:
2219 char *tmp = (char *) alloca (3 + strlen (version_string));
2220 sprintf (tmp, "\"%s\"", version_string);
2221 buf = tmp;
2222 break;
2225 case T_CONST:
2226 buf = hp->value.cpval;
2227 break;
2229 case T_SPECLINE:
2231 char *tmp = (char *) alloca (10);
2232 sprintf (tmp, "%d", ip->lineno);
2233 buf = tmp;
2234 break;
2237 case T_DATE:
2238 case T_TIME:
2240 char *tmp = (char *) alloca (20);
2242 if (timebuf == NULL) {
2243 t = time (0);
2244 timebuf = localtime (&t);
2246 if (hp->type == T_DATE)
2247 sprintf (tmp, "\"%s %2d %4d\"", monthnames[timebuf->tm_mon],
2248 timebuf->tm_mday, timebuf->tm_year + 1900);
2249 else
2250 sprintf (tmp, "\"%02d:%02d:%02d\"", timebuf->tm_hour, timebuf->tm_min,
2251 timebuf->tm_sec);
2252 buf = tmp;
2253 break;
2256 case T_SPEC_DEFINED:
2257 buf = " 0 "; /* Assume symbol is not defined */
2258 ip = &instack[indepth];
2259 SKIP_WHITE_SPACE (ip->bufp);
2260 if (*ip->bufp == '(') {
2261 paren++;
2262 ip->bufp++; /* Skip over the paren */
2263 SKIP_WHITE_SPACE (ip->bufp);
2266 if (!is_idstart (*ip->bufp))
2267 goto oops;
2269 HASHNODE *hp = lookup (ip->bufp, -1, -1);
2271 if (hp && hp->type != T_UNUSED && hp->type != T_SPEC_DEFINED)
2272 buf = " 1 ";
2274 while (is_idchar (*ip->bufp))
2275 ++ip->bufp;
2276 SKIP_WHITE_SPACE (ip->bufp);
2277 if (paren) {
2278 if (*ip->bufp != ')')
2279 goto oops;
2280 ++ip->bufp;
2282 break;
2284 oops:
2286 error ("`defined' must be followed by ident or (ident)");
2287 break;
2289 default:
2290 error ("cccp error: invalid special hash type"); /* time for gdb */
2291 abort ();
2293 len = strlen (buf);
2294 check_expand (op, len);
2295 memcpy (op->bufp, buf, len);
2296 op->bufp += len;
2300 /* Routines to handle #directives */
2303 * Process include file by reading it in and calling rescan.
2304 * Expects to see "fname" or <fname> on the input.
2306 static void
2307 do_include (buf, limit, op)
2308 U_CHAR *buf, *limit;
2309 FILE_BUF *op;
2311 U_CHAR *fbeg, *fend; /* Beginning and end of fname */
2313 struct file_name_list *stackp = include; /* Chain of dirs to search */
2314 struct file_name_list dsp[1]; /* First in chain, if #include "..." */
2315 int flen;
2317 int retried = 0; /* Have already tried macro
2318 expanding the include line*/
2319 FILE_BUF trybuf; /* It got expanded into here */
2320 int system_header_p = 0; /* 0 for "...", 1 for <...> */
2322 get_filename:
2324 fbeg = buf;
2325 SKIP_WHITE_SPACE (fbeg);
2326 /* Discard trailing whitespace so we can easily see
2327 if we have parsed all the significant chars we were given. */
2328 while (limit != fbeg && is_nvspace (limit[-1])) limit--;
2330 switch (*fbeg++) {
2331 case '\"':
2332 fend = fbeg;
2333 while (fend != limit && *fend != '\"')
2334 fend++;
2335 if (*fend == '\"' && fend + 1 == limit) {
2336 FILE_BUF *fp;
2338 /* We have "filename". Figure out directory this source
2339 file is coming from and put it on the front of the list. */
2341 /* If -I- was specified, don't search current dir, only spec'd ones. */
2342 if (ignore_srcdir) break;
2344 for (fp = &instack[indepth]; fp >= instack; fp--)
2346 size_t n;
2347 const char *ep, *nam;
2349 if ((nam = fp->fname) != NULL) {
2350 /* Found a named file. Figure out dir of the file,
2351 and put it in front of the search list. */
2352 dsp[0].next = stackp;
2353 stackp = dsp;
2354 ep = strrchr (nam, '/');
2355 if (ep != NULL) {
2356 char *f;
2357 n = ep - nam;
2358 f = (char *) alloca (n + 1);
2359 strncpy (f, nam, n);
2360 f[n] = '\0';
2361 dsp[0].fname = f;
2362 if (n > max_include_len) max_include_len = n;
2363 } else {
2364 dsp[0].fname = 0; /* Current directory */
2366 break;
2369 break;
2371 goto fail;
2373 case '<':
2374 fend = fbeg;
2375 while (fend != limit && *fend != '>') fend++;
2376 if (*fend == '>' && fend + 1 == limit) {
2377 system_header_p = 1;
2378 /* If -I-, start with the first -I dir after the -I-. */
2379 if (first_bracket_include)
2380 stackp = first_bracket_include;
2381 break;
2383 goto fail;
2385 default:
2386 fail:
2387 if (retried) {
2388 error ("#include expects \"fname\" or <fname>");
2389 return;
2390 } else {
2391 trybuf = expand_to_temp_buffer (buf, limit, 0);
2392 buf = (U_CHAR *) alloca (trybuf.bufp - trybuf.buf + 1);
2393 memcpy (buf, trybuf.buf, trybuf.bufp - trybuf.buf);
2394 limit = buf + (trybuf.bufp - trybuf.buf);
2395 free (trybuf.buf);
2396 retried++;
2397 goto get_filename;
2401 flen = fend - fbeg;
2402 process_include (stackp, fbeg, flen, system_header_p, op);
2405 static void
2406 do_include_next (buf, limit, op)
2407 U_CHAR *buf, *limit;
2408 FILE_BUF *op;
2410 U_CHAR *fbeg, *fend; /* Beginning and end of fname */
2412 struct file_name_list *stackp; /* Chain of dirs to search */
2413 int flen;
2415 int retried = 0; /* Have already tried macro
2416 expanding the include line*/
2417 FILE_BUF trybuf; /* It got expanded into here */
2418 int system_header_p = 0; /* 0 for "...", 1 for <...> */
2420 /* Treat as plain #include if we don't know where to start
2421 looking. */
2422 stackp = instack[indepth].next_header_dir;
2423 if (stackp == 0)
2425 do_include (buf, limit, op);
2426 return;
2429 get_filename:
2431 fbeg = buf;
2432 SKIP_WHITE_SPACE (fbeg);
2433 /* Discard trailing whitespace so we can easily see
2434 if we have parsed all the significant chars we were given. */
2435 while (limit != fbeg && is_nvspace (limit[-1])) limit--;
2437 switch (*fbeg++) {
2438 case '\"':
2439 fend = fbeg;
2440 while (fend != limit && *fend != '\"')
2441 fend++;
2442 if (*fend == '\"' && fend + 1 == limit)
2443 break;
2444 goto fail;
2446 case '<':
2447 fend = fbeg;
2448 while (fend != limit && *fend != '>') fend++;
2449 if (*fend == '>' && fend + 1 == limit) {
2450 system_header_p = 1;
2451 break;
2453 goto fail;
2455 default:
2456 fail:
2457 if (retried) {
2458 error ("#include expects \"fname\" or <fname>");
2459 return;
2460 } else {
2461 trybuf = expand_to_temp_buffer (buf, limit, 0);
2462 buf = (U_CHAR *) alloca (trybuf.bufp - trybuf.buf + 1);
2463 memcpy (buf, trybuf.buf, trybuf.bufp - trybuf.buf);
2464 limit = buf + (trybuf.bufp - trybuf.buf);
2465 free (trybuf.buf);
2466 retried++;
2467 goto get_filename;
2471 flen = fend - fbeg;
2472 process_include (stackp, fbeg, flen, system_header_p, op);
2475 static void
2476 process_include (stackp, fbeg, flen, system_header_p, op)
2477 struct file_name_list *stackp;
2478 const U_CHAR *fbeg;
2479 int flen;
2480 int system_header_p;
2481 FILE_BUF *op;
2483 char *fname;
2484 int f = -1; /* file number */
2486 fname = (char *) alloca (max_include_len + flen + 2);
2487 /* + 2 above for slash and terminating null. */
2489 /* If specified file name is absolute, just open it. */
2491 if (IS_ABSOLUTE_PATHNAME (fbeg)) {
2492 strncpy (fname, (const char *)fbeg, flen);
2493 fname[flen] = 0;
2494 f = open (fname, O_RDONLY, 0666);
2495 } else {
2496 /* Search directory path, trying to open the file.
2497 Copy each filename tried into FNAME. */
2499 for (; stackp; stackp = stackp->next) {
2500 if (stackp->fname) {
2501 strcpy (fname, stackp->fname);
2502 strcat (fname, "/");
2503 fname[strlen (fname) + flen] = 0;
2504 } else {
2505 fname[0] = 0;
2507 strncat (fname, (const char *)fbeg, flen);
2508 if ((f = open (fname, O_RDONLY, 0666)) >= 0)
2509 break;
2513 if (f < 0) {
2514 strncpy (fname, (const char *)fbeg, flen);
2515 fname[flen] = 0;
2516 if (deps_missing_files
2517 && print_deps > (system_header_p || (system_include_depth > 0))) {
2519 /* If requested as a system header, assume it belongs in
2520 the first system header directory. */
2521 if (first_bracket_include)
2522 stackp = first_bracket_include;
2523 else
2524 stackp = include;
2526 if (!system_header_p || IS_ABSOLUTE_PATHNAME (fbeg) || !stackp->fname)
2527 deps_add_dep (deps, fname);
2528 else {
2529 char *p;
2530 int len = strlen(stackp->fname);
2532 p = (char *) alloca (len + flen + 2);
2533 memcpy (p, stackp->fname, len);
2534 p[len++] = '/';
2535 memcpy (p + len, fbeg, flen);
2536 len += flen;
2537 p[len] = '\0';
2538 deps_add_dep (deps, p);
2540 } else if (print_deps
2541 && print_deps <= (system_header_p
2542 || (system_include_depth > 0)))
2543 warning ("no include path in which to find %.*s", flen, fbeg);
2544 else
2545 error_from_errno (fname);
2547 } else {
2549 /* Check to see if this include file is a once-only include file.
2550 If so, give up. */
2552 struct file_name_list* ptr;
2554 for (ptr = dont_repeat_files; ptr; ptr = ptr->next) {
2555 if (!strcmp (ptr->fname, fname)) {
2556 close (f);
2557 return; /* This file was once'd. */
2561 for (ptr = all_include_files; ptr; ptr = ptr->next) {
2562 if (!strcmp (ptr->fname, fname))
2563 break; /* This file was included before. */
2566 if (ptr == 0) {
2567 /* This is the first time for this file. */
2568 /* Add it to list of files included. */
2570 ptr = (struct file_name_list *) xmalloc (sizeof (struct file_name_list));
2571 ptr->next = all_include_files;
2572 all_include_files = ptr;
2573 ptr->fname = xstrdup (fname);
2575 /* For -M, add this file to the dependencies. */
2576 if (print_deps > (system_header_p || (system_include_depth > 0)))
2577 deps_add_dep (deps, fname);
2580 if (system_header_p)
2581 system_include_depth++;
2583 /* Actually process the file. */
2584 finclude (f, fname, stackp->next, op);
2586 if (system_header_p)
2587 system_include_depth--;
2589 close (f);
2593 /* Replace all CR NL, NL CR and CR sequences with NL. */
2595 static void
2596 fixup_newlines (fp)
2597 FILE_BUF *fp;
2599 U_CHAR *p, *q, *end;
2601 if (fp->length <= 0)
2602 return;
2604 end = fp->buf + fp->length;
2605 *end = '\r';
2606 p = (U_CHAR *) strchr ((const char *) fp->buf, '\r');
2607 *end = '\0';
2608 if (p == end)
2609 return;
2611 if (p > fp->buf && p[-1] == '\n')
2612 p--;
2613 q = p;
2614 while (p < end)
2615 switch (*p)
2617 default:
2618 *q++ = *p++;
2619 break;
2620 case '\n':
2621 case '\r':
2622 p += 1 + (p[0] + p[1] == '\n' + '\r');
2623 *q++ = '\n';
2624 break;
2627 fp->length = q - fp->buf;
2630 /* Process the contents of include file FNAME, already open on descriptor F,
2631 with output to OP. */
2633 static void
2634 finclude (f, fname, nhd, op)
2635 int f;
2636 const char *fname;
2637 struct file_name_list *nhd;
2638 FILE_BUF *op;
2640 int st_mode;
2641 long st_size;
2642 long i;
2643 FILE_BUF *fp; /* For input stack frame */
2645 CHECK_DEPTH (return;);
2647 if (file_size_and_mode (f, &st_mode, &st_size))
2648 goto nope;
2650 fp = &instack[indepth + 1];
2651 memset (fp, 0, sizeof (FILE_BUF));
2652 fp->fname = fname;
2653 fp->length = 0;
2654 fp->lineno = 1;
2655 fp->if_stack = if_stack;
2656 fp->next_header_dir = nhd;
2658 if (S_ISREG (st_mode)) {
2659 fp->buf = (U_CHAR *) xmalloc (st_size + 2);
2660 fp->bufp = fp->buf;
2662 /* Read the file contents, knowing that st_size is an upper bound
2663 on the number of bytes we can read. */
2664 while (st_size > 0) {
2665 i = read (f, fp->buf + fp->length, st_size);
2666 if (i <= 0) {
2667 if (i == 0) break;
2668 goto nope;
2670 fp->length += i;
2671 st_size -= i;
2674 else {
2675 /* Cannot count its file size before reading. */
2677 U_CHAR *bufp;
2678 U_CHAR *basep;
2679 int bsize = 2000;
2681 st_size = 0;
2682 basep = (U_CHAR *) xmalloc (bsize + 2);
2683 bufp = basep;
2685 for (;;) {
2686 i = read (f, bufp, bsize - st_size);
2687 if (i < 0)
2688 goto nope; /* error! */
2689 if (i == 0)
2690 break; /* End of file */
2691 st_size += i;
2692 bufp += i;
2693 if (bsize == st_size) { /* Buffer is full! */
2694 bsize *= 2;
2695 basep = (U_CHAR *) xrealloc (basep, bsize + 2);
2696 bufp = basep + st_size; /* May have moved */
2699 fp->buf = basep;
2700 fp->bufp = fp->buf;
2701 fp->length = st_size;
2703 close (f);
2704 fixup_newlines (fp);
2706 /* Make sure data ends with a newline. And put a null after it. */
2708 if (fp->length > 0 && fp->buf[fp->length-1] != '\n')
2709 fp->buf[fp->length++] = '\n';
2710 fp->buf[fp->length] = '\0';
2712 indepth++;
2713 output_line_command (fp, op, 0, enter_file);
2714 rescan (op, 0);
2715 indepth--;
2716 instack[indepth].lineno++;
2717 instack[indepth].bufp++; /* Skip the new line. */
2718 output_line_command (&instack[indepth], op, 0, leave_file);
2719 free (fp->buf);
2720 return;
2722 nope:
2723 perror_with_name (fname);
2724 close (f);
2728 /* Process a #define command.
2729 BUF points to the contents of the #define command, as a continguous string.
2730 LIMIT points to the first character past the end of the definition.
2731 KEYWORD is the keyword-table entry for #define. */
2733 static void
2734 do_define (buf, limit, op)
2735 U_CHAR *buf, *limit;
2736 FILE_BUF *op ATTRIBUTE_UNUSED;
2738 U_CHAR *bp; /* temp ptr into input buffer */
2739 U_CHAR *symname; /* remember where symbol name starts */
2740 int sym_length; /* and how long it is */
2742 DEFINITION *defn;
2743 int arglengths = 0; /* Accumulate lengths of arg names
2744 plus number of args. */
2745 int hashcode;
2747 bp = buf;
2749 while (is_nvspace (*bp))
2750 bp++;
2752 symname = bp; /* remember where it starts */
2753 while (is_idchar (*bp) && bp < limit) {
2754 bp++;
2756 sym_length = bp - symname;
2757 if (sym_length == 0)
2759 error ("invalid macro name");
2760 return;
2762 else if (!is_idstart (*symname)) {
2763 U_CHAR *msg; /* what pain... */
2764 msg = (U_CHAR *) alloca (sym_length + 1);
2765 memcpy (msg, symname, sym_length);
2766 msg[sym_length] = 0;
2767 error ("invalid macro name `%s'", msg);
2768 return;
2769 } else {
2770 if (! strncmp ((const char *)symname, "defined", 7) && sym_length == 7)
2772 error ("\"defined\" cannot be used as a macro name");
2773 return;
2777 /* lossage will occur if identifiers or control keywords are broken
2778 across lines using backslash. This is not the right place to take
2779 care of that. */
2781 if (*bp == '(') {
2782 struct arglist *arg_ptrs = NULL;
2783 int argno = 0;
2785 bp++; /* skip '(' */
2786 SKIP_WHITE_SPACE (bp);
2788 /* Loop over macro argument names. */
2789 while (*bp != ')') {
2790 struct arglist *temp;
2792 temp = (struct arglist *) alloca (sizeof (struct arglist));
2793 temp->name = bp;
2794 temp->next = arg_ptrs;
2795 temp->argno = argno++;
2796 arg_ptrs = temp;
2798 if (!is_idstart (*bp))
2799 warning ("parameter name starts with a digit in #define");
2801 /* Find the end of the arg name. */
2802 while (is_idchar (*bp)) {
2803 bp++;
2805 temp->length = bp - temp->name;
2806 arglengths += temp->length + 2;
2807 SKIP_WHITE_SPACE (bp);
2808 if (temp->length == 0 || (*bp != ',' && *bp != ')')) {
2809 error ("badly punctuated parameter list in #define");
2810 return;
2812 if (*bp == ',') {
2813 bp++;
2814 SKIP_WHITE_SPACE (bp);
2816 if (bp >= limit) {
2817 error ("unterminated parameter list in #define");
2818 return;
2822 ++bp; /* skip paren */
2823 while (is_nvspace (*bp) && bp < limit) /* and leading whitespace */
2824 ++bp;
2825 /* now everything from bp before limit is the definition. */
2826 defn = collect_expansion (bp, limit, argno, arg_ptrs);
2828 /* Now set defn->argnames to the result of concatenating
2829 the argument names in reverse order
2830 with comma-space between them. */
2832 struct arglist *temp;
2833 int i = 0;
2834 U_CHAR *tmp = (U_CHAR *) xmalloc (arglengths + 1);
2836 for (temp = arg_ptrs; temp; temp = temp->next) {
2837 memcpy (&tmp[i], temp->name, temp->length);
2838 i += temp->length;
2839 if (temp->next != 0) {
2840 tmp[i++] = ',';
2841 tmp[i++] = ' ';
2844 tmp[i] = 0;
2845 defn->argnames = tmp;
2848 } else {
2849 /* simple expansion or empty definition; skip leading whitespace */
2850 while (is_nvspace (*bp) && bp < limit)
2851 ++bp;
2852 /* now everything from bp before limit is the definition. */
2853 defn = collect_expansion (bp, limit, -1, 0);
2854 defn->argnames = (const U_CHAR *) "";
2857 hashcode = hashf (symname, sym_length, HASHSIZE);
2860 HASHNODE *hp;
2861 if ((hp = lookup (symname, sym_length, hashcode)) == NULL)
2862 hp = install (symname, sym_length, T_MACRO, hashcode);
2863 else {
2864 if (hp->type != T_MACRO || compare_defs (defn, hp->value.defn))
2865 warning ("\"%.*s\" redefined", sym_length, symname);
2867 /* Replace the old definition. */
2868 hp->type = T_MACRO;
2871 hp->value.defn = defn;
2876 * return zero if two DEFINITIONs are isomorphic
2878 static int
2879 compare_defs (d1, d2)
2880 DEFINITION *d1, *d2;
2882 struct reflist *a1, *a2;
2883 U_CHAR *p1 = d1->expansion;
2884 U_CHAR *p2 = d2->expansion;
2885 int first = 1;
2887 if (d1->nargs != d2->nargs)
2888 return 1;
2889 if (strcmp ((const char *)d1->argnames, (const char *)d2->argnames))
2890 return 1;
2891 for (a1 = d1->pattern, a2 = d2->pattern; a1 && a2;
2892 a1 = a1->next, a2 = a2->next) {
2893 if (!((a1->nchars == a2->nchars
2894 && ! strncmp ((const char *)p1, (const char *)p2, a1->nchars))
2895 || ! comp_def_part (first, p1, a1->nchars, p2, a2->nchars, 0))
2896 || a1->argno != a2->argno
2897 || a1->stringify != a2->stringify
2898 || a1->raw_before != a2->raw_before
2899 || a1->raw_after != a2->raw_after)
2900 return 1;
2901 first = 0;
2902 p1 += a1->nchars;
2903 p2 += a2->nchars;
2905 if (a1 != a2)
2906 return 1;
2907 if (comp_def_part (first, p1, d1->length - (p1 - d1->expansion),
2908 p2, d2->length - (p2 - d2->expansion), 1))
2909 return 1;
2910 return 0;
2913 /* Return 1 if two parts of two macro definitions are effectively different.
2914 One of the parts starts at BEG1 and has LEN1 chars;
2915 the other has LEN2 chars at BEG2.
2916 Any sequence of whitespace matches any other sequence of whitespace.
2917 FIRST means these parts are the first of a macro definition;
2918 so ignore leading whitespace entirely.
2919 LAST means these parts are the last of a macro definition;
2920 so ignore trailing whitespace entirely. */
2921 static int
2922 comp_def_part (first, beg1, len1, beg2, len2, last)
2923 int first;
2924 const U_CHAR *beg1, *beg2;
2925 int len1, len2;
2926 int last;
2928 const U_CHAR *end1 = beg1 + len1;
2929 const U_CHAR *end2 = beg2 + len2;
2930 if (first) {
2931 while (beg1 != end1 && is_space (*beg1)) beg1++;
2932 while (beg2 != end2 && is_space (*beg2)) beg2++;
2934 if (last) {
2935 while (beg1 != end1 && is_space (end1[-1])) end1--;
2936 while (beg2 != end2 && is_space (end2[-1])) end2--;
2938 while (beg1 != end1 && beg2 != end2) {
2939 if (is_space (*beg1) && is_space (*beg2)) {
2940 while (beg1 != end1 && is_space (*beg1)) beg1++;
2941 while (beg2 != end2 && is_space (*beg2)) beg2++;
2942 } else if (*beg1 == *beg2) {
2943 beg1++; beg2++;
2944 } else break;
2946 return (beg1 != end1) || (beg2 != end2);
2949 /* Read a replacement list for a macro with parameters.
2950 Build the DEFINITION structure.
2951 Reads characters of text starting at BUF until LIMIT.
2952 ARGLIST specifies the formal parameters to look for
2953 in the text of the definition; NARGS is the number of args
2954 in that list, or -1 for a macro name that wants no argument list.
2955 MACRONAME is the macro name itself (so we can avoid recursive expansion)
2956 and NAMELEN is its length in characters.
2958 Note that comments and backslash-newlines have already been deleted
2959 from the argument. */
2961 /* Leading and trailing Space, Tab, etc. are converted to markers
2962 Newline Space, Newline Tab, etc.
2963 Newline Space makes a space in the final output
2964 but is discarded if stringified. (Newline Tab is similar but
2965 makes a Tab instead.)
2967 If there is no trailing whitespace, a Newline Space is added at the end
2968 to prevent concatenation that would be contrary to the standard. */
2970 static DEFINITION *
2971 collect_expansion (buf, end, nargs, arglist)
2972 U_CHAR *buf, *end;
2973 int nargs;
2974 struct arglist *arglist;
2976 DEFINITION *defn;
2977 U_CHAR *p, *limit, *lastp, *exp_p;
2978 struct reflist *endpat = NULL;
2979 /* Pointer to first nonspace after last ## seen. */
2980 U_CHAR *concat = 0;
2981 /* Pointer to first nonspace after last single-# seen. */
2982 U_CHAR *stringify = 0;
2983 int maxsize;
2984 int expected_delimiter = '\0';
2986 /* Scan thru the replacement list, ignoring comments and quoted
2987 strings, picking up on the macro calls. It does a linear search
2988 thru the arg list on every potential symbol. Profiling might say
2989 that something smarter should happen. */
2991 if (end < buf)
2992 abort ();
2994 /* Find the beginning of the trailing whitespace. */
2995 /* Find end of leading whitespace. */
2996 limit = end;
2997 p = buf;
2998 while (p < limit && is_space (limit[-1])) limit--;
2999 while (p < limit && is_space (*p)) p++;
3001 /* Allocate space for the text in the macro definition.
3002 Leading and trailing whitespace chars need 2 bytes each.
3003 Each other input char may or may not need 1 byte,
3004 so this is an upper bound.
3005 The extra 2 are for invented trailing newline-marker and final null. */
3006 maxsize = (sizeof (DEFINITION)
3007 + 2 * (end - limit) + 2 * (p - buf)
3008 + (limit - p) + 3);
3009 defn = (DEFINITION *) xcalloc (1, maxsize);
3011 defn->nargs = nargs;
3012 exp_p = defn->expansion = (U_CHAR *) defn + sizeof (DEFINITION);
3013 lastp = exp_p;
3015 p = buf;
3017 /* Convert leading whitespace to Newline-markers. */
3018 while (p < limit && is_space (*p)) {
3019 *exp_p++ = '\n';
3020 *exp_p++ = *p++;
3023 /* Process the main body of the definition. */
3024 while (p < limit) {
3025 int skipped_arg = 0;
3026 U_CHAR c = *p++;
3028 *exp_p++ = c;
3030 /* In -traditional mode, recognize arguments inside strings and
3031 and character constants, and ignore special properties of #.
3032 Arguments inside strings are considered "stringified", but no
3033 extra quote marks are supplied. */
3034 switch (c) {
3035 case '\'':
3036 case '\"':
3037 if (expected_delimiter != '\0') {
3038 if (c == expected_delimiter)
3039 expected_delimiter = '\0';
3040 } else
3041 expected_delimiter = c;
3042 break;
3044 case '\\':
3045 /* Backslash quotes delimiters and itself, but not macro args. */
3046 if (expected_delimiter != 0 && p < limit
3047 && (*p == expected_delimiter || *p == '\\')) {
3048 *exp_p++ = *p++;
3049 continue;
3051 break;
3053 case '/':
3054 if (expected_delimiter != '\0') /* No comments inside strings. */
3055 break;
3056 if (*p == '*') {
3057 /* If we find a comment that wasn't removed by handle_directive,
3058 this must be -traditional. So replace the comment with
3059 nothing at all. */
3060 exp_p--;
3061 p += 1;
3062 while (p < limit && !(p[-2] == '*' && p[-1] == '/'))
3063 p++;
3065 break;
3068 if (is_idchar (c) && nargs > 0) {
3069 U_CHAR *id_beg = p - 1;
3070 int id_len;
3072 --exp_p;
3073 while (p != limit && is_idchar (*p)) p++;
3074 id_len = p - id_beg;
3076 if (is_idstart (c)) {
3077 struct arglist *arg;
3079 for (arg = arglist; arg != NULL; arg = arg->next) {
3080 struct reflist *tpat;
3082 if (arg->name[0] == c
3083 && arg->length == id_len
3084 && strncmp ((const char *)arg->name,
3085 (const char *)id_beg, id_len) == 0) {
3086 /* make a pat node for this arg and append it to the end of
3087 the pat list */
3088 tpat = (struct reflist *) xmalloc (sizeof (struct reflist));
3089 tpat->next = NULL;
3090 tpat->raw_before = concat == id_beg;
3091 tpat->raw_after = 0;
3092 tpat->stringify = expected_delimiter != '\0';
3094 if (endpat == NULL)
3095 defn->pattern = tpat;
3096 else
3097 endpat->next = tpat;
3098 endpat = tpat;
3100 tpat->argno = arg->argno;
3101 tpat->nchars = exp_p - lastp;
3103 U_CHAR *p1 = p;
3104 SKIP_WHITE_SPACE (p1);
3105 if (p1 + 2 <= limit && p1[0] == '#' && p1[1] == '#')
3106 tpat->raw_after = 1;
3108 lastp = exp_p; /* place to start copying from next time */
3109 skipped_arg = 1;
3110 break;
3115 /* If this was not a macro arg, copy it into the expansion. */
3116 if (! skipped_arg) {
3117 U_CHAR *lim1 = p;
3118 p = id_beg;
3119 while (p != lim1)
3120 *exp_p++ = *p++;
3121 if (stringify == id_beg)
3122 error ("# operator should be followed by a macro argument name");
3127 if (limit < end) {
3128 /* Convert trailing whitespace to Newline-markers. */
3129 while (limit < end && is_space (*limit)) {
3130 *exp_p++ = '\n';
3131 *exp_p++ = *limit++;
3134 *exp_p = '\0';
3136 defn->length = exp_p - defn->expansion;
3138 /* Crash now if we overrun the allocated size. */
3139 if (defn->length + 1 > maxsize)
3140 abort ();
3142 return defn;
3146 * interpret #line command. Remembers previously seen fnames
3147 * in its very own hash table.
3149 #define FNAME_HASHSIZE 37
3150 static void
3151 do_line (buf, limit, op)
3152 U_CHAR *buf, *limit;
3153 FILE_BUF *op;
3155 U_CHAR *bp;
3156 FILE_BUF *ip = &instack[indepth];
3157 FILE_BUF tem;
3158 int new_lineno;
3159 enum file_change_code file_change = same_file;
3161 /* Expand any macros. */
3162 tem = expand_to_temp_buffer (buf, limit, 0);
3164 /* Point to macroexpanded line, which is null-terminated now. */
3165 bp = tem.buf;
3166 SKIP_WHITE_SPACE (bp);
3168 if (!ISDIGIT (*bp)) {
3169 error ("invalid format #line command");
3170 return;
3173 /* The Newline at the end of this line remains to be processed.
3174 To put the next line at the specified line number,
3175 we must store a line number now that is one less. */
3176 new_lineno = atoi ((const char *)bp);
3178 /* skip over the line number. */
3179 while (ISDIGIT (*bp))
3180 bp++;
3182 SKIP_WHITE_SPACE (bp);
3184 if (*bp == '\"') {
3185 static HASHNODE *fname_table[FNAME_HASHSIZE];
3186 HASHNODE *hp, **hash_bucket;
3187 U_CHAR *fname;
3188 int fname_length;
3190 fname = ++bp;
3192 while (*bp && *bp != '\"')
3193 bp++;
3194 if (*bp != '\"') {
3195 error ("invalid format #line command");
3196 return;
3199 fname_length = bp - fname;
3201 bp++;
3202 SKIP_WHITE_SPACE (bp);
3203 if (*bp) {
3204 if (*bp == '1')
3205 file_change = enter_file;
3206 else if (*bp == '2')
3207 file_change = leave_file;
3208 else {
3209 error ("invalid format #line command");
3210 return;
3213 bp++;
3214 SKIP_WHITE_SPACE (bp);
3215 if (*bp) {
3216 error ("invalid format #line command");
3217 return;
3221 hash_bucket =
3222 &fname_table[hashf (fname, fname_length, FNAME_HASHSIZE)];
3223 for (hp = *hash_bucket; hp != NULL; hp = hp->next)
3224 if (hp->length == fname_length &&
3225 strncmp (hp->value.cpval, (const char *)fname, fname_length) == 0) {
3226 ip->fname = hp->value.cpval;
3227 break;
3229 if (hp == 0) {
3230 char *q;
3231 /* Didn't find it; cons up a new one. */
3232 hp = (HASHNODE *) xcalloc (1, sizeof (HASHNODE) + fname_length + 1);
3233 hp->next = *hash_bucket;
3234 *hash_bucket = hp;
3236 hp->length = fname_length;
3237 ip->fname = hp->value.cpval = q = ((char *) hp) + sizeof (HASHNODE);
3238 memcpy (q, fname, fname_length);
3240 } else if (*bp) {
3241 error ("invalid format #line command");
3242 return;
3245 ip->lineno = new_lineno;
3246 output_line_command (ip, op, 0, file_change);
3247 ip->bufp++; /* Skip the new line. */
3248 check_expand (op, ip->length - (ip->bufp - ip->buf));
3252 * remove all definitions of symbol from symbol table.
3253 * according to un*x /lib/cpp, it is not an error to undef
3254 * something that has no definitions, so it isn't one here either.
3256 static void
3257 do_undef (buf, limit, op)
3258 U_CHAR *buf;
3259 U_CHAR *limit ATTRIBUTE_UNUSED;
3260 FILE_BUF *op ATTRIBUTE_UNUSED;
3262 HASHNODE *hp;
3264 SKIP_WHITE_SPACE (buf);
3266 if (! strncmp ((const char *)buf, "defined", 7) && ! is_idchar (buf[7]))
3267 warning ("undefining `defined'");
3269 while ((hp = lookup (buf, -1, -1)) != NULL) {
3270 if (hp->type != T_MACRO)
3271 warning ("undefining `%s'", hp->name);
3272 delete_macro (hp);
3276 /* Read the tokens of the answer into the macro pool. Only commit the
3277 memory if we intend it as permanent storage, i.e. the #assert case.
3278 Returns 0 on success. */
3280 static int
3281 parse_answer (buf, limit, answerp, type)
3282 const unsigned char *buf, *limit;
3283 struct answer **answerp;
3284 int type;
3286 const unsigned char *start;
3288 /* Skip leading whitespace. */
3289 if (buf < limit && *buf == ' ')
3290 buf++;
3292 /* Parentheses are optional here. */
3293 if (buf == limit && type == T_UNASSERT)
3294 return 0;
3296 if (buf == limit || *buf++ != '(')
3298 if (type == T_IF)
3299 return 0;
3301 error ("missing '(' after predicate");
3302 return 1;
3305 /* Drop whitespace at start. */
3306 while (buf < limit && *buf == ' ')
3307 buf++;
3309 start = buf;
3310 while (buf < limit && *buf != ')')
3311 buf++;
3313 if (buf == limit)
3315 error ("missing ')' to complete answer");
3316 return 1;
3319 if (buf == start)
3321 error ("predicate's answer is empty");
3322 return 1;
3325 if ((type == T_ASSERT || type == T_UNASSERT) && buf + 1 != limit)
3327 error ("extra text at end of directive");
3328 return 1;
3331 /* Lose trailing whitespace. */
3332 if (buf[-1] == ' ')
3333 buf--;
3335 *answerp = (struct answer *) xmalloc (sizeof (struct answer));
3336 (*answerp)->answer = start;
3337 (*answerp)->len = buf - start;
3339 return 0;
3342 /* Parses an assertion, returning a pointer to the hash node of the
3343 predicate, or 0 on error. If an answer was supplied, it is placed
3344 in ANSWERP, otherwise it is set to 0. */
3345 static HASHNODE *
3346 parse_assertion (buf, limit, answerp, type)
3347 const unsigned char *buf, *limit;
3348 struct answer **answerp;
3349 int type;
3351 HASHNODE *result = 0;
3352 const unsigned char *climit;
3353 unsigned char *bp, *symname = canonicalize_text (buf, limit, &climit);
3354 unsigned int len;
3356 bp = symname;
3357 if (bp < climit && is_idstart (*bp))
3360 bp++;
3361 while (bp < climit && is_idchar (*bp));
3363 len = bp - symname;
3365 *answerp = 0;
3366 if (len == 0)
3368 if (symname == climit)
3369 error ("assertion without predicate");
3370 else
3371 error ("predicate must be an identifier");
3373 /* Unfortunately, because of the way we handle #if, we don't avoid
3374 macro expansion in answers. This is not easy to fix. */
3375 else if (parse_answer (bp, climit, answerp, type) == 0)
3377 unsigned char *sym = alloca (len + 1);
3378 int hashcode;
3380 /* Prefix '#' to get it out of macro namespace. */
3381 sym[0] = '#';
3382 memcpy (sym + 1, symname, len);
3384 hashcode = hashf (sym, len + 1, HASHSIZE);
3385 result = lookup (sym, len + 1, hashcode);
3386 if (result == 0)
3387 result = install (sym, len + 1, T_UNUSED, hashcode);
3390 return result;
3393 /* Test an assertion within a preprocessor conditional. Returns zero
3394 on error or failure, one on success. */
3396 test_assertion (pbuf)
3397 unsigned char **pbuf; /* NUL-terminated. */
3399 unsigned char *buf = *pbuf;
3400 unsigned char *limit = buf + strlen ((char *) buf);
3401 struct answer *answer;
3402 HASHNODE *node;
3403 int result = 0;
3405 node = parse_assertion (buf, limit, &answer, T_IF);
3406 if (node)
3408 result = (node->type == T_ASSERT &&
3409 (answer == 0 || *find_answer (node, answer) != 0));
3411 /* Yuk. We update pbuf to point after the assertion test.
3412 First, move past the identifier. */
3413 if (is_space (*buf))
3414 buf++;
3415 while (is_idchar (*buf))
3416 buf++;
3417 /* If we have an answer, we need to move past the parentheses. */
3418 if (answer)
3419 while (*buf++ != ')')
3421 *pbuf = buf;
3424 return result;
3427 /* Handle a #error directive. */
3428 static void
3429 do_error (buf, limit, op)
3430 U_CHAR *buf;
3431 U_CHAR *limit;
3432 FILE_BUF *op ATTRIBUTE_UNUSED;
3434 error ("#error%.*s", (int) (limit - buf), buf);
3437 /* Handle a #warning directive. */
3438 static void
3439 do_warning (buf, limit, op)
3440 U_CHAR *buf;
3441 U_CHAR *limit;
3442 FILE_BUF *op ATTRIBUTE_UNUSED;
3444 warning ("#warning%.*s", (int) (limit - buf), buf);
3447 /* Handle a #assert directive. */
3448 static void
3449 do_assert (buf, limit, op)
3450 U_CHAR *buf;
3451 U_CHAR *limit;
3452 FILE_BUF *op ATTRIBUTE_UNUSED;
3454 struct answer *new_answer;
3455 HASHNODE *node;
3457 node = parse_assertion (buf, limit, &new_answer, T_ASSERT);
3458 if (node)
3460 /* Place the new answer in the answer list. First check there
3461 is not a duplicate. */
3462 new_answer->next = 0;
3463 if (node->type == T_ASSERT)
3465 if (*find_answer (node, new_answer))
3467 free (new_answer);
3468 warning ("\"%s\" re-asserted", node->name + 1);
3469 return;
3471 new_answer->next = node->value.answers;
3473 node->type = T_ASSERT;
3474 node->value.answers = new_answer;
3478 /* Function body to be provided later. */
3479 static void
3480 do_unassert (buf, limit, op)
3481 U_CHAR *buf;
3482 U_CHAR *limit;
3483 FILE_BUF *op ATTRIBUTE_UNUSED;
3485 HASHNODE *node;
3486 struct answer *answer;
3488 node = parse_assertion (buf, limit, &answer, T_UNASSERT);
3489 /* It isn't an error to #unassert something that isn't asserted. */
3490 if (node)
3492 if (node->type == T_ASSERT)
3494 if (answer)
3496 struct answer **p = find_answer (node, answer), *temp;
3498 /* Remove the answer from the list. */
3499 temp = *p;
3500 if (temp)
3501 *p = temp->next;
3503 /* Did we free the last answer? */
3504 if (node->value.answers == 0)
3505 delete_macro (node);
3507 else
3508 delete_macro (node);
3511 free (answer);
3515 /* Returns a pointer to the pointer to the answer in the answer chain,
3516 or a pointer to NULL if the answer is not in the chain. */
3517 static struct answer **
3518 find_answer (node, candidate)
3519 HASHNODE *node;
3520 const struct answer *candidate;
3522 struct answer **result;
3524 for (result = &node->value.answers; *result; result = &(*result)->next)
3526 struct answer *answer = *result;
3528 if (answer->len == candidate->len
3529 && !memcmp (answer->answer, candidate->answer, answer->len))
3530 break;
3533 return result;
3536 /* Return a malloced buffer with leading and trailing whitespace
3537 removed, and all instances of internal whitespace reduced to a
3538 single space. */
3539 static unsigned char *
3540 canonicalize_text (buf, limit, climit)
3541 const unsigned char *buf, *limit, **climit;
3543 unsigned int len = limit - buf;
3544 unsigned char *result = (unsigned char *) xmalloc (len), *dest;
3546 for (dest = result; buf < limit;)
3548 if (! is_space (*buf))
3549 *dest++ = *buf++;
3550 else
3552 while (++buf < limit && is_space (*buf))
3554 if (dest != result && buf != limit)
3555 *dest++ = ' ';
3559 *climit = dest;
3560 return result;
3564 * handle #if command by
3565 * 1) inserting special `defined' keyword into the hash table
3566 * that gets turned into 0 or 1 by special_symbol (thus,
3567 * if the luser has a symbol called `defined' already, it won't
3568 * work inside the #if command)
3569 * 2) rescan the input into a temporary output buffer
3570 * 3) pass the output buffer to the yacc parser and collect a value
3571 * 4) clean up the mess left from steps 1 and 2.
3572 * 5) call conditional_skip to skip til the next #endif (etc.),
3573 * or not, depending on the value from step 3.
3575 static void
3576 do_if (buf, limit, op)
3577 U_CHAR *buf, *limit;
3578 FILE_BUF *op ATTRIBUTE_UNUSED;
3580 int value;
3581 FILE_BUF *ip = &instack[indepth];
3583 value = eval_if_expression (buf, limit - buf);
3584 conditional_skip (ip, value == 0, T_IF);
3588 * handle a #elif directive by not changing if_stack either.
3589 * see the comment above do_else.
3591 static void
3592 do_elif (buf, limit, op)
3593 U_CHAR *buf, *limit;
3594 FILE_BUF *op;
3596 int value;
3597 FILE_BUF *ip = &instack[indepth];
3599 if (if_stack == instack[indepth].if_stack) {
3600 error ("#elif not within a conditional");
3601 return;
3602 } else {
3603 if (if_stack->type != T_IF && if_stack->type != T_ELIF) {
3604 error ("#elif after #else");
3605 fprintf (stderr, " (matches line %d", if_stack->lineno);
3606 if (if_stack->fname != NULL && ip->fname != NULL &&
3607 strcmp (if_stack->fname, ip->fname) != 0)
3608 fprintf (stderr, ", file %s", if_stack->fname);
3609 fprintf (stderr, ")\n");
3611 if_stack->type = T_ELIF;
3614 if (if_stack->if_succeeded)
3615 skip_if_group (ip, 0);
3616 else {
3617 value = eval_if_expression (buf, limit - buf);
3618 if (value == 0)
3619 skip_if_group (ip, 0);
3620 else {
3621 ++if_stack->if_succeeded; /* continue processing input */
3622 output_line_command (ip, op, 1, same_file);
3628 * evaluate a #if expression in BUF, of length LENGTH,
3629 * then parse the result as a C expression and return the value as an int.
3631 static int
3632 eval_if_expression (buf, length)
3633 const U_CHAR *buf;
3634 int length;
3636 FILE_BUF temp_obuf;
3637 HASHNODE *save_defined;
3638 int value;
3640 save_defined = install (U"defined", -1, T_SPEC_DEFINED, -1);
3641 temp_obuf = expand_to_temp_buffer (buf, buf + length, 0);
3642 delete_macro (save_defined); /* clean up special symbol */
3644 value = parse_c_expression ((const char *)temp_obuf.buf);
3646 free (temp_obuf.buf);
3648 return value;
3652 * routine to handle ifdef/ifndef. Try to look up the symbol,
3653 * then do or don't skip to the #endif/#else/#elif depending
3654 * on what directive is actually being processed.
3656 static void
3657 do_xifdef (buf, limit, type)
3658 U_CHAR *buf, *limit;
3659 enum node_type type;
3661 int skip;
3662 FILE_BUF *ip = &instack[indepth];
3663 U_CHAR *end;
3665 /* Discard leading and trailing whitespace. */
3666 SKIP_WHITE_SPACE (buf);
3667 while (limit != buf && is_nvspace (limit[-1])) limit--;
3669 /* Find the end of the identifier at the beginning. */
3670 for (end = buf; is_idchar (*end); end++);
3672 if (end == buf)
3673 skip = (type == T_IFDEF);
3674 else
3675 skip = (lookup (buf, end-buf, -1) == NULL) ^ (type == T_IFNDEF);
3677 conditional_skip (ip, skip, T_IF);
3680 static void
3681 do_ifdef (buf, limit, op)
3682 U_CHAR *buf, *limit;
3683 FILE_BUF *op ATTRIBUTE_UNUSED;
3685 do_xifdef (buf, limit, T_IFDEF);
3688 static void
3689 do_ifndef (buf, limit, op)
3690 U_CHAR *buf, *limit;
3691 FILE_BUF *op ATTRIBUTE_UNUSED;
3693 do_xifdef (buf, limit, T_IFNDEF);
3697 * push TYPE on stack; then, if SKIP is nonzero, skip ahead.
3699 static void
3700 conditional_skip (ip, skip, type)
3701 FILE_BUF *ip;
3702 int skip;
3703 enum node_type type;
3705 IF_STACK_FRAME *temp;
3707 temp = (IF_STACK_FRAME *) xcalloc (1, sizeof (IF_STACK_FRAME));
3708 temp->fname = ip->fname;
3709 temp->lineno = ip->lineno;
3710 temp->next = if_stack;
3711 if_stack = temp;
3713 if_stack->type = type;
3715 if (skip != 0) {
3716 skip_if_group (ip, 0);
3717 return;
3718 } else {
3719 ++if_stack->if_succeeded;
3720 output_line_command (ip, &outbuf, 1, same_file);
3725 * skip to #endif, #else, or #elif. adjust line numbers, etc.
3726 * leaves input ptr at the sharp sign found.
3727 * If ANY is nonzero, return at next directive of any sort.
3729 static void
3730 skip_if_group (ip, any)
3731 FILE_BUF *ip;
3732 int any;
3734 U_CHAR *bp = ip->bufp, *cp;
3735 U_CHAR *endb = ip->buf + ip->length;
3736 const struct directive *kt;
3737 IF_STACK_FRAME *save_if_stack = if_stack; /* don't pop past here */
3738 U_CHAR *beg_of_line = bp;
3740 while (bp < endb) {
3741 switch (*bp++) {
3742 case '/': /* possible comment */
3743 if (*bp == '\\' && bp[1] == '\n')
3744 newline_fix (bp);
3745 if (*bp == '*') {
3746 ip->bufp = ++bp;
3747 bp = skip_to_end_of_comment (ip, &ip->lineno);
3749 break;
3750 case '\"':
3751 case '\'':
3752 bp = skip_quoted_string (bp - 1, endb, ip->lineno, &ip->lineno, 0, 0);
3753 break;
3754 case '\\':
3755 /* Char after backslash loses its special meaning. */
3756 if (bp < endb) {
3757 if (*bp == '\n')
3758 ++ip->lineno; /* But do update the line-count. */
3759 bp++;
3761 break;
3762 case '\n':
3763 ++ip->lineno;
3764 beg_of_line = bp;
3765 break;
3766 case '#':
3767 ip->bufp = bp - 1;
3769 /* # keyword: a # must be first nonblank char on the line */
3770 if (beg_of_line == 0)
3771 break;
3772 /* Scan from start of line, skipping whitespace, comments
3773 and backslash-newlines, and see if we reach this #.
3774 If not, this # is not special. */
3775 bp = beg_of_line;
3776 while (1) {
3777 if (is_nvspace (*bp))
3778 bp++;
3779 else if (*bp == '\\' && bp[1] == '\n')
3780 bp += 2;
3781 else if (*bp == '/' && bp[1] == '*') {
3782 bp += 2;
3783 while (!(*bp == '*' && bp[1] == '/')) {
3784 if (*bp == '\n')
3785 ip->lineno++;
3786 bp++;
3788 bp += 2;
3790 else break;
3792 if (bp != ip->bufp) {
3793 bp = ip->bufp + 1; /* Reset bp to after the #. */
3794 break;
3797 bp = ip->bufp + 1; /* Point after '#'. */
3799 /* Skip whitespace and \-newline. */
3800 while (1) {
3801 if (is_nvspace (*bp))
3802 bp++;
3803 else if (*bp == '\\' && bp[1] == '\n')
3804 bp += 2;
3805 else if (*bp == '/' && bp[1] == '*') {
3806 bp += 2;
3807 while (!(*bp == '*' && bp[1] == '/'))
3808 bp++;
3809 bp += 2;
3811 else break;
3814 cp = bp;
3816 /* Now find end of directive name.
3817 If we encounter a backslash-newline, exchange it with any following
3818 symbol-constituents so that we end up with a contiguous name. */
3820 while (1) {
3821 if (is_idchar (*bp))
3822 bp++;
3823 else {
3824 if (*bp == '\\' && bp[1] == '\n')
3825 name_newline_fix (bp);
3826 if (is_idchar (*bp))
3827 bp++;
3828 else break;
3832 for (kt = directive_table; kt->length >= 0; kt++) {
3833 IF_STACK_FRAME *temp;
3834 if (strncmp ((const char *)cp, kt->name, kt->length) == 0
3835 && !is_idchar (cp[kt->length])) {
3837 /* If we are asked to return on next directive,
3838 do so now. */
3839 if (any)
3840 return;
3842 switch (kt->type) {
3843 case T_IF:
3844 case T_IFDEF:
3845 case T_IFNDEF:
3846 temp = (IF_STACK_FRAME *) xcalloc (1, sizeof (IF_STACK_FRAME));
3847 temp->next = if_stack;
3848 if_stack = temp;
3849 temp->lineno = ip->lineno;
3850 temp->fname = ip->fname;
3851 temp->type = kt->type;
3852 break;
3853 case T_ELSE:
3854 case T_ENDIF:
3855 case T_ELIF:
3856 if (if_stack == instack[indepth].if_stack) {
3857 error ("#%s not within a conditional", kt->name);
3858 break;
3860 else if (if_stack == save_if_stack)
3861 return; /* found what we came for */
3863 if (kt->type != T_ENDIF) {
3864 if (if_stack->type == T_ELSE)
3865 error ("#else or #elif after #else");
3866 if_stack->type = kt->type;
3867 break;
3870 temp = if_stack;
3871 if_stack = if_stack->next;
3872 free (temp);
3873 break;
3875 default:
3876 /* Anything else is ignored. */
3877 break;
3879 break;
3884 ip->bufp = bp;
3885 /* after this returns, rescan will exit because ip->bufp
3886 now points to the end of the buffer.
3887 rescan is responsible for the error message also. */
3891 * handle a #else directive. Do this by just continuing processing
3892 * without changing if_stack ; this is so that the error message
3893 * for missing #endif's etc. will point to the original #if. It
3894 * is possible that something different would be better.
3896 static void
3897 do_else (buf, limit, op)
3898 U_CHAR *buf ATTRIBUTE_UNUSED;
3899 U_CHAR *limit ATTRIBUTE_UNUSED;
3900 FILE_BUF *op;
3902 FILE_BUF *ip = &instack[indepth];
3904 if (if_stack == instack[indepth].if_stack) {
3905 error ("#else not within a conditional");
3906 return;
3907 } else {
3908 if (if_stack->type != T_IF && if_stack->type != T_ELIF) {
3909 error ("#else after #else");
3910 fprintf (stderr, " (matches line %d", if_stack->lineno);
3911 if (strcmp (if_stack->fname, ip->fname) != 0)
3912 fprintf (stderr, ", file %s", if_stack->fname);
3913 fprintf (stderr, ")\n");
3915 if_stack->type = T_ELSE;
3918 if (if_stack->if_succeeded)
3919 skip_if_group (ip, 0);
3920 else {
3921 ++if_stack->if_succeeded; /* continue processing input */
3922 output_line_command (ip, op, 1, same_file);
3927 * unstack after #endif command
3929 static void
3930 do_endif (buf, limit, op)
3931 U_CHAR *buf ATTRIBUTE_UNUSED;
3932 U_CHAR *limit ATTRIBUTE_UNUSED;
3933 FILE_BUF *op;
3935 if (if_stack == instack[indepth].if_stack)
3936 error ("unbalanced #endif");
3937 else {
3938 IF_STACK_FRAME *temp = if_stack;
3939 if_stack = if_stack->next;
3940 free (temp);
3941 output_line_command (&instack[indepth], op, 1, same_file);
3946 * Skip a comment, assuming the input ptr immediately follows the
3947 * initial slash-star. Bump line counter as necessary.
3948 * (The canonical line counter is &ip->lineno).
3949 * Don't use this routine (or the next one) if bumping the line
3950 * counter is not sufficient to deal with newlines in the string.
3952 static U_CHAR *
3953 skip_to_end_of_comment (ip, line_counter)
3954 FILE_BUF *ip;
3955 int *line_counter; /* place to remember newlines, or NULL */
3957 U_CHAR *limit = ip->buf + ip->length;
3958 U_CHAR *bp = ip->bufp;
3959 FILE_BUF *op = &outbuf; /* JF */
3960 int output = put_out_comments && !line_counter;
3962 /* JF this line_counter stuff is a crock to make sure the
3963 comment is only put out once, no matter how many times
3964 the comment is skipped. It almost works */
3965 if (output) {
3966 *op->bufp++ = '/';
3967 *op->bufp++ = '*';
3969 while (bp < limit) {
3970 if (output)
3971 *op->bufp++ = *bp;
3972 switch (*bp++) {
3973 case '/':
3974 if (warn_comments && bp < limit && *bp == '*')
3975 warning("`/*' within comment");
3976 break;
3977 case '\n':
3978 if (line_counter != NULL)
3979 ++*line_counter;
3980 if (output)
3981 ++op->lineno;
3982 break;
3983 case '*':
3984 if (*bp == '\\' && bp[1] == '\n')
3985 newline_fix (bp);
3986 if (*bp == '/') {
3987 if (output)
3988 *op->bufp++ = '/';
3989 ip->bufp = ++bp;
3990 return bp;
3992 break;
3995 ip->bufp = bp;
3996 return bp;
4000 * Skip over a quoted string. BP points to the opening quote.
4001 * Returns a pointer after the closing quote. Don't go past LIMIT.
4002 * START_LINE is the line number of the starting point (but it need
4003 * not be valid if the starting point is inside a macro expansion).
4005 * The input stack state is not changed.
4007 * If COUNT_NEWLINES is nonzero, it points to an int to increment
4008 * for each newline passed.
4010 * If BACKSLASH_NEWLINES_P is nonzero, store 1 thru it
4011 * if we pass a backslash-newline.
4013 * If EOFP is nonzero, set *EOFP to 1 if the string is unterminated.
4015 static U_CHAR *
4016 skip_quoted_string (bp, limit, start_line, count_newlines, backslash_newlines_p, eofp)
4017 const U_CHAR *bp;
4018 const U_CHAR *limit;
4019 int start_line;
4020 int *count_newlines;
4021 int *backslash_newlines_p;
4022 int *eofp;
4024 U_CHAR c, match;
4026 match = *bp++;
4027 while (1) {
4028 if (bp >= limit) {
4029 error_with_line (line_for_error (start_line),
4030 "unterminated string or character constant");
4031 if (eofp)
4032 *eofp = 1;
4033 break;
4035 c = *bp++;
4036 if (c == '\\') {
4037 while (*bp == '\\' && bp[1] == '\n') {
4038 if (backslash_newlines_p)
4039 *backslash_newlines_p = 1;
4040 if (count_newlines)
4041 ++*count_newlines;
4042 bp += 2;
4044 if (*bp == '\n' && count_newlines) {
4045 if (backslash_newlines_p)
4046 *backslash_newlines_p = 1;
4047 ++*count_newlines;
4049 bp++;
4050 } else if (c == '\n') {
4051 /* Unterminated strings and character constants are 'legal'. */
4052 bp--; /* Don't consume the newline. */
4053 if (eofp)
4054 *eofp = 1;
4055 break;
4056 } else if (c == match)
4057 break;
4059 return (U_CHAR *) bp;
4063 * write out a #line command, for instance, after an #include file.
4064 * If CONDITIONAL is nonzero, we can omit the #line if it would
4065 * appear to be a no-op, and we can output a few newlines instead
4066 * if we want to increase the line number by a small amount.
4067 * FILE_CHANGE says whether we are entering a file, leaving, or neither.
4070 static void
4071 output_line_command (ip, op, conditional, file_change)
4072 FILE_BUF *ip, *op;
4073 int conditional;
4074 enum file_change_code file_change;
4076 int len;
4077 char line_cmd_buf[500];
4079 if (no_line_commands
4080 || ip->fname == NULL
4081 || no_output) {
4082 op->lineno = ip->lineno;
4083 return;
4086 if (conditional) {
4087 if (ip->lineno == op->lineno)
4088 return;
4090 /* If the inherited line number is a little too small,
4091 output some newlines instead of a #line command. */
4092 if (ip->lineno > op->lineno && ip->lineno < op->lineno + 8) {
4093 check_expand (op, 10);
4094 while (ip->lineno > op->lineno) {
4095 *op->bufp++ = '\n';
4096 op->lineno++;
4098 return;
4102 sprintf (line_cmd_buf, "# %d \"%s\"", ip->lineno, ip->fname);
4103 if (file_change != same_file)
4104 strcat (line_cmd_buf, file_change == enter_file ? " 1" : " 2");
4105 if (system_include_depth > 0)
4106 strcat (line_cmd_buf, " 3");
4107 len = strlen (line_cmd_buf);
4108 line_cmd_buf[len++] = '\n';
4109 check_expand (op, len + 1);
4110 if (op->bufp > op->buf && op->bufp[-1] != '\n')
4111 *op->bufp++ = '\n';
4112 memcpy (op->bufp, line_cmd_buf, len);
4113 op->bufp += len;
4114 op->lineno = ip->lineno;
4118 /* Expand a macro call.
4119 HP points to the symbol that is the macro being called.
4120 Put the result of expansion onto the input stack
4121 so that subsequent input by our caller will use it.
4123 If macro wants arguments, caller has already verified that
4124 an argument list follows; arguments come from the input stack. */
4126 static void
4127 macroexpand (hp, op)
4128 HASHNODE *hp;
4129 FILE_BUF *op;
4131 int nargs;
4132 DEFINITION *defn = hp->value.defn;
4133 U_CHAR *xbuf;
4134 int xbuf_len;
4135 int start_line = instack[indepth].lineno;
4137 CHECK_DEPTH (return;);
4139 /* it might not actually be a macro. */
4140 if (hp->type != T_MACRO) {
4141 special_symbol (hp, op);
4142 return;
4145 nargs = defn->nargs;
4147 if (nargs >= 0) {
4148 int i;
4149 struct argdata *args;
4150 const char *parse_error = 0;
4152 args = (struct argdata *) alloca ((nargs + 1) * sizeof (struct argdata));
4154 for (i = 0; i < nargs; i++) {
4155 args[i].raw = args[i].expanded = (U_CHAR *) "";
4156 args[i].raw_length = args[i].expand_length
4157 = args[i].stringified_length = 0;
4158 args[i].free1 = args[i].free2 = 0;
4161 /* Parse all the macro args that are supplied. I counts them.
4162 The first NARGS args are stored in ARGS.
4163 The rest are discarded. */
4164 i = 0;
4165 do {
4166 /* Discard the open-parenthesis or comma before the next arg. */
4167 ++instack[indepth].bufp;
4168 parse_error
4169 = macarg ((i < nargs || (nargs == 0 && i == 0)) ? &args[i] : 0);
4170 if (parse_error)
4172 error_with_line (line_for_error (start_line), "%s", parse_error);
4173 break;
4175 i++;
4176 } while (*instack[indepth].bufp != ')');
4178 /* If we got one arg but it was just whitespace, call that 0 args. */
4179 if (i == 1) {
4180 const U_CHAR *bp = args[0].raw;
4181 const U_CHAR *lim = bp + args[0].raw_length;
4182 while (bp != lim && is_space (*bp)) bp++;
4183 if (bp == lim)
4184 i = 0;
4187 if (nargs == 0 && i > 0)
4188 error ("arguments given to macro `%s'", hp->name);
4189 else if (i < nargs) {
4190 /* traditional C allows foo() if foo wants one argument. */
4191 if (nargs == 1 && i == 0)
4193 else if (i == 0)
4194 error ("no args to macro `%s'", hp->name);
4195 else if (i == 1)
4196 error ("only 1 arg to macro `%s'", hp->name);
4197 else
4198 error ("only %d args to macro `%s'", i, hp->name);
4199 } else if (i > nargs)
4200 error ("too many (%d) args to macro `%s'", i, hp->name);
4202 /* Swallow the closeparen. */
4203 ++instack[indepth].bufp;
4205 /* If macro wants zero args, we parsed the arglist for checking only.
4206 Read directly from the macro definition. */
4207 if (nargs == 0) {
4208 xbuf = defn->expansion;
4209 xbuf_len = defn->length;
4210 } else {
4211 U_CHAR *exp = defn->expansion;
4212 int offset; /* offset in expansion,
4213 copied a piece at a time */
4214 int totlen; /* total amount of exp buffer filled so far */
4216 struct reflist *ap;
4218 /* Macro really takes args. Compute the expansion of this call. */
4220 /* Compute length in characters of the macro's expansion. */
4221 xbuf_len = defn->length;
4222 for (ap = defn->pattern; ap != NULL; ap = ap->next) {
4223 if (ap->stringify)
4224 xbuf_len += args[ap->argno].stringified_length;
4225 else
4226 xbuf_len += args[ap->argno].raw_length;
4229 xbuf = (U_CHAR *) xmalloc (xbuf_len + 1);
4231 /* Generate in XBUF the complete expansion
4232 with arguments substituted in.
4233 TOTLEN is the total size generated so far.
4234 OFFSET is the index in the definition
4235 of where we are copying from. */
4236 offset = totlen = 0;
4237 for (ap = defn->pattern; ap != NULL; ap = ap->next) {
4238 struct argdata *arg = &args[ap->argno];
4240 for (i = 0; i < ap->nchars; i++)
4241 xbuf[totlen++] = exp[offset++];
4243 if (ap->stringify != 0) {
4244 int arglen = arg->raw_length;
4245 int escaped = 0;
4246 int in_string = 0;
4247 int c;
4248 i = 0;
4249 while (i < arglen
4250 && (c = arg->raw[i], is_space (c)))
4251 i++;
4252 while (i < arglen
4253 && (c = arg->raw[arglen - 1], is_space (c)))
4254 arglen--;
4255 for (; i < arglen; i++) {
4256 c = arg->raw[i];
4258 /* Special markers Newline Space
4259 generate nothing for a stringified argument. */
4260 if (c == '\n' && arg->raw[i+1] != '\n') {
4261 i++;
4262 continue;
4265 /* Internal sequences of whitespace are replaced by one space
4266 except within an string or char token. */
4267 if (! in_string
4268 && (c == '\n' ? arg->raw[i+1] == '\n' : is_space (c))) {
4269 while (1) {
4270 /* Note that Newline Space does occur within whitespace
4271 sequences; consider it part of the sequence. */
4272 if (c == '\n' && is_space (arg->raw[i+1]))
4273 i += 2;
4274 else if (c != '\n' && is_space (c))
4275 i++;
4276 else break;
4277 c = arg->raw[i];
4279 i--;
4280 c = ' ';
4283 if (escaped)
4284 escaped = 0;
4285 else {
4286 if (c == '\\')
4287 escaped = 1;
4288 if (in_string) {
4289 if (c == in_string)
4290 in_string = 0;
4291 } else if (c == '\"' || c == '\'')
4292 in_string = c;
4295 /* Escape these chars */
4296 if (c == '\"' || (in_string && c == '\\'))
4297 xbuf[totlen++] = '\\';
4298 if (ISPRINT (c))
4299 xbuf[totlen++] = c;
4300 else {
4301 sprintf ((char *) &xbuf[totlen], "\\%03o", (unsigned int) c);
4302 totlen += 4;
4305 } else {
4306 const U_CHAR *p1 = arg->raw;
4307 const U_CHAR *l1 = p1 + arg->raw_length;
4309 if (ap->raw_before) {
4310 while (p1 != l1 && is_space (*p1)) p1++;
4311 while (p1 != l1 && is_idchar (*p1))
4312 xbuf[totlen++] = *p1++;
4313 /* Delete any no-reexpansion marker that follows
4314 an identifier at the beginning of the argument
4315 if the argument is concatenated with what precedes it. */
4316 if (p1[0] == '\n' && p1[1] == '-')
4317 p1 += 2;
4319 if (ap->raw_after) {
4320 /* Arg is concatenated after: delete trailing whitespace,
4321 whitespace markers, and no-reexpansion markers. */
4322 while (p1 != l1) {
4323 if (is_space (l1[-1])) l1--;
4324 else if (l1[-1] == '-') {
4325 const U_CHAR *p2 = l1 - 1;
4326 /* If a `-' is preceded by an odd number of newlines then it
4327 and the last newline are a no-reexpansion marker. */
4328 while (p2 != p1 && p2[-1] == '\n') p2--;
4329 if ((l1 - 1 - p2) & 1) {
4330 l1 -= 2;
4332 else break;
4334 else break;
4337 memmove (xbuf + totlen, p1, l1 - p1);
4338 totlen += l1 - p1;
4341 if (totlen > xbuf_len)
4342 abort ();
4345 /* if there is anything left of the definition
4346 after handling the arg list, copy that in too. */
4348 for (i = offset; i < defn->length; i++)
4349 xbuf[totlen++] = exp[i];
4351 xbuf[totlen] = 0;
4352 xbuf_len = totlen;
4354 for (i = 0; i < nargs; i++) {
4355 if (args[i].free1 != 0)
4356 free (args[i].free1);
4357 if (args[i].free2 != 0)
4358 free (args[i].free2);
4361 } else {
4362 xbuf = defn->expansion;
4363 xbuf_len = defn->length;
4366 /* Now put the expansion on the input stack
4367 so our caller will commence reading from it. */
4369 FILE_BUF *ip2;
4371 ip2 = &instack[++indepth];
4373 ip2->fname = 0;
4374 ip2->lineno = 0;
4375 ip2->buf = xbuf;
4376 ip2->length = xbuf_len;
4377 ip2->bufp = xbuf;
4378 ip2->free_ptr = (nargs > 0) ? xbuf : 0;
4379 ip2->macro = hp;
4380 ip2->if_stack = if_stack;
4385 * Parse a macro argument and store the info on it into *ARGPTR.
4386 * Return nonzero to indicate a syntax error.
4389 static const char *
4390 macarg (argptr)
4391 struct argdata *argptr;
4393 FILE_BUF *ip = &instack[indepth];
4394 int paren = 0;
4395 int newlines = 0;
4396 int comments = 0;
4398 /* Try to parse as much of the argument as exists at this
4399 input stack level. */
4400 U_CHAR *bp = macarg1 (ip->bufp, ip->buf + ip->length,
4401 &paren, &newlines, &comments);
4403 /* If we find the end of the argument at this level,
4404 set up *ARGPTR to point at it in the input stack. */
4405 if (!(ip->fname != 0 && (newlines != 0 || comments != 0))
4406 && bp != ip->buf + ip->length) {
4407 if (argptr != 0) {
4408 argptr->raw = ip->bufp;
4409 argptr->raw_length = bp - ip->bufp;
4411 ip->bufp = bp;
4412 } else {
4413 /* This input stack level ends before the macro argument does.
4414 We must pop levels and keep parsing.
4415 Therefore, we must allocate a temporary buffer and copy
4416 the macro argument into it. */
4417 int bufsize = bp - ip->bufp;
4418 int extra = newlines;
4419 U_CHAR *buffer = (U_CHAR *) xmalloc (bufsize + extra + 1);
4420 int final_start = 0;
4422 memcpy (buffer, ip->bufp, bufsize);
4423 ip->bufp = bp;
4424 ip->lineno += newlines;
4426 while (bp == ip->buf + ip->length) {
4427 if (instack[indepth].macro == 0) {
4428 free (buffer);
4429 return "unterminated macro call";
4431 ip->macro->type = T_MACRO;
4432 if (ip->free_ptr)
4433 free (ip->free_ptr);
4434 ip = &instack[--indepth];
4435 newlines = 0;
4436 comments = 0;
4437 bp = macarg1 (ip->bufp, ip->buf + ip->length, &paren,
4438 &newlines, &comments);
4439 final_start = bufsize;
4440 bufsize += bp - ip->bufp;
4441 extra += newlines;
4442 buffer = (U_CHAR *) xrealloc (buffer, bufsize + extra + 1);
4443 memcpy (buffer + bufsize - (bp - ip->bufp), ip->bufp, bp - ip->bufp);
4444 ip->bufp = bp;
4445 ip->lineno += newlines;
4448 /* Now, if arg is actually wanted, record its raw form,
4449 discarding comments and duplicating newlines in whatever
4450 part of it did not come from a macro expansion.
4451 EXTRA space has been preallocated for duplicating the newlines.
4452 FINAL_START is the index of the start of that part. */
4453 if (argptr != 0) {
4454 argptr->raw = buffer;
4455 argptr->raw_length = bufsize;
4456 argptr->free1 = buffer;
4457 argptr->newlines = newlines;
4458 argptr->comments = comments;
4459 if ((newlines || comments) && ip->fname != 0)
4460 argptr->raw_length
4461 = final_start +
4462 discard_comments (argptr->raw + final_start,
4463 argptr->raw_length - final_start,
4464 newlines);
4465 argptr->raw[argptr->raw_length] = 0;
4466 if (argptr->raw_length > bufsize + extra)
4467 abort ();
4471 /* If we are not discarding this argument,
4472 macroexpand it and compute its length as stringified.
4473 All this info goes into *ARGPTR. */
4475 if (argptr != 0) {
4476 FILE_BUF obuf;
4477 const U_CHAR *buf, *lim;
4478 int totlen;
4480 obuf = expand_to_temp_buffer (argptr->raw,
4481 argptr->raw + argptr->raw_length,
4484 argptr->expanded = obuf.buf;
4485 argptr->expand_length = obuf.length;
4486 argptr->free2 = obuf.buf;
4488 buf = argptr->raw;
4489 lim = buf + argptr->raw_length;
4491 totlen = 0;
4492 while (buf != lim) {
4493 U_CHAR c = *buf++;
4494 totlen++;
4495 /* Internal sequences of whitespace are replaced by one space
4496 in most cases, but not always. So count all the whitespace
4497 in case we need to keep it all. */
4498 if (c == '\"' || c == '\\') /* escape these chars */
4499 totlen++;
4500 else if (!ISPRINT (c))
4501 totlen += 3;
4503 argptr->stringified_length = totlen;
4505 return 0;
4508 /* Scan text from START (inclusive) up to LIMIT (exclusive),
4509 counting parens in *DEPTHPTR,
4510 and return if reach LIMIT
4511 or before a `)' that would make *DEPTHPTR negative
4512 or before a comma when *DEPTHPTR is zero.
4513 Single and double quotes are matched and termination
4514 is inhibited within them. Comments also inhibit it.
4515 Value returned is pointer to stopping place.
4517 Increment *NEWLINES each time a newline is passed.
4518 Set *COMMENTS to 1 if a comment is seen. */
4520 static U_CHAR *
4521 macarg1 (start, limit, depthptr, newlines, comments)
4522 U_CHAR *start;
4523 const U_CHAR *limit;
4524 int *depthptr, *newlines, *comments;
4526 U_CHAR *bp = start;
4528 while (bp < limit) {
4529 switch (*bp) {
4530 case '(':
4531 (*depthptr)++;
4532 break;
4533 case ')':
4534 if (--(*depthptr) < 0)
4535 return bp;
4536 break;
4537 case '\\':
4538 /* Traditionally, backslash makes following char not special. */
4539 if (bp + 1 < limit)
4541 bp++;
4542 /* But count source lines anyway. */
4543 if (*bp == '\n')
4544 ++*newlines;
4546 break;
4547 case '\n':
4548 ++*newlines;
4549 break;
4550 case '/':
4551 if (bp[1] == '\\' && bp[2] == '\n')
4552 newline_fix (bp + 1);
4553 if (bp[1] != '*' || bp + 1 >= limit)
4554 break;
4555 *comments = 1;
4556 bp += 2;
4557 while (bp + 1 < limit) {
4558 if (bp[0] == '*'
4559 && bp[1] == '\\' && bp[2] == '\n')
4560 newline_fix (bp + 1);
4561 if (bp[0] == '*' && bp[1] == '/')
4562 break;
4563 if (*bp == '\n') ++*newlines;
4564 bp++;
4566 bp += 1;
4567 break;
4568 case '\'':
4569 case '\"':
4571 int quotec;
4572 for (quotec = *bp++; bp + 1 < limit && *bp != quotec; bp++) {
4573 if (*bp == '\\') {
4574 bp++;
4575 if (*bp == '\n')
4576 ++*newlines;
4577 while (*bp == '\\' && bp[1] == '\n') {
4578 bp += 2;
4580 } else if (*bp == '\n') {
4581 ++*newlines;
4582 if (quotec == '\'')
4583 break;
4587 break;
4588 case ',':
4589 if ((*depthptr) == 0)
4590 return bp;
4591 break;
4593 bp++;
4596 return bp;
4599 /* Discard comments and duplicate newlines
4600 in the string of length LENGTH at START,
4601 except inside of string constants.
4602 The string is copied into itself with its beginning staying fixed.
4604 NEWLINES is the number of newlines that must be duplicated.
4605 We assume that that much extra space is available past the end
4606 of the string. */
4608 static int
4609 discard_comments (start, length, newlines)
4610 U_CHAR *start;
4611 int length;
4612 int newlines;
4614 U_CHAR *ibp;
4615 U_CHAR *obp;
4616 const U_CHAR *limit;
4617 int c;
4619 /* If we have newlines to duplicate, copy everything
4620 that many characters up. Then, in the second part,
4621 we will have room to insert the newlines
4622 while copying down.
4623 NEWLINES may actually be too large, because it counts
4624 newlines in string constants, and we don't duplicate those.
4625 But that does no harm. */
4626 if (newlines > 0) {
4627 ibp = start + length;
4628 obp = ibp + newlines;
4629 limit = start;
4630 while (limit != ibp)
4631 *--obp = *--ibp;
4634 ibp = start + newlines;
4635 limit = start + length + newlines;
4636 obp = start;
4638 while (ibp < limit) {
4639 *obp++ = c = *ibp++;
4640 switch (c) {
4641 case '\n':
4642 /* Duplicate the newline. */
4643 *obp++ = '\n';
4644 break;
4646 case '\\':
4647 if (*ibp == '\n') {
4648 obp--;
4649 ibp++;
4651 break;
4653 case '/':
4654 if (*ibp == '\\' && ibp[1] == '\n')
4655 newline_fix (ibp);
4656 /* Delete any comment. */
4657 if (ibp[0] != '*' || ibp + 1 >= limit)
4658 break;
4659 obp--;
4660 ibp++;
4661 while (ibp + 1 < limit) {
4662 if (ibp[0] == '*'
4663 && ibp[1] == '\\' && ibp[2] == '\n')
4664 newline_fix (ibp + 1);
4665 if (ibp[0] == '*' && ibp[1] == '/')
4666 break;
4667 ibp++;
4669 ibp += 2;
4670 break;
4672 case '\'':
4673 case '\"':
4674 /* Notice and skip strings, so that we don't
4675 think that comments start inside them,
4676 and so we don't duplicate newlines in them. */
4678 int quotec = c;
4679 while (ibp < limit) {
4680 *obp++ = c = *ibp++;
4681 if (c == quotec)
4682 break;
4683 if (c == '\n' && quotec == '\'')
4684 break;
4685 if (c == '\\' && ibp < limit) {
4686 while (*ibp == '\\' && ibp[1] == '\n')
4687 ibp += 2;
4688 *obp++ = *ibp++;
4692 break;
4696 return obp - start;
4700 /* Core error handling routine. */
4701 static void
4702 v_message (mtype, line, msgid, ap)
4703 enum msgtype mtype;
4704 int line;
4705 const char *msgid;
4706 va_list ap;
4708 const char *fname = 0;
4709 int i;
4711 if (mtype == MT_WARNING && inhibit_warnings)
4712 return;
4714 for (i = indepth; i >= 0; i--)
4715 if (instack[i].fname != NULL) {
4716 if (line == 0)
4717 line = instack[i].lineno;
4718 fname = instack[i].fname;
4719 break;
4722 if (fname)
4723 fprintf (stderr, "%s:%d: ", fname, line);
4724 else
4725 fprintf (stderr, "%s: ", progname);
4727 if (mtype == MT_WARNING)
4728 fputs (_("warning: "), stderr);
4730 vfprintf (stderr, _(msgid), ap);
4731 putc ('\n', stderr);
4733 if (mtype == MT_ERROR)
4734 errors++;
4738 * error - print error message and increment count of errors.
4740 void
4741 error VPARAMS ((const char *msgid, ...))
4743 VA_OPEN(ap, msgid);
4744 VA_FIXEDARG (ap, const char *, msgid);
4746 v_message (MT_ERROR, 0, msgid, ap);
4747 VA_CLOSE (ap);
4750 void
4751 error_with_line VPARAMS ((int line, const char *msgid, ...))
4753 VA_OPEN(ap, msgid);
4754 VA_FIXEDARG (ap, int, line);
4755 VA_FIXEDARG (ap, const char *, msgid);
4757 v_message (MT_ERROR, line, msgid, ap);
4758 VA_CLOSE (ap);
4761 /* Error including a message from `errno'. */
4762 void
4763 error_from_errno (name)
4764 const char *name;
4766 error ("%s: %s", name, strerror (errno));
4769 /* Print error message but don't count it. */
4770 void
4771 warning VPARAMS ((const char *msgid, ...))
4773 VA_OPEN(ap, msgid);
4774 VA_FIXEDARG (ap, const char *, msgid);
4776 v_message (MT_WARNING, 0, msgid, ap);
4777 VA_CLOSE (ap);
4780 void
4781 fatal VPARAMS ((const char *msgid, ...))
4783 VA_OPEN(ap, msgid);
4784 VA_FIXEDARG (ap, const char *, msgid);
4786 v_message (MT_FATAL, 0, msgid, ap);
4787 VA_CLOSE (ap);
4788 exit (FATAL_EXIT_CODE);
4791 /* More 'friendly' abort that prints the location at which we died. */
4792 void
4793 fancy_abort (line, func)
4794 int line;
4795 const char *func;
4797 fatal ("internal error in %s, at tradcpp.c:%d\n\
4798 Please submit a full bug report.\n\
4799 See %s for instructions.", func, line, GCCBUGURL);
4802 void
4803 perror_with_name (name)
4804 const char *name;
4806 fprintf (stderr, "%s: %s: %s\n", progname, name, strerror (errno));
4807 errors++;
4810 void
4811 pfatal_with_name (name)
4812 const char *name;
4814 perror_with_name (name);
4815 exit (FATAL_EXIT_CODE);
4818 /* Return the line at which an error occurred.
4819 The error is not necessarily associated with the current spot
4820 in the input stack, so LINE says where. LINE will have been
4821 copied from ip->lineno for the current input level.
4822 If the current level is for a file, we return LINE.
4823 But if the current level is not for a file, LINE is meaningless.
4824 In that case, we return the lineno of the innermost file. */
4825 static int
4826 line_for_error (line)
4827 int line;
4829 int i;
4830 int line1 = line;
4832 for (i = indepth; i >= 0; ) {
4833 if (instack[i].fname != 0)
4834 return line1;
4835 i--;
4836 if (i < 0)
4837 return 0;
4838 line1 = instack[i].lineno;
4840 return 0;
4844 * If OBUF doesn't have NEEDED bytes after OPTR, make it bigger.
4846 * As things stand, nothing is ever placed in the output buffer to be
4847 * removed again except when it's KNOWN to be part of an identifier,
4848 * so flushing and moving down everything left, instead of expanding,
4849 * should work ok.
4852 static void
4853 grow_outbuf (obuf, needed)
4854 FILE_BUF *obuf;
4855 int needed;
4857 U_CHAR *p;
4858 int minsize;
4860 if (obuf->length - (obuf->bufp - obuf->buf) > needed)
4861 return;
4863 /* Make it at least twice as big as it is now. */
4864 obuf->length *= 2;
4865 /* Make it have at least 150% of the free space we will need. */
4866 minsize = (3 * needed) / 2 + (obuf->bufp - obuf->buf);
4867 if (minsize > obuf->length)
4868 obuf->length = minsize;
4870 p = (U_CHAR *) xrealloc (obuf->buf, obuf->length);
4871 obuf->bufp = p + (obuf->bufp - obuf->buf);
4872 obuf->buf = p;
4875 /* Symbol table for macro names and special symbols */
4878 * install a name in the main hash table, even if it is already there.
4879 * name stops with first non alphanumeric, except leading '#'.
4880 * caller must check against redefinition if that is desired.
4881 * delete_macro () removes things installed by install () in fifo order.
4882 * this is important because of the `defined' special symbol used
4883 * in #if, and also if pushdef/popdef directives are ever implemented.
4885 * If LEN is >= 0, it is the length of the name.
4886 * Otherwise, compute the length by scanning the entire name.
4888 * If HASH is >= 0, it is the precomputed hash code.
4889 * Otherwise, compute the hash code.
4891 * caller must set the value, if any is desired.
4893 static HASHNODE *
4894 install (name, len, type, hash)
4895 const U_CHAR *name;
4896 int len;
4897 enum node_type type;
4898 int hash;
4899 /* watch out here if sizeof (U_CHAR *) != sizeof (int) */
4901 HASHNODE *hp;
4902 int bucket;
4903 const U_CHAR *p;
4904 U_CHAR *q;
4906 if (len < 0) {
4907 p = name;
4908 while (is_idchar (*p))
4909 p++;
4910 len = p - name;
4913 if (hash < 0)
4914 hash = hashf (name, len, HASHSIZE);
4916 hp = (HASHNODE *) xmalloc (sizeof (HASHNODE) + len + 1);
4917 bucket = hash;
4918 hp->bucket_hdr = &hashtab[bucket];
4919 hp->next = hashtab[bucket];
4920 hashtab[bucket] = hp;
4921 hp->prev = NULL;
4922 if (hp->next != NULL)
4923 hp->next->prev = hp;
4924 hp->type = type;
4925 hp->length = len;
4926 hp->name = q = ((U_CHAR *) hp) + sizeof (HASHNODE);
4927 memcpy (q, name, len);
4928 q[len] = 0;
4929 return hp;
4933 * find the most recent hash node for name name (ending with first
4934 * non-identifier char) installed by install
4936 * If LEN is >= 0, it is the length of the name.
4937 * Otherwise, compute the length by scanning the entire name.
4939 * If HASH is >= 0, it is the precomputed hash code.
4940 * Otherwise, compute the hash code.
4942 HASHNODE *
4943 lookup (name, len, hash)
4944 const U_CHAR *name;
4945 int len;
4946 int hash;
4948 const U_CHAR *bp;
4949 HASHNODE *bucket;
4951 if (len < 0) {
4952 for (bp = name; is_idchar (*bp); bp++) ;
4953 len = bp - name;
4956 if (hash < 0)
4957 hash = hashf (name, len, HASHSIZE);
4959 bucket = hashtab[hash];
4960 while (bucket) {
4961 if (bucket->length == len
4962 && strncmp ((const char *)bucket->name, (const char *)name, len) == 0)
4963 return bucket;
4964 bucket = bucket->next;
4966 return NULL;
4970 * Delete a hash node. Some weirdness to free junk from macros.
4971 * More such weirdness will have to be added if you define more hash
4972 * types that need it.
4975 /* Note that the DEFINITION of a macro is removed from the hash table
4976 but its storage is not freed. This would be a storage leak
4977 except that it is not reasonable to keep undefining and redefining
4978 large numbers of macros many times.
4979 In any case, this is necessary, because a macro can be #undef'd
4980 in the middle of reading the arguments to a call to it.
4981 If #undef freed the DEFINITION, that would crash. */
4982 static void
4983 delete_macro (hp)
4984 HASHNODE *hp;
4987 if (hp->prev != NULL)
4988 hp->prev->next = hp->next;
4989 if (hp->next != NULL)
4990 hp->next->prev = hp->prev;
4992 /* make sure that the bucket chain header that
4993 the deleted guy was on points to the right thing afterwards. */
4994 if (hp == *hp->bucket_hdr)
4995 *hp->bucket_hdr = hp->next;
4997 free (hp);
5001 * return hash function on name. must be compatible with the one
5002 * computed a step at a time, elsewhere
5004 static int
5005 hashf (name, len, hashsize)
5006 const U_CHAR *name;
5007 int len;
5008 int hashsize;
5010 int r = 0;
5012 while (len--)
5013 r = HASHSTEP (r, *name++);
5015 return MAKE_POS (r) % hashsize;
5018 /* Dump all macro definitions as #defines to stdout. */
5020 static void
5021 dump_all_macros ()
5023 int bucket;
5025 for (bucket = 0; bucket < HASHSIZE; bucket++) {
5026 HASHNODE *hp;
5028 for (hp = hashtab[bucket]; hp; hp= hp->next) {
5029 if (hp->type == T_MACRO) {
5030 DEFINITION *defn = hp->value.defn;
5031 struct reflist *ap;
5032 int offset;
5033 int concat;
5036 /* Print the definition of the macro HP. */
5038 printf ("#define %s", hp->name);
5039 if (defn->nargs >= 0) {
5040 int i;
5042 printf ("(");
5043 for (i = 0; i < defn->nargs; i++) {
5044 dump_arg_n (defn, i);
5045 if (i + 1 < defn->nargs)
5046 printf (", ");
5048 printf (")");
5051 printf (" ");
5053 offset = 0;
5054 concat = 0;
5055 for (ap = defn->pattern; ap != NULL; ap = ap->next) {
5056 dump_defn_1 (defn->expansion, offset, ap->nchars);
5057 if (ap->nchars != 0)
5058 concat = 0;
5059 offset += ap->nchars;
5060 if (ap->stringify)
5061 printf (" #");
5062 if (ap->raw_before && !concat)
5063 printf (" ## ");
5064 concat = 0;
5065 dump_arg_n (defn, ap->argno);
5066 if (ap->raw_after) {
5067 printf (" ## ");
5068 concat = 1;
5071 dump_defn_1 (defn->expansion, offset, defn->length - offset);
5072 printf ("\n");
5078 /* Output to stdout a substring of a macro definition.
5079 BASE is the beginning of the definition.
5080 Output characters START thru LENGTH.
5081 Discard newlines outside of strings, thus
5082 converting funny-space markers to ordinary spaces. */
5083 static void
5084 dump_defn_1 (base, start, length)
5085 const U_CHAR *base;
5086 int start;
5087 int length;
5089 const U_CHAR *p = base + start;
5090 const U_CHAR *limit = base + start + length;
5092 while (p < limit) {
5093 if (*p != '\n')
5094 putchar (*p);
5095 else if (*p == '\"' || *p =='\'') {
5096 const U_CHAR *p1 = skip_quoted_string (p, limit, 0, 0, 0, 0);
5097 fwrite (p, p1 - p, 1, stdout);
5098 p = p1 - 1;
5100 p++;
5104 /* Print the name of argument number ARGNUM of macro definition DEFN.
5105 Recall that DEFN->argnames contains all the arg names
5106 concatenated in reverse order with comma-space in between. */
5107 static void
5108 dump_arg_n (defn, argnum)
5109 DEFINITION *defn;
5110 int argnum;
5112 const U_CHAR *p = defn->argnames;
5113 while (argnum + 1 < defn->nargs) {
5114 p = (const U_CHAR *) strchr ((const char *)p, ' ') + 1;
5115 argnum++;
5118 while (*p && *p != ',') {
5119 putchar (*p);
5120 p++;
5124 /* Initialize the built-in macros. */
5125 #define DSC(x) U x, sizeof x - 1
5126 #define install_spec(name, type) \
5127 install(DSC(name), type, -1);
5128 #define install_value(name, val) do { \
5129 hp = install(DSC(name), T_CONST, -1); hp->value.cpval = val; \
5130 } while (0)
5131 static void
5132 initialize_builtins ()
5134 HASHNODE *hp;
5136 install_spec ("__BASE_FILE__", T_BASE_FILE);
5137 install_spec ("__DATE__", T_DATE);
5138 install_spec ("__FILE__", T_FILE);
5139 install_spec ("__TIME__", T_TIME);
5140 install_spec ("__VERSION__", T_VERSION);
5141 install_spec ("__INCLUDE_LEVEL__", T_INCLUDE_LEVEL);
5142 install_spec ("__LINE__", T_SPECLINE);
5144 if (flag_signed_char == 0)
5145 install_value ("__CHAR_UNSIGNED__", "1");
5147 #undef DSC
5148 #undef install_spec
5149 #undef install_value
5151 /* Common handler of command line directives -U, -D and -A. */
5152 static void
5153 run_directive (str, len, type)
5154 const char *str;
5155 size_t len;
5156 enum node_type type;
5158 const struct directive *kt;
5159 FILE_BUF *ip = &instack[++indepth];
5160 ip->fname = "*command line*";
5162 ip->buf = ip->bufp = (U_CHAR *) str;
5163 ip->length = len;
5164 ip->lineno = 1;
5165 ip->macro = 0;
5166 ip->free_ptr = 0;
5167 ip->if_stack = if_stack;
5169 for (kt = directive_table; kt->type != type; kt++)
5172 (*kt->func) ((U_CHAR *) str, (U_CHAR *) str + len, NULL);
5173 --indepth;
5176 /* Handle the -D option. If STR is just an identifier, define it with
5177 * value 1. If STR has anything after the identifier, then it should
5178 * be identifier-space-definition. */
5179 static void
5180 make_definition (str)
5181 const char *str;
5183 char *buf, *p;
5184 size_t count;
5186 /* Copy the entire option so we can modify it.
5187 Change the first "=" in the string to a space. If there is none,
5188 tack " 1" on the end. */
5190 /* Length including the null. */
5191 count = strlen (str);
5192 buf = (char *) alloca (count + 2);
5193 memcpy (buf, str, count);
5195 p = strchr (str, '=');
5196 if (p)
5197 buf[p - str] = ' ';
5198 else
5200 buf[count++] = ' ';
5201 buf[count++] = '1';
5204 run_directive (buf, count, T_DEFINE);
5207 /* Handle the -U option. */
5208 static void
5209 make_undef (str)
5210 const char *str;
5212 run_directive (str, strlen (str), T_UNDEF);
5215 /* Handles the #assert (-A) and #unassert (-A-) command line options. */
5216 static void
5217 make_assertion (str)
5218 const char *str;
5220 enum node_type type = T_ASSERT;
5221 size_t count;
5222 const char *p;
5224 if (*str == '-')
5226 str++;
5227 type = T_UNASSERT;
5230 count = strlen (str);
5231 p = strchr (str, '=');
5232 if (p)
5234 /* Copy the entire option so we can modify it. Change the first
5235 "=" in the string to a '(', and tack a ')' on the end. */
5236 char *buf = (char *) alloca (count + 1);
5238 memcpy (buf, str, count);
5239 buf[p - str] = '(';
5240 buf[count++] = ')';
5241 str = buf;
5244 run_directive (str, count, type);
5247 /* Get the file-mode and data size of the file open on FD
5248 and store them in *MODE_POINTER and *SIZE_POINTER. */
5250 static int
5251 file_size_and_mode (fd, mode_pointer, size_pointer)
5252 int fd;
5253 int *mode_pointer;
5254 long *size_pointer;
5256 struct stat sbuf;
5258 if (fstat (fd, &sbuf) < 0) return -1;
5259 if (mode_pointer) *mode_pointer = sbuf.st_mode;
5260 if (size_pointer) *size_pointer = sbuf.st_size;
5261 return 0;