* cfgloop.c (flow_loop_entry_edges_find): Fix typo.
[official-gcc.git] / gcc / tradcpp.c
blob11c5fd3c5c0733b40dc418e2c6aff375b4ad755f
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 warn if slash-star appears in a comment. */
91 int warn_comments;
93 /* Nonzero causes output not to be done,
94 but directives such as #define that have side effects
95 are still obeyed. */
97 int no_output;
99 /* Value of __USER_LABEL_PREFIX__. Target-dependent, also controlled
100 by -f(no-)leading-underscore. */
101 static const char *user_label_prefix;
103 /* I/O buffer structure.
104 The `fname' field is nonzero for source files and #include files
105 and for the dummy text used for -D and -U.
106 It is zero for rescanning results of macro expansion
107 and for expanding macro arguments. */
108 #define INPUT_STACK_MAX 200
109 struct file_name_list;
110 struct file_buf {
111 const char *fname;
112 int lineno;
113 int length;
114 U_CHAR *buf;
115 U_CHAR *bufp;
116 /* Macro that this level is the expansion of.
117 Included so that we can reenable the macro
118 at the end of this level. */
119 struct hashnode *macro;
120 /* Value of if_stack at start of this file.
121 Used to prohibit unmatched #endif (etc) in an include file. */
122 struct if_stack *if_stack;
123 /* Object to be freed at end of input at this level. */
124 U_CHAR *free_ptr;
125 /* Position to start scanning for #include_next in this file. */
126 struct file_name_list *next_header_dir;
127 } instack[INPUT_STACK_MAX];
129 typedef struct file_buf FILE_BUF;
131 /* Current nesting level of input sources.
132 `instack[indepth]' is the level currently being read. */
133 int indepth = -1;
134 #define CHECK_DEPTH(code) \
135 if (indepth >= (INPUT_STACK_MAX - 1)) \
137 error_with_line (line_for_error (instack[indepth].lineno), \
138 "macro or #include recursion too deep"); \
139 code; \
142 /* Current depth in #include directives that use <...>. */
143 int system_include_depth = 0;
145 /* The output buffer. Its LENGTH field is the amount of room allocated
146 for the buffer, not the number of chars actually present. To get
147 that, subtract outbuf.buf from outbuf.bufp. */
149 #define OUTBUF_SIZE 10 /* initial size of output buffer */
150 FILE_BUF outbuf;
152 /* Grow output buffer OBUF points at
153 so it can hold at least NEEDED more chars. */
155 #define check_expand(OBUF, NEEDED) do { \
156 if ((OBUF)->length - ((OBUF)->bufp - (OBUF)->buf) <= (NEEDED)) \
157 grow_outbuf ((OBUF), (NEEDED)); \
158 } while (0)
160 struct file_name_list
162 struct file_name_list *next;
163 const char *fname;
166 struct file_name_list *include = 0; /* First dir to search */
167 /* First dir to search for <file> */
168 struct file_name_list *first_bracket_include = 0;
169 struct file_name_list *last_include = 0; /* Last in chain */
171 /* List of included files that contained #once. */
172 struct file_name_list *dont_repeat_files = 0;
174 /* List of other included files. */
175 struct file_name_list *all_include_files = 0;
177 /* Structure allocated for every #define. For a simple replacement
178 such as
179 #define foo bar ,
180 nargs = -1, the `pattern' list is null, and the expansion is just
181 the replacement text. Nargs = 0 means a functionlike macro with no args,
182 e.g.,
183 #define getchar() getc (stdin) .
184 When there are args, the expansion is the replacement text with the
185 args squashed out, and the reflist is a list describing how to
186 build the output from the input: e.g., "3 chars, then the 1st arg,
187 then 9 chars, then the 3rd arg, then 0 chars, then the 2nd arg".
188 The chars here come from the expansion. Whatever is left of the
189 expansion after the last arg-occurrence is copied after that arg.
190 Note that the reflist can be arbitrarily long---
191 its length depends on the number of times the arguments appear in
192 the replacement text, not how many args there are. Example:
193 #define f(x) x+x+x+x+x+x+x would have replacement text "++++++" and
194 pattern list
195 { (0, 1), (1, 1), (1, 1), ..., (1, 1), NULL }
196 where (x, y) means (nchars, argno). */
198 typedef struct definition DEFINITION;
199 struct definition {
200 int nargs;
201 int length; /* length of expansion string */
202 U_CHAR *expansion;
203 struct reflist {
204 struct reflist *next;
205 char stringify; /* nonzero if this arg was preceded by a
206 # operator. */
207 char raw_before; /* Nonzero if a ## operator before arg. */
208 char raw_after; /* Nonzero if a ## operator after arg. */
209 int nchars; /* Number of literal chars to copy before
210 this arg occurrence. */
211 int argno; /* Number of arg to substitute (origin-0) */
212 } *pattern;
213 /* Names of macro args, concatenated in reverse order
214 with comma-space between them.
215 The only use of this is that we warn on redefinition
216 if this differs between the old and new definitions. */
217 const U_CHAR *argnames;
220 /* Chained list of answers to an assertion. */
221 struct answer
223 struct answer *next;
224 const unsigned char *answer;
225 size_t len;
228 /* different kinds of things that can appear in the value field
229 of a hash node. Actually, this may be useless now. */
230 union hashval {
231 const char *cpval;
232 DEFINITION *defn;
233 struct answer *answers;
236 /* The structure of a node in the hash table. The hash table
237 has entries for all tokens defined by #define commands (type T_MACRO),
238 plus some special tokens like __LINE__ (these each have their own
239 type, and the appropriate code is run when that type of node is seen.
240 It does not contain control words like "#define", which are recognized
241 by a separate piece of code. */
243 /* different flavors of hash nodes --- also used in keyword table */
244 enum node_type {
245 T_DEFINE = 1, /* `#define' */
246 T_INCLUDE, /* `#include' */
247 T_INCLUDE_NEXT,/* `#include_next' */
248 T_IFDEF, /* `#ifdef' */
249 T_IFNDEF, /* `#ifndef' */
250 T_IF, /* `#if' */
251 T_ELSE, /* `#else' */
252 T_ELIF, /* `#elif' */
253 T_UNDEF, /* `#undef' */
254 T_LINE, /* `#line' */
255 T_ENDIF, /* `#endif' */
256 T_ERROR, /* `#error' */
257 T_WARNING, /* `#warning' */
258 T_ASSERT, /* `#assert' */
259 T_UNASSERT, /* `#unassert' */
260 T_SPECLINE, /* special symbol `__LINE__' */
261 T_DATE, /* `__DATE__' */
262 T_FILE, /* `__FILE__' */
263 T_BASE_FILE, /* `__BASE_FILE__' */
264 T_INCLUDE_LEVEL, /* `__INCLUDE_LEVEL__' */
265 T_VERSION, /* `__VERSION__' */
266 T_TIME, /* `__TIME__' */
267 T_CONST, /* Constant value, used by `__STDC__' */
268 T_MACRO, /* macro defined by `#define' */
269 T_SPEC_DEFINED, /* special `defined' macro for use in #if statements */
270 T_UNUSED /* Used for something not defined. */
273 struct hashnode {
274 struct hashnode *next; /* double links for easy deletion */
275 struct hashnode *prev;
276 struct hashnode **bucket_hdr; /* also, a back pointer to this node's hash
277 chain is kept, in case the node is the head
278 of the chain and gets deleted. */
279 enum node_type type; /* type of special token */
280 int length; /* length of token, for quick comparison */
281 U_CHAR *name; /* the actual name */
282 union hashval value; /* pointer to expansion, or whatever */
285 typedef struct hashnode HASHNODE;
287 static HASHNODE *parse_assertion PARAMS ((const unsigned char *,
288 const unsigned char *,
289 struct answer **, int));
290 static struct answer **find_answer PARAMS ((HASHNODE *,
291 const struct answer *));
292 static int parse_answer PARAMS ((const unsigned char *, const unsigned char *,
293 struct answer **, int));
294 static unsigned char *canonicalize_text PARAMS ((const unsigned char *,
295 const unsigned char *,
296 const unsigned char **));
298 /* Some definitions for the hash table. The hash function MUST be
299 computed as shown in hashf () below. That is because the rescan
300 loop computes the hash value `on the fly' for most tokens,
301 in order to avoid the overhead of a lot of procedure calls to
302 the hashf () function. Hashf () only exists for the sake of
303 politeness, for use when speed isn't so important. */
305 #define HASHSIZE 1403
306 HASHNODE *hashtab[HASHSIZE];
307 #define HASHSTEP(old, c) ((old << 2) + c)
308 #define MAKE_POS(v) (v & 0x7fffffff) /* make number positive */
310 /* `struct directive' defines one #-directive, including how to handle it. */
312 struct directive {
313 const int length; /* Length of name */
314 void (*const func) PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
315 /* Function to handle directive */
316 const char *const name; /* Name of directive */
317 const enum node_type type; /* Code which describes which directive. */
320 /* Last arg to output_line_command. */
321 enum file_change_code {same_file, enter_file, leave_file};
323 /* This structure represents one parsed argument in a macro call.
324 `raw' points to the argument text as written (`raw_length' is its length).
325 `expanded' points to the argument's macro-expansion
326 (its length is `expand_length').
327 `stringified_length' is the length the argument would have
328 if stringified.
329 `free1' and `free2', if nonzero, point to blocks to be freed
330 when the macro argument data is no longer needed. */
332 struct argdata {
333 U_CHAR *raw, *expanded;
334 int raw_length, expand_length;
335 int stringified_length;
336 U_CHAR *free1, *free2;
337 char newlines;
338 char comments;
341 /* The arglist structure is built by do_define to tell
342 collect_definition where the argument names begin. That
343 is, for a define like "#define f(x,y,z) foo+x-bar*y", the arglist
344 would contain pointers to the strings x, y, and z.
345 Collect_definition would then build a DEFINITION node,
346 with reflist nodes pointing to the places x, y, and z had
347 appeared. So the arglist is just convenience data passed
348 between these two routines. It is not kept around after
349 the current #define has been processed and entered into the
350 hash table. */
352 struct arglist {
353 struct arglist *next;
354 U_CHAR *name;
355 int length;
356 int argno;
359 /* Function prototypes. */
361 static void do_define PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
362 static void do_error PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
363 static void do_warning PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
364 static void do_line PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
365 static void do_include PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
366 static void do_include_next PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
367 static void do_undef PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
368 static void do_if PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
369 static void do_ifdef PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
370 static void do_ifndef PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
371 static void do_else PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
372 static void do_elif PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
373 static void do_endif PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
374 static void do_assert PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
375 static void do_unassert PARAMS ((U_CHAR *, U_CHAR *, FILE_BUF *));
376 static void do_xifdef PARAMS ((U_CHAR *, U_CHAR *, enum node_type));
378 static struct hashnode *install PARAMS ((const U_CHAR *, int, enum node_type, int));
379 static int hashf PARAMS ((const U_CHAR *, int, int));
380 static int compare_defs PARAMS ((DEFINITION *, DEFINITION *));
381 static int comp_def_part PARAMS ((int, const U_CHAR *, int,
382 const U_CHAR *, int, int));
383 static void delete_macro PARAMS ((HASHNODE *));
385 /* First arg to v_message. */
386 enum msgtype { MT_WARNING = 0, MT_ERROR, MT_FATAL };
387 static void v_message PARAMS ((enum msgtype mtype, int line,
388 const char *msgid, va_list ap))
389 ATTRIBUTE_PRINTF (3, 0);
391 static int line_for_error PARAMS ((int));
393 /* We know perfectly well which file this is, so we don't need to
394 use __FILE__. */
395 #undef abort
396 #if (GCC_VERSION >= 2007)
397 #define abort() fancy_abort(__LINE__, __FUNCTION__)
398 #else
399 #define abort() fancy_abort(__LINE__, 0);
400 #endif
402 static void macroexpand PARAMS ((HASHNODE *, FILE_BUF *));
403 static void special_symbol PARAMS ((HASHNODE *, FILE_BUF *));
404 static void dump_all_macros PARAMS ((void));
405 static void dump_defn_1 PARAMS ((const U_CHAR *, int, int));
406 static void dump_arg_n PARAMS ((DEFINITION *, int));
407 static void conditional_skip PARAMS ((FILE_BUF *, int, enum node_type));
408 static void skip_if_group PARAMS ((FILE_BUF *, int));
409 static void output_line_command PARAMS ((FILE_BUF *, FILE_BUF *,
410 int, enum file_change_code));
412 static int eval_if_expression PARAMS ((const U_CHAR *, int));
414 static void output_deps PARAMS ((void));
415 static void initialize_builtins PARAMS ((void));
416 static void run_directive PARAMS ((const char *, size_t,
417 enum node_type));
418 static void make_definition PARAMS ((const char *));
419 static void make_undef PARAMS ((const char *));
420 static void make_assertion PARAMS ((const char *));
422 static void grow_outbuf PARAMS ((FILE_BUF *, int));
423 static int handle_directive PARAMS ((FILE_BUF *, FILE_BUF *));
424 static void process_include PARAMS ((struct file_name_list *,
425 const U_CHAR *, int, int, FILE_BUF *));
426 static void finclude PARAMS ((int, const char *,
427 struct file_name_list *, FILE_BUF *));
428 static void init_dependency_output PARAMS ((void));
429 static void rescan PARAMS ((FILE_BUF *, int));
430 static void newline_fix PARAMS ((U_CHAR *));
431 static void name_newline_fix PARAMS ((U_CHAR *));
432 static U_CHAR *macarg1 PARAMS ((U_CHAR *, const U_CHAR *, int *,
433 int *, int *));
434 static const char *macarg PARAMS ((struct argdata *));
435 static int discard_comments PARAMS ((U_CHAR *, int, int));
436 static int file_size_and_mode PARAMS ((int, int *, long *));
438 static U_CHAR *skip_to_end_of_comment PARAMS ((FILE_BUF *, int *));
439 static U_CHAR *skip_quoted_string PARAMS ((const U_CHAR *, const U_CHAR *,
440 int, int *, int *, int *));
442 int main PARAMS ((int, char **));
444 /* Convenience. Write U"string" to get an unsigned string constant. */
445 #define U (const unsigned char *)
447 /* Here is the actual list of #-directives, most-often-used first. */
449 static const struct directive directive_table[] = {
450 { 6, do_define, "define", T_DEFINE },
451 { 7, do_include, "include", T_INCLUDE },
452 { 5, do_endif, "endif", T_ENDIF },
453 { 5, do_ifdef, "ifdef", T_IFDEF },
454 { 2, do_if, "if", T_IF, },
455 { 4, do_else, "else", T_ELSE },
456 { 6, do_ifndef, "ifndef", T_IFNDEF },
457 { 5, do_undef, "undef", T_UNDEF },
458 { 4, do_line, "line", T_LINE },
459 { 4, do_elif, "elif", T_ELIF },
460 { 5, do_error, "error", T_ERROR },
461 { 7, do_warning, "warning", T_WARNING },
462 { 12, do_include_next, "include_next", T_INCLUDE_NEXT },
463 { 6, do_assert, "assert", T_ASSERT },
464 { 8, do_unassert,"unassert",T_UNASSERT},
465 { -1, 0, "", T_UNUSED},
468 #define SKIP_WHITE_SPACE(p) do { while (is_nvspace(*p)) p++; } while (0)
469 #define SKIP_ALL_WHITE_SPACE(p) do { while (is_space(*p)) p++; } while (0)
471 int errors = 0; /* Error counter for exit code */
473 static FILE_BUF expand_to_temp_buffer PARAMS ((const U_CHAR *, const U_CHAR *, int));
474 static DEFINITION *collect_expansion PARAMS ((U_CHAR *, U_CHAR *, int,
475 struct arglist *));
477 /* Stack of conditionals currently in progress
478 (including both successful and failing conditionals). */
480 struct if_stack {
481 struct if_stack *next; /* for chaining to the next stack frame */
482 const char *fname; /* copied from input when frame is made */
483 int lineno; /* similarly */
484 int if_succeeded; /* true if a leg of this if-group
485 has been passed through rescan */
486 enum node_type type; /* type of last directive seen in this group */
488 typedef struct if_stack IF_STACK_FRAME;
489 IF_STACK_FRAME *if_stack = NULL;
491 /* Nonzero means -I- has been seen,
492 so don't look for #include "foo" the source-file directory. */
493 int ignore_srcdir;
495 /* Pending directives. */
496 enum pending_dir_t {PD_NONE = 0, PD_DEFINE, PD_UNDEF, PD_ASSERTION, PD_FILE};
498 typedef struct pending_dir pending_dir;
499 struct pending_dir
501 const char *arg;
502 enum pending_dir_t type;
506 main (argc, argv)
507 int argc;
508 char **argv;
510 int st_mode;
511 long st_size;
512 const char *in_fname, *out_fname;
513 int f, i;
514 FILE_BUF *fp;
515 pending_dir *pend = (pending_dir *) xcalloc (argc, sizeof (pending_dir));
516 int no_standard_includes = 0;
518 hex_init ();
520 #ifdef RLIMIT_STACK
521 /* Get rid of any avoidable limit on stack size. */
523 struct rlimit rlim;
525 /* Set the stack limit huge so that alloca (particularly stringtab
526 * in dbxread.c) does not fail. */
527 getrlimit (RLIMIT_STACK, &rlim);
528 rlim.rlim_cur = rlim.rlim_max;
529 setrlimit (RLIMIT_STACK, &rlim);
531 #endif /* RLIMIT_STACK defined */
533 progname = argv[0];
535 in_fname = NULL;
536 out_fname = NULL;
538 no_line_commands = 0;
539 dump_macros = 0;
540 no_output = 0;
542 max_include_len = cpp_GCC_INCLUDE_DIR_len + 7; /* ??? */
544 gcc_init_libintl ();
546 /* It's simplest to just create this struct whether or not it will
547 be needed. */
548 deps = deps_init ();
550 /* Process switches and find input file name. */
552 for (i = 1; i < argc; i++) {
553 if (argv[i][0] != '-') {
554 if (out_fname != NULL)
555 fatal ("usage: %s [switches] input output", argv[0]);
556 else if (in_fname != NULL)
557 out_fname = argv[i];
558 else
559 in_fname = argv[i];
560 } else {
561 int c = argv[i][1];
563 switch (c) {
564 case 'E':
565 case '$':
566 break; /* Ignore for compatibility with ISO/extended cpp. */
568 case 'l':
569 if (!strcmp (argv[i], "-lang-c++")
570 || !strcmp (argv[i], "-lang-objc++"))
571 fatal ("-traditional is not supported in C++");
572 else if (!strcmp (argv[i], "-lang-c89"))
573 fatal ("-traditional and -ansi are mutually exclusive");
574 else if (!strcmp (argv[i], "-lang-objc"))
575 pend[i].type = PD_DEFINE, pend[i].arg = "__OBJC__";
576 else if (!strcmp (argv[i], "-lang-asm"))
577 pend[i].type = PD_DEFINE, pend[i].arg = "__ASSEMBLER__";
578 else if (!strcmp (argv[i], "-lang-fortran"))
579 pend[i].type = PD_DEFINE, pend[i].arg = "_LANGUAGE_FORTRAN";
580 /* All other possibilities ignored. */
581 break;
583 case 'i':
584 if (!strcmp (argv[i], "-include"))
586 if (i + 1 == argc)
587 fatal ("filename missing after -i option");
588 else
589 pend[i].type = PD_FILE, pend[i].arg = argv[i + 1], i++;
591 else if (!strcmp (argv[i], "-iprefix"))
592 i++; /* Ignore for compatibility */
593 else if (!strcmp (argv[i], "-isystem")
594 || !strcmp (argv[i], "-iwithprefix")
595 || !strcmp (argv[i], "-iwithprefixbefore")
596 || !strcmp (argv[i], "-idirafter"))
597 goto add_include; /* best we can do */
599 break;
601 case 'o':
602 if (out_fname != NULL)
603 fatal ("output filename specified twice");
604 if (i + 1 == argc)
605 fatal ("filename missing after -o option");
606 out_fname = argv[++i];
607 if (!strcmp (out_fname, "-"))
608 out_fname = "";
609 break;
611 case 'w':
612 inhibit_warnings = 1;
613 break;
615 case 'W':
616 if (!strcmp (argv[i], "-Wcomments"))
617 warn_comments = 1;
618 else if (!strcmp (argv[i], "-Wcomment"))
619 warn_comments = 1;
620 else if (!strcmp (argv[i], "-Wall")) {
621 warn_comments = 1;
623 break;
625 case 'f':
626 if (!strcmp (argv[i], "-fleading-underscore"))
627 user_label_prefix = "_";
628 else if (!strcmp (argv[i], "-fno-leading-underscore"))
629 user_label_prefix = "";
630 break;
632 case 'M':
634 char *p = NULL;
636 /* -MD and -MMD for tradcpp are deprecated and undocumented
637 (use -M or -MM with -MF instead), and probably should be
638 removed with the next major GCC version. For the moment
639 we allow these for the benefit of Automake 1.4, which
640 uses these when dependency tracking is enabled. Automake
641 1.5 will fix this. */
642 if (!strncmp (argv[i], "-MD", 3)) {
643 p = argv[i] + 3;
644 print_deps = 2;
645 } else if (!strncmp (argv[i], "-MMD", 4)) {
646 p = argv[i] + 4;
647 print_deps = 1;
648 } else if (!strcmp (argv[i], "-M")) {
649 print_deps = 2;
650 } else if (!strcmp (argv[i], "-MM")) {
651 print_deps = 1;
652 } else if (!strcmp (argv[i], "-MG")) {
653 deps_missing_files = 1;
654 } else if (!strcmp (argv[i], "-MF")) {
655 p = argv[i] + 3;
656 } else if (!strcmp (argv[i], "-MP")) {
657 print_deps_phony_targets = 1;
658 } else if (!strcmp (argv[i], "-MQ") || !strcmp (argv[i], "-MT")) {
659 /* Add a target. -MQ quotes for Make. */
660 const char *tgt = argv[i] + 3;
661 int quoted = argv[i][2] == 'Q';
663 if (*tgt == '\0' && i + 1 == argc)
664 fatal ("target missing after %s option", argv[i]);
665 else
667 if (*tgt == '\0')
668 tgt = argv[++i];
670 deps_add_target (deps, tgt, quoted);
674 if (p) {
675 if (*p)
676 deps_file = p;
677 else if (i + 1 == argc)
678 fatal ("filename missing after %s option", argv[i]);
679 else
680 deps_file = argv[++i];
683 break;
685 case 'd':
686 dump_macros = 1;
687 no_output = 1;
688 break;
690 case 'v':
691 fprintf (stderr, "GNU traditional CPP version %s\n", version_string);
692 break;
694 case 'D':
695 case 'U':
696 case 'A':
698 char *p;
700 if (argv[i][2] != 0)
701 p = argv[i] + 2;
702 else if (i + 1 == argc)
703 fatal ("macro name missing after -%c option", c);
704 else
705 p = argv[++i];
707 if (c == 'D')
708 pend[i].type = PD_DEFINE;
709 else if (c == 'U')
710 pend[i].type = PD_UNDEF;
711 else
712 pend[i].type = PD_ASSERTION;
713 pend[i].arg = p;
715 break;
717 case 'C':
718 put_out_comments = 1;
719 break;
721 case 'p':
722 if (!strcmp (argv[i], "-pedantic"))
723 fatal ("-pedantic and -traditional are mutually exclusive");
724 break;
726 case 't':
727 if (!strcmp (argv[i], "-trigraphs"))
728 fatal ("-trigraphs and -traditional are mutually exclusive");
729 break;
731 case 'P':
732 no_line_commands = 1;
733 break;
735 case 'I': /* Add directory to path for includes. */
736 add_include:
738 struct file_name_list *dirtmp;
740 if (! ignore_srcdir && !strcmp (argv[i] + 2, "-"))
741 ignore_srcdir = 1;
742 else {
743 dirtmp = (struct file_name_list *)
744 xmalloc (sizeof (struct file_name_list));
745 dirtmp->next = 0; /* New one goes on the end */
746 if (include == 0)
747 include = dirtmp;
748 else
749 last_include->next = dirtmp;
750 last_include = dirtmp; /* Tail follows the last one */
751 if (argv[i][1] == 'I' && argv[i][2] != 0)
752 dirtmp->fname = argv[i] + 2;
753 else if (i + 1 == argc)
754 fatal ("directory name missing after -I option");
755 else
756 dirtmp->fname = argv[++i];
757 if (strlen (dirtmp->fname) > max_include_len)
758 max_include_len = strlen (dirtmp->fname);
759 if (ignore_srcdir && first_bracket_include == 0)
760 first_bracket_include = dirtmp;
763 break;
765 case 'n':
766 /* -nostdinc causes no default include directories.
767 You must specify all include-file directories with -I. */
768 no_standard_includes = 1;
769 break;
771 case '\0': /* JF handle '-' as file name meaning stdin or stdout */
772 if (in_fname == NULL) {
773 in_fname = "";
774 break;
775 } else if (out_fname == NULL) {
776 out_fname = "";
777 break;
778 } /* else fall through into error */
780 default:
781 fatal ("invalid option `%s'", argv[i]);
786 init_dependency_output ();
788 /* After checking the environment variables, check if -M or -MM has
789 not been specified, but other -M options have. */
790 if (print_deps == 0
791 && (deps_missing_files || deps_file || print_deps_phony_targets))
792 fatal ("you must additionally specify either -M or -MM");
794 if (user_label_prefix == 0)
795 user_label_prefix = USER_LABEL_PREFIX;
797 if (print_deps)
799 /* Set the default target (if there is none already), and
800 the dependency on the main file. */
801 deps_add_default_target (deps, in_fname);
803 deps_add_dep (deps, in_fname);
806 /* Install __LINE__, etc. Must follow option processing. */
807 initialize_builtins ();
809 /* Do defines specified with -D and undefines specified with -U. */
810 for (i = 1; i < argc; i++)
811 if (pend[i].type == PD_DEFINE)
812 make_definition (pend[i].arg);
813 else if (pend[i].type == PD_UNDEF)
814 make_undef (pend[i].arg);
815 else if (pend[i].type == PD_ASSERTION)
816 make_assertion (pend[i].arg);
818 /* Unless -fnostdinc,
819 tack on the standard include file dirs to the specified list */
820 if (!no_standard_includes) {
821 const struct default_include *di;
822 struct file_name_list *old_last_include = last_include;
823 struct file_name_list *dirtmp;
824 for (di = cpp_include_defaults; di->fname; di++) {
825 if (di->cplusplus)
826 continue;
827 dirtmp = (struct file_name_list *)
828 xmalloc (sizeof (struct file_name_list));
829 dirtmp->next = 0; /* New one goes on the end */
830 if (include == 0)
831 include = dirtmp;
832 else
833 last_include->next = dirtmp;
834 last_include = dirtmp; /* Tail follows the last one */
835 dirtmp->fname = di->fname;
836 if (strlen (dirtmp->fname) > max_include_len)
837 max_include_len = strlen (dirtmp->fname);
840 if (ignore_srcdir && first_bracket_include == 0)
841 first_bracket_include = old_last_include->next;
844 /* Initialize output buffer */
846 outbuf.buf = (U_CHAR *) xmalloc (OUTBUF_SIZE);
847 outbuf.bufp = outbuf.buf;
848 outbuf.length = OUTBUF_SIZE;
850 /* Scan the -i files before the main input.
851 Much like #including them, but with no_output set
852 so that only their macro definitions matter. */
854 no_output++;
855 indepth++;
856 for (i = 1; i < argc; i++)
857 if (pend[i].type == PD_FILE)
859 int fd = open (pend[i].arg, O_RDONLY, 0666);
860 if (fd < 0)
862 perror_with_name (pend[i].arg);
863 return FATAL_EXIT_CODE;
866 /* For -M, add this file to the dependencies. */
867 if (print_deps)
868 deps_add_dep (deps, pend[i].arg);
870 finclude (fd, pend[i].arg, 0, &outbuf);
872 indepth--;
873 no_output--;
875 /* Pending directives no longer needed. */
876 free ((PTR) pend);
878 /* Create an input stack level for the main input file
879 and copy the entire contents of the file into it. */
881 fp = &instack[++indepth];
883 /* JF check for stdin */
884 if (in_fname == NULL || *in_fname == 0) {
885 in_fname = "";
886 f = 0;
887 } else if ((f = open (in_fname, O_RDONLY, 0666)) < 0)
888 goto sys_error;
890 if (file_size_and_mode (f, &st_mode, &st_size))
891 goto sys_error;
892 fp->fname = in_fname;
893 fp->lineno = 1;
894 /* JF all this is mine about reading pipes and ttys */
895 if (!S_ISREG (st_mode)) {
896 /* Read input from a file that is not a normal disk file.
897 We cannot preallocate a buffer with the correct size,
898 so we must read in the file a piece at the time and make it bigger. */
899 int size;
900 int bsize;
901 int cnt;
902 U_CHAR *bufp;
904 bsize = 2000;
905 size = 0;
906 fp->buf = (U_CHAR *) xmalloc (bsize + 2);
907 bufp = fp->buf;
908 for (;;) {
909 cnt = read (f, bufp, bsize - size);
910 if (cnt < 0) goto sys_error; /* error! */
911 if (cnt == 0) break; /* End of file */
912 size += cnt;
913 bufp += cnt;
914 if (bsize == size) { /* Buffer is full! */
915 bsize *= 2;
916 fp->buf = (U_CHAR *) xrealloc (fp->buf, bsize + 2);
917 bufp = fp->buf + size; /* May have moved */
920 fp->length = size;
921 } else {
922 /* Read a file whose size we can determine in advance.
923 For the sake of VMS, st_size is just an upper bound. */
924 long i;
925 fp->length = 0;
926 fp->buf = (U_CHAR *) xmalloc (st_size + 2);
928 while (st_size > 0) {
929 i = read (f, fp->buf + fp->length, st_size);
930 if (i <= 0) {
931 if (i == 0) break;
932 goto sys_error;
934 fp->length += i;
935 st_size -= i;
938 fp->bufp = fp->buf;
939 fp->if_stack = if_stack;
941 /* Make sure data ends with a newline. And put a null after it. */
943 if (fp->length > 0 && fp->buf[fp->length-1] != '\n')
944 fp->buf[fp->length++] = '\n';
945 fp->buf[fp->length] = '\0';
947 /* Now that we know the input file is valid, open the output. */
949 if (!out_fname || !strcmp (out_fname, ""))
950 out_fname = "stdout";
951 else if (! freopen (out_fname, "w", stdout))
952 pfatal_with_name (out_fname);
954 output_line_command (fp, &outbuf, 0, same_file);
956 /* Scan the input, processing macros and directives. */
958 rescan (&outbuf, 0);
960 /* Now we have processed the entire input
961 Write whichever kind of output has been requested. */
964 if (dump_macros)
965 dump_all_macros ();
966 else if (! inhibit_output)
967 if (write (fileno (stdout), outbuf.buf, outbuf.bufp - outbuf.buf) < 0)
968 fatal ("I/O error on output");
970 /* Don't write the deps file if preprocessing has failed. */
971 if (print_deps && errors == 0)
972 output_deps ();
974 /* Destruct the deps object. */
975 deps_free (deps);
977 if (ferror (stdout))
978 fatal ("I/O error on output");
980 if (errors)
981 exit (FATAL_EXIT_CODE);
982 exit (SUCCESS_EXIT_CODE);
984 sys_error:
985 pfatal_with_name (in_fname);
988 /* Set up dependency-file output. */
989 static void
990 init_dependency_output ()
992 char *spec, *s, *output_file;
994 /* Either of two environment variables can specify output of deps.
995 Its value is either "OUTPUT_FILE" or "OUTPUT_FILE DEPS_TARGET",
996 where OUTPUT_FILE is the file to write deps info to
997 and DEPS_TARGET is the target to mention in the deps. */
999 if (print_deps == 0)
1001 spec = getenv ("DEPENDENCIES_OUTPUT");
1002 if (spec)
1003 print_deps = 1;
1004 else
1006 spec = getenv ("SUNPRO_DEPENDENCIES");
1007 if (spec)
1008 print_deps = 2;
1009 else
1010 return;
1013 /* Find the space before the DEPS_TARGET, if there is one. */
1014 s = strchr (spec, ' ');
1015 if (s)
1017 /* Let the caller perform MAKE quoting. */
1018 deps_add_target (deps, s + 1, 0);
1019 output_file = (char *) xmalloc (s - spec + 1);
1020 memcpy (output_file, spec, s - spec);
1021 output_file[s - spec] = 0;
1023 else
1024 output_file = spec;
1026 /* Command line overrides environment variables. */
1027 if (deps_file == 0)
1028 deps_file = output_file;
1029 deps_append = 1;
1032 /* If dependencies go to standard output, or -MG is used, we should
1033 suppress output. The user may be requesting other stuff to
1034 stdout, with -dM, -v etc. We let them shoot themselves in the
1035 foot. */
1036 if (deps_file == 0 || deps_missing_files)
1037 inhibit_output = 1;
1040 /* Use mkdeps.c to output dependency information. */
1041 static void
1042 output_deps ()
1044 /* Stream on which to print the dependency information. */
1045 FILE *deps_stream = 0;
1046 const char *const deps_mode = deps_append ? "a" : "w";
1048 if (deps_file == 0)
1049 deps_stream = stdout;
1050 else
1052 deps_stream = fopen (deps_file, deps_mode);
1053 if (deps_stream == 0)
1055 error_from_errno (deps_file);
1056 return;
1060 deps_write (deps, deps_stream, 72);
1062 if (print_deps_phony_targets)
1063 deps_phony_targets (deps, deps_stream);
1065 /* Don't close stdout. */
1066 if (deps_file)
1068 if (ferror (deps_stream) || fclose (deps_stream) != 0)
1069 fatal ("I/O error on output");
1073 /* Move all backslash-newline pairs out of embarrassing places.
1074 Exchange all such pairs following BP
1075 with any potentially-embarrasing characters that follow them.
1076 Potentially-embarrassing characters are / and *
1077 (because a backslash-newline inside a comment delimiter
1078 would cause it not to be recognized). */
1079 static void
1080 newline_fix (bp)
1081 U_CHAR *bp;
1083 U_CHAR *p = bp;
1084 int count = 0;
1086 /* First count the backslash-newline pairs here. */
1088 while (*p++ == '\\' && *p++ == '\n')
1089 count++;
1091 p = bp + count * 2;
1093 /* Exit if what follows the backslash-newlines is not embarrassing. */
1095 if (count == 0 || (*p != '/' && *p != '*'))
1096 return;
1098 /* Copy all potentially embarrassing characters
1099 that follow the backslash-newline pairs
1100 down to where the pairs originally started. */
1102 while (*p == '*' || *p == '/')
1103 *bp++ = *p++;
1105 /* Now write the same number of pairs after the embarrassing chars. */
1106 while (count-- > 0) {
1107 *bp++ = '\\';
1108 *bp++ = '\n';
1112 /* Like newline_fix but for use within a directive-name.
1113 Move any backslash-newlines up past any following symbol constituents. */
1114 static void
1115 name_newline_fix (bp)
1116 U_CHAR *bp;
1118 U_CHAR *p = bp;
1119 int count = 0;
1121 /* First count the backslash-newline pairs here. */
1123 while (*p++ == '\\' && *p++ == '\n')
1124 count++;
1126 p = bp + count * 2;
1128 /* What follows the backslash-newlines is not embarrassing. */
1130 if (count == 0 || !is_idchar (*p))
1131 return;
1133 /* Copy all potentially embarrassing characters
1134 that follow the backslash-newline pairs
1135 down to where the pairs originally started. */
1137 while (is_idchar (*p))
1138 *bp++ = *p++;
1140 /* Now write the same number of pairs after the embarrassing chars. */
1141 while (count-- > 0) {
1142 *bp++ = '\\';
1143 *bp++ = '\n';
1148 * The main loop of the program.
1150 * Read characters from the input stack, transferring them to the
1151 * output buffer OP.
1153 * Macros are expanded and push levels on the input stack.
1154 * At the end of such a level it is popped off and we keep reading.
1155 * At the end of any other kind of level, we return.
1156 * #-directives are handled, except within macros.
1158 * If OUTPUT_MARKS is nonzero, keep Newline markers found in the input
1159 * and insert them when appropriate. This is set while scanning macro
1160 * arguments before substitution. It is zero when scanning for final output.
1161 * There are three types of Newline markers:
1162 * * Newline - follows a macro name that was not expanded
1163 * because it appeared inside an expansion of the same macro.
1164 * This marker prevents future expansion of that identifier.
1165 * When the input is rescanned into the final output, these are deleted.
1166 * These are also deleted by ## concatenation.
1167 * * Newline Space (or Newline and any other whitespace character)
1168 * stands for a place that tokens must be separated or whitespace
1169 * is otherwise desirable, but where the ANSI standard specifies there
1170 * is no whitespace. This marker turns into a Space (or whichever other
1171 * whitespace char appears in the marker) in the final output,
1172 * but it turns into nothing in an argument that is stringified with #.
1173 * Such stringified arguments are the only place where the ANSI standard
1174 * specifies with precision that whitespace may not appear.
1176 * During this function, IP->bufp is kept cached in IBP for speed of access.
1177 * Likewise, OP->bufp is kept in OBP. Before calling a subroutine
1178 * IBP, IP and OBP must be copied back to memory. IP and IBP are
1179 * copied back with the RECACHE macro. OBP must be copied back from OP->bufp
1180 * explicitly, and before RECACHE, since RECACHE uses OBP.
1183 static void
1184 rescan (op, output_marks)
1185 FILE_BUF *op;
1186 int output_marks;
1188 /* Character being scanned in main loop. */
1189 U_CHAR c;
1191 /* Length of pending accumulated identifier. */
1192 int ident_length = 0;
1194 /* Hash code of pending accumulated identifier. */
1195 int hash = 0;
1197 /* Current input level (&instack[indepth]). */
1198 FILE_BUF *ip;
1200 /* Pointer for scanning input. */
1201 U_CHAR *ibp;
1203 /* Pointer to end of input. End of scan is controlled by LIMIT. */
1204 U_CHAR *limit;
1206 /* Pointer for storing output. */
1207 U_CHAR *obp;
1209 /* REDO_CHAR is nonzero if we are processing an identifier
1210 after backing up over the terminating character.
1211 Sometimes we process an identifier without backing up over
1212 the terminating character, if the terminating character
1213 is not special. Backing up is done so that the terminating character
1214 will be dispatched on again once the identifier is dealt with. */
1215 int redo_char = 0;
1217 /* 1 if within an identifier inside of which a concatenation
1218 marker (Newline -) has been seen. */
1219 int concatenated = 0;
1221 /* While scanning a comment or a string constant,
1222 this records the line it started on, for error messages. */
1223 int start_line;
1225 /* Record position of last `real' newline. */
1226 U_CHAR *beg_of_line;
1228 /* This has to be a global bacause of RECACHE. */
1229 U_CHAR *obufp_before_macroname = NULL;
1231 /* Pop the innermost input stack level, assuming it is a macro expansion. */
1233 #define POPMACRO \
1234 do { ip->macro->type = T_MACRO; \
1235 if (ip->free_ptr) free (ip->free_ptr); \
1236 --indepth; } while (0)
1238 /* Reload `rescan's local variables that describe the current
1239 level of the input stack. */
1241 #define RECACHE \
1242 do { ip = &instack[indepth]; \
1243 ibp = ip->bufp; \
1244 limit = ip->buf + ip->length; \
1245 op->bufp = obp; \
1246 check_expand (op, limit - ibp); \
1247 beg_of_line = 0; \
1248 obufp_before_macroname += op->bufp - obp; \
1249 obp = op->bufp; } while (0)
1251 if (no_output && instack[indepth].fname != 0)
1252 skip_if_group (&instack[indepth], 1);
1254 obp = op->bufp;
1255 RECACHE;
1256 beg_of_line = ibp;
1258 /* Our caller must always put a null after the end of
1259 the input at each input stack level. */
1260 if (*limit != 0)
1261 abort ();
1263 while (1) {
1264 c = *ibp++;
1265 *obp++ = c;
1267 switch (c) {
1268 case '\\':
1269 if (ibp >= limit)
1270 break;
1271 if (*ibp == '\n') {
1272 /* Always merge lines ending with backslash-newline,
1273 even in middle of identifier. */
1274 ++ibp;
1275 ++ip->lineno;
1276 --obp; /* remove backslash from obuf */
1277 break;
1279 /* Otherwise, backslash suppresses specialness of following char,
1280 so copy it here to prevent the switch from seeing it.
1281 But first get any pending identifier processed. */
1282 if (ident_length > 0)
1283 goto specialchar;
1284 *obp++ = *ibp++;
1285 break;
1287 case '#':
1288 /* If this is expanding a macro definition, don't recognize
1289 preprocessor directives. */
1290 if (ip->macro != 0)
1291 goto randomchar;
1292 if (ident_length)
1293 goto specialchar;
1295 /* # keyword: a # must be the first char on the line */
1296 if (beg_of_line == 0)
1297 goto randomchar;
1298 if (beg_of_line + 1 != ibp)
1299 goto randomchar;
1301 /* This # can start a directive. */
1303 --obp; /* Don't copy the '#' */
1305 ip->bufp = ibp;
1306 op->bufp = obp;
1307 if (! handle_directive (ip, op)) {
1308 #ifdef USE_C_ALLOCA
1309 alloca (0);
1310 #endif
1311 /* Not a known directive: treat it as ordinary text.
1312 IP, OP, IBP, etc. have not been changed. */
1313 if (no_output && instack[indepth].fname) {
1314 /* If not generating expanded output,
1315 what we do with ordinary text is skip it.
1316 Discard everything until next # directive. */
1317 skip_if_group (&instack[indepth], 1);
1318 RECACHE;
1319 beg_of_line = ibp;
1320 break;
1322 ++obp; /* Copy the '#' after all */
1323 goto randomchar;
1325 #ifdef USE_C_ALLOCA
1326 alloca (0);
1327 #endif
1328 /* A # directive has been successfully processed. */
1329 /* If not generating expanded output, ignore everything until
1330 next # directive. */
1331 if (no_output && instack[indepth].fname)
1332 skip_if_group (&instack[indepth], 1);
1333 obp = op->bufp;
1334 RECACHE;
1335 beg_of_line = ibp;
1336 break;
1338 case '\"': /* skip quoted string */
1339 case '\'':
1340 /* A single quoted string is treated like a double -- some
1341 programs (e.g., troff) are perverse this way */
1343 if (ident_length)
1344 goto specialchar;
1346 start_line = ip->lineno;
1348 /* Skip ahead to a matching quote. */
1350 while (1) {
1351 if (ibp >= limit) {
1352 if (ip->macro != 0) {
1353 /* try harder: this string crosses a macro expansion boundary */
1354 POPMACRO;
1355 RECACHE;
1356 continue;
1358 break;
1360 *obp++ = *ibp;
1361 switch (*ibp++) {
1362 case '\n':
1363 ++ip->lineno;
1364 ++op->lineno;
1365 /* Traditionally, end of line ends a string constant with no error.
1366 So exit the loop and record the new line. */
1367 beg_of_line = ibp;
1368 goto while2end;
1370 case '\\':
1371 if (ibp >= limit)
1372 break;
1373 if (*ibp == '\n') {
1374 /* Backslash newline is replaced by nothing at all,
1375 but keep the line counts correct. */
1376 --obp;
1377 ++ibp;
1378 ++ip->lineno;
1379 } else {
1380 /* ANSI stupidly requires that in \\ the second \
1381 is *not* prevented from combining with a newline. */
1382 while (*ibp == '\\' && ibp[1] == '\n') {
1383 ibp += 2;
1384 ++ip->lineno;
1386 *obp++ = *ibp++;
1388 break;
1390 case '\"':
1391 case '\'':
1392 if (ibp[-1] == c)
1393 goto while2end;
1394 break;
1397 while2end:
1398 break;
1400 case '/':
1401 if (*ibp == '\\' && ibp[1] == '\n')
1402 newline_fix (ibp);
1403 /* Don't look for comments inside a macro definition. */
1404 if (ip->macro != 0)
1405 goto randomchar;
1406 /* A comment constitutes white space, so it can terminate an identifier.
1407 Process the identifier, if any. */
1408 if (ident_length)
1409 goto specialchar;
1411 if (*ibp != '*')
1412 goto randomchar;
1414 /* We have a comment. Skip it, optionally copying it to output. */
1416 start_line = ip->lineno;
1418 ++ibp; /* Skip the star. */
1420 /* In K+R C, a comment is equivalent to nothing. Note that we
1421 already output the slash; we might not want it. */
1422 if (! put_out_comments)
1423 obp--;
1424 else
1425 *obp++ = '*';
1428 U_CHAR *before_bp = ibp;
1430 while (ibp < limit) {
1431 switch (*ibp++) {
1432 case '/':
1433 if (warn_comments && ibp < limit && *ibp == '*')
1434 warning("`/*' within comment");
1435 break;
1436 case '*':
1437 if (*ibp == '\\' && ibp[1] == '\n')
1438 newline_fix (ibp);
1439 if (ibp >= limit || *ibp == '/')
1440 goto comment_end;
1441 break;
1442 case '\n':
1443 ++ip->lineno;
1444 /* Copy the newline into the output buffer, in order to
1445 avoid the pain of a #line every time a multiline comment
1446 is seen. */
1447 if (!put_out_comments)
1448 *obp++ = '\n';
1449 ++op->lineno;
1452 comment_end:
1454 if (ibp >= limit)
1455 error_with_line (line_for_error (start_line),
1456 "unterminated comment");
1457 else {
1458 ibp++;
1459 if (put_out_comments) {
1460 memcpy (obp, before_bp, ibp - before_bp);
1461 obp += ibp - before_bp;
1465 break;
1467 case '0': case '1': case '2': case '3': case '4':
1468 case '5': case '6': case '7': case '8': case '9':
1469 /* If digit is not part of identifier, it starts a number,
1470 which means that following letters are not an identifier.
1471 "0x5" does not refer to an identifier "x5".
1472 So copy all alphanumerics that follow without accumulating
1473 as an identifier. Periods also, for sake of "3.e7". */
1475 if (ident_length == 0) {
1476 while (ibp < limit) {
1477 while (ibp < limit && ibp[0] == '\\' && ibp[1] == '\n') {
1478 ++ip->lineno;
1479 ibp += 2;
1481 c = *ibp++;
1482 if (! ISIDNUM (c) && c != '.') {
1483 --ibp;
1484 break;
1486 *obp++ = c;
1487 /* A sign can be part of a preprocessing number
1488 if it follows an e. */
1489 if (c == 'e' || c == 'E') {
1490 while (ibp < limit && ibp[0] == '\\' && ibp[1] == '\n') {
1491 ++ip->lineno;
1492 ibp += 2;
1494 if (ibp < limit && (*ibp == '+' || *ibp == '-')) {
1495 *obp++ = *ibp++;
1496 /* Traditional C does not let the token go past the sign. */
1497 break;
1501 break;
1503 /* fall through */
1505 case '_':
1506 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
1507 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
1508 case 'm': case 'n': case 'o': case 'p': case 'q': case 'r':
1509 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
1510 case 'y': case 'z':
1511 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
1512 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
1513 case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R':
1514 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
1515 case 'Y': case 'Z':
1516 ident_length++;
1517 /* Compute step of hash function, to avoid a proc call on every token */
1518 hash = HASHSTEP (hash, c);
1519 break;
1521 case '\n':
1522 /* If reprocessing a macro expansion, newline is a special marker. */
1523 if (ip->macro != 0) {
1524 /* Newline White is a "funny space" to separate tokens that are
1525 supposed to be separate but without space between.
1526 Here White means any horizontal whitespace character.
1527 Newline - marks a recursive macro use that is not
1528 supposed to be expandable. */
1530 if (*ibp == '-') {
1531 /* Newline - inhibits expansion of preceding token.
1532 If expanding a macro arg, we keep the newline -.
1533 In final output, it is deleted. */
1534 if (! concatenated) {
1535 ident_length = 0;
1536 hash = 0;
1538 ibp++;
1539 if (!output_marks) {
1540 obp--;
1541 } else {
1542 /* If expanding a macro arg, keep the newline -. */
1543 *obp++ = '-';
1545 } else if (is_space (*ibp)) {
1546 /* Newline Space does not prevent expansion of preceding token
1547 so expand the preceding token and then come back. */
1548 if (ident_length > 0)
1549 goto specialchar;
1551 /* If generating final output, newline space makes a space. */
1552 if (!output_marks) {
1553 obp[-1] = *ibp++;
1554 /* And Newline Newline makes a newline, so count it. */
1555 if (obp[-1] == '\n')
1556 op->lineno++;
1557 } else {
1558 /* If expanding a macro arg, keep the newline space.
1559 If the arg gets stringified, newline space makes nothing. */
1560 *obp++ = *ibp++;
1562 } else abort (); /* Newline followed by something random? */
1563 break;
1566 /* If there is a pending identifier, handle it and come back here. */
1567 if (ident_length > 0)
1568 goto specialchar;
1570 beg_of_line = ibp;
1572 /* Update the line counts and output a #line if necessary. */
1573 ++ip->lineno;
1574 ++op->lineno;
1575 if (ip->lineno != op->lineno) {
1576 op->bufp = obp;
1577 output_line_command (ip, op, 1, same_file);
1578 check_expand (op, ip->length - (ip->bufp - ip->buf));
1579 obp = op->bufp;
1581 break;
1583 /* Come here either after (1) a null character that is part of the input
1584 or (2) at the end of the input, because there is a null there. */
1585 case 0:
1586 if (ibp <= limit)
1587 /* Our input really contains a null character. */
1588 goto randomchar;
1590 /* At end of a macro-expansion level, pop it and read next level. */
1591 if (ip->macro != 0) {
1592 obp--;
1593 ibp--;
1594 /* If we have an identifier that ends here, process it now, so
1595 we get the right error for recursion. */
1596 if (ident_length && ! is_idchar (*instack[indepth - 1].bufp)) {
1597 redo_char = 1;
1598 goto randomchar;
1600 POPMACRO;
1601 RECACHE;
1602 break;
1605 /* If we don't have a pending identifier,
1606 return at end of input. */
1607 if (ident_length == 0) {
1608 obp--;
1609 ibp--;
1610 op->bufp = obp;
1611 ip->bufp = ibp;
1612 goto ending;
1615 /* If we do have a pending identifier, just consider this null
1616 a special character and arrange to dispatch on it again.
1617 The second time, IDENT_LENGTH will be zero so we will return. */
1619 /* Fall through */
1621 specialchar:
1623 /* Handle the case of a character such as /, ', " or null
1624 seen following an identifier. Back over it so that
1625 after the identifier is processed the special char
1626 will be dispatched on again. */
1628 ibp--;
1629 obp--;
1630 redo_char = 1;
1632 default:
1634 randomchar:
1636 if (ident_length > 0) {
1637 HASHNODE *hp;
1639 /* We have just seen an identifier end. If it's a macro, expand it.
1641 IDENT_LENGTH is the length of the identifier
1642 and HASH is its hash code.
1644 The identifier has already been copied to the output,
1645 so if it is a macro we must remove it.
1647 If REDO_CHAR is 0, the char that terminated the identifier
1648 has been skipped in the output and the input.
1649 OBP-IDENT_LENGTH-1 points to the identifier.
1650 If the identifier is a macro, we must back over the terminator.
1652 If REDO_CHAR is 1, the terminating char has already been
1653 backed over. OBP-IDENT_LENGTH points to the identifier. */
1655 for (hp = hashtab[MAKE_POS (hash) % HASHSIZE]; hp != NULL;
1656 hp = hp->next) {
1658 if (hp->length == ident_length) {
1659 /* obufp_before_macroname is used only in this block,
1660 but it has to be global because of RECACHE. */
1661 int op_lineno_before_macroname;
1662 int i = ident_length;
1663 U_CHAR *p = hp->name;
1664 U_CHAR *q = obp - i;
1666 if (! redo_char)
1667 q--;
1669 do { /* All this to avoid a strncmp () */
1670 if (*p++ != *q++)
1671 goto hashcollision;
1672 } while (--i);
1674 /* We found a use of a macro name.
1675 see if the context shows it is a macro call. */
1677 /* Back up over terminating character if not already done. */
1678 if (! redo_char) {
1679 ibp--;
1680 obp--;
1683 obufp_before_macroname = obp - ident_length;
1684 op_lineno_before_macroname = op->lineno;
1686 /* If macro wants an arglist, verify that a '(' follows.
1687 first skip all whitespace, copying it to the output
1688 after the macro name. Then, if there is no '(',
1689 decide this is not a macro call and leave things that way. */
1690 if (hp->type == T_MACRO && hp->value.defn->nargs >= 0)
1692 while (1) {
1693 /* Scan forward over whitespace, copying it to the output. */
1694 if (ibp == limit && ip->macro != 0) {
1695 POPMACRO;
1696 RECACHE;
1698 /* A comment: copy it unchanged or discard it. */
1699 else if (*ibp == '/' && ibp+1 != limit && ibp[1] == '*') {
1700 if (put_out_comments) {
1701 *obp++ = '/';
1702 *obp++ = '*';
1704 ibp += 2;
1705 while (ibp + 1 != limit
1706 && !(ibp[0] == '*' && ibp[1] == '/')) {
1707 /* We need not worry about newline-marks,
1708 since they are never found in comments. */
1709 if (*ibp == '\n') {
1710 /* Newline in a file. Count it. */
1711 ++ip->lineno;
1712 ++op->lineno;
1714 if (put_out_comments)
1715 *obp++ = *ibp++;
1716 else
1717 ibp++;
1719 ibp += 2;
1720 if (put_out_comments) {
1721 *obp++ = '*';
1722 *obp++ = '/';
1725 else if (is_space (*ibp)) {
1726 *obp++ = *ibp++;
1727 if (ibp[-1] == '\n') {
1728 if (ip->macro == 0) {
1729 /* Newline in a file. Count it. */
1730 ++ip->lineno;
1731 ++op->lineno;
1732 } else if (!output_marks) {
1733 /* A newline mark, and we don't want marks
1734 in the output. If it is newline-hyphen,
1735 discard it entirely. Otherwise, it is
1736 newline-whitechar, so keep the whitechar. */
1737 obp--;
1738 if (*ibp == '-')
1739 ibp++;
1740 else {
1741 if (*ibp == '\n')
1742 ++op->lineno;
1743 *obp++ = *ibp++;
1745 } else {
1746 /* A newline mark; copy both chars to the output. */
1747 *obp++ = *ibp++;
1751 else break;
1753 if (*ibp != '(')
1754 break;
1757 /* This is now known to be a macro call.
1758 Discard the macro name from the output,
1759 along with any following whitespace just copied. */
1760 obp = obufp_before_macroname;
1761 op->lineno = op_lineno_before_macroname;
1763 /* Expand the macro, reading arguments as needed,
1764 and push the expansion on the input stack. */
1765 ip->bufp = ibp;
1766 op->bufp = obp;
1767 macroexpand (hp, op);
1769 /* Reexamine input stack, since macroexpand has pushed
1770 a new level on it. */
1771 obp = op->bufp;
1772 RECACHE;
1773 break;
1775 hashcollision:
1777 } /* End hash-table-search loop */
1778 ident_length = hash = 0; /* Stop collecting identifier */
1779 redo_char = 0;
1780 concatenated = 0;
1781 } /* End if (ident_length > 0) */
1782 } /* End switch */
1783 } /* End per-char loop */
1785 /* Come here to return -- but first give an error message
1786 if there was an unterminated successful conditional. */
1787 ending:
1788 if (if_stack != ip->if_stack) {
1789 const char *str;
1790 switch (if_stack->type) {
1791 case T_IF:
1792 str = "if";
1793 break;
1794 case T_IFDEF:
1795 str = "ifdef";
1796 break;
1797 case T_IFNDEF:
1798 str = "ifndef";
1799 break;
1800 case T_ELSE:
1801 str = "else";
1802 break;
1803 case T_ELIF:
1804 str = "elif";
1805 break;
1806 default:
1807 abort ();
1809 error_with_line (line_for_error (if_stack->lineno),
1810 "unterminated #%s conditional", str);
1812 if_stack = ip->if_stack;
1816 * Rescan a string into a temporary buffer and return the result
1817 * as a FILE_BUF. Note this function returns a struct, not a pointer.
1819 * OUTPUT_MARKS nonzero means keep Newline markers found in the input
1820 * and insert such markers when appropriate. See `rescan' for details.
1821 * OUTPUT_MARKS is 1 for macroexpanding a macro argument separately
1822 * before substitution; it is 0 for other uses.
1824 static FILE_BUF
1825 expand_to_temp_buffer (buf, limit, output_marks)
1826 const U_CHAR *buf, *limit;
1827 int output_marks;
1829 FILE_BUF *ip;
1830 FILE_BUF obuf;
1831 int length = limit - buf;
1832 U_CHAR *buf1;
1833 int odepth = indepth;
1835 if (length < 0)
1836 abort ();
1838 /* Set up the input on the input stack. */
1840 buf1 = (U_CHAR *) alloca (length + 1);
1842 const U_CHAR *p1 = buf;
1843 U_CHAR *p2 = buf1;
1845 while (p1 != limit)
1846 *p2++ = *p1++;
1848 buf1[length] = 0;
1850 /* Set up to receive the output. */
1852 obuf.length = length * 2 + 100; /* Usually enough. Why be stingy? */
1853 obuf.bufp = obuf.buf = (U_CHAR *) xmalloc (obuf.length);
1854 obuf.fname = 0;
1855 obuf.macro = 0;
1856 obuf.free_ptr = 0;
1858 CHECK_DEPTH ({return obuf;});
1860 ++indepth;
1862 ip = &instack[indepth];
1863 ip->fname = 0;
1864 ip->macro = 0;
1865 ip->free_ptr = 0;
1866 ip->length = length;
1867 ip->buf = ip->bufp = buf1;
1868 ip->if_stack = if_stack;
1870 ip->lineno = obuf.lineno = 1;
1872 /* Scan the input, create the output. */
1874 rescan (&obuf, output_marks);
1876 /* Pop input stack to original state. */
1877 --indepth;
1879 if (indepth != odepth)
1880 abort ();
1882 /* Record the output. */
1883 obuf.length = obuf.bufp - obuf.buf;
1885 return obuf;
1889 * Process a # directive. Expects IP->bufp to point to the '#', as in
1890 * `#define foo bar'. Passes to the command handler
1891 * (do_define, do_include, etc.): the addresses of the 1st and
1892 * last chars of the command (starting immediately after the #
1893 * keyword), plus op and the keyword table pointer. If the command
1894 * contains comments it is copied into a temporary buffer sans comments
1895 * and the temporary buffer is passed to the command handler instead.
1896 * Likewise for backslash-newlines.
1898 * Returns nonzero if this was a known # directive.
1899 * Otherwise, returns zero, without advancing the input pointer.
1902 static int
1903 handle_directive (ip, op)
1904 FILE_BUF *ip, *op;
1906 U_CHAR *bp, *cp;
1907 const struct directive *kt;
1908 int ident_length;
1909 U_CHAR *resume_p;
1911 /* Nonzero means we must copy the entire command
1912 to get rid of comments or backslash-newlines. */
1913 int copy_command = 0;
1915 U_CHAR *ident, *after_ident;
1917 bp = ip->bufp;
1918 /* Skip whitespace and \-newline. */
1919 while (1) {
1920 if (is_nvspace (*bp))
1921 bp++;
1922 else if (*bp == '/' && (newline_fix (bp + 1), bp[1]) == '*') {
1923 ip->bufp = bp;
1924 skip_to_end_of_comment (ip, &ip->lineno);
1925 bp = ip->bufp;
1926 } else if (*bp == '\\' && bp[1] == '\n') {
1927 bp += 2; ip->lineno++;
1928 } else break;
1931 /* Now find end of directive name.
1932 If we encounter a backslash-newline, exchange it with any following
1933 symbol-constituents so that we end up with a contiguous name. */
1935 cp = bp;
1936 while (1) {
1937 if (is_idchar (*cp))
1938 cp++;
1939 else {
1940 if (*cp == '\\' && cp[1] == '\n')
1941 name_newline_fix (cp);
1942 if (is_idchar (*cp))
1943 cp++;
1944 else break;
1947 ident_length = cp - bp;
1948 ident = bp;
1949 after_ident = cp;
1951 /* A line of just `#' becomes blank. */
1953 if (ident_length == 0 && *after_ident == '\n') {
1954 ip->bufp = after_ident;
1955 return 1;
1959 * Decode the keyword and call the appropriate expansion
1960 * routine, after moving the input pointer up to the next line.
1962 for (kt = directive_table; kt->length > 0; kt++) {
1963 if (kt->length == ident_length
1964 && !strncmp (kt->name, (const char *)ident, ident_length)) {
1965 U_CHAR *buf;
1966 U_CHAR *limit = ip->buf + ip->length;
1967 int unterminated = 0;
1969 /* Nonzero means do not delete comments within the directive.
1970 #define needs this to detect traditional token paste. */
1971 int keep_comments = kt->type == T_DEFINE;
1973 /* Find the end of this command (first newline not backslashed
1974 and not in a string or comment).
1975 Set COPY_COMMAND if the command must be copied
1976 (it contains a backslash-newline or a comment). */
1978 buf = bp = after_ident;
1979 while (bp < limit) {
1980 U_CHAR c = *bp++;
1981 switch (c) {
1982 case '\\':
1983 if (bp < limit) {
1984 if (*bp == '\n') {
1985 ip->lineno++;
1986 copy_command = 1;
1988 bp++;
1990 break;
1992 case '\'':
1993 case '\"':
1994 bp = skip_quoted_string (bp - 1, limit, ip->lineno, &ip->lineno, &copy_command, &unterminated);
1995 if (unterminated) {
1996 /* Traditional preprocessing permits unterminated strings. */
1997 ip->bufp = bp;
1998 goto endloop1;
2000 break;
2002 /* <...> is special for #include. */
2003 case '<':
2004 if (kt->type != T_INCLUDE)
2005 break;
2006 while (*bp && *bp != '>') bp++;
2007 break;
2009 case '/':
2010 if (*bp == '\\' && bp[1] == '\n')
2011 newline_fix (bp);
2012 if (*bp == '*') {
2013 U_CHAR *obp = bp - 1;
2014 ip->bufp = bp + 1;
2015 skip_to_end_of_comment (ip, &ip->lineno);
2016 bp = ip->bufp;
2017 /* No need to copy the command because of a comment at the end;
2018 just don't include the comment in the directive. */
2019 if (bp == limit || *bp == '\n') {
2020 bp = obp;
2021 goto endloop1;
2023 /* Don't remove the comments if this is #define. */
2024 if (! keep_comments)
2025 copy_command++;
2027 break;
2029 case '\n':
2030 --bp; /* Point to the newline */
2031 ip->bufp = bp;
2032 goto endloop1;
2035 ip->bufp = bp;
2037 endloop1:
2038 resume_p = ip->bufp;
2039 /* BP is the end of the directive.
2040 RESUME_P is the next interesting data after the directive.
2041 A comment may come between. */
2043 if (copy_command) {
2044 U_CHAR *xp = buf;
2045 /* Need to copy entire command into temp buffer before dispatching */
2047 cp = (U_CHAR *) alloca (bp - buf + 5); /* room for cmd plus
2048 some slop */
2049 buf = cp;
2051 /* Copy to the new buffer, deleting comments
2052 and backslash-newlines (and whitespace surrounding the latter). */
2054 while (xp < bp) {
2055 U_CHAR c = *xp++;
2056 *cp++ = c;
2058 switch (c) {
2059 case '\n':
2060 break;
2062 /* <...> is special for #include. */
2063 case '<':
2064 if (kt->type != T_INCLUDE)
2065 break;
2066 while (xp < bp && c != '>') {
2067 c = *xp++;
2068 if (c == '\\' && xp < bp && *xp == '\n')
2069 xp++, ip->lineno++;
2070 else
2071 *cp++ = c;
2073 break;
2075 case '\\':
2076 if (*xp == '\n') {
2077 xp++;
2078 cp--;
2079 if (cp != buf && is_space (cp[-1])) {
2080 while (cp != buf && is_space(cp[-1])) cp--;
2081 cp++;
2082 SKIP_WHITE_SPACE (xp);
2083 } else if (is_nvspace (*xp)) {
2084 *cp++ = *xp++;
2085 SKIP_WHITE_SPACE (xp);
2087 } else {
2088 *cp++ = *xp++;
2090 break;
2092 case '\'':
2093 case '\"':
2095 const U_CHAR *bp1
2096 = skip_quoted_string (xp - 1, limit, ip->lineno, 0, 0, 0);
2097 while (xp != bp1)
2098 *cp++ = *xp++;
2100 break;
2102 case '/':
2103 if (*xp == '*') {
2104 ip->bufp = xp + 1;
2105 skip_to_end_of_comment (ip, 0);
2106 if (keep_comments)
2107 while (xp != ip->bufp)
2108 *cp++ = *xp++;
2109 /* Delete the slash. */
2110 else
2111 cp--;
2112 xp = ip->bufp;
2117 /* Null-terminate the copy. */
2119 *cp = 0;
2121 else
2122 cp = bp;
2124 ip->bufp = resume_p;
2126 /* Call the appropriate command handler. buf now points to
2127 either the appropriate place in the input buffer, or to
2128 the temp buffer if it was necessary to make one. cp
2129 points to the first char after the contents of the (possibly
2130 copied) command, in either case. */
2131 (*kt->func) (buf, cp, op);
2132 check_expand (op, ip->length - (ip->bufp - ip->buf));
2134 return 1;
2138 return 0;
2141 static const char *const
2142 monthnames[] = {"Jan", "Feb", "Mar", "Apr", "May", "Jun",
2143 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"};
2146 * expand things like __FILE__. Place the expansion into the output
2147 * buffer *without* rescanning.
2149 static void
2150 special_symbol (hp, op)
2151 HASHNODE *hp;
2152 FILE_BUF *op;
2154 const char *buf;
2155 time_t t;
2156 int i, len;
2157 int true_indepth;
2158 FILE_BUF *ip = NULL;
2159 static struct tm *timebuf = NULL;
2161 int paren = 0; /* For special `defined' keyword */
2163 for (i = indepth; i >= 0; i--)
2164 if (instack[i].fname != NULL) {
2165 ip = &instack[i];
2166 break;
2168 if (ip == NULL)
2169 fatal ("not in any file?!");
2171 switch (hp->type) {
2172 case T_FILE:
2173 case T_BASE_FILE:
2175 const char *string;
2176 if (hp->type == T_FILE)
2177 string = ip->fname;
2178 else
2179 string = instack[0].fname;
2181 if (string)
2183 char *tmp = (char *) alloca (3 + strlen (string));
2184 sprintf (tmp, "\"%s\"", string);
2185 buf = tmp;
2187 else
2188 buf = "";
2190 break;
2193 case T_INCLUDE_LEVEL:
2195 char *tmp = (char *) alloca (8); /* Eigth bytes ought to be more than enough */
2196 true_indepth = 0;
2197 for (i = indepth; i >= 0; i--)
2198 if (instack[i].fname != NULL)
2199 true_indepth++;
2201 sprintf (tmp, "%d", true_indepth - 1);
2202 buf = tmp;
2203 break;
2206 case T_VERSION:
2208 char *tmp = (char *) alloca (3 + strlen (version_string));
2209 sprintf (tmp, "\"%s\"", version_string);
2210 buf = tmp;
2211 break;
2214 case T_CONST:
2215 buf = hp->value.cpval;
2216 break;
2218 case T_SPECLINE:
2220 char *tmp = (char *) alloca (10);
2221 sprintf (tmp, "%d", ip->lineno);
2222 buf = tmp;
2223 break;
2226 case T_DATE:
2227 case T_TIME:
2229 char *tmp = (char *) alloca (20);
2231 if (timebuf == NULL) {
2232 t = time (0);
2233 timebuf = localtime (&t);
2235 if (hp->type == T_DATE)
2236 sprintf (tmp, "\"%s %2d %4d\"", monthnames[timebuf->tm_mon],
2237 timebuf->tm_mday, timebuf->tm_year + 1900);
2238 else
2239 sprintf (tmp, "\"%02d:%02d:%02d\"", timebuf->tm_hour, timebuf->tm_min,
2240 timebuf->tm_sec);
2241 buf = tmp;
2242 break;
2245 case T_SPEC_DEFINED:
2246 buf = " 0 "; /* Assume symbol is not defined */
2247 ip = &instack[indepth];
2248 SKIP_WHITE_SPACE (ip->bufp);
2249 if (*ip->bufp == '(') {
2250 paren++;
2251 ip->bufp++; /* Skip over the paren */
2252 SKIP_WHITE_SPACE (ip->bufp);
2255 if (!is_idstart (*ip->bufp))
2256 goto oops;
2258 HASHNODE *hp = lookup (ip->bufp, -1, -1);
2260 if (hp && hp->type != T_UNUSED && hp->type != T_SPEC_DEFINED)
2261 buf = " 1 ";
2263 while (is_idchar (*ip->bufp))
2264 ++ip->bufp;
2265 SKIP_WHITE_SPACE (ip->bufp);
2266 if (paren) {
2267 if (*ip->bufp != ')')
2268 goto oops;
2269 ++ip->bufp;
2271 break;
2273 oops:
2275 error ("`defined' must be followed by ident or (ident)");
2276 break;
2278 default:
2279 error ("cccp error: invalid special hash type"); /* time for gdb */
2280 abort ();
2282 len = strlen (buf);
2283 check_expand (op, len);
2284 memcpy (op->bufp, buf, len);
2285 op->bufp += len;
2289 /* Routines to handle #directives */
2292 * Process include file by reading it in and calling rescan.
2293 * Expects to see "fname" or <fname> on the input.
2295 static void
2296 do_include (buf, limit, op)
2297 U_CHAR *buf, *limit;
2298 FILE_BUF *op;
2300 U_CHAR *fbeg, *fend; /* Beginning and end of fname */
2302 struct file_name_list *stackp = include; /* Chain of dirs to search */
2303 struct file_name_list dsp[1]; /* First in chain, if #include "..." */
2304 int flen;
2306 int retried = 0; /* Have already tried macro
2307 expanding the include line*/
2308 FILE_BUF trybuf; /* It got expanded into here */
2309 int system_header_p = 0; /* 0 for "...", 1 for <...> */
2311 get_filename:
2313 fbeg = buf;
2314 SKIP_WHITE_SPACE (fbeg);
2315 /* Discard trailing whitespace so we can easily see
2316 if we have parsed all the significant chars we were given. */
2317 while (limit != fbeg && is_nvspace (limit[-1])) limit--;
2319 switch (*fbeg++) {
2320 case '\"':
2321 fend = fbeg;
2322 while (fend != limit && *fend != '\"')
2323 fend++;
2324 if (*fend == '\"' && fend + 1 == limit) {
2325 FILE_BUF *fp;
2327 /* We have "filename". Figure out directory this source
2328 file is coming from and put it on the front of the list. */
2330 /* If -I- was specified, don't search current dir, only spec'd ones. */
2331 if (ignore_srcdir) break;
2333 for (fp = &instack[indepth]; fp >= instack; fp--)
2335 size_t n;
2336 const char *ep, *nam;
2338 if ((nam = fp->fname) != NULL) {
2339 /* Found a named file. Figure out dir of the file,
2340 and put it in front of the search list. */
2341 dsp[0].next = stackp;
2342 stackp = dsp;
2343 ep = strrchr (nam, '/');
2344 if (ep != NULL) {
2345 char *f;
2346 n = ep - nam;
2347 f = (char *) alloca (n + 1);
2348 strncpy (f, nam, n);
2349 f[n] = '\0';
2350 dsp[0].fname = f;
2351 if (n > max_include_len) max_include_len = n;
2352 } else {
2353 dsp[0].fname = 0; /* Current directory */
2355 break;
2358 break;
2360 goto fail;
2362 case '<':
2363 fend = fbeg;
2364 while (fend != limit && *fend != '>') fend++;
2365 if (*fend == '>' && fend + 1 == limit) {
2366 system_header_p = 1;
2367 /* If -I-, start with the first -I dir after the -I-. */
2368 if (first_bracket_include)
2369 stackp = first_bracket_include;
2370 break;
2372 goto fail;
2374 default:
2375 fail:
2376 if (retried) {
2377 error ("#include expects \"fname\" or <fname>");
2378 return;
2379 } else {
2380 trybuf = expand_to_temp_buffer (buf, limit, 0);
2381 buf = (U_CHAR *) alloca (trybuf.bufp - trybuf.buf + 1);
2382 memcpy (buf, trybuf.buf, trybuf.bufp - trybuf.buf);
2383 limit = buf + (trybuf.bufp - trybuf.buf);
2384 free (trybuf.buf);
2385 retried++;
2386 goto get_filename;
2390 flen = fend - fbeg;
2391 process_include (stackp, fbeg, flen, system_header_p, op);
2394 static void
2395 do_include_next (buf, limit, op)
2396 U_CHAR *buf, *limit;
2397 FILE_BUF *op;
2399 U_CHAR *fbeg, *fend; /* Beginning and end of fname */
2401 struct file_name_list *stackp; /* Chain of dirs to search */
2402 int flen;
2404 int retried = 0; /* Have already tried macro
2405 expanding the include line*/
2406 FILE_BUF trybuf; /* It got expanded into here */
2407 int system_header_p = 0; /* 0 for "...", 1 for <...> */
2409 /* Treat as plain #include if we don't know where to start
2410 looking. */
2411 stackp = instack[indepth].next_header_dir;
2412 if (stackp == 0)
2414 do_include (buf, limit, op);
2415 return;
2418 get_filename:
2420 fbeg = buf;
2421 SKIP_WHITE_SPACE (fbeg);
2422 /* Discard trailing whitespace so we can easily see
2423 if we have parsed all the significant chars we were given. */
2424 while (limit != fbeg && is_nvspace (limit[-1])) limit--;
2426 switch (*fbeg++) {
2427 case '\"':
2428 fend = fbeg;
2429 while (fend != limit && *fend != '\"')
2430 fend++;
2431 if (*fend == '\"' && fend + 1 == limit)
2432 break;
2433 goto fail;
2435 case '<':
2436 fend = fbeg;
2437 while (fend != limit && *fend != '>') fend++;
2438 if (*fend == '>' && fend + 1 == limit) {
2439 system_header_p = 1;
2440 break;
2442 goto fail;
2444 default:
2445 fail:
2446 if (retried) {
2447 error ("#include expects \"fname\" or <fname>");
2448 return;
2449 } else {
2450 trybuf = expand_to_temp_buffer (buf, limit, 0);
2451 buf = (U_CHAR *) alloca (trybuf.bufp - trybuf.buf + 1);
2452 memcpy (buf, trybuf.buf, trybuf.bufp - trybuf.buf);
2453 limit = buf + (trybuf.bufp - trybuf.buf);
2454 free (trybuf.buf);
2455 retried++;
2456 goto get_filename;
2460 flen = fend - fbeg;
2461 process_include (stackp, fbeg, flen, system_header_p, op);
2464 static void
2465 process_include (stackp, fbeg, flen, system_header_p, op)
2466 struct file_name_list *stackp;
2467 const U_CHAR *fbeg;
2468 int flen;
2469 int system_header_p;
2470 FILE_BUF *op;
2472 char *fname;
2473 int f = -1; /* file number */
2475 fname = (char *) alloca (max_include_len + flen + 2);
2476 /* + 2 above for slash and terminating null. */
2478 /* If specified file name is absolute, just open it. */
2480 if (IS_ABSOLUTE_PATHNAME (fbeg)) {
2481 strncpy (fname, (const char *)fbeg, flen);
2482 fname[flen] = 0;
2483 f = open (fname, O_RDONLY, 0666);
2484 } else {
2485 /* Search directory path, trying to open the file.
2486 Copy each filename tried into FNAME. */
2488 for (; stackp; stackp = stackp->next) {
2489 if (stackp->fname) {
2490 strcpy (fname, stackp->fname);
2491 strcat (fname, "/");
2492 fname[strlen (fname) + flen] = 0;
2493 } else {
2494 fname[0] = 0;
2496 strncat (fname, (const char *)fbeg, flen);
2497 if ((f = open (fname, O_RDONLY, 0666)) >= 0)
2498 break;
2502 if (f < 0) {
2503 strncpy (fname, (const char *)fbeg, flen);
2504 fname[flen] = 0;
2505 if (deps_missing_files
2506 && print_deps > (system_header_p || (system_include_depth > 0))) {
2508 /* If requested as a system header, assume it belongs in
2509 the first system header directory. */
2510 if (first_bracket_include)
2511 stackp = first_bracket_include;
2512 else
2513 stackp = include;
2515 if (!system_header_p || IS_ABSOLUTE_PATHNAME (fbeg) || !stackp->fname)
2516 deps_add_dep (deps, fname);
2517 else {
2518 char *p;
2519 int len = strlen(stackp->fname);
2521 p = (char *) alloca (len + flen + 2);
2522 memcpy (p, stackp->fname, len);
2523 p[len++] = '/';
2524 memcpy (p + len, fbeg, flen);
2525 len += flen;
2526 p[len] = '\0';
2527 deps_add_dep (deps, p);
2529 } else if (print_deps
2530 && print_deps <= (system_header_p
2531 || (system_include_depth > 0)))
2532 warning ("no include path in which to find %.*s", flen, fbeg);
2533 else
2534 error_from_errno (fname);
2536 } else {
2538 /* Check to see if this include file is a once-only include file.
2539 If so, give up. */
2541 struct file_name_list* ptr;
2543 for (ptr = dont_repeat_files; ptr; ptr = ptr->next) {
2544 if (!strcmp (ptr->fname, fname)) {
2545 close (f);
2546 return; /* This file was once'd. */
2550 for (ptr = all_include_files; ptr; ptr = ptr->next) {
2551 if (!strcmp (ptr->fname, fname))
2552 break; /* This file was included before. */
2555 if (ptr == 0) {
2556 /* This is the first time for this file. */
2557 /* Add it to list of files included. */
2559 ptr = (struct file_name_list *) xmalloc (sizeof (struct file_name_list));
2560 ptr->next = all_include_files;
2561 all_include_files = ptr;
2562 ptr->fname = xstrdup (fname);
2564 /* For -M, add this file to the dependencies. */
2565 if (print_deps > (system_header_p || (system_include_depth > 0)))
2566 deps_add_dep (deps, fname);
2569 if (system_header_p)
2570 system_include_depth++;
2572 /* Actually process the file. */
2573 finclude (f, fname, stackp->next, op);
2575 if (system_header_p)
2576 system_include_depth--;
2578 close (f);
2582 /* Process the contents of include file FNAME, already open on descriptor F,
2583 with output to OP. */
2585 static void
2586 finclude (f, fname, nhd, op)
2587 int f;
2588 const char *fname;
2589 struct file_name_list *nhd;
2590 FILE_BUF *op;
2592 int st_mode;
2593 long st_size;
2594 long i;
2595 FILE_BUF *fp; /* For input stack frame */
2597 CHECK_DEPTH (return;);
2599 if (file_size_and_mode (f, &st_mode, &st_size))
2600 goto nope;
2602 fp = &instack[indepth + 1];
2603 memset (fp, 0, sizeof (FILE_BUF));
2604 fp->fname = fname;
2605 fp->length = 0;
2606 fp->lineno = 1;
2607 fp->if_stack = if_stack;
2608 fp->next_header_dir = nhd;
2610 if (S_ISREG (st_mode)) {
2611 fp->buf = (U_CHAR *) xmalloc (st_size + 2);
2612 fp->bufp = fp->buf;
2614 /* Read the file contents, knowing that st_size is an upper bound
2615 on the number of bytes we can read. */
2616 while (st_size > 0) {
2617 i = read (f, fp->buf + fp->length, st_size);
2618 if (i <= 0) {
2619 if (i == 0) break;
2620 goto nope;
2622 fp->length += i;
2623 st_size -= i;
2626 else {
2627 /* Cannot count its file size before reading. */
2629 U_CHAR *bufp;
2630 U_CHAR *basep;
2631 int bsize = 2000;
2633 st_size = 0;
2634 basep = (U_CHAR *) xmalloc (bsize + 2);
2635 bufp = basep;
2637 for (;;) {
2638 i = read (f, bufp, bsize - st_size);
2639 if (i < 0)
2640 goto nope; /* error! */
2641 if (i == 0)
2642 break; /* End of file */
2643 st_size += i;
2644 bufp += i;
2645 if (bsize == st_size) { /* Buffer is full! */
2646 bsize *= 2;
2647 basep = (U_CHAR *) xrealloc (basep, bsize + 2);
2648 bufp = basep + st_size; /* May have moved */
2651 fp->buf = basep;
2652 fp->bufp = fp->buf;
2653 fp->length = st_size;
2655 close (f);
2657 /* Make sure data ends with a newline. And put a null after it. */
2659 if (fp->length > 0 && fp->buf[fp->length-1] != '\n')
2660 fp->buf[fp->length++] = '\n';
2661 fp->buf[fp->length] = '\0';
2663 indepth++;
2664 output_line_command (fp, op, 0, enter_file);
2665 rescan (op, 0);
2666 indepth--;
2667 instack[indepth].lineno++;
2668 instack[indepth].bufp++; /* Skip the new line. */
2669 output_line_command (&instack[indepth], op, 0, leave_file);
2670 free (fp->buf);
2671 return;
2673 nope:
2674 perror_with_name (fname);
2675 close (f);
2679 /* Process a #define command.
2680 BUF points to the contents of the #define command, as a continguous string.
2681 LIMIT points to the first character past the end of the definition.
2682 KEYWORD is the keyword-table entry for #define. */
2684 static void
2685 do_define (buf, limit, op)
2686 U_CHAR *buf, *limit;
2687 FILE_BUF *op ATTRIBUTE_UNUSED;
2689 U_CHAR *bp; /* temp ptr into input buffer */
2690 U_CHAR *symname; /* remember where symbol name starts */
2691 int sym_length; /* and how long it is */
2693 DEFINITION *defn;
2694 int arglengths = 0; /* Accumulate lengths of arg names
2695 plus number of args. */
2696 int hashcode;
2698 bp = buf;
2700 while (is_nvspace (*bp))
2701 bp++;
2703 symname = bp; /* remember where it starts */
2704 while (is_idchar (*bp) && bp < limit) {
2705 bp++;
2707 sym_length = bp - symname;
2708 if (sym_length == 0)
2710 error ("invalid macro name");
2711 return;
2713 else if (!is_idstart (*symname)) {
2714 U_CHAR *msg; /* what pain... */
2715 msg = (U_CHAR *) alloca (sym_length + 1);
2716 memcpy (msg, symname, sym_length);
2717 msg[sym_length] = 0;
2718 error ("invalid macro name `%s'", msg);
2719 return;
2720 } else {
2721 if (! strncmp ((const char *)symname, "defined", 7) && sym_length == 7)
2723 error ("\"defined\" cannot be used as a macro name");
2724 return;
2728 /* lossage will occur if identifiers or control keywords are broken
2729 across lines using backslash. This is not the right place to take
2730 care of that. */
2732 if (*bp == '(') {
2733 struct arglist *arg_ptrs = NULL;
2734 int argno = 0;
2736 bp++; /* skip '(' */
2737 SKIP_WHITE_SPACE (bp);
2739 /* Loop over macro argument names. */
2740 while (*bp != ')') {
2741 struct arglist *temp;
2743 temp = (struct arglist *) alloca (sizeof (struct arglist));
2744 temp->name = bp;
2745 temp->next = arg_ptrs;
2746 temp->argno = argno++;
2747 arg_ptrs = temp;
2749 if (!is_idstart (*bp))
2750 warning ("parameter name starts with a digit in #define");
2752 /* Find the end of the arg name. */
2753 while (is_idchar (*bp)) {
2754 bp++;
2756 temp->length = bp - temp->name;
2757 arglengths += temp->length + 2;
2758 SKIP_WHITE_SPACE (bp);
2759 if (temp->length == 0 || (*bp != ',' && *bp != ')')) {
2760 error ("badly punctuated parameter list in #define");
2761 return;
2763 if (*bp == ',') {
2764 bp++;
2765 SKIP_WHITE_SPACE (bp);
2767 if (bp >= limit) {
2768 error ("unterminated parameter list in #define");
2769 return;
2773 ++bp; /* skip paren */
2774 while (is_nvspace (*bp) && bp < limit) /* and leading whitespace */
2775 ++bp;
2776 /* now everything from bp before limit is the definition. */
2777 defn = collect_expansion (bp, limit, argno, arg_ptrs);
2779 /* Now set defn->argnames to the result of concatenating
2780 the argument names in reverse order
2781 with comma-space between them. */
2783 struct arglist *temp;
2784 int i = 0;
2785 U_CHAR *tmp = (U_CHAR *) xmalloc (arglengths + 1);
2787 for (temp = arg_ptrs; temp; temp = temp->next) {
2788 memcpy (&tmp[i], temp->name, temp->length);
2789 i += temp->length;
2790 if (temp->next != 0) {
2791 tmp[i++] = ',';
2792 tmp[i++] = ' ';
2795 tmp[i] = 0;
2796 defn->argnames = tmp;
2799 } else {
2800 /* simple expansion or empty definition; skip leading whitespace */
2801 while (is_nvspace (*bp) && bp < limit)
2802 ++bp;
2803 /* now everything from bp before limit is the definition. */
2804 defn = collect_expansion (bp, limit, -1, 0);
2805 defn->argnames = (const U_CHAR *) "";
2808 hashcode = hashf (symname, sym_length, HASHSIZE);
2811 HASHNODE *hp;
2812 if ((hp = lookup (symname, sym_length, hashcode)) == NULL)
2813 hp = install (symname, sym_length, T_MACRO, hashcode);
2814 else {
2815 if (hp->type != T_MACRO || compare_defs (defn, hp->value.defn))
2816 warning ("\"%.*s\" redefined", sym_length, symname);
2818 /* Replace the old definition. */
2819 hp->type = T_MACRO;
2822 hp->value.defn = defn;
2827 * return zero if two DEFINITIONs are isomorphic
2829 static int
2830 compare_defs (d1, d2)
2831 DEFINITION *d1, *d2;
2833 struct reflist *a1, *a2;
2834 U_CHAR *p1 = d1->expansion;
2835 U_CHAR *p2 = d2->expansion;
2836 int first = 1;
2838 if (d1->nargs != d2->nargs)
2839 return 1;
2840 if (strcmp ((const char *)d1->argnames, (const char *)d2->argnames))
2841 return 1;
2842 for (a1 = d1->pattern, a2 = d2->pattern; a1 && a2;
2843 a1 = a1->next, a2 = a2->next) {
2844 if (!((a1->nchars == a2->nchars
2845 && ! strncmp ((const char *)p1, (const char *)p2, a1->nchars))
2846 || ! comp_def_part (first, p1, a1->nchars, p2, a2->nchars, 0))
2847 || a1->argno != a2->argno
2848 || a1->stringify != a2->stringify
2849 || a1->raw_before != a2->raw_before
2850 || a1->raw_after != a2->raw_after)
2851 return 1;
2852 first = 0;
2853 p1 += a1->nchars;
2854 p2 += a2->nchars;
2856 if (a1 != a2)
2857 return 1;
2858 if (comp_def_part (first, p1, d1->length - (p1 - d1->expansion),
2859 p2, d2->length - (p2 - d2->expansion), 1))
2860 return 1;
2861 return 0;
2864 /* Return 1 if two parts of two macro definitions are effectively different.
2865 One of the parts starts at BEG1 and has LEN1 chars;
2866 the other has LEN2 chars at BEG2.
2867 Any sequence of whitespace matches any other sequence of whitespace.
2868 FIRST means these parts are the first of a macro definition;
2869 so ignore leading whitespace entirely.
2870 LAST means these parts are the last of a macro definition;
2871 so ignore trailing whitespace entirely. */
2872 static int
2873 comp_def_part (first, beg1, len1, beg2, len2, last)
2874 int first;
2875 const U_CHAR *beg1, *beg2;
2876 int len1, len2;
2877 int last;
2879 const U_CHAR *end1 = beg1 + len1;
2880 const U_CHAR *end2 = beg2 + len2;
2881 if (first) {
2882 while (beg1 != end1 && is_space (*beg1)) beg1++;
2883 while (beg2 != end2 && is_space (*beg2)) beg2++;
2885 if (last) {
2886 while (beg1 != end1 && is_space (end1[-1])) end1--;
2887 while (beg2 != end2 && is_space (end2[-1])) end2--;
2889 while (beg1 != end1 && beg2 != end2) {
2890 if (is_space (*beg1) && is_space (*beg2)) {
2891 while (beg1 != end1 && is_space (*beg1)) beg1++;
2892 while (beg2 != end2 && is_space (*beg2)) beg2++;
2893 } else if (*beg1 == *beg2) {
2894 beg1++; beg2++;
2895 } else break;
2897 return (beg1 != end1) || (beg2 != end2);
2900 /* Read a replacement list for a macro with parameters.
2901 Build the DEFINITION structure.
2902 Reads characters of text starting at BUF until LIMIT.
2903 ARGLIST specifies the formal parameters to look for
2904 in the text of the definition; NARGS is the number of args
2905 in that list, or -1 for a macro name that wants no argument list.
2906 MACRONAME is the macro name itself (so we can avoid recursive expansion)
2907 and NAMELEN is its length in characters.
2909 Note that comments and backslash-newlines have already been deleted
2910 from the argument. */
2912 /* Leading and trailing Space, Tab, etc. are converted to markers
2913 Newline Space, Newline Tab, etc.
2914 Newline Space makes a space in the final output
2915 but is discarded if stringified. (Newline Tab is similar but
2916 makes a Tab instead.)
2918 If there is no trailing whitespace, a Newline Space is added at the end
2919 to prevent concatenation that would be contrary to the standard. */
2921 static DEFINITION *
2922 collect_expansion (buf, end, nargs, arglist)
2923 U_CHAR *buf, *end;
2924 int nargs;
2925 struct arglist *arglist;
2927 DEFINITION *defn;
2928 U_CHAR *p, *limit, *lastp, *exp_p;
2929 struct reflist *endpat = NULL;
2930 /* Pointer to first nonspace after last ## seen. */
2931 U_CHAR *concat = 0;
2932 /* Pointer to first nonspace after last single-# seen. */
2933 U_CHAR *stringify = 0;
2934 int maxsize;
2935 int expected_delimiter = '\0';
2937 /* Scan thru the replacement list, ignoring comments and quoted
2938 strings, picking up on the macro calls. It does a linear search
2939 thru the arg list on every potential symbol. Profiling might say
2940 that something smarter should happen. */
2942 if (end < buf)
2943 abort ();
2945 /* Find the beginning of the trailing whitespace. */
2946 /* Find end of leading whitespace. */
2947 limit = end;
2948 p = buf;
2949 while (p < limit && is_space (limit[-1])) limit--;
2950 while (p < limit && is_space (*p)) p++;
2952 /* Allocate space for the text in the macro definition.
2953 Leading and trailing whitespace chars need 2 bytes each.
2954 Each other input char may or may not need 1 byte,
2955 so this is an upper bound.
2956 The extra 2 are for invented trailing newline-marker and final null. */
2957 maxsize = (sizeof (DEFINITION)
2958 + 2 * (end - limit) + 2 * (p - buf)
2959 + (limit - p) + 3);
2960 defn = (DEFINITION *) xcalloc (1, maxsize);
2962 defn->nargs = nargs;
2963 exp_p = defn->expansion = (U_CHAR *) defn + sizeof (DEFINITION);
2964 lastp = exp_p;
2966 p = buf;
2968 /* Convert leading whitespace to Newline-markers. */
2969 while (p < limit && is_space (*p)) {
2970 *exp_p++ = '\n';
2971 *exp_p++ = *p++;
2974 /* Process the main body of the definition. */
2975 while (p < limit) {
2976 int skipped_arg = 0;
2977 U_CHAR c = *p++;
2979 *exp_p++ = c;
2981 /* In -traditional mode, recognize arguments inside strings and
2982 and character constants, and ignore special properties of #.
2983 Arguments inside strings are considered "stringified", but no
2984 extra quote marks are supplied. */
2985 switch (c) {
2986 case '\'':
2987 case '\"':
2988 if (expected_delimiter != '\0') {
2989 if (c == expected_delimiter)
2990 expected_delimiter = '\0';
2991 } else
2992 expected_delimiter = c;
2993 break;
2995 case '\\':
2996 /* Backslash quotes delimiters and itself, but not macro args. */
2997 if (expected_delimiter != 0 && p < limit
2998 && (*p == expected_delimiter || *p == '\\')) {
2999 *exp_p++ = *p++;
3000 continue;
3002 break;
3004 case '/':
3005 if (expected_delimiter != '\0') /* No comments inside strings. */
3006 break;
3007 if (*p == '*') {
3008 /* If we find a comment that wasn't removed by handle_directive,
3009 this must be -traditional. So replace the comment with
3010 nothing at all. */
3011 exp_p--;
3012 p += 1;
3013 while (p < limit && !(p[-2] == '*' && p[-1] == '/'))
3014 p++;
3016 break;
3019 if (is_idchar (c) && nargs > 0) {
3020 U_CHAR *id_beg = p - 1;
3021 int id_len;
3023 --exp_p;
3024 while (p != limit && is_idchar (*p)) p++;
3025 id_len = p - id_beg;
3027 if (is_idstart (c)) {
3028 struct arglist *arg;
3030 for (arg = arglist; arg != NULL; arg = arg->next) {
3031 struct reflist *tpat;
3033 if (arg->name[0] == c
3034 && arg->length == id_len
3035 && strncmp ((const char *)arg->name,
3036 (const char *)id_beg, id_len) == 0) {
3037 /* make a pat node for this arg and append it to the end of
3038 the pat list */
3039 tpat = (struct reflist *) xmalloc (sizeof (struct reflist));
3040 tpat->next = NULL;
3041 tpat->raw_before = concat == id_beg;
3042 tpat->raw_after = 0;
3043 tpat->stringify = expected_delimiter != '\0';
3045 if (endpat == NULL)
3046 defn->pattern = tpat;
3047 else
3048 endpat->next = tpat;
3049 endpat = tpat;
3051 tpat->argno = arg->argno;
3052 tpat->nchars = exp_p - lastp;
3054 U_CHAR *p1 = p;
3055 SKIP_WHITE_SPACE (p1);
3056 if (p1 + 2 <= limit && p1[0] == '#' && p1[1] == '#')
3057 tpat->raw_after = 1;
3059 lastp = exp_p; /* place to start copying from next time */
3060 skipped_arg = 1;
3061 break;
3066 /* If this was not a macro arg, copy it into the expansion. */
3067 if (! skipped_arg) {
3068 U_CHAR *lim1 = p;
3069 p = id_beg;
3070 while (p != lim1)
3071 *exp_p++ = *p++;
3072 if (stringify == id_beg)
3073 error ("# operator should be followed by a macro argument name");
3078 if (limit < end) {
3079 /* Convert trailing whitespace to Newline-markers. */
3080 while (limit < end && is_space (*limit)) {
3081 *exp_p++ = '\n';
3082 *exp_p++ = *limit++;
3085 *exp_p = '\0';
3087 defn->length = exp_p - defn->expansion;
3089 /* Crash now if we overrun the allocated size. */
3090 if (defn->length + 1 > maxsize)
3091 abort ();
3093 return defn;
3097 * interpret #line command. Remembers previously seen fnames
3098 * in its very own hash table.
3100 #define FNAME_HASHSIZE 37
3101 static void
3102 do_line (buf, limit, op)
3103 U_CHAR *buf, *limit;
3104 FILE_BUF *op;
3106 U_CHAR *bp;
3107 FILE_BUF *ip = &instack[indepth];
3108 FILE_BUF tem;
3109 int new_lineno;
3110 enum file_change_code file_change = same_file;
3112 /* Expand any macros. */
3113 tem = expand_to_temp_buffer (buf, limit, 0);
3115 /* Point to macroexpanded line, which is null-terminated now. */
3116 bp = tem.buf;
3117 SKIP_WHITE_SPACE (bp);
3119 if (!ISDIGIT (*bp)) {
3120 error ("invalid format #line command");
3121 return;
3124 /* The Newline at the end of this line remains to be processed.
3125 To put the next line at the specified line number,
3126 we must store a line number now that is one less. */
3127 new_lineno = atoi ((const char *)bp);
3129 /* skip over the line number. */
3130 while (ISDIGIT (*bp))
3131 bp++;
3133 SKIP_WHITE_SPACE (bp);
3135 if (*bp == '\"') {
3136 static HASHNODE *fname_table[FNAME_HASHSIZE];
3137 HASHNODE *hp, **hash_bucket;
3138 U_CHAR *fname;
3139 int fname_length;
3141 fname = ++bp;
3143 while (*bp && *bp != '\"')
3144 bp++;
3145 if (*bp != '\"') {
3146 error ("invalid format #line command");
3147 return;
3150 fname_length = bp - fname;
3152 bp++;
3153 SKIP_WHITE_SPACE (bp);
3154 if (*bp) {
3155 if (*bp == '1')
3156 file_change = enter_file;
3157 else if (*bp == '2')
3158 file_change = leave_file;
3159 else {
3160 error ("invalid format #line command");
3161 return;
3164 bp++;
3165 SKIP_WHITE_SPACE (bp);
3166 if (*bp) {
3167 error ("invalid format #line command");
3168 return;
3172 hash_bucket =
3173 &fname_table[hashf (fname, fname_length, FNAME_HASHSIZE)];
3174 for (hp = *hash_bucket; hp != NULL; hp = hp->next)
3175 if (hp->length == fname_length &&
3176 strncmp (hp->value.cpval, (const char *)fname, fname_length) == 0) {
3177 ip->fname = hp->value.cpval;
3178 break;
3180 if (hp == 0) {
3181 char *q;
3182 /* Didn't find it; cons up a new one. */
3183 hp = (HASHNODE *) xcalloc (1, sizeof (HASHNODE) + fname_length + 1);
3184 hp->next = *hash_bucket;
3185 *hash_bucket = hp;
3187 hp->length = fname_length;
3188 ip->fname = hp->value.cpval = q = ((char *) hp) + sizeof (HASHNODE);
3189 memcpy (q, fname, fname_length);
3191 } else if (*bp) {
3192 error ("invalid format #line command");
3193 return;
3196 ip->lineno = new_lineno;
3197 output_line_command (ip, op, 0, file_change);
3198 ip->bufp++; /* Skip the new line. */
3199 check_expand (op, ip->length - (ip->bufp - ip->buf));
3203 * remove all definitions of symbol from symbol table.
3204 * according to un*x /lib/cpp, it is not an error to undef
3205 * something that has no definitions, so it isn't one here either.
3207 static void
3208 do_undef (buf, limit, op)
3209 U_CHAR *buf;
3210 U_CHAR *limit ATTRIBUTE_UNUSED;
3211 FILE_BUF *op ATTRIBUTE_UNUSED;
3213 HASHNODE *hp;
3215 SKIP_WHITE_SPACE (buf);
3217 if (! strncmp ((const char *)buf, "defined", 7) && ! is_idchar (buf[7]))
3218 warning ("undefining `defined'");
3220 while ((hp = lookup (buf, -1, -1)) != NULL) {
3221 if (hp->type != T_MACRO)
3222 warning ("undefining `%s'", hp->name);
3223 delete_macro (hp);
3227 /* Read the tokens of the answer into the macro pool. Only commit the
3228 memory if we intend it as permanent storage, i.e. the #assert case.
3229 Returns 0 on success. */
3231 static int
3232 parse_answer (buf, limit, answerp, type)
3233 const unsigned char *buf, *limit;
3234 struct answer **answerp;
3235 int type;
3237 const unsigned char *start;
3239 /* Skip leading whitespace. */
3240 if (buf < limit && *buf == ' ')
3241 buf++;
3243 /* Parentheses are optional here. */
3244 if (buf == limit && type == T_UNASSERT)
3245 return 0;
3247 if (buf == limit || *buf++ != '(')
3249 if (type == T_IF)
3250 return 0;
3252 error ("missing '(' after predicate");
3253 return 1;
3256 /* Drop whitespace at start. */
3257 while (buf < limit && *buf == ' ')
3258 buf++;
3260 start = buf;
3261 while (buf < limit && *buf != ')')
3262 buf++;
3264 if (buf == limit)
3266 error ("missing ')' to complete answer");
3267 return 1;
3270 if (buf == start)
3272 error ("predicate's answer is empty");
3273 return 1;
3276 if ((type == T_ASSERT || type == T_UNASSERT) && buf + 1 != limit)
3278 error ("extra text at end of directive");
3279 return 1;
3282 /* Lose trailing whitespace. */
3283 if (buf[-1] == ' ')
3284 buf--;
3286 *answerp = (struct answer *) xmalloc (sizeof (struct answer));
3287 (*answerp)->answer = start;
3288 (*answerp)->len = buf - start;
3290 return 0;
3293 /* Parses an assertion, returning a pointer to the hash node of the
3294 predicate, or 0 on error. If an answer was supplied, it is placed
3295 in ANSWERP, otherwise it is set to 0. */
3296 static HASHNODE *
3297 parse_assertion (buf, limit, answerp, type)
3298 const unsigned char *buf, *limit;
3299 struct answer **answerp;
3300 int type;
3302 HASHNODE *result = 0;
3303 const unsigned char *climit;
3304 unsigned char *bp, *symname = canonicalize_text (buf, limit, &climit);
3305 unsigned int len;
3307 bp = symname;
3308 if (bp < climit && is_idstart (*bp))
3311 bp++;
3312 while (bp < climit && is_idchar (*bp));
3314 len = bp - symname;
3316 *answerp = 0;
3317 if (len == 0)
3319 if (symname == climit)
3320 error ("assertion without predicate");
3321 else
3322 error ("predicate must be an identifier");
3324 /* Unfortunately, because of the way we handle #if, we don't avoid
3325 macro expansion in answers. This is not easy to fix. */
3326 else if (parse_answer (bp, climit, answerp, type) == 0)
3328 unsigned char *sym = alloca (len + 1);
3329 int hashcode;
3331 /* Prefix '#' to get it out of macro namespace. */
3332 sym[0] = '#';
3333 memcpy (sym + 1, symname, len);
3335 hashcode = hashf (sym, len + 1, HASHSIZE);
3336 result = lookup (sym, len + 1, hashcode);
3337 if (result == 0)
3338 result = install (sym, len + 1, T_UNUSED, hashcode);
3341 return result;
3344 /* Test an assertion within a preprocessor conditional. Returns zero
3345 on error or failure, one on success. */
3347 test_assertion (pbuf)
3348 unsigned char **pbuf; /* NUL-terminated. */
3350 unsigned char *buf = *pbuf;
3351 unsigned char *limit = buf + strlen ((char *) buf);
3352 struct answer *answer;
3353 HASHNODE *node;
3354 int result = 0;
3356 node = parse_assertion (buf, limit, &answer, T_IF);
3357 if (node)
3359 result = (node->type == T_ASSERT &&
3360 (answer == 0 || *find_answer (node, answer) != 0));
3362 /* Yuk. We update pbuf to point after the assertion test.
3363 First, move past the identifier. */
3364 if (is_space (*buf))
3365 buf++;
3366 while (is_idchar (*buf))
3367 buf++;
3368 /* If we have an answer, we need to move past the parentheses. */
3369 if (answer)
3370 while (*buf++ != ')')
3372 *pbuf = buf;
3375 return result;
3378 /* Handle a #error directive. */
3379 static void
3380 do_error (buf, limit, op)
3381 U_CHAR *buf;
3382 U_CHAR *limit;
3383 FILE_BUF *op ATTRIBUTE_UNUSED;
3385 error ("#error%.*s", (int) (limit - buf), buf);
3388 /* Handle a #warning directive. */
3389 static void
3390 do_warning (buf, limit, op)
3391 U_CHAR *buf;
3392 U_CHAR *limit;
3393 FILE_BUF *op ATTRIBUTE_UNUSED;
3395 warning ("#warning%.*s", (int) (limit - buf), buf);
3398 /* Handle a #assert directive. */
3399 static void
3400 do_assert (buf, limit, op)
3401 U_CHAR *buf;
3402 U_CHAR *limit;
3403 FILE_BUF *op ATTRIBUTE_UNUSED;
3405 struct answer *new_answer;
3406 HASHNODE *node;
3408 node = parse_assertion (buf, limit, &new_answer, T_ASSERT);
3409 if (node)
3411 /* Place the new answer in the answer list. First check there
3412 is not a duplicate. */
3413 new_answer->next = 0;
3414 if (node->type == T_ASSERT)
3416 if (*find_answer (node, new_answer))
3418 free (new_answer);
3419 warning ("\"%s\" re-asserted", node->name + 1);
3420 return;
3422 new_answer->next = node->value.answers;
3424 node->type = T_ASSERT;
3425 node->value.answers = new_answer;
3429 /* Function body to be provided later. */
3430 static void
3431 do_unassert (buf, limit, op)
3432 U_CHAR *buf;
3433 U_CHAR *limit;
3434 FILE_BUF *op ATTRIBUTE_UNUSED;
3436 HASHNODE *node;
3437 struct answer *answer;
3439 node = parse_assertion (buf, limit, &answer, T_UNASSERT);
3440 /* It isn't an error to #unassert something that isn't asserted. */
3441 if (node)
3443 if (node->type == T_ASSERT)
3445 if (answer)
3447 struct answer **p = find_answer (node, answer), *temp;
3449 /* Remove the answer from the list. */
3450 temp = *p;
3451 if (temp)
3452 *p = temp->next;
3454 /* Did we free the last answer? */
3455 if (node->value.answers == 0)
3456 delete_macro (node);
3458 else
3459 delete_macro (node);
3462 free (answer);
3466 /* Returns a pointer to the pointer to the answer in the answer chain,
3467 or a pointer to NULL if the answer is not in the chain. */
3468 static struct answer **
3469 find_answer (node, candidate)
3470 HASHNODE *node;
3471 const struct answer *candidate;
3473 struct answer **result;
3475 for (result = &node->value.answers; *result; result = &(*result)->next)
3477 struct answer *answer = *result;
3479 if (answer->len == candidate->len
3480 && !memcmp (answer->answer, candidate->answer, answer->len))
3481 break;
3484 return result;
3487 /* Return a malloced buffer with leading and trailing whitespace
3488 removed, and all instances of internal whitespace reduced to a
3489 single space. */
3490 static unsigned char *
3491 canonicalize_text (buf, limit, climit)
3492 const unsigned char *buf, *limit, **climit;
3494 unsigned int len = limit - buf;
3495 unsigned char *result = (unsigned char *) xmalloc (len), *dest;
3497 for (dest = result; buf < limit;)
3499 if (! is_space (*buf))
3500 *dest++ = *buf++;
3501 else
3503 while (++buf < limit && is_space (*buf))
3505 if (dest != result && buf != limit)
3506 *dest++ = ' ';
3510 *climit = dest;
3511 return result;
3515 * handle #if command by
3516 * 1) inserting special `defined' keyword into the hash table
3517 * that gets turned into 0 or 1 by special_symbol (thus,
3518 * if the luser has a symbol called `defined' already, it won't
3519 * work inside the #if command)
3520 * 2) rescan the input into a temporary output buffer
3521 * 3) pass the output buffer to the yacc parser and collect a value
3522 * 4) clean up the mess left from steps 1 and 2.
3523 * 5) call conditional_skip to skip til the next #endif (etc.),
3524 * or not, depending on the value from step 3.
3526 static void
3527 do_if (buf, limit, op)
3528 U_CHAR *buf, *limit;
3529 FILE_BUF *op ATTRIBUTE_UNUSED;
3531 int value;
3532 FILE_BUF *ip = &instack[indepth];
3534 value = eval_if_expression (buf, limit - buf);
3535 conditional_skip (ip, value == 0, T_IF);
3539 * handle a #elif directive by not changing if_stack either.
3540 * see the comment above do_else.
3542 static void
3543 do_elif (buf, limit, op)
3544 U_CHAR *buf, *limit;
3545 FILE_BUF *op;
3547 int value;
3548 FILE_BUF *ip = &instack[indepth];
3550 if (if_stack == instack[indepth].if_stack) {
3551 error ("#elif not within a conditional");
3552 return;
3553 } else {
3554 if (if_stack->type != T_IF && if_stack->type != T_ELIF) {
3555 error ("#elif after #else");
3556 fprintf (stderr, " (matches line %d", if_stack->lineno);
3557 if (if_stack->fname != NULL && ip->fname != NULL &&
3558 strcmp (if_stack->fname, ip->fname) != 0)
3559 fprintf (stderr, ", file %s", if_stack->fname);
3560 fprintf (stderr, ")\n");
3562 if_stack->type = T_ELIF;
3565 if (if_stack->if_succeeded)
3566 skip_if_group (ip, 0);
3567 else {
3568 value = eval_if_expression (buf, limit - buf);
3569 if (value == 0)
3570 skip_if_group (ip, 0);
3571 else {
3572 ++if_stack->if_succeeded; /* continue processing input */
3573 output_line_command (ip, op, 1, same_file);
3579 * evaluate a #if expression in BUF, of length LENGTH,
3580 * then parse the result as a C expression and return the value as an int.
3582 static int
3583 eval_if_expression (buf, length)
3584 const U_CHAR *buf;
3585 int length;
3587 FILE_BUF temp_obuf;
3588 HASHNODE *save_defined;
3589 int value;
3591 save_defined = install (U"defined", -1, T_SPEC_DEFINED, -1);
3592 temp_obuf = expand_to_temp_buffer (buf, buf + length, 0);
3593 delete_macro (save_defined); /* clean up special symbol */
3595 value = parse_c_expression ((const char *)temp_obuf.buf);
3597 free (temp_obuf.buf);
3599 return value;
3603 * routine to handle ifdef/ifndef. Try to look up the symbol,
3604 * then do or don't skip to the #endif/#else/#elif depending
3605 * on what directive is actually being processed.
3607 static void
3608 do_xifdef (buf, limit, type)
3609 U_CHAR *buf, *limit;
3610 enum node_type type;
3612 int skip;
3613 FILE_BUF *ip = &instack[indepth];
3614 U_CHAR *end;
3616 /* Discard leading and trailing whitespace. */
3617 SKIP_WHITE_SPACE (buf);
3618 while (limit != buf && is_nvspace (limit[-1])) limit--;
3620 /* Find the end of the identifier at the beginning. */
3621 for (end = buf; is_idchar (*end); end++);
3623 if (end == buf)
3624 skip = (type == T_IFDEF);
3625 else
3626 skip = (lookup (buf, end-buf, -1) == NULL) ^ (type == T_IFNDEF);
3628 conditional_skip (ip, skip, T_IF);
3631 static void
3632 do_ifdef (buf, limit, op)
3633 U_CHAR *buf, *limit;
3634 FILE_BUF *op ATTRIBUTE_UNUSED;
3636 do_xifdef (buf, limit, T_IFDEF);
3639 static void
3640 do_ifndef (buf, limit, op)
3641 U_CHAR *buf, *limit;
3642 FILE_BUF *op ATTRIBUTE_UNUSED;
3644 do_xifdef (buf, limit, T_IFNDEF);
3648 * push TYPE on stack; then, if SKIP is nonzero, skip ahead.
3650 static void
3651 conditional_skip (ip, skip, type)
3652 FILE_BUF *ip;
3653 int skip;
3654 enum node_type type;
3656 IF_STACK_FRAME *temp;
3658 temp = (IF_STACK_FRAME *) xcalloc (1, sizeof (IF_STACK_FRAME));
3659 temp->fname = ip->fname;
3660 temp->lineno = ip->lineno;
3661 temp->next = if_stack;
3662 if_stack = temp;
3664 if_stack->type = type;
3666 if (skip != 0) {
3667 skip_if_group (ip, 0);
3668 return;
3669 } else {
3670 ++if_stack->if_succeeded;
3671 output_line_command (ip, &outbuf, 1, same_file);
3676 * skip to #endif, #else, or #elif. adjust line numbers, etc.
3677 * leaves input ptr at the sharp sign found.
3678 * If ANY is nonzero, return at next directive of any sort.
3680 static void
3681 skip_if_group (ip, any)
3682 FILE_BUF *ip;
3683 int any;
3685 U_CHAR *bp = ip->bufp, *cp;
3686 U_CHAR *endb = ip->buf + ip->length;
3687 const struct directive *kt;
3688 IF_STACK_FRAME *save_if_stack = if_stack; /* don't pop past here */
3689 U_CHAR *beg_of_line = bp;
3691 while (bp < endb) {
3692 switch (*bp++) {
3693 case '/': /* possible comment */
3694 if (*bp == '\\' && bp[1] == '\n')
3695 newline_fix (bp);
3696 if (*bp == '*') {
3697 ip->bufp = ++bp;
3698 bp = skip_to_end_of_comment (ip, &ip->lineno);
3700 break;
3701 case '\"':
3702 case '\'':
3703 bp = skip_quoted_string (bp - 1, endb, ip->lineno, &ip->lineno, 0, 0);
3704 break;
3705 case '\\':
3706 /* Char after backslash loses its special meaning. */
3707 if (bp < endb) {
3708 if (*bp == '\n')
3709 ++ip->lineno; /* But do update the line-count. */
3710 bp++;
3712 break;
3713 case '\n':
3714 ++ip->lineno;
3715 beg_of_line = bp;
3716 break;
3717 case '#':
3718 ip->bufp = bp - 1;
3720 /* # keyword: a # must be first nonblank char on the line */
3721 if (beg_of_line == 0)
3722 break;
3723 /* Scan from start of line, skipping whitespace, comments
3724 and backslash-newlines, and see if we reach this #.
3725 If not, this # is not special. */
3726 bp = beg_of_line;
3727 while (1) {
3728 if (is_nvspace (*bp))
3729 bp++;
3730 else if (*bp == '\\' && bp[1] == '\n')
3731 bp += 2;
3732 else if (*bp == '/' && bp[1] == '*') {
3733 bp += 2;
3734 while (!(*bp == '*' && bp[1] == '/')) {
3735 if (*bp == '\n')
3736 ip->lineno++;
3737 bp++;
3739 bp += 2;
3741 else break;
3743 if (bp != ip->bufp) {
3744 bp = ip->bufp + 1; /* Reset bp to after the #. */
3745 break;
3748 bp = ip->bufp + 1; /* Point after '#'. */
3750 /* Skip whitespace and \-newline. */
3751 while (1) {
3752 if (is_nvspace (*bp))
3753 bp++;
3754 else if (*bp == '\\' && bp[1] == '\n')
3755 bp += 2;
3756 else if (*bp == '/' && bp[1] == '*') {
3757 bp += 2;
3758 while (!(*bp == '*' && bp[1] == '/'))
3759 bp++;
3760 bp += 2;
3762 else break;
3765 cp = bp;
3767 /* Now find end of directive name.
3768 If we encounter a backslash-newline, exchange it with any following
3769 symbol-constituents so that we end up with a contiguous name. */
3771 while (1) {
3772 if (is_idchar (*bp))
3773 bp++;
3774 else {
3775 if (*bp == '\\' && bp[1] == '\n')
3776 name_newline_fix (bp);
3777 if (is_idchar (*bp))
3778 bp++;
3779 else break;
3783 for (kt = directive_table; kt->length >= 0; kt++) {
3784 IF_STACK_FRAME *temp;
3785 if (strncmp ((const char *)cp, kt->name, kt->length) == 0
3786 && !is_idchar (cp[kt->length])) {
3788 /* If we are asked to return on next directive,
3789 do so now. */
3790 if (any)
3791 return;
3793 switch (kt->type) {
3794 case T_IF:
3795 case T_IFDEF:
3796 case T_IFNDEF:
3797 temp = (IF_STACK_FRAME *) xcalloc (1, sizeof (IF_STACK_FRAME));
3798 temp->next = if_stack;
3799 if_stack = temp;
3800 temp->lineno = ip->lineno;
3801 temp->fname = ip->fname;
3802 temp->type = kt->type;
3803 break;
3804 case T_ELSE:
3805 case T_ENDIF:
3806 case T_ELIF:
3807 if (if_stack == instack[indepth].if_stack) {
3808 error ("#%s not within a conditional", kt->name);
3809 break;
3811 else if (if_stack == save_if_stack)
3812 return; /* found what we came for */
3814 if (kt->type != T_ENDIF) {
3815 if (if_stack->type == T_ELSE)
3816 error ("#else or #elif after #else");
3817 if_stack->type = kt->type;
3818 break;
3821 temp = if_stack;
3822 if_stack = if_stack->next;
3823 free (temp);
3824 break;
3826 default:
3827 /* Anything else is ignored. */
3828 break;
3830 break;
3835 ip->bufp = bp;
3836 /* after this returns, rescan will exit because ip->bufp
3837 now points to the end of the buffer.
3838 rescan is responsible for the error message also. */
3842 * handle a #else directive. Do this by just continuing processing
3843 * without changing if_stack ; this is so that the error message
3844 * for missing #endif's etc. will point to the original #if. It
3845 * is possible that something different would be better.
3847 static void
3848 do_else (buf, limit, op)
3849 U_CHAR *buf ATTRIBUTE_UNUSED;
3850 U_CHAR *limit ATTRIBUTE_UNUSED;
3851 FILE_BUF *op;
3853 FILE_BUF *ip = &instack[indepth];
3855 if (if_stack == instack[indepth].if_stack) {
3856 error ("#else not within a conditional");
3857 return;
3858 } else {
3859 if (if_stack->type != T_IF && if_stack->type != T_ELIF) {
3860 error ("#else after #else");
3861 fprintf (stderr, " (matches line %d", if_stack->lineno);
3862 if (strcmp (if_stack->fname, ip->fname) != 0)
3863 fprintf (stderr, ", file %s", if_stack->fname);
3864 fprintf (stderr, ")\n");
3866 if_stack->type = T_ELSE;
3869 if (if_stack->if_succeeded)
3870 skip_if_group (ip, 0);
3871 else {
3872 ++if_stack->if_succeeded; /* continue processing input */
3873 output_line_command (ip, op, 1, same_file);
3878 * unstack after #endif command
3880 static void
3881 do_endif (buf, limit, op)
3882 U_CHAR *buf ATTRIBUTE_UNUSED;
3883 U_CHAR *limit ATTRIBUTE_UNUSED;
3884 FILE_BUF *op;
3886 if (if_stack == instack[indepth].if_stack)
3887 error ("unbalanced #endif");
3888 else {
3889 IF_STACK_FRAME *temp = if_stack;
3890 if_stack = if_stack->next;
3891 free (temp);
3892 output_line_command (&instack[indepth], op, 1, same_file);
3897 * Skip a comment, assuming the input ptr immediately follows the
3898 * initial slash-star. Bump line counter as necessary.
3899 * (The canonical line counter is &ip->lineno).
3900 * Don't use this routine (or the next one) if bumping the line
3901 * counter is not sufficient to deal with newlines in the string.
3903 static U_CHAR *
3904 skip_to_end_of_comment (ip, line_counter)
3905 FILE_BUF *ip;
3906 int *line_counter; /* place to remember newlines, or NULL */
3908 U_CHAR *limit = ip->buf + ip->length;
3909 U_CHAR *bp = ip->bufp;
3910 FILE_BUF *op = &outbuf; /* JF */
3911 int output = put_out_comments && !line_counter;
3913 /* JF this line_counter stuff is a crock to make sure the
3914 comment is only put out once, no matter how many times
3915 the comment is skipped. It almost works */
3916 if (output) {
3917 *op->bufp++ = '/';
3918 *op->bufp++ = '*';
3920 while (bp < limit) {
3921 if (output)
3922 *op->bufp++ = *bp;
3923 switch (*bp++) {
3924 case '/':
3925 if (warn_comments && bp < limit && *bp == '*')
3926 warning("`/*' within comment");
3927 break;
3928 case '\n':
3929 if (line_counter != NULL)
3930 ++*line_counter;
3931 if (output)
3932 ++op->lineno;
3933 break;
3934 case '*':
3935 if (*bp == '\\' && bp[1] == '\n')
3936 newline_fix (bp);
3937 if (*bp == '/') {
3938 if (output)
3939 *op->bufp++ = '/';
3940 ip->bufp = ++bp;
3941 return bp;
3943 break;
3946 ip->bufp = bp;
3947 return bp;
3951 * Skip over a quoted string. BP points to the opening quote.
3952 * Returns a pointer after the closing quote. Don't go past LIMIT.
3953 * START_LINE is the line number of the starting point (but it need
3954 * not be valid if the starting point is inside a macro expansion).
3956 * The input stack state is not changed.
3958 * If COUNT_NEWLINES is nonzero, it points to an int to increment
3959 * for each newline passed.
3961 * If BACKSLASH_NEWLINES_P is nonzero, store 1 thru it
3962 * if we pass a backslash-newline.
3964 * If EOFP is nonzero, set *EOFP to 1 if the string is unterminated.
3966 static U_CHAR *
3967 skip_quoted_string (bp, limit, start_line, count_newlines, backslash_newlines_p, eofp)
3968 const U_CHAR *bp;
3969 const U_CHAR *limit;
3970 int start_line;
3971 int *count_newlines;
3972 int *backslash_newlines_p;
3973 int *eofp;
3975 U_CHAR c, match;
3977 match = *bp++;
3978 while (1) {
3979 if (bp >= limit) {
3980 error_with_line (line_for_error (start_line),
3981 "unterminated string or character constant");
3982 if (eofp)
3983 *eofp = 1;
3984 break;
3986 c = *bp++;
3987 if (c == '\\') {
3988 while (*bp == '\\' && bp[1] == '\n') {
3989 if (backslash_newlines_p)
3990 *backslash_newlines_p = 1;
3991 if (count_newlines)
3992 ++*count_newlines;
3993 bp += 2;
3995 if (*bp == '\n' && count_newlines) {
3996 if (backslash_newlines_p)
3997 *backslash_newlines_p = 1;
3998 ++*count_newlines;
4000 bp++;
4001 } else if (c == '\n') {
4002 /* Unterminated strings and character constants are 'legal'. */
4003 bp--; /* Don't consume the newline. */
4004 if (eofp)
4005 *eofp = 1;
4006 break;
4007 } else if (c == match)
4008 break;
4010 return (U_CHAR *) bp;
4014 * write out a #line command, for instance, after an #include file.
4015 * If CONDITIONAL is nonzero, we can omit the #line if it would
4016 * appear to be a no-op, and we can output a few newlines instead
4017 * if we want to increase the line number by a small amount.
4018 * FILE_CHANGE says whether we are entering a file, leaving, or neither.
4021 static void
4022 output_line_command (ip, op, conditional, file_change)
4023 FILE_BUF *ip, *op;
4024 int conditional;
4025 enum file_change_code file_change;
4027 int len;
4028 char line_cmd_buf[500];
4030 if (no_line_commands
4031 || ip->fname == NULL
4032 || no_output) {
4033 op->lineno = ip->lineno;
4034 return;
4037 if (conditional) {
4038 if (ip->lineno == op->lineno)
4039 return;
4041 /* If the inherited line number is a little too small,
4042 output some newlines instead of a #line command. */
4043 if (ip->lineno > op->lineno && ip->lineno < op->lineno + 8) {
4044 check_expand (op, 10);
4045 while (ip->lineno > op->lineno) {
4046 *op->bufp++ = '\n';
4047 op->lineno++;
4049 return;
4053 sprintf (line_cmd_buf, "# %d \"%s\"", ip->lineno, ip->fname);
4054 if (file_change != same_file)
4055 strcat (line_cmd_buf, file_change == enter_file ? " 1" : " 2");
4056 if (system_include_depth > 0)
4057 strcat (line_cmd_buf, " 3");
4058 len = strlen (line_cmd_buf);
4059 line_cmd_buf[len++] = '\n';
4060 check_expand (op, len + 1);
4061 if (op->bufp > op->buf && op->bufp[-1] != '\n')
4062 *op->bufp++ = '\n';
4063 memcpy (op->bufp, line_cmd_buf, len);
4064 op->bufp += len;
4065 op->lineno = ip->lineno;
4069 /* Expand a macro call.
4070 HP points to the symbol that is the macro being called.
4071 Put the result of expansion onto the input stack
4072 so that subsequent input by our caller will use it.
4074 If macro wants arguments, caller has already verified that
4075 an argument list follows; arguments come from the input stack. */
4077 static void
4078 macroexpand (hp, op)
4079 HASHNODE *hp;
4080 FILE_BUF *op;
4082 int nargs;
4083 DEFINITION *defn = hp->value.defn;
4084 U_CHAR *xbuf;
4085 int xbuf_len;
4086 int start_line = instack[indepth].lineno;
4088 CHECK_DEPTH (return;);
4090 /* it might not actually be a macro. */
4091 if (hp->type != T_MACRO) {
4092 special_symbol (hp, op);
4093 return;
4096 nargs = defn->nargs;
4098 if (nargs >= 0) {
4099 int i;
4100 struct argdata *args;
4101 const char *parse_error = 0;
4103 args = (struct argdata *) alloca ((nargs + 1) * sizeof (struct argdata));
4105 for (i = 0; i < nargs; i++) {
4106 args[i].raw = args[i].expanded = (U_CHAR *) "";
4107 args[i].raw_length = args[i].expand_length
4108 = args[i].stringified_length = 0;
4109 args[i].free1 = args[i].free2 = 0;
4112 /* Parse all the macro args that are supplied. I counts them.
4113 The first NARGS args are stored in ARGS.
4114 The rest are discarded. */
4115 i = 0;
4116 do {
4117 /* Discard the open-parenthesis or comma before the next arg. */
4118 ++instack[indepth].bufp;
4119 parse_error
4120 = macarg ((i < nargs || (nargs == 0 && i == 0)) ? &args[i] : 0);
4121 if (parse_error)
4123 error_with_line (line_for_error (start_line), "%s", parse_error);
4124 break;
4126 i++;
4127 } while (*instack[indepth].bufp != ')');
4129 /* If we got one arg but it was just whitespace, call that 0 args. */
4130 if (i == 1) {
4131 const U_CHAR *bp = args[0].raw;
4132 const U_CHAR *lim = bp + args[0].raw_length;
4133 while (bp != lim && is_space (*bp)) bp++;
4134 if (bp == lim)
4135 i = 0;
4138 if (nargs == 0 && i > 0)
4139 error ("arguments given to macro `%s'", hp->name);
4140 else if (i < nargs) {
4141 /* traditional C allows foo() if foo wants one argument. */
4142 if (nargs == 1 && i == 0)
4144 else if (i == 0)
4145 error ("no args to macro `%s'", hp->name);
4146 else if (i == 1)
4147 error ("only 1 arg to macro `%s'", hp->name);
4148 else
4149 error ("only %d args to macro `%s'", i, hp->name);
4150 } else if (i > nargs)
4151 error ("too many (%d) args to macro `%s'", i, hp->name);
4153 /* Swallow the closeparen. */
4154 ++instack[indepth].bufp;
4156 /* If macro wants zero args, we parsed the arglist for checking only.
4157 Read directly from the macro definition. */
4158 if (nargs == 0) {
4159 xbuf = defn->expansion;
4160 xbuf_len = defn->length;
4161 } else {
4162 U_CHAR *exp = defn->expansion;
4163 int offset; /* offset in expansion,
4164 copied a piece at a time */
4165 int totlen; /* total amount of exp buffer filled so far */
4167 struct reflist *ap;
4169 /* Macro really takes args. Compute the expansion of this call. */
4171 /* Compute length in characters of the macro's expansion. */
4172 xbuf_len = defn->length;
4173 for (ap = defn->pattern; ap != NULL; ap = ap->next) {
4174 if (ap->stringify)
4175 xbuf_len += args[ap->argno].stringified_length;
4176 else
4177 xbuf_len += args[ap->argno].raw_length;
4180 xbuf = (U_CHAR *) xmalloc (xbuf_len + 1);
4182 /* Generate in XBUF the complete expansion
4183 with arguments substituted in.
4184 TOTLEN is the total size generated so far.
4185 OFFSET is the index in the definition
4186 of where we are copying from. */
4187 offset = totlen = 0;
4188 for (ap = defn->pattern; ap != NULL; ap = ap->next) {
4189 struct argdata *arg = &args[ap->argno];
4191 for (i = 0; i < ap->nchars; i++)
4192 xbuf[totlen++] = exp[offset++];
4194 if (ap->stringify != 0) {
4195 int arglen = arg->raw_length;
4196 int escaped = 0;
4197 int in_string = 0;
4198 int c;
4199 i = 0;
4200 while (i < arglen
4201 && (c = arg->raw[i], is_space (c)))
4202 i++;
4203 while (i < arglen
4204 && (c = arg->raw[arglen - 1], is_space (c)))
4205 arglen--;
4206 for (; i < arglen; i++) {
4207 c = arg->raw[i];
4209 /* Special markers Newline Space
4210 generate nothing for a stringified argument. */
4211 if (c == '\n' && arg->raw[i+1] != '\n') {
4212 i++;
4213 continue;
4216 /* Internal sequences of whitespace are replaced by one space
4217 except within an string or char token. */
4218 if (! in_string
4219 && (c == '\n' ? arg->raw[i+1] == '\n' : is_space (c))) {
4220 while (1) {
4221 /* Note that Newline Space does occur within whitespace
4222 sequences; consider it part of the sequence. */
4223 if (c == '\n' && is_space (arg->raw[i+1]))
4224 i += 2;
4225 else if (c != '\n' && is_space (c))
4226 i++;
4227 else break;
4228 c = arg->raw[i];
4230 i--;
4231 c = ' ';
4234 if (escaped)
4235 escaped = 0;
4236 else {
4237 if (c == '\\')
4238 escaped = 1;
4239 if (in_string) {
4240 if (c == in_string)
4241 in_string = 0;
4242 } else if (c == '\"' || c == '\'')
4243 in_string = c;
4246 /* Escape these chars */
4247 if (c == '\"' || (in_string && c == '\\'))
4248 xbuf[totlen++] = '\\';
4249 if (ISPRINT (c))
4250 xbuf[totlen++] = c;
4251 else {
4252 sprintf ((char *) &xbuf[totlen], "\\%03o", (unsigned int) c);
4253 totlen += 4;
4256 } else {
4257 const U_CHAR *p1 = arg->raw;
4258 const U_CHAR *l1 = p1 + arg->raw_length;
4260 if (ap->raw_before) {
4261 while (p1 != l1 && is_space (*p1)) p1++;
4262 while (p1 != l1 && is_idchar (*p1))
4263 xbuf[totlen++] = *p1++;
4264 /* Delete any no-reexpansion marker that follows
4265 an identifier at the beginning of the argument
4266 if the argument is concatenated with what precedes it. */
4267 if (p1[0] == '\n' && p1[1] == '-')
4268 p1 += 2;
4270 if (ap->raw_after) {
4271 /* Arg is concatenated after: delete trailing whitespace,
4272 whitespace markers, and no-reexpansion markers. */
4273 while (p1 != l1) {
4274 if (is_space (l1[-1])) l1--;
4275 else if (l1[-1] == '-') {
4276 const U_CHAR *p2 = l1 - 1;
4277 /* If a `-' is preceded by an odd number of newlines then it
4278 and the last newline are a no-reexpansion marker. */
4279 while (p2 != p1 && p2[-1] == '\n') p2--;
4280 if ((l1 - 1 - p2) & 1) {
4281 l1 -= 2;
4283 else break;
4285 else break;
4288 memmove (xbuf + totlen, p1, l1 - p1);
4289 totlen += l1 - p1;
4292 if (totlen > xbuf_len)
4293 abort ();
4296 /* if there is anything left of the definition
4297 after handling the arg list, copy that in too. */
4299 for (i = offset; i < defn->length; i++)
4300 xbuf[totlen++] = exp[i];
4302 xbuf[totlen] = 0;
4303 xbuf_len = totlen;
4305 for (i = 0; i < nargs; i++) {
4306 if (args[i].free1 != 0)
4307 free (args[i].free1);
4308 if (args[i].free2 != 0)
4309 free (args[i].free2);
4312 } else {
4313 xbuf = defn->expansion;
4314 xbuf_len = defn->length;
4317 /* Now put the expansion on the input stack
4318 so our caller will commence reading from it. */
4320 FILE_BUF *ip2;
4322 ip2 = &instack[++indepth];
4324 ip2->fname = 0;
4325 ip2->lineno = 0;
4326 ip2->buf = xbuf;
4327 ip2->length = xbuf_len;
4328 ip2->bufp = xbuf;
4329 ip2->free_ptr = (nargs > 0) ? xbuf : 0;
4330 ip2->macro = hp;
4331 ip2->if_stack = if_stack;
4336 * Parse a macro argument and store the info on it into *ARGPTR.
4337 * Return nonzero to indicate a syntax error.
4340 static const char *
4341 macarg (argptr)
4342 struct argdata *argptr;
4344 FILE_BUF *ip = &instack[indepth];
4345 int paren = 0;
4346 int newlines = 0;
4347 int comments = 0;
4349 /* Try to parse as much of the argument as exists at this
4350 input stack level. */
4351 U_CHAR *bp = macarg1 (ip->bufp, ip->buf + ip->length,
4352 &paren, &newlines, &comments);
4354 /* If we find the end of the argument at this level,
4355 set up *ARGPTR to point at it in the input stack. */
4356 if (!(ip->fname != 0 && (newlines != 0 || comments != 0))
4357 && bp != ip->buf + ip->length) {
4358 if (argptr != 0) {
4359 argptr->raw = ip->bufp;
4360 argptr->raw_length = bp - ip->bufp;
4362 ip->bufp = bp;
4363 } else {
4364 /* This input stack level ends before the macro argument does.
4365 We must pop levels and keep parsing.
4366 Therefore, we must allocate a temporary buffer and copy
4367 the macro argument into it. */
4368 int bufsize = bp - ip->bufp;
4369 int extra = newlines;
4370 U_CHAR *buffer = (U_CHAR *) xmalloc (bufsize + extra + 1);
4371 int final_start = 0;
4373 memcpy (buffer, ip->bufp, bufsize);
4374 ip->bufp = bp;
4375 ip->lineno += newlines;
4377 while (bp == ip->buf + ip->length) {
4378 if (instack[indepth].macro == 0) {
4379 free (buffer);
4380 return "unterminated macro call";
4382 ip->macro->type = T_MACRO;
4383 if (ip->free_ptr)
4384 free (ip->free_ptr);
4385 ip = &instack[--indepth];
4386 newlines = 0;
4387 comments = 0;
4388 bp = macarg1 (ip->bufp, ip->buf + ip->length, &paren,
4389 &newlines, &comments);
4390 final_start = bufsize;
4391 bufsize += bp - ip->bufp;
4392 extra += newlines;
4393 buffer = (U_CHAR *) xrealloc (buffer, bufsize + extra + 1);
4394 memcpy (buffer + bufsize - (bp - ip->bufp), ip->bufp, bp - ip->bufp);
4395 ip->bufp = bp;
4396 ip->lineno += newlines;
4399 /* Now, if arg is actually wanted, record its raw form,
4400 discarding comments and duplicating newlines in whatever
4401 part of it did not come from a macro expansion.
4402 EXTRA space has been preallocated for duplicating the newlines.
4403 FINAL_START is the index of the start of that part. */
4404 if (argptr != 0) {
4405 argptr->raw = buffer;
4406 argptr->raw_length = bufsize;
4407 argptr->free1 = buffer;
4408 argptr->newlines = newlines;
4409 argptr->comments = comments;
4410 if ((newlines || comments) && ip->fname != 0)
4411 argptr->raw_length
4412 = final_start +
4413 discard_comments (argptr->raw + final_start,
4414 argptr->raw_length - final_start,
4415 newlines);
4416 argptr->raw[argptr->raw_length] = 0;
4417 if (argptr->raw_length > bufsize + extra)
4418 abort ();
4422 /* If we are not discarding this argument,
4423 macroexpand it and compute its length as stringified.
4424 All this info goes into *ARGPTR. */
4426 if (argptr != 0) {
4427 FILE_BUF obuf;
4428 const U_CHAR *buf, *lim;
4429 int totlen;
4431 obuf = expand_to_temp_buffer (argptr->raw,
4432 argptr->raw + argptr->raw_length,
4435 argptr->expanded = obuf.buf;
4436 argptr->expand_length = obuf.length;
4437 argptr->free2 = obuf.buf;
4439 buf = argptr->raw;
4440 lim = buf + argptr->raw_length;
4442 totlen = 0;
4443 while (buf != lim) {
4444 U_CHAR c = *buf++;
4445 totlen++;
4446 /* Internal sequences of whitespace are replaced by one space
4447 in most cases, but not always. So count all the whitespace
4448 in case we need to keep it all. */
4449 if (c == '\"' || c == '\\') /* escape these chars */
4450 totlen++;
4451 else if (!ISPRINT (c))
4452 totlen += 3;
4454 argptr->stringified_length = totlen;
4456 return 0;
4459 /* Scan text from START (inclusive) up to LIMIT (exclusive),
4460 counting parens in *DEPTHPTR,
4461 and return if reach LIMIT
4462 or before a `)' that would make *DEPTHPTR negative
4463 or before a comma when *DEPTHPTR is zero.
4464 Single and double quotes are matched and termination
4465 is inhibited within them. Comments also inhibit it.
4466 Value returned is pointer to stopping place.
4468 Increment *NEWLINES each time a newline is passed.
4469 Set *COMMENTS to 1 if a comment is seen. */
4471 static U_CHAR *
4472 macarg1 (start, limit, depthptr, newlines, comments)
4473 U_CHAR *start;
4474 const U_CHAR *limit;
4475 int *depthptr, *newlines, *comments;
4477 U_CHAR *bp = start;
4479 while (bp < limit) {
4480 switch (*bp) {
4481 case '(':
4482 (*depthptr)++;
4483 break;
4484 case ')':
4485 if (--(*depthptr) < 0)
4486 return bp;
4487 break;
4488 case '\\':
4489 /* Traditionally, backslash makes following char not special. */
4490 if (bp + 1 < limit)
4492 bp++;
4493 /* But count source lines anyway. */
4494 if (*bp == '\n')
4495 ++*newlines;
4497 break;
4498 case '\n':
4499 ++*newlines;
4500 break;
4501 case '/':
4502 if (bp[1] == '\\' && bp[2] == '\n')
4503 newline_fix (bp + 1);
4504 if (bp[1] != '*' || bp + 1 >= limit)
4505 break;
4506 *comments = 1;
4507 bp += 2;
4508 while (bp + 1 < limit) {
4509 if (bp[0] == '*'
4510 && bp[1] == '\\' && bp[2] == '\n')
4511 newline_fix (bp + 1);
4512 if (bp[0] == '*' && bp[1] == '/')
4513 break;
4514 if (*bp == '\n') ++*newlines;
4515 bp++;
4517 bp += 1;
4518 break;
4519 case '\'':
4520 case '\"':
4522 int quotec;
4523 for (quotec = *bp++; bp + 1 < limit && *bp != quotec; bp++) {
4524 if (*bp == '\\') {
4525 bp++;
4526 if (*bp == '\n')
4527 ++*newlines;
4528 while (*bp == '\\' && bp[1] == '\n') {
4529 bp += 2;
4531 } else if (*bp == '\n') {
4532 ++*newlines;
4533 if (quotec == '\'')
4534 break;
4538 break;
4539 case ',':
4540 if ((*depthptr) == 0)
4541 return bp;
4542 break;
4544 bp++;
4547 return bp;
4550 /* Discard comments and duplicate newlines
4551 in the string of length LENGTH at START,
4552 except inside of string constants.
4553 The string is copied into itself with its beginning staying fixed.
4555 NEWLINES is the number of newlines that must be duplicated.
4556 We assume that that much extra space is available past the end
4557 of the string. */
4559 static int
4560 discard_comments (start, length, newlines)
4561 U_CHAR *start;
4562 int length;
4563 int newlines;
4565 U_CHAR *ibp;
4566 U_CHAR *obp;
4567 const U_CHAR *limit;
4568 int c;
4570 /* If we have newlines to duplicate, copy everything
4571 that many characters up. Then, in the second part,
4572 we will have room to insert the newlines
4573 while copying down.
4574 NEWLINES may actually be too large, because it counts
4575 newlines in string constants, and we don't duplicate those.
4576 But that does no harm. */
4577 if (newlines > 0) {
4578 ibp = start + length;
4579 obp = ibp + newlines;
4580 limit = start;
4581 while (limit != ibp)
4582 *--obp = *--ibp;
4585 ibp = start + newlines;
4586 limit = start + length + newlines;
4587 obp = start;
4589 while (ibp < limit) {
4590 *obp++ = c = *ibp++;
4591 switch (c) {
4592 case '\n':
4593 /* Duplicate the newline. */
4594 *obp++ = '\n';
4595 break;
4597 case '\\':
4598 if (*ibp == '\n') {
4599 obp--;
4600 ibp++;
4602 break;
4604 case '/':
4605 if (*ibp == '\\' && ibp[1] == '\n')
4606 newline_fix (ibp);
4607 /* Delete any comment. */
4608 if (ibp[0] != '*' || ibp + 1 >= limit)
4609 break;
4610 obp--;
4611 ibp++;
4612 while (ibp + 1 < limit) {
4613 if (ibp[0] == '*'
4614 && ibp[1] == '\\' && ibp[2] == '\n')
4615 newline_fix (ibp + 1);
4616 if (ibp[0] == '*' && ibp[1] == '/')
4617 break;
4618 ibp++;
4620 ibp += 2;
4621 break;
4623 case '\'':
4624 case '\"':
4625 /* Notice and skip strings, so that we don't
4626 think that comments start inside them,
4627 and so we don't duplicate newlines in them. */
4629 int quotec = c;
4630 while (ibp < limit) {
4631 *obp++ = c = *ibp++;
4632 if (c == quotec)
4633 break;
4634 if (c == '\n' && quotec == '\'')
4635 break;
4636 if (c == '\\' && ibp < limit) {
4637 while (*ibp == '\\' && ibp[1] == '\n')
4638 ibp += 2;
4639 *obp++ = *ibp++;
4643 break;
4647 return obp - start;
4651 /* Core error handling routine. */
4652 static void
4653 v_message (mtype, line, msgid, ap)
4654 enum msgtype mtype;
4655 int line;
4656 const char *msgid;
4657 va_list ap;
4659 const char *fname = 0;
4660 int i;
4662 if (mtype == MT_WARNING && inhibit_warnings)
4663 return;
4665 for (i = indepth; i >= 0; i--)
4666 if (instack[i].fname != NULL) {
4667 if (line == 0)
4668 line = instack[i].lineno;
4669 fname = instack[i].fname;
4670 break;
4673 if (fname)
4674 fprintf (stderr, "%s:%d: ", fname, line);
4675 else
4676 fprintf (stderr, "%s: ", progname);
4678 if (mtype == MT_WARNING)
4679 fputs (_("warning: "), stderr);
4681 vfprintf (stderr, _(msgid), ap);
4682 putc ('\n', stderr);
4684 if (mtype == MT_ERROR)
4685 errors++;
4689 * error - print error message and increment count of errors.
4691 void
4692 error VPARAMS ((const char *msgid, ...))
4694 VA_OPEN(ap, msgid);
4695 VA_FIXEDARG (ap, const char *, msgid);
4697 v_message (MT_ERROR, 0, msgid, ap);
4698 VA_CLOSE (ap);
4701 void
4702 error_with_line VPARAMS ((int line, const char *msgid, ...))
4704 VA_OPEN(ap, msgid);
4705 VA_FIXEDARG (ap, int, line);
4706 VA_FIXEDARG (ap, const char *, msgid);
4708 v_message (MT_ERROR, line, msgid, ap);
4709 VA_CLOSE (ap);
4712 /* Error including a message from `errno'. */
4713 void
4714 error_from_errno (name)
4715 const char *name;
4717 error ("%s: %s", name, strerror (errno));
4720 /* Print error message but don't count it. */
4721 void
4722 warning VPARAMS ((const char *msgid, ...))
4724 VA_OPEN(ap, msgid);
4725 VA_FIXEDARG (ap, const char *, msgid);
4727 v_message (MT_WARNING, 0, msgid, ap);
4728 VA_CLOSE (ap);
4731 void
4732 fatal VPARAMS ((const char *msgid, ...))
4734 VA_OPEN(ap, msgid);
4735 VA_FIXEDARG (ap, const char *, msgid);
4737 v_message (MT_FATAL, 0, msgid, ap);
4738 VA_CLOSE (ap);
4739 exit (FATAL_EXIT_CODE);
4742 /* More 'friendly' abort that prints the location at which we died. */
4743 void
4744 fancy_abort (line, func)
4745 int line;
4746 const char *func;
4748 fatal ("internal error in %s, at tradcpp.c:%d\n\
4749 Please submit a full bug report.\n\
4750 See %s for instructions.", func, line, GCCBUGURL);
4753 void
4754 perror_with_name (name)
4755 const char *name;
4757 fprintf (stderr, "%s: %s: %s\n", progname, name, strerror (errno));
4758 errors++;
4761 void
4762 pfatal_with_name (name)
4763 const char *name;
4765 perror_with_name (name);
4766 exit (FATAL_EXIT_CODE);
4769 /* Return the line at which an error occurred.
4770 The error is not necessarily associated with the current spot
4771 in the input stack, so LINE says where. LINE will have been
4772 copied from ip->lineno for the current input level.
4773 If the current level is for a file, we return LINE.
4774 But if the current level is not for a file, LINE is meaningless.
4775 In that case, we return the lineno of the innermost file. */
4776 static int
4777 line_for_error (line)
4778 int line;
4780 int i;
4781 int line1 = line;
4783 for (i = indepth; i >= 0; ) {
4784 if (instack[i].fname != 0)
4785 return line1;
4786 i--;
4787 if (i < 0)
4788 return 0;
4789 line1 = instack[i].lineno;
4791 return 0;
4795 * If OBUF doesn't have NEEDED bytes after OPTR, make it bigger.
4797 * As things stand, nothing is ever placed in the output buffer to be
4798 * removed again except when it's KNOWN to be part of an identifier,
4799 * so flushing and moving down everything left, instead of expanding,
4800 * should work ok.
4803 static void
4804 grow_outbuf (obuf, needed)
4805 FILE_BUF *obuf;
4806 int needed;
4808 U_CHAR *p;
4809 int minsize;
4811 if (obuf->length - (obuf->bufp - obuf->buf) > needed)
4812 return;
4814 /* Make it at least twice as big as it is now. */
4815 obuf->length *= 2;
4816 /* Make it have at least 150% of the free space we will need. */
4817 minsize = (3 * needed) / 2 + (obuf->bufp - obuf->buf);
4818 if (minsize > obuf->length)
4819 obuf->length = minsize;
4821 p = (U_CHAR *) xrealloc (obuf->buf, obuf->length);
4822 obuf->bufp = p + (obuf->bufp - obuf->buf);
4823 obuf->buf = p;
4826 /* Symbol table for macro names and special symbols */
4829 * install a name in the main hash table, even if it is already there.
4830 * name stops with first non alphanumeric, except leading '#'.
4831 * caller must check against redefinition if that is desired.
4832 * delete_macro () removes things installed by install () in fifo order.
4833 * this is important because of the `defined' special symbol used
4834 * in #if, and also if pushdef/popdef directives are ever implemented.
4836 * If LEN is >= 0, it is the length of the name.
4837 * Otherwise, compute the length by scanning the entire name.
4839 * If HASH is >= 0, it is the precomputed hash code.
4840 * Otherwise, compute the hash code.
4842 * caller must set the value, if any is desired.
4844 static HASHNODE *
4845 install (name, len, type, hash)
4846 const U_CHAR *name;
4847 int len;
4848 enum node_type type;
4849 int hash;
4850 /* watch out here if sizeof (U_CHAR *) != sizeof (int) */
4852 HASHNODE *hp;
4853 int bucket;
4854 const U_CHAR *p;
4855 U_CHAR *q;
4857 if (len < 0) {
4858 p = name;
4859 while (is_idchar (*p))
4860 p++;
4861 len = p - name;
4864 if (hash < 0)
4865 hash = hashf (name, len, HASHSIZE);
4867 hp = (HASHNODE *) xmalloc (sizeof (HASHNODE) + len + 1);
4868 bucket = hash;
4869 hp->bucket_hdr = &hashtab[bucket];
4870 hp->next = hashtab[bucket];
4871 hashtab[bucket] = hp;
4872 hp->prev = NULL;
4873 if (hp->next != NULL)
4874 hp->next->prev = hp;
4875 hp->type = type;
4876 hp->length = len;
4877 hp->name = q = ((U_CHAR *) hp) + sizeof (HASHNODE);
4878 memcpy (q, name, len);
4879 q[len] = 0;
4880 return hp;
4884 * find the most recent hash node for name name (ending with first
4885 * non-identifier char) installed by install
4887 * If LEN is >= 0, it is the length of the name.
4888 * Otherwise, compute the length by scanning the entire name.
4890 * If HASH is >= 0, it is the precomputed hash code.
4891 * Otherwise, compute the hash code.
4893 HASHNODE *
4894 lookup (name, len, hash)
4895 const U_CHAR *name;
4896 int len;
4897 int hash;
4899 const U_CHAR *bp;
4900 HASHNODE *bucket;
4902 if (len < 0) {
4903 for (bp = name; is_idchar (*bp); bp++) ;
4904 len = bp - name;
4907 if (hash < 0)
4908 hash = hashf (name, len, HASHSIZE);
4910 bucket = hashtab[hash];
4911 while (bucket) {
4912 if (bucket->length == len
4913 && strncmp ((const char *)bucket->name, (const char *)name, len) == 0)
4914 return bucket;
4915 bucket = bucket->next;
4917 return NULL;
4921 * Delete a hash node. Some weirdness to free junk from macros.
4922 * More such weirdness will have to be added if you define more hash
4923 * types that need it.
4926 /* Note that the DEFINITION of a macro is removed from the hash table
4927 but its storage is not freed. This would be a storage leak
4928 except that it is not reasonable to keep undefining and redefining
4929 large numbers of macros many times.
4930 In any case, this is necessary, because a macro can be #undef'd
4931 in the middle of reading the arguments to a call to it.
4932 If #undef freed the DEFINITION, that would crash. */
4933 static void
4934 delete_macro (hp)
4935 HASHNODE *hp;
4938 if (hp->prev != NULL)
4939 hp->prev->next = hp->next;
4940 if (hp->next != NULL)
4941 hp->next->prev = hp->prev;
4943 /* make sure that the bucket chain header that
4944 the deleted guy was on points to the right thing afterwards. */
4945 if (hp == *hp->bucket_hdr)
4946 *hp->bucket_hdr = hp->next;
4948 free (hp);
4952 * return hash function on name. must be compatible with the one
4953 * computed a step at a time, elsewhere
4955 static int
4956 hashf (name, len, hashsize)
4957 const U_CHAR *name;
4958 int len;
4959 int hashsize;
4961 int r = 0;
4963 while (len--)
4964 r = HASHSTEP (r, *name++);
4966 return MAKE_POS (r) % hashsize;
4969 /* Dump all macro definitions as #defines to stdout. */
4971 static void
4972 dump_all_macros ()
4974 int bucket;
4976 for (bucket = 0; bucket < HASHSIZE; bucket++) {
4977 HASHNODE *hp;
4979 for (hp = hashtab[bucket]; hp; hp= hp->next) {
4980 if (hp->type == T_MACRO) {
4981 DEFINITION *defn = hp->value.defn;
4982 struct reflist *ap;
4983 int offset;
4984 int concat;
4987 /* Print the definition of the macro HP. */
4989 printf ("#define %s", hp->name);
4990 if (defn->nargs >= 0) {
4991 int i;
4993 printf ("(");
4994 for (i = 0; i < defn->nargs; i++) {
4995 dump_arg_n (defn, i);
4996 if (i + 1 < defn->nargs)
4997 printf (", ");
4999 printf (")");
5002 printf (" ");
5004 offset = 0;
5005 concat = 0;
5006 for (ap = defn->pattern; ap != NULL; ap = ap->next) {
5007 dump_defn_1 (defn->expansion, offset, ap->nchars);
5008 if (ap->nchars != 0)
5009 concat = 0;
5010 offset += ap->nchars;
5011 if (ap->stringify)
5012 printf (" #");
5013 if (ap->raw_before && !concat)
5014 printf (" ## ");
5015 concat = 0;
5016 dump_arg_n (defn, ap->argno);
5017 if (ap->raw_after) {
5018 printf (" ## ");
5019 concat = 1;
5022 dump_defn_1 (defn->expansion, offset, defn->length - offset);
5023 printf ("\n");
5029 /* Output to stdout a substring of a macro definition.
5030 BASE is the beginning of the definition.
5031 Output characters START thru LENGTH.
5032 Discard newlines outside of strings, thus
5033 converting funny-space markers to ordinary spaces. */
5034 static void
5035 dump_defn_1 (base, start, length)
5036 const U_CHAR *base;
5037 int start;
5038 int length;
5040 const U_CHAR *p = base + start;
5041 const U_CHAR *limit = base + start + length;
5043 while (p < limit) {
5044 if (*p != '\n')
5045 putchar (*p);
5046 else if (*p == '\"' || *p =='\'') {
5047 const U_CHAR *p1 = skip_quoted_string (p, limit, 0, 0, 0, 0);
5048 fwrite (p, p1 - p, 1, stdout);
5049 p = p1 - 1;
5051 p++;
5055 /* Print the name of argument number ARGNUM of macro definition DEFN.
5056 Recall that DEFN->argnames contains all the arg names
5057 concatenated in reverse order with comma-space in between. */
5058 static void
5059 dump_arg_n (defn, argnum)
5060 DEFINITION *defn;
5061 int argnum;
5063 const U_CHAR *p = defn->argnames;
5064 while (argnum + 1 < defn->nargs) {
5065 p = (const U_CHAR *) strchr ((const char *)p, ' ') + 1;
5066 argnum++;
5069 while (*p && *p != ',') {
5070 putchar (*p);
5071 p++;
5075 /* Initialize the built-in macros. */
5076 #define DSC(x) U x, sizeof x - 1
5077 #define install_spec(name, type) \
5078 install(DSC(name), type, -1);
5079 #define install_value(name, val) \
5080 hp = install(DSC(name), T_CONST, -1); hp->value.cpval = val;
5081 static void
5082 initialize_builtins ()
5084 HASHNODE *hp;
5086 install_spec ("__BASE_FILE__", T_BASE_FILE);
5087 install_spec ("__DATE__", T_DATE);
5088 install_spec ("__FILE__", T_FILE);
5089 install_spec ("__TIME__", T_TIME);
5090 install_spec ("__VERSION__", T_VERSION);
5091 install_spec ("__INCLUDE_LEVEL__", T_INCLUDE_LEVEL);
5092 install_spec ("__LINE__", T_SPECLINE);
5094 #ifndef NO_BUILTIN_SIZE_TYPE
5095 install_value ("__SIZE_TYPE__", SIZE_TYPE);
5096 #endif
5097 #ifndef NO_BUILTIN_PTRDIFF_TYPE
5098 install_value ("__PTRDIFF_TYPE__", PTRDIFF_TYPE);
5099 #endif
5100 #ifndef NO_BUILTIN_WCHAR_TYPE
5101 install_value ("__WCHAR_TYPE__", WCHAR_TYPE);
5102 #endif
5103 #ifndef NO_BUILTIN_WINT_TYPE
5104 install_value ("__WINT_TYPE__", WINT_TYPE);
5105 #endif
5106 install_value ("__REGISTER_PREFIX__", REGISTER_PREFIX);
5107 install_value ("__USER_LABEL_PREFIX__", user_label_prefix);
5109 #undef DSC
5110 #undef install_spec
5111 #undef install_value
5113 /* Common handler of command line directives -U, -D and -A. */
5114 static void
5115 run_directive (str, len, type)
5116 const char *str;
5117 size_t len;
5118 enum node_type type;
5120 const struct directive *kt;
5121 FILE_BUF *ip = &instack[++indepth];
5122 ip->fname = "*command line*";
5124 ip->buf = ip->bufp = (U_CHAR *) str;
5125 ip->length = len;
5126 ip->lineno = 1;
5127 ip->macro = 0;
5128 ip->free_ptr = 0;
5129 ip->if_stack = if_stack;
5131 for (kt = directive_table; kt->type != type; kt++)
5134 (*kt->func) ((U_CHAR *) str, (U_CHAR *) str + len, NULL);
5135 --indepth;
5138 /* Handle the -D option. If STR is just an identifier, define it with
5139 * value 1. If STR has anything after the identifier, then it should
5140 * be identifier-space-definition. */
5141 static void
5142 make_definition (str)
5143 const char *str;
5145 char *buf, *p;
5146 size_t count;
5148 /* Copy the entire option so we can modify it.
5149 Change the first "=" in the string to a space. If there is none,
5150 tack " 1" on the end. */
5152 /* Length including the null. */
5153 count = strlen (str);
5154 buf = (char *) alloca (count + 2);
5155 memcpy (buf, str, count);
5157 p = strchr (str, '=');
5158 if (p)
5159 buf[p - str] = ' ';
5160 else
5162 buf[count++] = ' ';
5163 buf[count++] = '1';
5166 run_directive (buf, count, T_DEFINE);
5169 /* Handle the -U option. */
5170 static void
5171 make_undef (str)
5172 const char *str;
5174 run_directive (str, strlen (str), T_UNDEF);
5177 /* Handles the #assert (-A) and #unassert (-A-) command line options. */
5178 static void
5179 make_assertion (str)
5180 const char *str;
5182 enum node_type type = T_ASSERT;
5183 size_t count;
5184 const char *p;
5186 if (*str == '-')
5188 str++;
5189 type = T_UNASSERT;
5192 count = strlen (str);
5193 p = strchr (str, '=');
5194 if (p)
5196 /* Copy the entire option so we can modify it. Change the first
5197 "=" in the string to a '(', and tack a ')' on the end. */
5198 char *buf = (char *) alloca (count + 1);
5200 memcpy (buf, str, count);
5201 buf[p - str] = '(';
5202 buf[count++] = ')';
5203 str = buf;
5206 run_directive (str, count, type);
5209 /* Get the file-mode and data size of the file open on FD
5210 and store them in *MODE_POINTER and *SIZE_POINTER. */
5212 static int
5213 file_size_and_mode (fd, mode_pointer, size_pointer)
5214 int fd;
5215 int *mode_pointer;
5216 long *size_pointer;
5218 struct stat sbuf;
5220 if (fstat (fd, &sbuf) < 0) return -1;
5221 if (mode_pointer) *mode_pointer = sbuf.st_mode;
5222 if (size_pointer) *size_pointer = sbuf.st_size;
5223 return 0;