xref: /openbsd/gnu/gcc/gcc/protoize.c (revision 404b540a)
1 /* Protoize program - Original version by Ron Guilmette (rfg@segfault.us.com).
2    Copyright (C) 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3    1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
4 
5 This file is part of GCC.
6 
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 2, or (at your option) any later
10 version.
11 
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
15 for more details.
16 
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING.  If not, write to the Free
19 Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
20 02110-1301, USA.  */
21 
22 #include "config.h"
23 #include "system.h"
24 #include "coretypes.h"
25 #include "tm.h"
26 #include "intl.h"
27 #include "cppdefault.h"
28 
29 #include <setjmp.h>
30 #include <signal.h>
31 #if ! defined( SIGCHLD ) && defined( SIGCLD )
32 #  define SIGCHLD SIGCLD
33 #endif
34 #ifdef HAVE_UNISTD_H
35 #include <unistd.h>
36 #endif
37 #include "version.h"
38 
39 /* Include getopt.h for the sake of getopt_long.  */
40 #include "getopt.h"
41 
42 /* Macro to see if the path elements match.  */
43 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
44 #define IS_SAME_PATH_CHAR(a,b) (TOUPPER (a) == TOUPPER (b))
45 #else
46 #define IS_SAME_PATH_CHAR(a,b) ((a) == (b))
47 #endif
48 
49 /* Macro to see if the paths match.  */
50 #define IS_SAME_PATH(a,b) (FILENAME_CMP (a, b) == 0)
51 
52 /* Suffix for aux-info files.  */
53 #ifdef __MSDOS__
54 #define AUX_INFO_SUFFIX "X"
55 #else
56 #define AUX_INFO_SUFFIX ".X"
57 #endif
58 
59 /* Suffix for saved files.  */
60 #ifdef __MSDOS__
61 #define SAVE_SUFFIX "sav"
62 #else
63 #define SAVE_SUFFIX ".save"
64 #endif
65 
66 /* Suffix for renamed C++ files.  */
67 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
68 #define CPLUS_FILE_SUFFIX "cc"
69 #else
70 #define CPLUS_FILE_SUFFIX "C"
71 #endif
72 
73 static void usage (void) ATTRIBUTE_NORETURN;
74 static void aux_info_corrupted (void) ATTRIBUTE_NORETURN;
75 static void declare_source_confusing (const char *) ATTRIBUTE_NORETURN;
76 static const char *shortpath (const char *, const char *);
77 static void notice (const char *, ...) ATTRIBUTE_PRINTF_1;
78 static char *savestring (const char *, unsigned int);
79 static char *dupnstr (const char *, size_t);
80 static int safe_read (int, void *, int);
81 static void safe_write (int, void *, int, const char *);
82 static void save_pointers (void);
83 static void restore_pointers (void);
84 static int is_id_char (int);
85 static int in_system_include_dir (const char *);
86 static int directory_specified_p (const char *);
87 static int file_excluded_p (const char *);
88 static char *unexpand_if_needed (const char *);
89 static char *abspath (const char *, const char *);
90 static void check_aux_info (int);
91 static const char *find_corresponding_lparen (const char *);
92 static int referenced_file_is_newer (const char *, time_t);
93 static void save_def_or_dec (const char *, int);
94 static void munge_compile_params (const char *);
95 static int gen_aux_info_file (const char *);
96 static void process_aux_info_file (const char *, int, int);
97 static int identify_lineno (const char *);
98 static void check_source (int, const char *);
99 static const char *seek_to_line (int);
100 static const char *forward_to_next_token_char (const char *);
101 static void output_bytes (const char *, size_t);
102 static void output_string (const char *);
103 static void output_up_to (const char *);
104 static int other_variable_style_function (const char *);
105 static const char *find_rightmost_formals_list (const char *);
106 static void do_cleaning (char *, const char *);
107 static const char *careful_find_l_paren (const char *);
108 static void do_processing (void);
109 
110 /* Look for these where the `const' qualifier is intentionally cast aside.  */
111 #define NONCONST
112 
113 /* Define a default place to find the SYSCALLS.X file.  */
114 
115 #ifndef UNPROTOIZE
116 
117 #ifndef STANDARD_EXEC_PREFIX
118 #define STANDARD_EXEC_PREFIX "/usr/local/lib/gcc-lib/"
119 #endif /* !defined STANDARD_EXEC_PREFIX */
120 
121 static const char * const standard_exec_prefix = STANDARD_EXEC_PREFIX;
122 static const char * const target_machine = DEFAULT_TARGET_MACHINE;
123 static const char * const target_version = DEFAULT_TARGET_VERSION;
124 
125 #endif /* !defined (UNPROTOIZE) */
126 
127 /* Suffix of aux_info files.  */
128 
129 static const char * const aux_info_suffix = AUX_INFO_SUFFIX;
130 
131 /* String to attach to filenames for saved versions of original files.  */
132 
133 static const char * const save_suffix = SAVE_SUFFIX;
134 
135 #ifndef UNPROTOIZE
136 
137 /* String to attach to C filenames renamed to C++.  */
138 
139 static const char * const cplus_suffix = CPLUS_FILE_SUFFIX;
140 
141 /* File name of the file which contains descriptions of standard system
142    routines.  Note that we never actually do anything with this file per se,
143    but we do read in its corresponding aux_info file.  */
144 
145 static const char syscalls_filename[] = "SYSCALLS.c";
146 
147 /* Default place to find the above file.  */
148 
149 static const char * default_syscalls_dir;
150 
151 /* Variable to hold the complete absolutized filename of the SYSCALLS.c.X
152    file.  */
153 
154 static char * syscalls_absolute_filename;
155 
156 #endif /* !defined (UNPROTOIZE) */
157 
158 /* Type of the structure that holds information about macro unexpansions.  */
159 
160 struct unexpansion_struct {
161   const char *const expanded;
162   const char *const contracted;
163 };
164 typedef struct unexpansion_struct unexpansion;
165 
166 /* A table of conversions that may need to be made for some (stupid) older
167    operating systems where these types are preprocessor macros rather than
168    typedefs (as they really ought to be).
169 
170    WARNING: The contracted forms must be as small (or smaller) as the
171    expanded forms, or else havoc will ensue.  */
172 
173 static const unexpansion unexpansions[] = {
174   { "struct _iobuf", "FILE" },
175   { 0, 0 }
176 };
177 
178 /* The number of "primary" slots in the hash tables for filenames and for
179    function names.  This can be as big or as small as you like, except that
180    it must be a power of two.  */
181 
182 #define HASH_TABLE_SIZE		(1 << 9)
183 
184 /* Bit mask to use when computing hash values.  */
185 
186 static const int hash_mask = (HASH_TABLE_SIZE - 1);
187 
188 
189 /* Datatype for lists of directories or filenames.  */
190 struct string_list
191 {
192   const char *name;
193   struct string_list *next;
194 };
195 
196 static struct string_list *string_list_cons (const char *,
197 					     struct string_list *);
198 
199 /* List of directories in which files should be converted.  */
200 
201 struct string_list *directory_list;
202 
203 /* List of file names which should not be converted.
204    A file is excluded if the end of its name, following a /,
205    matches one of the names in this list.  */
206 
207 struct string_list *exclude_list;
208 
209 /* The name of the other style of variable-number-of-parameters functions
210    (i.e. the style that we want to leave unconverted because we don't yet
211    know how to convert them to this style.  This string is used in warning
212    messages.  */
213 
214 /* Also define here the string that we can search for in the parameter lists
215    taken from the .X files which will unambiguously indicate that we have
216    found a varargs style function.  */
217 
218 #ifdef UNPROTOIZE
219 static const char * const other_var_style = "stdarg";
220 #else /* !defined (UNPROTOIZE) */
221 static const char * const other_var_style = "varargs";
222 static const char *varargs_style_indicator = "va_alist";
223 #endif /* !defined (UNPROTOIZE) */
224 
225 /* The following two types are used to create hash tables.  In this program,
226    there are two hash tables which are used to store and quickly lookup two
227    different classes of strings.  The first type of strings stored in the
228    first hash table are absolute filenames of files which protoize needs to
229    know about.  The second type of strings (stored in the second hash table)
230    are function names.  It is this second class of strings which really
231    inspired the use of the hash tables, because there may be a lot of them.  */
232 
233 typedef struct hash_table_entry_struct hash_table_entry;
234 
235 /* Do some typedefs so that we don't have to write "struct" so often.  */
236 
237 typedef struct def_dec_info_struct def_dec_info;
238 typedef struct file_info_struct file_info;
239 typedef struct f_list_chain_item_struct f_list_chain_item;
240 
241 #ifndef UNPROTOIZE
242 static int is_syscalls_file (const file_info *);
243 static void rename_c_file (const hash_table_entry *);
244 static const def_dec_info *find_extern_def (const def_dec_info *,
245 					    const def_dec_info *);
246 static const def_dec_info *find_static_definition (const def_dec_info *);
247 static void connect_defs_and_decs (const hash_table_entry *);
248 static void add_local_decl (const def_dec_info *, const char *);
249 static void add_global_decls (const file_info *, const char *);
250 #endif /* ! UNPROTOIZE */
251 static int needs_to_be_converted (const file_info *);
252 static void visit_each_hash_node (const hash_table_entry *,
253 				  void (*)(const hash_table_entry *));
254 static hash_table_entry *add_symbol (hash_table_entry *, const char *);
255 static hash_table_entry *lookup (hash_table_entry *, const char *);
256 static void free_def_dec (def_dec_info *);
257 static file_info *find_file (const char *, int);
258 static void reverse_def_dec_list (const hash_table_entry *);
259 static void edit_fn_declaration (const def_dec_info *, const char *);
260 static int edit_formals_lists (const char *, unsigned int,
261 			       const def_dec_info *);
262 static void edit_fn_definition (const def_dec_info *, const char *);
263 static void scan_for_missed_items (const file_info *);
264 static void edit_file (const hash_table_entry *);
265 
266 /* In the struct below, note that the "_info" field has two different uses
267    depending on the type of hash table we are in (i.e. either the filenames
268    hash table or the function names hash table).  In the filenames hash table
269    the info fields of the entries point to the file_info struct which is
270    associated with each filename (1 per filename).  In the function names
271    hash table, the info field points to the head of a singly linked list of
272    def_dec_info entries which are all defs or decs of the function whose
273    name is pointed to by the "symbol" field.  Keeping all of the defs/decs
274    for a given function name on a special list specifically for that function
275    name makes it quick and easy to find out all of the important information
276    about a given (named) function.  */
277 
278 struct hash_table_entry_struct {
279   hash_table_entry *		hash_next;	/* -> to secondary entries */
280   const char *			symbol;		/* -> to the hashed string */
281   union {
282     const def_dec_info *	_ddip;
283     file_info *			_fip;
284   } _info;
285 };
286 #define ddip _info._ddip
287 #define fip _info._fip
288 
289 /* Define a type specifically for our two hash tables.  */
290 
291 typedef hash_table_entry hash_table[HASH_TABLE_SIZE];
292 
293 /* The following struct holds all of the important information about any
294    single filename (e.g. file) which we need to know about.  */
295 
296 struct file_info_struct {
297   const hash_table_entry *	hash_entry; /* -> to associated hash entry */
298   const def_dec_info *		defs_decs;  /* -> to chain of defs/decs */
299   time_t			mtime;      /* Time of last modification.  */
300 };
301 
302 /* Due to the possibility that functions may return pointers to functions,
303    (which may themselves have their own parameter lists) and due to the
304    fact that returned pointers-to-functions may be of type "pointer-to-
305    function-returning-pointer-to-function" (ad nauseum) we have to keep
306    an entire chain of ANSI style formal parameter lists for each function.
307 
308    Normally, for any given function, there will only be one formals list
309    on the chain, but you never know.
310 
311    Note that the head of each chain of formals lists is pointed to by the
312    `f_list_chain' field of the corresponding def_dec_info record.
313 
314    For any given chain, the item at the head of the chain is the *leftmost*
315    parameter list seen in the actual C language function declaration.  If
316    there are other members of the chain, then these are linked in left-to-right
317    order from the head of the chain.  */
318 
319 struct f_list_chain_item_struct {
320   const f_list_chain_item *	chain_next;	/* -> to next item on chain */
321   const char *			formals_list;	/* -> to formals list string */
322 };
323 
324 /* The following struct holds all of the important information about any
325    single function definition or declaration which we need to know about.
326    Note that for unprotoize we don't need to know very much because we
327    never even create records for stuff that we don't intend to convert
328    (like for instance defs and decs which are already in old K&R format
329    and "implicit" function declarations).  */
330 
331 struct def_dec_info_struct {
332   const def_dec_info *	next_in_file;	/* -> to rest of chain for file */
333   file_info *        	file;		/* -> file_info for containing file */
334   int        		line;		/* source line number of def/dec */
335   const char *		ansi_decl;	/* -> left end of ansi decl */
336   hash_table_entry *	hash_entry;	/* -> hash entry for function name */
337   unsigned int        	is_func_def;	/* = 0 means this is a declaration */
338   const def_dec_info *	next_for_func;	/* -> to rest of chain for func name */
339   unsigned int		f_list_count;	/* count of formals lists we expect */
340   char			prototyped;	/* = 0 means already prototyped */
341 #ifndef UNPROTOIZE
342   const f_list_chain_item * f_list_chain;	/* -> chain of formals lists */
343   const def_dec_info *	definition;	/* -> def/dec containing related def */
344   char	        	is_static;	/* = 0 means visibility is "extern"  */
345   char			is_implicit;	/* != 0 for implicit func decl's */
346   char			written;	/* != 0 means written for implicit */
347 #else /* !defined (UNPROTOIZE) */
348   const char *		formal_names;	/* -> to list of names of formals */
349   const char *		formal_decls;	/* -> to string of formal declarations */
350 #endif /* !defined (UNPROTOIZE) */
351 };
352 
353 /* Pointer to the tail component of the filename by which this program was
354    invoked.  Used everywhere in error and warning messages.  */
355 
356 static const char *pname;
357 
358 /* Error counter.  Will be nonzero if we should give up at the next convenient
359    stopping point.  */
360 
361 static int errors = 0;
362 
363 /* Option flags.  */
364 /* ??? The variables are not marked static because some of them have
365    the same names as gcc variables declared in options.h.  */
366 /* ??? These comments should say what the flag mean as well as the options
367    that set them.  */
368 
369 /* File name to use for running gcc.  Allows GCC 2 to be named
370    something other than gcc.  */
371 static const char *compiler_file_name = "gcc";
372 
373 int version_flag = 0;		/* Print our version number.  */
374 int quiet_flag = 0;		/* Don't print messages normally.  */
375 int nochange_flag = 0;		/* Don't convert, just say what files
376 				   we would have converted.  */
377 int nosave_flag = 0;		/* Don't save the old version.  */
378 int keep_flag = 0;		/* Don't delete the .X files.  */
379 static const char ** compile_params = 0;	/* Option string for gcc.  */
380 #ifdef UNPROTOIZE
381 static const char *indent_string = "     ";	/* Indentation for newly
382 						   inserted parm decls.  */
383 #else /* !defined (UNPROTOIZE) */
384 int local_flag = 0;		/* Insert new local decls (when?).  */
385 int global_flag = 0;		/* set by -g option */
386 int cplusplus_flag = 0;		/* Rename converted files to *.C.  */
387 static const char *nondefault_syscalls_dir = 0; /* Dir to look for
388 						   SYSCALLS.c.X in.  */
389 #endif /* !defined (UNPROTOIZE) */
390 
391 /* An index into the compile_params array where we should insert the source
392    file name when we are ready to exec the C compiler.  A zero value indicates
393    that we have not yet called munge_compile_params.  */
394 
395 static int input_file_name_index = 0;
396 
397 /* An index into the compile_params array where we should insert the filename
398    for the aux info file, when we run the C compiler.  */
399 static int aux_info_file_name_index = 0;
400 
401 /* Count of command line arguments which were "filename" arguments.  */
402 
403 static int n_base_source_files = 0;
404 
405 /* Points to a malloc'ed list of pointers to all of the filenames of base
406    source files which were specified on the command line.  */
407 
408 static const char **base_source_filenames;
409 
410 /* Line number of the line within the current aux_info file that we
411    are currently processing.  Used for error messages in case the prototypes
412    info file is corrupted somehow.  */
413 
414 static int current_aux_info_lineno;
415 
416 /* Pointer to the name of the source file currently being converted.  */
417 
418 static const char *convert_filename;
419 
420 /* Pointer to relative root string (taken from aux_info file) which indicates
421    where directory the user was in when he did the compilation step that
422    produced the containing aux_info file.  */
423 
424 static const char *invocation_filename;
425 
426 /* Pointer to the base of the input buffer that holds the original text for the
427    source file currently being converted.  */
428 
429 static const char *orig_text_base;
430 
431 /* Pointer to the byte just beyond the end of the input buffer that holds the
432    original text for the source file currently being converted.  */
433 
434 static const char *orig_text_limit;
435 
436 /* Pointer to the base of the input buffer that holds the cleaned text for the
437    source file currently being converted.  */
438 
439 static const char *clean_text_base;
440 
441 /* Pointer to the byte just beyond the end of the input buffer that holds the
442    cleaned text for the source file currently being converted.  */
443 
444 static const char *clean_text_limit;
445 
446 /* Pointer to the last byte in the cleaned text buffer that we have already
447    (virtually) copied to the output buffer (or decided to ignore).  */
448 
449 static const char * clean_read_ptr;
450 
451 /* Pointer to the base of the output buffer that holds the replacement text
452    for the source file currently being converted.  */
453 
454 static char *repl_text_base;
455 
456 /* Pointer to the byte just beyond the end of the output buffer that holds the
457    replacement text for the source file currently being converted.  */
458 
459 static char *repl_text_limit;
460 
461 /* Pointer to the last byte which has been stored into the output buffer.
462    The next byte to be stored should be stored just past where this points
463    to.  */
464 
465 static char * repl_write_ptr;
466 
467 /* Pointer into the cleaned text buffer for the source file we are currently
468    converting.  This points to the first character of the line that we last
469    did a "seek_to_line" to (see below).  */
470 
471 static const char *last_known_line_start;
472 
473 /* Number of the line (in the cleaned text buffer) that we last did a
474    "seek_to_line" to.  Will be one if we just read a new source file
475    into the cleaned text buffer.  */
476 
477 static int last_known_line_number;
478 
479 /* The filenames hash table.  */
480 
481 static hash_table filename_primary;
482 
483 /* The function names hash table.  */
484 
485 static hash_table function_name_primary;
486 
487 /* The place to keep the recovery address which is used only in cases where
488    we get hopelessly confused by something in the cleaned original text.  */
489 
490 static jmp_buf source_confusion_recovery;
491 
492 /* A pointer to the current directory filename (used by abspath).  */
493 
494 static char *cwd_buffer;
495 
496 /* A place to save the read pointer until we are sure that an individual
497    attempt at editing will succeed.  */
498 
499 static const char * saved_clean_read_ptr;
500 
501 /* A place to save the write pointer until we are sure that an individual
502    attempt at editing will succeed.  */
503 
504 static char * saved_repl_write_ptr;
505 
506 /* Translate and output an error message.  */
507 static void
notice(const char * cmsgid,...)508 notice (const char *cmsgid, ...)
509 {
510   va_list ap;
511 
512   va_start (ap, cmsgid);
513   vfprintf (stderr, _(cmsgid), ap);
514   va_end (ap);
515 }
516 
517 
518 /* Make a copy of a string INPUT with size SIZE.  */
519 
520 static char *
savestring(const char * input,unsigned int size)521 savestring (const char *input, unsigned int size)
522 {
523   char *output = xmalloc (size + 1);
524   strcpy (output, input);
525   return output;
526 }
527 
528 
529 /* Make a duplicate of the first N bytes of a given string in a newly
530    allocated area.  */
531 
532 static char *
dupnstr(const char * s,size_t n)533 dupnstr (const char *s, size_t n)
534 {
535   char *ret_val = xmalloc (n + 1);
536 
537   strncpy (ret_val, s, n);
538   ret_val[n] = '\0';
539   return ret_val;
540 }
541 
542 /* Read LEN bytes at PTR from descriptor DESC, for file FILENAME,
543    retrying if necessary.  Return the actual number of bytes read.  */
544 
545 static int
safe_read(int desc,void * ptr,int len)546 safe_read (int desc, void *ptr, int len)
547 {
548   int left = len;
549   while (left > 0) {
550     int nchars = read (desc, ptr, left);
551     if (nchars < 0)
552       {
553 #ifdef EINTR
554 	if (errno == EINTR)
555 	  continue;
556 #endif
557 	return nchars;
558       }
559     if (nchars == 0)
560       break;
561     /* Arithmetic on void pointers is a gcc extension.  */
562     ptr = (char *) ptr + nchars;
563     left -= nchars;
564   }
565   return len - left;
566 }
567 
568 /* Write LEN bytes at PTR to descriptor DESC,
569    retrying if necessary, and treating any real error as fatal.  */
570 
571 static void
safe_write(int desc,void * ptr,int len,const char * out_fname)572 safe_write (int desc, void *ptr, int len, const char *out_fname)
573 {
574   while (len > 0) {
575     int written = write (desc, ptr, len);
576     if (written < 0)
577       {
578 	int errno_val = errno;
579 #ifdef EINTR
580 	if (errno_val == EINTR)
581 	  continue;
582 #endif
583 	notice ("%s: error writing file '%s': %s\n",
584 		pname, shortpath (NULL, out_fname), xstrerror (errno_val));
585 	return;
586       }
587     /* Arithmetic on void pointers is a gcc extension.  */
588     ptr = (char *) ptr + written;
589     len -= written;
590   }
591 }
592 
593 /* Get setup to recover in case the edit we are about to do goes awry.  */
594 
595 static void
save_pointers(void)596 save_pointers (void)
597 {
598   saved_clean_read_ptr = clean_read_ptr;
599   saved_repl_write_ptr = repl_write_ptr;
600 }
601 
602 /* Call this routine to recover our previous state whenever something looks
603    too confusing in the source code we are trying to edit.  */
604 
605 static void
restore_pointers(void)606 restore_pointers (void)
607 {
608   clean_read_ptr = saved_clean_read_ptr;
609   repl_write_ptr = saved_repl_write_ptr;
610 }
611 
612 /* Return true if the given character is a valid identifier character.  */
613 
614 static int
is_id_char(int ch)615 is_id_char (int ch)
616 {
617   return (ISIDNUM (ch) || (ch == '$'));
618 }
619 
620 /* Give a message indicating the proper way to invoke this program and then
621    exit with nonzero status.  */
622 
623 static void
usage(void)624 usage (void)
625 {
626 #ifdef UNPROTOIZE
627   notice ("%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ filename ... ]'\n",
628 	  pname, pname);
629 #else /* !defined (UNPROTOIZE) */
630   notice ("%s: usage '%s [ -VqfnkNlgC ] [ -B <dirname> ] [ filename ... ]'\n",
631 	  pname, pname);
632 #endif /* !defined (UNPROTOIZE) */
633   exit (FATAL_EXIT_CODE);
634 }
635 
636 /* Return true if the given filename (assumed to be an absolute filename)
637    designates a file residing anywhere beneath any one of the "system"
638    include directories.  */
639 
640 static int
in_system_include_dir(const char * path)641 in_system_include_dir (const char *path)
642 {
643   const struct default_include *p;
644 
645   gcc_assert (IS_ABSOLUTE_PATH (path));
646 
647   for (p = cpp_include_defaults; p->fname; p++)
648     if (!strncmp (path, p->fname, strlen (p->fname))
649 	&& IS_DIR_SEPARATOR (path[strlen (p->fname)]))
650       return 1;
651   return 0;
652 }
653 
654 #if 0
655 /* Return true if the given filename designates a file that the user has
656    read access to and for which the user has write access to the containing
657    directory.  */
658 
659 static int
file_could_be_converted(const char * path)660 file_could_be_converted (const char *path)
661 {
662   char *const dir_name = alloca (strlen (path) + 1);
663 
664   if (access (path, R_OK))
665     return 0;
666 
667   {
668     char *dir_last_slash;
669 
670     strcpy (dir_name, path);
671     dir_last_slash = strrchr (dir_name, DIR_SEPARATOR);
672 #ifdef DIR_SEPARATOR_2
673     {
674       char *slash;
675 
676       slash = strrchr (dir_last_slash ? dir_last_slash : dir_name,
677 		       DIR_SEPARATOR_2);
678       if (slash)
679 	dir_last_slash = slash;
680     }
681 #endif
682     gcc_assert (dir_last_slash);
683     *dir_last_slash = '\0';
684   }
685 
686   if (access (path, W_OK))
687     return 0;
688 
689   return 1;
690 }
691 
692 /* Return true if the given filename designates a file that we are allowed
693    to modify.  Files which we should not attempt to modify are (a) "system"
694    include files, and (b) files which the user doesn't have write access to,
695    and (c) files which reside in directories which the user doesn't have
696    write access to.  Unless requested to be quiet, give warnings about
697    files that we will not try to convert for one reason or another.  An
698    exception is made for "system" include files, which we never try to
699    convert and for which we don't issue the usual warnings.  */
700 
701 static int
file_normally_convertible(const char * path)702 file_normally_convertible (const char *path)
703 {
704   char *const dir_name = alloca (strlen (path) + 1);
705 
706   if (in_system_include_dir (path))
707     return 0;
708 
709   {
710     char *dir_last_slash;
711 
712     strcpy (dir_name, path);
713     dir_last_slash = strrchr (dir_name, DIR_SEPARATOR);
714 #ifdef DIR_SEPARATOR_2
715     {
716       char *slash;
717 
718       slash = strrchr (dir_last_slash ? dir_last_slash : dir_name,
719 		       DIR_SEPARATOR_2);
720       if (slash)
721 	dir_last_slash = slash;
722     }
723 #endif
724     gcc_assert (dir_last_slash);
725     *dir_last_slash = '\0';
726   }
727 
728   if (access (path, R_OK))
729     {
730       if (!quiet_flag)
731 	notice ("%s: warning: no read access for file '%s'\n",
732 		pname, shortpath (NULL, path));
733       return 0;
734     }
735 
736   if (access (path, W_OK))
737     {
738       if (!quiet_flag)
739 	notice ("%s: warning: no write access for file '%s'\n",
740 		pname, shortpath (NULL, path));
741       return 0;
742     }
743 
744   if (access (dir_name, W_OK))
745     {
746       if (!quiet_flag)
747 	notice ("%s: warning: no write access for dir containing '%s'\n",
748 		pname, shortpath (NULL, path));
749       return 0;
750     }
751 
752   return 1;
753 }
754 #endif /* 0 */
755 
756 #ifndef UNPROTOIZE
757 
758 /* Return true if the given file_info struct refers to the special SYSCALLS.c.X
759    file.  Return false otherwise.  */
760 
761 static int
is_syscalls_file(const file_info * fi_p)762 is_syscalls_file (const file_info *fi_p)
763 {
764   char const *f = fi_p->hash_entry->symbol;
765   size_t fl = strlen (f), sysl = sizeof (syscalls_filename) - 1;
766   return sysl <= fl  &&  strcmp (f + fl - sysl, syscalls_filename) == 0;
767 }
768 
769 #endif /* !defined (UNPROTOIZE) */
770 
771 /* Check to see if this file will need to have anything done to it on this
772    run.  If there is nothing in the given file which both needs conversion
773    and for which we have the necessary stuff to do the conversion, return
774    false.  Otherwise, return true.
775 
776    Note that (for protoize) it is only valid to call this function *after*
777    the connections between declarations and definitions have all been made
778    by connect_defs_and_decs.  */
779 
780 static int
needs_to_be_converted(const file_info * file_p)781 needs_to_be_converted (const file_info *file_p)
782 {
783   const def_dec_info *ddp;
784 
785 #ifndef UNPROTOIZE
786 
787   if (is_syscalls_file (file_p))
788     return 0;
789 
790 #endif /* !defined (UNPROTOIZE) */
791 
792   for (ddp = file_p->defs_decs; ddp; ddp = ddp->next_in_file)
793 
794     if (
795 
796 #ifndef UNPROTOIZE
797 
798       /* ... and if we a protoizing and this function is in old style ...  */
799       !ddp->prototyped
800       /* ... and if this a definition or is a decl with an associated def ...  */
801       && (ddp->is_func_def || (!ddp->is_func_def && ddp->definition))
802 
803 #else /* defined (UNPROTOIZE) */
804 
805       /* ... and if we are unprotoizing and this function is in new style ...  */
806       ddp->prototyped
807 
808 #endif /* defined (UNPROTOIZE) */
809       )
810 	  /* ... then the containing file needs converting.  */
811 	  return -1;
812   return 0;
813 }
814 
815 /* Return 1 if the file name NAME is in a directory
816    that should be converted.  */
817 
818 static int
directory_specified_p(const char * name)819 directory_specified_p (const char *name)
820 {
821   struct string_list *p;
822 
823   for (p = directory_list; p; p = p->next)
824     if (!strncmp (name, p->name, strlen (p->name))
825 	&& IS_DIR_SEPARATOR (name[strlen (p->name)]))
826       {
827 	const char *q = name + strlen (p->name) + 1;
828 
829 	/* If there are more slashes, it's in a subdir, so
830 	   this match doesn't count.  */
831 	while (*q++)
832 	  if (IS_DIR_SEPARATOR (*(q-1)))
833 	    goto lose;
834 	return 1;
835 
836       lose: ;
837       }
838 
839   return 0;
840 }
841 
842 /* Return 1 if the file named NAME should be excluded from conversion.  */
843 
844 static int
file_excluded_p(const char * name)845 file_excluded_p (const char *name)
846 {
847   struct string_list *p;
848   int len = strlen (name);
849 
850   for (p = exclude_list; p; p = p->next)
851     if (!strcmp (name + len - strlen (p->name), p->name)
852 	&& IS_DIR_SEPARATOR (name[len - strlen (p->name) - 1]))
853       return 1;
854 
855   return 0;
856 }
857 
858 /* Construct a new element of a string_list.
859    STRING is the new element value, and REST holds the remaining elements.  */
860 
861 static struct string_list *
string_list_cons(const char * string,struct string_list * rest)862 string_list_cons (const char *string, struct string_list *rest)
863 {
864   struct string_list *temp = xmalloc (sizeof (struct string_list));
865 
866   temp->next = rest;
867   temp->name = string;
868   return temp;
869 }
870 
871 /* ??? The GNU convention for mentioning function args in its comments
872    is to capitalize them.  So change "hash_tab_p" to HASH_TAB_P below.
873    Likewise for all the other functions.  */
874 
875 /* Given a hash table, apply some function to each node in the table. The
876    table to traverse is given as the "hash_tab_p" argument, and the
877    function to be applied to each node in the table is given as "func"
878    argument.  */
879 
880 static void
visit_each_hash_node(const hash_table_entry * hash_tab_p,void (* func)(const hash_table_entry *))881 visit_each_hash_node (const hash_table_entry *hash_tab_p,
882 		      void (*func) (const hash_table_entry *))
883 {
884   const hash_table_entry *primary;
885 
886   for (primary = hash_tab_p; primary < &hash_tab_p[HASH_TABLE_SIZE]; primary++)
887     if (primary->symbol)
888       {
889 	hash_table_entry *second;
890 
891 	(*func)(primary);
892 	for (second = primary->hash_next; second; second = second->hash_next)
893 	  (*func) (second);
894       }
895 }
896 
897 /* Initialize all of the fields of a new hash table entry, pointed
898    to by the "p" parameter.  Note that the space to hold the entry
899    is assumed to have already been allocated before this routine is
900    called.  */
901 
902 static hash_table_entry *
add_symbol(hash_table_entry * p,const char * s)903 add_symbol (hash_table_entry *p, const char *s)
904 {
905   p->hash_next = NULL;
906   p->symbol = xstrdup (s);
907   p->ddip = NULL;
908   p->fip = NULL;
909   return p;
910 }
911 
912 /* Look for a particular function name or filename in the particular
913    hash table indicated by "hash_tab_p".  If the name is not in the
914    given hash table, add it.  Either way, return a pointer to the
915    hash table entry for the given name.  */
916 
917 static hash_table_entry *
lookup(hash_table_entry * hash_tab_p,const char * search_symbol)918 lookup (hash_table_entry *hash_tab_p, const char *search_symbol)
919 {
920   int hash_value = 0;
921   const char *search_symbol_char_p = search_symbol;
922   hash_table_entry *p;
923 
924   while (*search_symbol_char_p)
925     hash_value += *search_symbol_char_p++;
926   hash_value &= hash_mask;
927   p = &hash_tab_p[hash_value];
928   if (! p->symbol)
929       return add_symbol (p, search_symbol);
930   if (!strcmp (p->symbol, search_symbol))
931     return p;
932   while (p->hash_next)
933     {
934       p = p->hash_next;
935       if (!strcmp (p->symbol, search_symbol))
936 	return p;
937     }
938   p->hash_next = xmalloc (sizeof (hash_table_entry));
939   p = p->hash_next;
940   return add_symbol (p, search_symbol);
941 }
942 
943 /* Throw a def/dec record on the junk heap.
944 
945    Also, since we are not using this record anymore, free up all of the
946    stuff it pointed to.  */
947 
948 static void
free_def_dec(def_dec_info * p)949 free_def_dec (def_dec_info *p)
950 {
951   free ((NONCONST void *) p->ansi_decl);
952 
953 #ifndef UNPROTOIZE
954   {
955     const f_list_chain_item * curr;
956     const f_list_chain_item * next;
957 
958     for (curr = p->f_list_chain; curr; curr = next)
959       {
960 	next = curr->chain_next;
961 	free ((NONCONST void *) curr);
962       }
963   }
964 #endif /* !defined (UNPROTOIZE) */
965 
966   free (p);
967 }
968 
969 /* Unexpand as many macro symbols as we can find.
970 
971    If the given line must be unexpanded, make a copy of it in the heap and
972    return a pointer to the unexpanded copy.  Otherwise return NULL.  */
973 
974 static char *
unexpand_if_needed(const char * aux_info_line)975 unexpand_if_needed (const char *aux_info_line)
976 {
977   static char *line_buf = 0;
978   static int line_buf_size = 0;
979   const unexpansion *unexp_p;
980   int got_unexpanded = 0;
981   const char *s;
982   char *copy_p = line_buf;
983 
984   if (line_buf == 0)
985     {
986       line_buf_size = 1024;
987       line_buf = xmalloc (line_buf_size);
988     }
989 
990   copy_p = line_buf;
991 
992   /* Make a copy of the input string in line_buf, expanding as necessary.  */
993 
994   for (s = aux_info_line; *s != '\n'; )
995     {
996       for (unexp_p = unexpansions; unexp_p->expanded; unexp_p++)
997 	{
998 	  const char *in_p = unexp_p->expanded;
999 	  size_t len = strlen (in_p);
1000 
1001 	  if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
1002 	    {
1003 	      int size = strlen (unexp_p->contracted);
1004 	      got_unexpanded = 1;
1005 	      if (copy_p + size - line_buf >= line_buf_size)
1006 		{
1007 		  int offset = copy_p - line_buf;
1008 		  line_buf_size *= 2;
1009 		  line_buf_size += size;
1010 		  line_buf = xrealloc (line_buf, line_buf_size);
1011 		  copy_p = line_buf + offset;
1012 		}
1013 	      strcpy (copy_p, unexp_p->contracted);
1014 	      copy_p += size;
1015 
1016 	      /* Assume that there will not be another replacement required
1017 	         within the text just replaced.  */
1018 
1019 	      s += len;
1020 	      goto continue_outer;
1021 	    }
1022 	}
1023       if (copy_p - line_buf == line_buf_size)
1024 	{
1025 	  int offset = copy_p - line_buf;
1026 	  line_buf_size *= 2;
1027 	  line_buf = xrealloc (line_buf, line_buf_size);
1028 	  copy_p = line_buf + offset;
1029 	}
1030       *copy_p++ = *s++;
1031 continue_outer: ;
1032     }
1033   if (copy_p + 2 - line_buf >= line_buf_size)
1034     {
1035       int offset = copy_p - line_buf;
1036       line_buf_size *= 2;
1037       line_buf = xrealloc (line_buf, line_buf_size);
1038       copy_p = line_buf + offset;
1039     }
1040   *copy_p++ = '\n';
1041   *copy_p = '\0';
1042 
1043   return (got_unexpanded ? savestring (line_buf, copy_p - line_buf) : 0);
1044 }
1045 
1046 /* Return the absolutized filename for the given relative
1047    filename.  Note that if that filename is already absolute, it may
1048    still be returned in a modified form because this routine also
1049    eliminates redundant slashes and single dots and eliminates double
1050    dots to get a shortest possible filename from the given input
1051    filename.  The absolutization of relative filenames is made by
1052    assuming that the given filename is to be taken as relative to
1053    the first argument (cwd) or to the current directory if cwd is
1054    NULL.  */
1055 
1056 static char *
abspath(const char * cwd,const char * rel_filename)1057 abspath (const char *cwd, const char *rel_filename)
1058 {
1059   /* Setup the current working directory as needed.  */
1060   const char *const cwd2 = (cwd) ? cwd : cwd_buffer;
1061   char *const abs_buffer = alloca (strlen (cwd2) + strlen (rel_filename) + 2);
1062   char *endp = abs_buffer;
1063   char *outp, *inp;
1064 
1065   /* Copy the  filename (possibly preceded by the current working
1066      directory name) into the absolutization buffer.  */
1067 
1068   {
1069     const char *src_p;
1070 
1071     if (! IS_ABSOLUTE_PATH (rel_filename))
1072       {
1073 	src_p = cwd2;
1074 	while ((*endp++ = *src_p++))
1075 	  continue;
1076 	*(endp-1) = DIR_SEPARATOR;     		/* overwrite null */
1077       }
1078 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1079     else if (IS_DIR_SEPARATOR (rel_filename[0]))
1080       {
1081 	/* A path starting with a directory separator is considered absolute
1082 	   for dos based filesystems, but it's really not -- it's just the
1083 	   convention used throughout GCC and it works. However, in this
1084 	   case, we still need to prepend the drive spec from cwd_buffer.  */
1085 	*endp++ = cwd2[0];
1086 	*endp++ = cwd2[1];
1087       }
1088 #endif
1089     src_p = rel_filename;
1090     while ((*endp++ = *src_p++))
1091       continue;
1092   }
1093 
1094   /* Now make a copy of abs_buffer into abs_buffer, shortening the
1095      filename (by taking out slashes and dots) as we go.  */
1096 
1097   outp = inp = abs_buffer;
1098   *outp++ = *inp++;        	/* copy first slash */
1099 #if defined (apollo) || defined (_WIN32) || defined (__INTERIX)
1100   if (IS_DIR_SEPARATOR (inp[0]))
1101     *outp++ = *inp++;        	/* copy second slash */
1102 #endif
1103   for (;;)
1104     {
1105       if (!inp[0])
1106 	break;
1107       else if (IS_DIR_SEPARATOR (inp[0]) && IS_DIR_SEPARATOR (outp[-1]))
1108 	{
1109 	  inp++;
1110 	  continue;
1111 	}
1112       else if (inp[0] == '.' && IS_DIR_SEPARATOR (outp[-1]))
1113 	{
1114 	  if (!inp[1])
1115 	    break;
1116 	  else if (IS_DIR_SEPARATOR (inp[1]))
1117 	    {
1118 	      inp += 2;
1119 	      continue;
1120 	    }
1121 	  else if ((inp[1] == '.') && (inp[2] == 0
1122 	                               || IS_DIR_SEPARATOR (inp[2])))
1123 	    {
1124 	      inp += (IS_DIR_SEPARATOR (inp[2])) ? 3 : 2;
1125 	      outp -= 2;
1126 	      while (outp >= abs_buffer && ! IS_DIR_SEPARATOR (*outp))
1127 	      	outp--;
1128 	      if (outp < abs_buffer)
1129 		{
1130 		  /* Catch cases like /.. where we try to backup to a
1131 		     point above the absolute root of the logical file
1132 		     system.  */
1133 
1134 		  notice ("%s: invalid file name: %s\n",
1135 			  pname, rel_filename);
1136 		  exit (FATAL_EXIT_CODE);
1137 		}
1138 	      *++outp = '\0';
1139 	      continue;
1140 	    }
1141 	}
1142       *outp++ = *inp++;
1143     }
1144 
1145   /* On exit, make sure that there is a trailing null, and make sure that
1146      the last character of the returned string is *not* a slash.  */
1147 
1148   *outp = '\0';
1149   if (IS_DIR_SEPARATOR (outp[-1]))
1150     *--outp  = '\0';
1151 
1152   /* Make a copy (in the heap) of the stuff left in the absolutization
1153      buffer and return a pointer to the copy.  */
1154 
1155   return savestring (abs_buffer, outp - abs_buffer);
1156 }
1157 
1158 /* Given a filename (and possibly a directory name from which the filename
1159    is relative) return a string which is the shortest possible
1160    equivalent for the corresponding full (absolutized) filename.  The
1161    shortest possible equivalent may be constructed by converting the
1162    absolutized filename to be a relative filename (i.e. relative to
1163    the actual current working directory).  However if a relative filename
1164    is longer, then the full absolute filename is returned.
1165 
1166    KNOWN BUG:
1167 
1168    Note that "simple-minded" conversion of any given type of filename (either
1169    relative or absolute) may not result in a valid equivalent filename if any
1170    subpart of the original filename is actually a symbolic link.  */
1171 
1172 static const char *
shortpath(const char * cwd,const char * filename)1173 shortpath (const char *cwd, const char *filename)
1174 {
1175   char *rel_buffer;
1176   char *rel_buf_p;
1177   char *cwd_p = cwd_buffer;
1178   char *path_p;
1179   int unmatched_slash_count = 0;
1180   size_t filename_len = strlen (filename);
1181 
1182   path_p = abspath (cwd, filename);
1183   rel_buf_p = rel_buffer = xmalloc (filename_len);
1184 
1185   while (*cwd_p && IS_SAME_PATH_CHAR (*cwd_p, *path_p))
1186     {
1187       cwd_p++;
1188       path_p++;
1189     }
1190   if (!*cwd_p && (!*path_p || IS_DIR_SEPARATOR (*path_p)))
1191     {
1192       /* whole pwd matched */
1193       if (!*path_p)        	/* input *is* the current path! */
1194 	return ".";
1195       else
1196 	return ++path_p;
1197     }
1198   else
1199     {
1200       if (*path_p)
1201 	{
1202 	  --cwd_p;
1203 	  --path_p;
1204 	  while (! IS_DIR_SEPARATOR (*cwd_p))     /* backup to last slash */
1205 	    {
1206 	      --cwd_p;
1207 	      --path_p;
1208 	    }
1209 	  cwd_p++;
1210 	  path_p++;
1211 	  unmatched_slash_count++;
1212 	}
1213 
1214       /* Find out how many directory levels in cwd were *not* matched.  */
1215       while (*cwd_p++)
1216 	if (IS_DIR_SEPARATOR (*(cwd_p-1)))
1217 	  unmatched_slash_count++;
1218 
1219       /* Now we know how long the "short name" will be.
1220 	 Reject it if longer than the input.  */
1221       if (unmatched_slash_count * 3 + strlen (path_p) >= filename_len)
1222 	return filename;
1223 
1224       /* For each of them, put a `../' at the beginning of the short name.  */
1225       while (unmatched_slash_count--)
1226 	{
1227 	  /* Give up if the result gets to be longer
1228 	     than the absolute path name.  */
1229 	  if (rel_buffer + filename_len <= rel_buf_p + 3)
1230 	    return filename;
1231 	  *rel_buf_p++ = '.';
1232 	  *rel_buf_p++ = '.';
1233 	  *rel_buf_p++ = DIR_SEPARATOR;
1234 	}
1235 
1236       /* Then tack on the unmatched part of the desired file's name.  */
1237       do
1238 	{
1239 	  if (rel_buffer + filename_len <= rel_buf_p)
1240 	    return filename;
1241 	}
1242       while ((*rel_buf_p++ = *path_p++));
1243 
1244       --rel_buf_p;
1245       if (IS_DIR_SEPARATOR (*(rel_buf_p-1)))
1246 	*--rel_buf_p = '\0';
1247       return rel_buffer;
1248     }
1249 }
1250 
1251 /* Lookup the given filename in the hash table for filenames.  If it is a
1252    new one, then the hash table info pointer will be null.  In this case,
1253    we create a new file_info record to go with the filename, and we initialize
1254    that record with some reasonable values.  */
1255 
1256 /* FILENAME was const, but that causes a warning on AIX when calling stat.
1257    That is probably a bug in AIX, but might as well avoid the warning.  */
1258 
1259 static file_info *
find_file(const char * filename,int do_not_stat)1260 find_file (const char *filename, int do_not_stat)
1261 {
1262   hash_table_entry *hash_entry_p;
1263 
1264   hash_entry_p = lookup (filename_primary, filename);
1265   if (hash_entry_p->fip)
1266     return hash_entry_p->fip;
1267   else
1268     {
1269       struct stat stat_buf;
1270       file_info *file_p = xmalloc (sizeof (file_info));
1271 
1272       /* If we cannot get status on any given source file, give a warning
1273 	 and then just set its time of last modification to infinity.  */
1274 
1275       if (do_not_stat)
1276 	stat_buf.st_mtime = (time_t) 0;
1277       else
1278 	{
1279 	  if (stat (filename, &stat_buf) == -1)
1280 	    {
1281 	      int errno_val = errno;
1282 	      notice ("%s: %s: can't get status: %s\n",
1283 		      pname, shortpath (NULL, filename),
1284 		      xstrerror (errno_val));
1285 	      stat_buf.st_mtime = (time_t) -1;
1286 	    }
1287 	}
1288 
1289       hash_entry_p->fip = file_p;
1290       file_p->hash_entry = hash_entry_p;
1291       file_p->defs_decs = NULL;
1292       file_p->mtime = stat_buf.st_mtime;
1293       return file_p;
1294     }
1295 }
1296 
1297 /* Generate a fatal error because some part of the aux_info file is
1298    messed up.  */
1299 
1300 static void
aux_info_corrupted(void)1301 aux_info_corrupted (void)
1302 {
1303   notice ("\n%s: fatal error: aux info file corrupted at line %d\n",
1304 	  pname, current_aux_info_lineno);
1305   exit (FATAL_EXIT_CODE);
1306 }
1307 
1308 /* ??? This comment is vague.  Say what the condition is for.  */
1309 /* Check to see that a condition is true.  This is kind of like an assert.  */
1310 
1311 static void
check_aux_info(int cond)1312 check_aux_info (int cond)
1313 {
1314   if (! cond)
1315     aux_info_corrupted ();
1316 }
1317 
1318 /* Given a pointer to the closing right parenthesis for a particular formals
1319    list (in an aux_info file) find the corresponding left parenthesis and
1320    return a pointer to it.  */
1321 
1322 static const char *
find_corresponding_lparen(const char * p)1323 find_corresponding_lparen (const char *p)
1324 {
1325   const char *q;
1326   int paren_depth;
1327 
1328   for (paren_depth = 1, q = p-1; paren_depth; q--)
1329     {
1330       switch (*q)
1331 	{
1332 	case ')':
1333 	  paren_depth++;
1334 	  break;
1335 	case '(':
1336 	  paren_depth--;
1337 	  break;
1338 	}
1339     }
1340   return ++q;
1341 }
1342 
1343 /* Given a line from  an aux info file, and a time at which the aux info
1344    file it came from was created, check to see if the item described in
1345    the line comes from a file which has been modified since the aux info
1346    file was created.  If so, return nonzero, else return zero.  */
1347 
1348 static int
referenced_file_is_newer(const char * l,time_t aux_info_mtime)1349 referenced_file_is_newer (const char *l, time_t aux_info_mtime)
1350 {
1351   const char *p;
1352   file_info *fi_p;
1353   char *filename;
1354 
1355   check_aux_info (l[0] == '/');
1356   check_aux_info (l[1] == '*');
1357   check_aux_info (l[2] == ' ');
1358 
1359   {
1360     const char *filename_start = p = l + 3;
1361 
1362     while (*p != ':'
1363 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1364 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1365 #endif
1366 	   )
1367       p++;
1368     filename = alloca ((size_t) (p - filename_start) + 1);
1369     strncpy (filename, filename_start, (size_t) (p - filename_start));
1370     filename[p-filename_start] = '\0';
1371   }
1372 
1373   /* Call find_file to find the file_info record associated with the file
1374      which contained this particular def or dec item.  Note that this call
1375      may cause a new file_info record to be created if this is the first time
1376      that we have ever known about this particular file.  */
1377 
1378   fi_p = find_file (abspath (invocation_filename, filename), 0);
1379 
1380   return (fi_p->mtime > aux_info_mtime);
1381 }
1382 
1383 /* Given a line of info from the aux_info file, create a new
1384    def_dec_info record to remember all of the important information about
1385    a function definition or declaration.
1386 
1387    Link this record onto the list of such records for the particular file in
1388    which it occurred in proper (descending) line number order (for now).
1389 
1390    If there is an identical record already on the list for the file, throw
1391    this one away.  Doing so takes care of the (useless and troublesome)
1392    duplicates which are bound to crop up due to multiple inclusions of any
1393    given individual header file.
1394 
1395    Finally, link the new def_dec record onto the list of such records
1396    pertaining to this particular function name.  */
1397 
1398 static void
save_def_or_dec(const char * l,int is_syscalls)1399 save_def_or_dec (const char *l, int is_syscalls)
1400 {
1401   const char *p;
1402   const char *semicolon_p;
1403   def_dec_info *def_dec_p = xmalloc (sizeof (def_dec_info));
1404 
1405 #ifndef UNPROTOIZE
1406   def_dec_p->written = 0;
1407 #endif /* !defined (UNPROTOIZE) */
1408 
1409   /* Start processing the line by picking off 5 pieces of information from
1410      the left hand end of the line.  These are filename, line number,
1411      new/old/implicit flag (new = ANSI prototype format), definition or
1412      declaration flag, and extern/static flag).  */
1413 
1414   check_aux_info (l[0] == '/');
1415   check_aux_info (l[1] == '*');
1416   check_aux_info (l[2] == ' ');
1417 
1418   {
1419     const char *filename_start = p = l + 3;
1420     char *filename;
1421 
1422     while (*p != ':'
1423 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1424 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1425 #endif
1426 	   )
1427       p++;
1428     filename = alloca ((size_t) (p - filename_start) + 1);
1429     strncpy (filename, filename_start, (size_t) (p - filename_start));
1430     filename[p-filename_start] = '\0';
1431 
1432     /* Call find_file to find the file_info record associated with the file
1433        which contained this particular def or dec item.  Note that this call
1434        may cause a new file_info record to be created if this is the first time
1435        that we have ever known about this particular file.
1436 
1437        Note that we started out by forcing all of the base source file names
1438        (i.e. the names of the aux_info files with the .X stripped off) into the
1439        filenames hash table, and we simultaneously setup file_info records for
1440        all of these base file names (even if they may be useless later).
1441        The file_info records for all of these "base" file names (properly)
1442        act as file_info records for the "original" (i.e. un-included) files
1443        which were submitted to gcc for compilation (when the -aux-info
1444        option was used).  */
1445 
1446     def_dec_p->file = find_file (abspath (invocation_filename, filename), is_syscalls);
1447   }
1448 
1449   {
1450     const char *line_number_start = ++p;
1451     char line_number[10];
1452 
1453     while (*p != ':'
1454 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1455 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1456 #endif
1457 	   )
1458       p++;
1459     strncpy (line_number, line_number_start, (size_t) (p - line_number_start));
1460     line_number[p-line_number_start] = '\0';
1461     def_dec_p->line = atoi (line_number);
1462   }
1463 
1464   /* Check that this record describes a new-style, old-style, or implicit
1465      definition or declaration.  */
1466 
1467   p++;	/* Skip over the `:'.  */
1468   check_aux_info ((*p == 'N') || (*p == 'O') || (*p == 'I'));
1469 
1470   /* Is this a new style (ANSI prototyped) definition or declaration? */
1471 
1472   def_dec_p->prototyped = (*p == 'N');
1473 
1474 #ifndef UNPROTOIZE
1475 
1476   /* Is this an implicit declaration? */
1477 
1478   def_dec_p->is_implicit = (*p == 'I');
1479 
1480 #endif /* !defined (UNPROTOIZE) */
1481 
1482   p++;
1483 
1484   check_aux_info ((*p == 'C') || (*p == 'F'));
1485 
1486   /* Is this item a function definition (F) or a declaration (C).  Note that
1487      we treat item taken from the syscalls file as though they were function
1488      definitions regardless of what the stuff in the file says.  */
1489 
1490   def_dec_p->is_func_def = ((*p++ == 'F') || is_syscalls);
1491 
1492 #ifndef UNPROTOIZE
1493   def_dec_p->definition = 0;	/* Fill this in later if protoizing.  */
1494 #endif /* !defined (UNPROTOIZE) */
1495 
1496   check_aux_info (*p++ == ' ');
1497   check_aux_info (*p++ == '*');
1498   check_aux_info (*p++ == '/');
1499   check_aux_info (*p++ == ' ');
1500 
1501 #ifdef UNPROTOIZE
1502   check_aux_info ((!strncmp (p, "static", 6)) || (!strncmp (p, "extern", 6)));
1503 #else /* !defined (UNPROTOIZE) */
1504   if (!strncmp (p, "static", 6))
1505     def_dec_p->is_static = -1;
1506   else if (!strncmp (p, "extern", 6))
1507     def_dec_p->is_static = 0;
1508   else
1509     check_aux_info (0);	/* Didn't find either `extern' or `static'.  */
1510 #endif /* !defined (UNPROTOIZE) */
1511 
1512   {
1513     const char *ansi_start = p;
1514 
1515     p += 6;	/* Pass over the "static" or "extern".  */
1516 
1517     /* We are now past the initial stuff.  Search forward from here to find
1518        the terminating semicolon that should immediately follow the entire
1519        ANSI format function declaration.  */
1520 
1521     while (*++p != ';')
1522       continue;
1523 
1524     semicolon_p = p;
1525 
1526     /* Make a copy of the ansi declaration part of the line from the aux_info
1527        file.  */
1528 
1529     def_dec_p->ansi_decl
1530       = dupnstr (ansi_start, (size_t) ((semicolon_p+1) - ansi_start));
1531 
1532     /* Backup and point at the final right paren of the final argument list.  */
1533 
1534     p--;
1535 
1536 #ifndef UNPROTOIZE
1537     def_dec_p->f_list_chain = NULL;
1538 #endif /* !defined (UNPROTOIZE) */
1539 
1540     while (p != ansi_start && (p[-1] == ' ' || p[-1] == '\t')) p--;
1541     if (*p != ')')
1542       {
1543 	free_def_dec (def_dec_p);
1544 	return;
1545       }
1546   }
1547 
1548   /* Now isolate a whole set of formal argument lists, one-by-one.  Normally,
1549      there will only be one list to isolate, but there could be more.  */
1550 
1551   def_dec_p->f_list_count = 0;
1552 
1553   for (;;)
1554     {
1555       const char *left_paren_p = find_corresponding_lparen (p);
1556 #ifndef UNPROTOIZE
1557       {
1558 	f_list_chain_item *cip = xmalloc (sizeof (f_list_chain_item));
1559 
1560 	cip->formals_list
1561 	  = dupnstr (left_paren_p + 1, (size_t) (p - (left_paren_p+1)));
1562 
1563 	/* Add the new chain item at the head of the current list.  */
1564 
1565 	cip->chain_next = def_dec_p->f_list_chain;
1566 	def_dec_p->f_list_chain = cip;
1567       }
1568 #endif /* !defined (UNPROTOIZE) */
1569       def_dec_p->f_list_count++;
1570 
1571       p = left_paren_p - 2;
1572 
1573       /* p must now point either to another right paren, or to the last
1574 	 character of the name of the function that was declared/defined.
1575 	 If p points to another right paren, then this indicates that we
1576 	 are dealing with multiple formals lists.  In that case, there
1577 	 really should be another right paren preceding this right paren.  */
1578 
1579       if (*p != ')')
1580 	break;
1581       else
1582 	check_aux_info (*--p == ')');
1583     }
1584 
1585 
1586   {
1587     const char *past_fn = p + 1;
1588 
1589     check_aux_info (*past_fn == ' ');
1590 
1591     /* Scan leftwards over the identifier that names the function.  */
1592 
1593     while (is_id_char (*p))
1594       p--;
1595     p++;
1596 
1597     /* p now points to the leftmost character of the function name.  */
1598 
1599     {
1600       char *fn_string = alloca (past_fn - p + 1);
1601 
1602       strncpy (fn_string, p, (size_t) (past_fn - p));
1603       fn_string[past_fn-p] = '\0';
1604       def_dec_p->hash_entry = lookup (function_name_primary, fn_string);
1605     }
1606   }
1607 
1608   /* Look at all of the defs and decs for this function name that we have
1609      collected so far.  If there is already one which is at the same
1610      line number in the same file, then we can discard this new def_dec_info
1611      record.
1612 
1613      As an extra assurance that any such pair of (nominally) identical
1614      function declarations are in fact identical, we also compare the
1615      ansi_decl parts of the lines from the aux_info files just to be on
1616      the safe side.
1617 
1618      This comparison will fail if (for instance) the user was playing
1619      messy games with the preprocessor which ultimately causes one
1620      function declaration in one header file to look differently when
1621      that file is included by two (or more) other files.  */
1622 
1623   {
1624     const def_dec_info *other;
1625 
1626     for (other = def_dec_p->hash_entry->ddip; other; other = other->next_for_func)
1627       {
1628 	if (def_dec_p->line == other->line && def_dec_p->file == other->file)
1629 	  {
1630 	    if (strcmp (def_dec_p->ansi_decl, other->ansi_decl))
1631 	      {
1632 	        notice ("%s:%d: declaration of function '%s' takes different forms\n",
1633 			def_dec_p->file->hash_entry->symbol,
1634 			def_dec_p->line,
1635 			def_dec_p->hash_entry->symbol);
1636 	        exit (FATAL_EXIT_CODE);
1637 	      }
1638 	    free_def_dec (def_dec_p);
1639 	    return;
1640 	  }
1641       }
1642   }
1643 
1644 #ifdef UNPROTOIZE
1645 
1646   /* If we are doing unprotoizing, we must now setup the pointers that will
1647      point to the K&R name list and to the K&R argument declarations list.
1648 
1649      Note that if this is only a function declaration, then we should not
1650      expect to find any K&R style formals list following the ANSI-style
1651      formals list.  This is because GCC knows that such information is
1652      useless in the case of function declarations (function definitions
1653      are a different story however).
1654 
1655      Since we are unprotoizing, we don't need any such lists anyway.
1656      All we plan to do is to delete all characters between ()'s in any
1657      case.  */
1658 
1659   def_dec_p->formal_names = NULL;
1660   def_dec_p->formal_decls = NULL;
1661 
1662   if (def_dec_p->is_func_def)
1663     {
1664       p = semicolon_p;
1665       check_aux_info (*++p == ' ');
1666       check_aux_info (*++p == '/');
1667       check_aux_info (*++p == '*');
1668       check_aux_info (*++p == ' ');
1669       check_aux_info (*++p == '(');
1670 
1671       {
1672 	const char *kr_names_start = ++p;   /* Point just inside '('.  */
1673 
1674 	while (*p++ != ')')
1675 	  continue;
1676 	p--;		/* point to closing right paren */
1677 
1678 	/* Make a copy of the K&R parameter names list.  */
1679 
1680 	def_dec_p->formal_names
1681 	  = dupnstr (kr_names_start, (size_t) (p - kr_names_start));
1682       }
1683 
1684       check_aux_info (*++p == ' ');
1685       p++;
1686 
1687       /* p now points to the first character of the K&R style declarations
1688 	 list (if there is one) or to the star-slash combination that ends
1689 	 the comment in which such lists get embedded.  */
1690 
1691       /* Make a copy of the K&R formal decls list and set the def_dec record
1692 	 to point to it.  */
1693 
1694       if (*p == '*')		/* Are there no K&R declarations? */
1695 	{
1696 	  check_aux_info (*++p == '/');
1697 	  def_dec_p->formal_decls = "";
1698 	}
1699       else
1700 	{
1701 	  const char *kr_decls_start = p;
1702 
1703 	  while (p[0] != '*' || p[1] != '/')
1704 	    p++;
1705 	  p--;
1706 
1707 	  check_aux_info (*p == ' ');
1708 
1709 	  def_dec_p->formal_decls
1710 	    = dupnstr (kr_decls_start, (size_t) (p - kr_decls_start));
1711 	}
1712 
1713       /* Handle a special case.  If we have a function definition marked as
1714 	 being in "old" style, and if its formal names list is empty, then
1715 	 it may actually have the string "void" in its real formals list
1716 	 in the original source code.  Just to make sure, we will get setup
1717 	 to convert such things anyway.
1718 
1719 	 This kludge only needs to be here because of an insurmountable
1720 	 problem with generating .X files.  */
1721 
1722       if (!def_dec_p->prototyped && !*def_dec_p->formal_names)
1723 	def_dec_p->prototyped = 1;
1724     }
1725 
1726   /* Since we are unprotoizing, if this item is already in old (K&R) style,
1727      we can just ignore it.  If that is true, throw away the itme now.  */
1728 
1729   if (!def_dec_p->prototyped)
1730     {
1731       free_def_dec (def_dec_p);
1732       return;
1733     }
1734 
1735 #endif /* defined (UNPROTOIZE) */
1736 
1737   /* Add this record to the head of the list of records pertaining to this
1738      particular function name.  */
1739 
1740   def_dec_p->next_for_func = def_dec_p->hash_entry->ddip;
1741   def_dec_p->hash_entry->ddip = def_dec_p;
1742 
1743   /* Add this new def_dec_info record to the sorted list of def_dec_info
1744      records for this file.  Note that we don't have to worry about duplicates
1745      (caused by multiple inclusions of header files) here because we have
1746      already eliminated duplicates above.  */
1747 
1748   if (!def_dec_p->file->defs_decs)
1749     {
1750       def_dec_p->file->defs_decs = def_dec_p;
1751       def_dec_p->next_in_file = NULL;
1752     }
1753   else
1754     {
1755       int line = def_dec_p->line;
1756       const def_dec_info *prev = NULL;
1757       const def_dec_info *curr = def_dec_p->file->defs_decs;
1758       const def_dec_info *next = curr->next_in_file;
1759 
1760       while (next && (line < curr->line))
1761 	{
1762 	  prev = curr;
1763 	  curr = next;
1764 	  next = next->next_in_file;
1765 	}
1766       if (line >= curr->line)
1767 	{
1768 	  def_dec_p->next_in_file = curr;
1769 	  if (prev)
1770 	    ((NONCONST def_dec_info *) prev)->next_in_file = def_dec_p;
1771 	  else
1772 	    def_dec_p->file->defs_decs = def_dec_p;
1773 	}
1774       else	/* assert (next == NULL); */
1775 	{
1776 	  ((NONCONST def_dec_info *) curr)->next_in_file = def_dec_p;
1777 	  /* assert (next == NULL); */
1778 	  def_dec_p->next_in_file = next;
1779 	}
1780     }
1781 }
1782 
1783 /* Set up the vector COMPILE_PARAMS which is the argument list for running GCC.
1784    Also set input_file_name_index and aux_info_file_name_index
1785    to the indices of the slots where the file names should go.  */
1786 
1787 /* We initialize the vector by  removing -g, -O, -S, -c, and -o options,
1788    and adding '-aux-info AUXFILE -S  -o /dev/null INFILE' at the end.  */
1789 
1790 static void
munge_compile_params(const char * params_list)1791 munge_compile_params (const char *params_list)
1792 {
1793   /* Build up the contents in a temporary vector
1794      that is so big that to has to be big enough.  */
1795   const char **temp_params
1796     = alloca ((strlen (params_list) + 8) * sizeof (char *));
1797   int param_count = 0;
1798   const char *param;
1799   struct stat st;
1800 
1801   temp_params[param_count++] = compiler_file_name;
1802   for (;;)
1803     {
1804       while (ISSPACE ((const unsigned char)*params_list))
1805 	params_list++;
1806       if (!*params_list)
1807 	break;
1808       param = params_list;
1809       while (*params_list && !ISSPACE ((const unsigned char)*params_list))
1810 	params_list++;
1811       if (param[0] != '-')
1812 	temp_params[param_count++]
1813 	  = dupnstr (param, (size_t) (params_list - param));
1814       else
1815 	{
1816 	  switch (param[1])
1817 	    {
1818 	    case 'g':
1819 	    case 'O':
1820 	    case 'S':
1821 	    case 'c':
1822 	      break;		/* Don't copy these.  */
1823 	    case 'o':
1824 	      while (ISSPACE ((const unsigned char)*params_list))
1825 		params_list++;
1826 	      while (*params_list
1827 		     && !ISSPACE ((const unsigned char)*params_list))
1828 		params_list++;
1829 	      break;
1830 	    default:
1831 	      temp_params[param_count++]
1832 		= dupnstr (param, (size_t) (params_list - param));
1833 	    }
1834 	}
1835       if (!*params_list)
1836 	break;
1837     }
1838   temp_params[param_count++] = "-aux-info";
1839 
1840   /* Leave room for the aux-info file name argument.  */
1841   aux_info_file_name_index = param_count;
1842   temp_params[param_count++] = NULL;
1843 
1844   temp_params[param_count++] = "-S";
1845   temp_params[param_count++] = "-o";
1846 
1847   if ((stat (HOST_BIT_BUCKET, &st) == 0)
1848       && (!S_ISDIR (st.st_mode))
1849       && (access (HOST_BIT_BUCKET, W_OK) == 0))
1850     temp_params[param_count++] = HOST_BIT_BUCKET;
1851   else
1852     /* FIXME: This is hardly likely to be right, if HOST_BIT_BUCKET is not
1853        writable.  But until this is rejigged to use make_temp_file(), this
1854        is the best we can do.  */
1855     temp_params[param_count++] = "/dev/null";
1856 
1857   /* Leave room for the input file name argument.  */
1858   input_file_name_index = param_count;
1859   temp_params[param_count++] = NULL;
1860   /* Terminate the list.  */
1861   temp_params[param_count++] = NULL;
1862 
1863   /* Make a copy of the compile_params in heap space.  */
1864 
1865   compile_params = xmalloc (sizeof (char *) * (param_count+1));
1866   memcpy (compile_params, temp_params, sizeof (char *) * param_count);
1867 }
1868 
1869 /* Do a recompilation for the express purpose of generating a new aux_info
1870    file to go with a specific base source file.
1871 
1872    The result is a boolean indicating success.  */
1873 
1874 static int
gen_aux_info_file(const char * base_filename)1875 gen_aux_info_file (const char *base_filename)
1876 {
1877   if (!input_file_name_index)
1878     munge_compile_params ("");
1879 
1880   /* Store the full source file name in the argument vector.  */
1881   compile_params[input_file_name_index] = shortpath (NULL, base_filename);
1882   /* Add .X to source file name to get aux-info file name.  */
1883   compile_params[aux_info_file_name_index] =
1884     concat (compile_params[input_file_name_index], aux_info_suffix, NULL);
1885 
1886   if (!quiet_flag)
1887     notice ("%s: compiling '%s'\n",
1888 	    pname, compile_params[input_file_name_index]);
1889 
1890   {
1891     char *errmsg_fmt, *errmsg_arg;
1892     int wait_status, pid;
1893 
1894     pid = pexecute (compile_params[0], (char * const *) compile_params,
1895 		    pname, NULL, &errmsg_fmt, &errmsg_arg,
1896 		    PEXECUTE_FIRST | PEXECUTE_LAST | PEXECUTE_SEARCH);
1897 
1898     if (pid == -1)
1899       {
1900 	int errno_val = errno;
1901 	fprintf (stderr, "%s: ", pname);
1902 	fprintf (stderr, errmsg_fmt, errmsg_arg);
1903 	fprintf (stderr, ": %s\n", xstrerror (errno_val));
1904 	return 0;
1905       }
1906 
1907     pid = pwait (pid, &wait_status, 0);
1908     if (pid == -1)
1909       {
1910 	notice ("%s: wait: %s\n", pname, xstrerror (errno));
1911 	return 0;
1912       }
1913     if (WIFSIGNALED (wait_status))
1914       {
1915 	notice ("%s: subprocess got fatal signal %d\n",
1916 		pname, WTERMSIG (wait_status));
1917 	return 0;
1918       }
1919     if (WIFEXITED (wait_status))
1920       {
1921 	if (WEXITSTATUS (wait_status) != 0)
1922 	  {
1923 	    notice ("%s: %s exited with status %d\n",
1924 		    pname, compile_params[0], WEXITSTATUS (wait_status));
1925 	    return 0;
1926 	  }
1927 	return 1;
1928       }
1929     gcc_unreachable ();
1930   }
1931 }
1932 
1933 /* Read in all of the information contained in a single aux_info file.
1934    Save all of the important stuff for later.  */
1935 
1936 static void
process_aux_info_file(const char * base_source_filename,int keep_it,int is_syscalls)1937 process_aux_info_file (const char *base_source_filename, int keep_it,
1938 		       int is_syscalls)
1939 {
1940   size_t base_len = strlen (base_source_filename);
1941   char * aux_info_filename = alloca (base_len + strlen (aux_info_suffix) + 1);
1942   char *aux_info_base;
1943   char *aux_info_limit;
1944   char *aux_info_relocated_name;
1945   const char *aux_info_second_line;
1946   time_t aux_info_mtime;
1947   size_t aux_info_size;
1948   int must_create;
1949 
1950   /* Construct the aux_info filename from the base source filename.  */
1951 
1952   strcpy (aux_info_filename, base_source_filename);
1953   strcat (aux_info_filename, aux_info_suffix);
1954 
1955   /* Check that the aux_info file exists and is readable.  If it does not
1956      exist, try to create it (once only).  */
1957 
1958   /* If file doesn't exist, set must_create.
1959      Likewise if it exists and we can read it but it is obsolete.
1960      Otherwise, report an error.  */
1961   must_create = 0;
1962 
1963   /* Come here with must_create set to 1 if file is out of date.  */
1964 start_over: ;
1965 
1966   if (access (aux_info_filename, R_OK) == -1)
1967     {
1968       if (errno == ENOENT)
1969 	{
1970 	  if (is_syscalls)
1971 	    {
1972 	      notice ("%s: warning: missing SYSCALLS file '%s'\n",
1973 		      pname, aux_info_filename);
1974 	      return;
1975 	    }
1976 	  must_create = 1;
1977 	}
1978       else
1979 	{
1980 	  int errno_val = errno;
1981 	  notice ("%s: can't read aux info file '%s': %s\n",
1982 		  pname, shortpath (NULL, aux_info_filename),
1983 		  xstrerror (errno_val));
1984 	  errors++;
1985 	  return;
1986 	}
1987     }
1988 #if 0 /* There is code farther down to take care of this.  */
1989   else
1990     {
1991       struct stat s1, s2;
1992       stat (aux_info_file_name, &s1);
1993       stat (base_source_file_name, &s2);
1994       if (s2.st_mtime > s1.st_mtime)
1995 	must_create = 1;
1996     }
1997 #endif /* 0 */
1998 
1999   /* If we need a .X file, create it, and verify we can read it.  */
2000   if (must_create)
2001     {
2002       if (!gen_aux_info_file (base_source_filename))
2003 	{
2004 	  errors++;
2005 	  return;
2006 	}
2007       if (access (aux_info_filename, R_OK) == -1)
2008 	{
2009 	  int errno_val = errno;
2010 	  notice ("%s: can't read aux info file '%s': %s\n",
2011 		  pname, shortpath (NULL, aux_info_filename),
2012 		  xstrerror (errno_val));
2013 	  errors++;
2014 	  return;
2015 	}
2016     }
2017 
2018   {
2019     struct stat stat_buf;
2020 
2021     /* Get some status information about this aux_info file.  */
2022 
2023     if (stat (aux_info_filename, &stat_buf) == -1)
2024       {
2025 	int errno_val = errno;
2026 	notice ("%s: can't get status of aux info file '%s': %s\n",
2027 		pname, shortpath (NULL, aux_info_filename),
2028 		xstrerror (errno_val));
2029 	errors++;
2030 	return;
2031       }
2032 
2033     /* Check on whether or not this aux_info file is zero length.  If it is,
2034        then just ignore it and return.  */
2035 
2036     if ((aux_info_size = stat_buf.st_size) == 0)
2037       return;
2038 
2039     /* Get the date/time of last modification for this aux_info file and
2040        remember it.  We will have to check that any source files that it
2041        contains information about are at least this old or older.  */
2042 
2043     aux_info_mtime = stat_buf.st_mtime;
2044 
2045     if (!is_syscalls)
2046       {
2047 	/* Compare mod time with the .c file; update .X file if obsolete.
2048 	   The code later on can fail to check the .c file
2049 	   if it did not directly define any functions.  */
2050 
2051 	if (stat (base_source_filename, &stat_buf) == -1)
2052 	  {
2053 	    int errno_val = errno;
2054 	    notice ("%s: can't get status of aux info file '%s': %s\n",
2055 		    pname, shortpath (NULL, base_source_filename),
2056 		    xstrerror (errno_val));
2057 	    errors++;
2058 	    return;
2059 	  }
2060 	if (stat_buf.st_mtime > aux_info_mtime)
2061 	  {
2062 	    must_create = 1;
2063 	    goto start_over;
2064 	  }
2065       }
2066   }
2067 
2068   {
2069     int aux_info_file;
2070     int fd_flags;
2071 
2072     /* Open the aux_info file.  */
2073 
2074     fd_flags = O_RDONLY;
2075 #ifdef O_BINARY
2076     /* Use binary mode to avoid having to deal with different EOL characters.  */
2077     fd_flags |= O_BINARY;
2078 #endif
2079     if ((aux_info_file = open (aux_info_filename, fd_flags, 0444 )) == -1)
2080       {
2081 	int errno_val = errno;
2082 	notice ("%s: can't open aux info file '%s' for reading: %s\n",
2083 		pname, shortpath (NULL, aux_info_filename),
2084 		xstrerror (errno_val));
2085 	return;
2086       }
2087 
2088     /* Allocate space to hold the aux_info file in memory.  */
2089 
2090     aux_info_base = xmalloc (aux_info_size + 1);
2091     aux_info_limit = aux_info_base + aux_info_size;
2092     *aux_info_limit = '\0';
2093 
2094     /* Read the aux_info file into memory.  */
2095 
2096     if (safe_read (aux_info_file, aux_info_base, aux_info_size) !=
2097 	(int) aux_info_size)
2098       {
2099 	int errno_val = errno;
2100 	notice ("%s: error reading aux info file '%s': %s\n",
2101 		pname, shortpath (NULL, aux_info_filename),
2102 		xstrerror (errno_val));
2103 	free (aux_info_base);
2104 	close (aux_info_file);
2105 	return;
2106       }
2107 
2108     /* Close the aux info file.  */
2109 
2110     if (close (aux_info_file))
2111       {
2112 	int errno_val = errno;
2113 	notice ("%s: error closing aux info file '%s': %s\n",
2114 		pname, shortpath (NULL, aux_info_filename),
2115 		xstrerror (errno_val));
2116 	free (aux_info_base);
2117 	close (aux_info_file);
2118 	return;
2119       }
2120   }
2121 
2122   /* Delete the aux_info file (unless requested not to).  If the deletion
2123      fails for some reason, don't even worry about it.  */
2124 
2125   if (must_create && !keep_it)
2126     if (unlink (aux_info_filename) == -1)
2127       {
2128 	int errno_val = errno;
2129 	notice ("%s: can't delete aux info file '%s': %s\n",
2130 		pname, shortpath (NULL, aux_info_filename),
2131 		xstrerror (errno_val));
2132       }
2133 
2134   /* Save a pointer into the first line of the aux_info file which
2135      contains the filename of the directory from which the compiler
2136      was invoked when the associated source file was compiled.
2137      This information is used later to help create complete
2138      filenames out of the (potentially) relative filenames in
2139      the aux_info file.  */
2140 
2141   {
2142     char *p = aux_info_base;
2143 
2144     while (*p != ':'
2145 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
2146 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
2147 #endif
2148 	   )
2149       p++;
2150     p++;
2151     while (*p == ' ')
2152       p++;
2153     invocation_filename = p;	/* Save a pointer to first byte of path.  */
2154     while (*p != ' ')
2155       p++;
2156     *p++ = DIR_SEPARATOR;
2157     *p++ = '\0';
2158     while (*p++ != '\n')
2159       continue;
2160     aux_info_second_line = p;
2161     aux_info_relocated_name = 0;
2162     if (! IS_ABSOLUTE_PATH (invocation_filename))
2163       {
2164 	/* INVOCATION_FILENAME is relative;
2165 	   append it to BASE_SOURCE_FILENAME's dir.  */
2166 	char *dir_end;
2167 	aux_info_relocated_name = xmalloc (base_len + (p-invocation_filename));
2168 	strcpy (aux_info_relocated_name, base_source_filename);
2169 	dir_end = strrchr (aux_info_relocated_name, DIR_SEPARATOR);
2170 #ifdef DIR_SEPARATOR_2
2171 	{
2172 	  char *slash;
2173 
2174 	  slash = strrchr (dir_end ? dir_end : aux_info_relocated_name,
2175 			   DIR_SEPARATOR_2);
2176 	  if (slash)
2177 	    dir_end = slash;
2178 	}
2179 #endif
2180 	if (dir_end)
2181 	  dir_end++;
2182 	else
2183 	  dir_end = aux_info_relocated_name;
2184 	strcpy (dir_end, invocation_filename);
2185 	invocation_filename = aux_info_relocated_name;
2186       }
2187   }
2188 
2189 
2190   {
2191     const char *aux_info_p;
2192 
2193     /* Do a pre-pass on the lines in the aux_info file, making sure that all
2194        of the source files referenced in there are at least as old as this
2195        aux_info file itself.  If not, go back and regenerate the aux_info
2196        file anew.  Don't do any of this for the syscalls file.  */
2197 
2198     if (!is_syscalls)
2199       {
2200 	current_aux_info_lineno = 2;
2201 
2202 	for (aux_info_p = aux_info_second_line; *aux_info_p; )
2203 	  {
2204 	    if (referenced_file_is_newer (aux_info_p, aux_info_mtime))
2205 	      {
2206 		free (aux_info_base);
2207 		free (aux_info_relocated_name);
2208 		if (keep_it && unlink (aux_info_filename) == -1)
2209 		  {
2210 		    int errno_val = errno;
2211 	            notice ("%s: can't delete file '%s': %s\n",
2212 			    pname, shortpath (NULL, aux_info_filename),
2213 			    xstrerror (errno_val));
2214 	            return;
2215 	          }
2216 		must_create = 1;
2217 	        goto start_over;
2218 	      }
2219 
2220 	    /* Skip over the rest of this line to start of next line.  */
2221 
2222 	    while (*aux_info_p != '\n')
2223 	      aux_info_p++;
2224 	    aux_info_p++;
2225 	    current_aux_info_lineno++;
2226 	  }
2227       }
2228 
2229     /* Now do the real pass on the aux_info lines.  Save their information in
2230        the in-core data base.  */
2231 
2232     current_aux_info_lineno = 2;
2233 
2234     for (aux_info_p = aux_info_second_line; *aux_info_p;)
2235       {
2236 	char *unexpanded_line = unexpand_if_needed (aux_info_p);
2237 
2238 	if (unexpanded_line)
2239 	  {
2240 	    save_def_or_dec (unexpanded_line, is_syscalls);
2241 	    free (unexpanded_line);
2242 	  }
2243 	else
2244 	  save_def_or_dec (aux_info_p, is_syscalls);
2245 
2246 	/* Skip over the rest of this line and get to start of next line.  */
2247 
2248 	while (*aux_info_p != '\n')
2249 	  aux_info_p++;
2250 	aux_info_p++;
2251 	current_aux_info_lineno++;
2252       }
2253   }
2254 
2255   free (aux_info_base);
2256   free (aux_info_relocated_name);
2257 }
2258 
2259 #ifndef UNPROTOIZE
2260 
2261 /* Check an individual filename for a .c suffix.  If the filename has this
2262    suffix, rename the file such that its suffix is changed to .C.  This
2263    function implements the -C option.  */
2264 
2265 static void
rename_c_file(const hash_table_entry * hp)2266 rename_c_file (const hash_table_entry *hp)
2267 {
2268   const char *filename = hp->symbol;
2269   int last_char_index = strlen (filename) - 1;
2270   char *const new_filename = alloca (strlen (filename)
2271 				     + strlen (cplus_suffix) + 1);
2272 
2273   /* Note that we don't care here if the given file was converted or not.  It
2274      is possible that the given file was *not* converted, simply because there
2275      was nothing in it which actually required conversion.  Even in this case,
2276      we want to do the renaming.  Note that we only rename files with the .c
2277      suffix (except for the syscalls file, which is left alone).  */
2278 
2279   if (filename[last_char_index] != 'c' || filename[last_char_index-1] != '.'
2280       || IS_SAME_PATH (syscalls_absolute_filename, filename))
2281     return;
2282 
2283   strcpy (new_filename, filename);
2284   strcpy (&new_filename[last_char_index], cplus_suffix);
2285 
2286   if (rename (filename, new_filename) == -1)
2287     {
2288       int errno_val = errno;
2289       notice ("%s: warning: can't rename file '%s' to '%s': %s\n",
2290 	      pname, shortpath (NULL, filename),
2291 	      shortpath (NULL, new_filename), xstrerror (errno_val));
2292       errors++;
2293       return;
2294     }
2295 }
2296 
2297 #endif /* !defined (UNPROTOIZE) */
2298 
2299 /* Take the list of definitions and declarations attached to a particular
2300    file_info node and reverse the order of the list.  This should get the
2301    list into an order such that the item with the lowest associated line
2302    number is nearest the head of the list.  When these lists are originally
2303    built, they are in the opposite order.  We want to traverse them in
2304    normal line number order later (i.e. lowest to highest) so reverse the
2305    order here.  */
2306 
2307 static void
reverse_def_dec_list(const hash_table_entry * hp)2308 reverse_def_dec_list (const hash_table_entry *hp)
2309 {
2310   file_info *file_p = hp->fip;
2311   def_dec_info *prev = NULL;
2312   def_dec_info *current = (def_dec_info *) file_p->defs_decs;
2313 
2314   if (!current)
2315     return;        		/* no list to reverse */
2316 
2317   prev = current;
2318   if (! (current = (def_dec_info *) current->next_in_file))
2319     return;        		/* can't reverse a single list element */
2320 
2321   prev->next_in_file = NULL;
2322 
2323   while (current)
2324     {
2325       def_dec_info *next = (def_dec_info *) current->next_in_file;
2326 
2327       current->next_in_file = prev;
2328       prev = current;
2329       current = next;
2330     }
2331 
2332   file_p->defs_decs = prev;
2333 }
2334 
2335 #ifndef UNPROTOIZE
2336 
2337 /* Find the (only?) extern definition for a particular function name, starting
2338    from the head of the linked list of entries for the given name.  If we
2339    cannot find an extern definition for the given function name, issue a
2340    warning and scrounge around for the next best thing, i.e. an extern
2341    function declaration with a prototype attached to it.  Note that we only
2342    allow such substitutions for extern declarations and never for static
2343    declarations.  That's because the only reason we allow them at all is
2344    to let un-prototyped function declarations for system-supplied library
2345    functions get their prototypes from our own extra SYSCALLS.c.X file which
2346    contains all of the correct prototypes for system functions.  */
2347 
2348 static const def_dec_info *
find_extern_def(const def_dec_info * head,const def_dec_info * user)2349 find_extern_def (const def_dec_info *head, const def_dec_info *user)
2350 {
2351   const def_dec_info *dd_p;
2352   const def_dec_info *extern_def_p = NULL;
2353   int conflict_noted = 0;
2354 
2355   /* Don't act too stupid here.  Somebody may try to convert an entire system
2356      in one swell fwoop (rather than one program at a time, as should be done)
2357      and in that case, we may find that there are multiple extern definitions
2358      of a given function name in the entire set of source files that we are
2359      converting.  If however one of these definitions resides in exactly the
2360      same source file as the reference we are trying to satisfy then in that
2361      case it would be stupid for us to fail to realize that this one definition
2362      *must* be the precise one we are looking for.
2363 
2364      To make sure that we don't miss an opportunity to make this "same file"
2365      leap of faith, we do a prescan of the list of records relating to the
2366      given function name, and we look (on this first scan) *only* for a
2367      definition of the function which is in the same file as the reference
2368      we are currently trying to satisfy.  */
2369 
2370   for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2371     if (dd_p->is_func_def && !dd_p->is_static && dd_p->file == user->file)
2372       return dd_p;
2373 
2374   /* Now, since we have not found a definition in the same file as the
2375      reference, we scan the list again and consider all possibilities from
2376      all files.  Here we may get conflicts with the things listed in the
2377      SYSCALLS.c.X file, but if that happens it only means that the source
2378      code being converted contains its own definition of a function which
2379      could have been supplied by libc.a.  In such cases, we should avoid
2380      issuing the normal warning, and defer to the definition given in the
2381      user's own code.  */
2382 
2383   for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2384     if (dd_p->is_func_def && !dd_p->is_static)
2385       {
2386 	if (!extern_def_p)	/* Previous definition? */
2387 	  extern_def_p = dd_p;	/* Remember the first definition found.  */
2388 	else
2389 	  {
2390 	    /* Ignore definition just found if it came from SYSCALLS.c.X.  */
2391 
2392 	    if (is_syscalls_file (dd_p->file))
2393 	      continue;
2394 
2395 	    /* Quietly replace the definition previously found with the one
2396 	       just found if the previous one was from SYSCALLS.c.X.  */
2397 
2398 	    if (is_syscalls_file (extern_def_p->file))
2399 	      {
2400 	        extern_def_p = dd_p;
2401 	        continue;
2402 	      }
2403 
2404 	    /* If we get here, then there is a conflict between two function
2405 	       declarations for the same function, both of which came from the
2406 	       user's own code.  */
2407 
2408 	    if (!conflict_noted)	/* first time we noticed? */
2409 	      {
2410 		conflict_noted = 1;
2411 		notice ("%s: conflicting extern definitions of '%s'\n",
2412 			pname, head->hash_entry->symbol);
2413 		if (!quiet_flag)
2414 		  {
2415 		    notice ("%s: declarations of '%s' will not be converted\n",
2416 			    pname, head->hash_entry->symbol);
2417 		    notice ("%s: conflict list for '%s' follows:\n",
2418 			    pname, head->hash_entry->symbol);
2419 		    fprintf (stderr, "%s:     %s(%d): %s\n",
2420 			     pname,
2421 			     shortpath (NULL, extern_def_p->file->hash_entry->symbol),
2422 			     extern_def_p->line, extern_def_p->ansi_decl);
2423 		  }
2424 	      }
2425 	    if (!quiet_flag)
2426 	      fprintf (stderr, "%s:     %s(%d): %s\n",
2427 		       pname,
2428 		       shortpath (NULL, dd_p->file->hash_entry->symbol),
2429 		       dd_p->line, dd_p->ansi_decl);
2430 	  }
2431       }
2432 
2433   /* We want to err on the side of caution, so if we found multiple conflicting
2434      definitions for the same function, treat this as being that same as if we
2435      had found no definitions (i.e. return NULL).  */
2436 
2437   if (conflict_noted)
2438     return NULL;
2439 
2440   if (!extern_def_p)
2441     {
2442       /* We have no definitions for this function so do the next best thing.
2443 	 Search for an extern declaration already in prototype form.  */
2444 
2445       for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2446 	if (!dd_p->is_func_def && !dd_p->is_static && dd_p->prototyped)
2447 	  {
2448 	    extern_def_p = dd_p;	/* save a pointer to the definition */
2449 	    if (!quiet_flag)
2450 	      notice ("%s: warning: using formals list from %s(%d) for function '%s'\n",
2451 		      pname,
2452 		      shortpath (NULL, dd_p->file->hash_entry->symbol),
2453 		      dd_p->line, dd_p->hash_entry->symbol);
2454 	    break;
2455 	  }
2456 
2457       /* Gripe about unprototyped function declarations that we found no
2458 	 corresponding definition (or other source of prototype information)
2459 	 for.
2460 
2461 	 Gripe even if the unprototyped declaration we are worried about
2462 	 exists in a file in one of the "system" include directories.  We
2463 	 can gripe about these because we should have at least found a
2464 	 corresponding (pseudo) definition in the SYSCALLS.c.X file.  If we
2465 	 didn't, then that means that the SYSCALLS.c.X file is missing some
2466 	 needed prototypes for this particular system.  That is worth telling
2467 	 the user about!  */
2468 
2469       if (!extern_def_p)
2470 	{
2471 	  const char *file = user->file->hash_entry->symbol;
2472 
2473 	  if (!quiet_flag)
2474 	    if (in_system_include_dir (file))
2475 	      {
2476 		/* Why copy this string into `needed' at all?
2477 		   Why not just use user->ansi_decl without copying?  */
2478 		char *needed = alloca (strlen (user->ansi_decl) + 1);
2479 	        char *p;
2480 
2481 	        strcpy (needed, user->ansi_decl);
2482 	        p = strstr (needed, user->hash_entry->symbol)
2483 	            + strlen (user->hash_entry->symbol) + 2;
2484 		/* Avoid having ??? in the string.  */
2485 		*p++ = '?';
2486 		*p++ = '?';
2487 		*p++ = '?';
2488 	        strcpy (p, ");");
2489 
2490 	        notice ("%s: %d: '%s' used but missing from SYSCALLS\n",
2491 			shortpath (NULL, file), user->line,
2492 			needed+7);	/* Don't print "extern " */
2493 	      }
2494 #if 0
2495 	    else
2496 	      notice ("%s: %d: warning: no extern definition for '%s'\n",
2497 		      shortpath (NULL, file), user->line,
2498 		      user->hash_entry->symbol);
2499 #endif
2500 	}
2501     }
2502   return extern_def_p;
2503 }
2504 
2505 /* Find the (only?) static definition for a particular function name in a
2506    given file.  Here we get the function-name and the file info indirectly
2507    from the def_dec_info record pointer which is passed in.  */
2508 
2509 static const def_dec_info *
find_static_definition(const def_dec_info * user)2510 find_static_definition (const def_dec_info *user)
2511 {
2512   const def_dec_info *head = user->hash_entry->ddip;
2513   const def_dec_info *dd_p;
2514   int num_static_defs = 0;
2515   const def_dec_info *static_def_p = NULL;
2516 
2517   for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2518     if (dd_p->is_func_def && dd_p->is_static && (dd_p->file == user->file))
2519       {
2520 	static_def_p = dd_p;	/* save a pointer to the definition */
2521 	num_static_defs++;
2522       }
2523   if (num_static_defs == 0)
2524     {
2525       if (!quiet_flag)
2526 	notice ("%s: warning: no static definition for '%s' in file '%s'\n",
2527 		pname, head->hash_entry->symbol,
2528 		shortpath (NULL, user->file->hash_entry->symbol));
2529     }
2530   else if (num_static_defs > 1)
2531     {
2532       notice ("%s: multiple static defs of '%s' in file '%s'\n",
2533 	      pname, head->hash_entry->symbol,
2534 	      shortpath (NULL, user->file->hash_entry->symbol));
2535       return NULL;
2536     }
2537   return static_def_p;
2538 }
2539 
2540 /* Find good prototype style formal argument lists for all of the function
2541    declarations which didn't have them before now.
2542 
2543    To do this we consider each function name one at a time.  For each function
2544    name, we look at the items on the linked list of def_dec_info records for
2545    that particular name.
2546 
2547    Somewhere on this list we should find one (and only one) def_dec_info
2548    record which represents the actual function definition, and this record
2549    should have a nice formal argument list already associated with it.
2550 
2551    Thus, all we have to do is to connect up all of the other def_dec_info
2552    records for this particular function name to the special one which has
2553    the full-blown formals list.
2554 
2555    Of course it is a little more complicated than just that.  See below for
2556    more details.  */
2557 
2558 static void
connect_defs_and_decs(const hash_table_entry * hp)2559 connect_defs_and_decs (const hash_table_entry *hp)
2560 {
2561   const def_dec_info *dd_p;
2562   const def_dec_info *extern_def_p = NULL;
2563   int first_extern_reference = 1;
2564 
2565   /* Traverse the list of definitions and declarations for this particular
2566      function name.  For each item on the list, if it is a function
2567      definition (either old style or new style) then GCC has already been
2568      kind enough to produce a prototype for us, and it is associated with
2569      the item already, so declare the item as its own associated "definition".
2570 
2571      Also, for each item which is only a function declaration, but which
2572      nonetheless has its own prototype already (obviously supplied by the user)
2573      declare the item as its own definition.
2574 
2575      Note that when/if there are multiple user-supplied prototypes already
2576      present for multiple declarations of any given function, these multiple
2577      prototypes *should* all match exactly with one another and with the
2578      prototype for the actual function definition.  We don't check for this
2579      here however, since we assume that the compiler must have already done
2580      this consistency checking when it was creating the .X files.  */
2581 
2582   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2583     if (dd_p->prototyped)
2584       ((NONCONST def_dec_info *) dd_p)->definition = dd_p;
2585 
2586   /* Traverse the list of definitions and declarations for this particular
2587      function name.  For each item on the list, if it is an extern function
2588      declaration and if it has no associated definition yet, go try to find
2589      the matching extern definition for the declaration.
2590 
2591      When looking for the matching function definition, warn the user if we
2592      fail to find one.
2593 
2594      If we find more that one function definition also issue a warning.
2595 
2596      Do the search for the matching definition only once per unique function
2597      name (and only when absolutely needed) so that we can avoid putting out
2598      redundant warning messages, and so that we will only put out warning
2599      messages when there is actually a reference (i.e. a declaration) for
2600      which we need to find a matching definition.  */
2601 
2602   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2603     if (!dd_p->is_func_def && !dd_p->is_static && !dd_p->definition)
2604       {
2605 	if (first_extern_reference)
2606 	  {
2607 	    extern_def_p = find_extern_def (hp->ddip, dd_p);
2608 	    first_extern_reference = 0;
2609 	  }
2610 	((NONCONST def_dec_info *) dd_p)->definition = extern_def_p;
2611       }
2612 
2613   /* Traverse the list of definitions and declarations for this particular
2614      function name.  For each item on the list, if it is a static function
2615      declaration and if it has no associated definition yet, go try to find
2616      the matching static definition for the declaration within the same file.
2617 
2618      When looking for the matching function definition, warn the user if we
2619      fail to find one in the same file with the declaration, and refuse to
2620      convert this kind of cross-file static function declaration.  After all,
2621      this is stupid practice and should be discouraged.
2622 
2623      We don't have to worry about the possibility that there is more than one
2624      matching function definition in the given file because that would have
2625      been flagged as an error by the compiler.
2626 
2627      Do the search for the matching definition only once per unique
2628      function-name/source-file pair (and only when absolutely needed) so that
2629      we can avoid putting out redundant warning messages, and so that we will
2630      only put out warning messages when there is actually a reference (i.e. a
2631      declaration) for which we actually need to find a matching definition.  */
2632 
2633   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2634     if (!dd_p->is_func_def && dd_p->is_static && !dd_p->definition)
2635       {
2636 	const def_dec_info *dd_p2;
2637 	const def_dec_info *static_def;
2638 
2639 	/* We have now found a single static declaration for which we need to
2640 	   find a matching definition.  We want to minimize the work (and the
2641 	   number of warnings), so we will find an appropriate (matching)
2642 	   static definition for this declaration, and then distribute it
2643 	   (as the definition for) any and all other static declarations
2644 	   for this function name which occur within the same file, and which
2645 	   do not already have definitions.
2646 
2647 	   Note that a trick is used here to prevent subsequent attempts to
2648 	   call find_static_definition for a given function-name & file
2649 	   if the first such call returns NULL.  Essentially, we convert
2650 	   these NULL return values to -1, and put the -1 into the definition
2651 	   field for each other static declaration from the same file which
2652 	   does not already have an associated definition.
2653 	   This makes these other static declarations look like they are
2654 	   actually defined already when the outer loop here revisits them
2655 	   later on.  Thus, the outer loop will skip over them.  Later, we
2656 	   turn the -1's back to NULL's.  */
2657 
2658 	((NONCONST def_dec_info *) dd_p)->definition =
2659 	  (static_def = find_static_definition (dd_p))
2660 	  ? static_def
2661 	  : (const def_dec_info *) -1;
2662 
2663 	for (dd_p2 = dd_p->next_for_func; dd_p2; dd_p2 = dd_p2->next_for_func)
2664 	  if (!dd_p2->is_func_def && dd_p2->is_static
2665 	      && !dd_p2->definition && (dd_p2->file == dd_p->file))
2666 	    ((NONCONST def_dec_info *) dd_p2)->definition = dd_p->definition;
2667       }
2668 
2669   /* Convert any dummy (-1) definitions we created in the step above back to
2670      NULL's (as they should be).  */
2671 
2672   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2673     if (dd_p->definition == (def_dec_info *) -1)
2674       ((NONCONST def_dec_info *) dd_p)->definition = NULL;
2675 }
2676 
2677 #endif /* !defined (UNPROTOIZE) */
2678 
2679 /* Give a pointer into the clean text buffer, return a number which is the
2680    original source line number that the given pointer points into.  */
2681 
2682 static int
identify_lineno(const char * clean_p)2683 identify_lineno (const char *clean_p)
2684 {
2685   int line_num = 1;
2686   const char *scan_p;
2687 
2688   for (scan_p = clean_text_base; scan_p <= clean_p; scan_p++)
2689     if (*scan_p == '\n')
2690       line_num++;
2691   return line_num;
2692 }
2693 
2694 /* Issue an error message and give up on doing this particular edit.  */
2695 
2696 static void
declare_source_confusing(const char * clean_p)2697 declare_source_confusing (const char *clean_p)
2698 {
2699   if (!quiet_flag)
2700     {
2701       if (clean_p == 0)
2702 	notice ("%s: %d: warning: source too confusing\n",
2703 		shortpath (NULL, convert_filename), last_known_line_number);
2704       else
2705 	notice ("%s: %d: warning: source too confusing\n",
2706 		shortpath (NULL, convert_filename),
2707 		identify_lineno (clean_p));
2708     }
2709   longjmp (source_confusion_recovery, 1);
2710 }
2711 
2712 /* Check that a condition which is expected to be true in the original source
2713    code is in fact true.  If not, issue an error message and give up on
2714    converting this particular source file.  */
2715 
2716 static void
check_source(int cond,const char * clean_p)2717 check_source (int cond, const char *clean_p)
2718 {
2719   if (!cond)
2720     declare_source_confusing (clean_p);
2721 }
2722 
2723 /* If we think of the in-core cleaned text buffer as a memory mapped
2724    file (with the variable last_known_line_start acting as sort of a
2725    file pointer) then we can imagine doing "seeks" on the buffer.  The
2726    following routine implements a kind of "seek" operation for the in-core
2727    (cleaned) copy of the source file.  When finished, it returns a pointer to
2728    the start of a given (numbered) line in the cleaned text buffer.
2729 
2730    Note that protoize only has to "seek" in the forward direction on the
2731    in-core cleaned text file buffers, and it never needs to back up.
2732 
2733    This routine is made a little bit faster by remembering the line number
2734    (and pointer value) supplied (and returned) from the previous "seek".
2735    This prevents us from always having to start all over back at the top
2736    of the in-core cleaned buffer again.  */
2737 
2738 static const char *
seek_to_line(int n)2739 seek_to_line (int n)
2740 {
2741   gcc_assert (n >= last_known_line_number);
2742 
2743   while (n > last_known_line_number)
2744     {
2745       while (*last_known_line_start != '\n')
2746 	check_source (++last_known_line_start < clean_text_limit, 0);
2747       last_known_line_start++;
2748       last_known_line_number++;
2749     }
2750   return last_known_line_start;
2751 }
2752 
2753 /* Given a pointer to a character in the cleaned text buffer, return a pointer
2754    to the next non-whitespace character which follows it.  */
2755 
2756 static const char *
forward_to_next_token_char(const char * ptr)2757 forward_to_next_token_char (const char *ptr)
2758 {
2759   for (++ptr; ISSPACE ((const unsigned char)*ptr);
2760        check_source (++ptr < clean_text_limit, 0))
2761     continue;
2762   return ptr;
2763 }
2764 
2765 /* Copy a chunk of text of length `len' and starting at `str' to the current
2766    output buffer.  Note that all attempts to add stuff to the current output
2767    buffer ultimately go through here.  */
2768 
2769 static void
output_bytes(const char * str,size_t len)2770 output_bytes (const char *str, size_t len)
2771 {
2772   if ((repl_write_ptr + 1) + len >= repl_text_limit)
2773     {
2774       size_t new_size = (repl_text_limit - repl_text_base) << 1;
2775       char *new_buf = xrealloc (repl_text_base, new_size);
2776 
2777       repl_write_ptr = new_buf + (repl_write_ptr - repl_text_base);
2778       repl_text_base = new_buf;
2779       repl_text_limit = new_buf + new_size;
2780     }
2781   memcpy (repl_write_ptr + 1, str, len);
2782   repl_write_ptr += len;
2783 }
2784 
2785 /* Copy all bytes (except the trailing null) of a null terminated string to
2786    the current output buffer.  */
2787 
2788 static void
output_string(const char * str)2789 output_string (const char *str)
2790 {
2791   output_bytes (str, strlen (str));
2792 }
2793 
2794 /* Copy some characters from the original text buffer to the current output
2795    buffer.
2796 
2797    This routine takes a pointer argument `p' which is assumed to be a pointer
2798    into the cleaned text buffer.  The bytes which are copied are the `original'
2799    equivalents for the set of bytes between the last value of `clean_read_ptr'
2800    and the argument value `p'.
2801 
2802    The set of bytes copied however, comes *not* from the cleaned text buffer,
2803    but rather from the direct counterparts of these bytes within the original
2804    text buffer.
2805 
2806    Thus, when this function is called, some bytes from the original text
2807    buffer (which may include original comments and preprocessing directives)
2808    will be copied into the  output buffer.
2809 
2810    Note that the request implied when this routine is called includes the
2811    byte pointed to by the argument pointer `p'.  */
2812 
2813 static void
output_up_to(const char * p)2814 output_up_to (const char *p)
2815 {
2816   size_t copy_length = (size_t) (p - clean_read_ptr);
2817   const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
2818 
2819   if (copy_length == 0)
2820     return;
2821 
2822   output_bytes (copy_start, copy_length);
2823   clean_read_ptr = p;
2824 }
2825 
2826 /* Given a pointer to a def_dec_info record which represents some form of
2827    definition of a function (perhaps a real definition, or in lieu of that
2828    perhaps just a declaration with a full prototype) return true if this
2829    function is one which we should avoid converting.  Return false
2830    otherwise.  */
2831 
2832 static int
other_variable_style_function(const char * ansi_header)2833 other_variable_style_function (const char *ansi_header)
2834 {
2835 #ifdef UNPROTOIZE
2836 
2837   /* See if we have a stdarg function, or a function which has stdarg style
2838      parameters or a stdarg style return type.  */
2839 
2840   return strstr (ansi_header, "...") != 0;
2841 
2842 #else /* !defined (UNPROTOIZE) */
2843 
2844   /* See if we have a varargs function, or a function which has varargs style
2845      parameters or a varargs style return type.  */
2846 
2847   const char *p;
2848   int len = strlen (varargs_style_indicator);
2849 
2850   for (p = ansi_header; p; )
2851     {
2852       const char *candidate;
2853 
2854       if ((candidate = strstr (p, varargs_style_indicator)) == 0)
2855 	return 0;
2856       else
2857 	if (!is_id_char (candidate[-1]) && !is_id_char (candidate[len]))
2858 	  return 1;
2859 	else
2860 	  p = candidate + 1;
2861     }
2862   return 0;
2863 #endif /* !defined (UNPROTOIZE) */
2864 }
2865 
2866 /* Do the editing operation specifically for a function "declaration".  Note
2867    that editing for function "definitions" are handled in a separate routine
2868    below.  */
2869 
2870 static void
edit_fn_declaration(const def_dec_info * def_dec_p,const char * volatile clean_text_p)2871 edit_fn_declaration (const def_dec_info *def_dec_p,
2872 		     const char *volatile clean_text_p)
2873 {
2874   const char *start_formals;
2875   const char *end_formals;
2876   const char *function_to_edit = def_dec_p->hash_entry->symbol;
2877   size_t func_name_len = strlen (function_to_edit);
2878   const char *end_of_fn_name;
2879 
2880 #ifndef UNPROTOIZE
2881 
2882   const f_list_chain_item *this_f_list_chain_item;
2883   const def_dec_info *definition = def_dec_p->definition;
2884 
2885   /* If we are protoizing, and if we found no corresponding definition for
2886      this particular function declaration, then just leave this declaration
2887      exactly as it is.  */
2888 
2889   if (!definition)
2890     return;
2891 
2892   /* If we are protoizing, and if the corresponding definition that we found
2893      for this particular function declaration defined an old style varargs
2894      function, then we want to issue a warning and just leave this function
2895      declaration unconverted.  */
2896 
2897   if (other_variable_style_function (definition->ansi_decl))
2898     {
2899       if (!quiet_flag)
2900 	notice ("%s: %d: warning: varargs function declaration not converted\n",
2901 		shortpath (NULL, def_dec_p->file->hash_entry->symbol),
2902 		def_dec_p->line);
2903       return;
2904     }
2905 
2906 #endif /* !defined (UNPROTOIZE) */
2907 
2908   /* Setup here to recover from confusing source code detected during this
2909      particular "edit".  */
2910 
2911   save_pointers ();
2912   if (setjmp (source_confusion_recovery))
2913     {
2914       restore_pointers ();
2915       notice ("%s: declaration of function '%s' not converted\n",
2916 	      pname, function_to_edit);
2917       return;
2918     }
2919 
2920   /* We are editing a function declaration.  The line number we did a seek to
2921      contains the comma or semicolon which follows the declaration.  Our job
2922      now is to scan backwards looking for the function name.  This name *must*
2923      be followed by open paren (ignoring whitespace, of course).  We need to
2924      replace everything between that open paren and the corresponding closing
2925      paren.  If we are protoizing, we need to insert the prototype-style
2926      formals lists.  If we are unprotoizing, we need to just delete everything
2927      between the pairs of opening and closing parens.  */
2928 
2929   /* First move up to the end of the line.  */
2930 
2931   while (*clean_text_p != '\n')
2932     check_source (++clean_text_p < clean_text_limit, 0);
2933   clean_text_p--;  /* Point to just before the newline character.  */
2934 
2935   /* Now we can scan backwards for the function name.  */
2936 
2937   do
2938     {
2939       for (;;)
2940 	{
2941 	  /* Scan leftwards until we find some character which can be
2942 	     part of an identifier.  */
2943 
2944 	  while (!is_id_char (*clean_text_p))
2945 	    check_source (--clean_text_p > clean_read_ptr, 0);
2946 
2947 	  /* Scan backwards until we find a char that cannot be part of an
2948 	     identifier.  */
2949 
2950 	  while (is_id_char (*clean_text_p))
2951 	    check_source (--clean_text_p > clean_read_ptr, 0);
2952 
2953 	  /* Having found an "id break", see if the following id is the one
2954 	     that we are looking for.  If so, then exit from this loop.  */
2955 
2956 	  if (!strncmp (clean_text_p+1, function_to_edit, func_name_len))
2957 	    {
2958 	      char ch = *(clean_text_p + 1 + func_name_len);
2959 
2960 	      /* Must also check to see that the name in the source text
2961 	         ends where it should (in order to prevent bogus matches
2962 	         on similar but longer identifiers.  */
2963 
2964 	      if (! is_id_char (ch))
2965 	        break;			/* exit from loop */
2966 	    }
2967 	}
2968 
2969       /* We have now found the first perfect match for the function name in
2970 	 our backward search.  This may or may not be the actual function
2971 	 name at the start of the actual function declaration (i.e. we could
2972 	 have easily been mislead).  We will try to avoid getting fooled too
2973 	 often by looking forward for the open paren which should follow the
2974 	 identifier we just found.  We ignore whitespace while hunting.  If
2975 	 the next non-whitespace byte we see is *not* an open left paren,
2976 	 then we must assume that we have been fooled and we start over
2977 	 again accordingly.  Note that there is no guarantee, that even if
2978 	 we do see the open paren, that we are in the right place.
2979 	 Programmers do the strangest things sometimes!  */
2980 
2981       end_of_fn_name = clean_text_p + strlen (def_dec_p->hash_entry->symbol);
2982       start_formals = forward_to_next_token_char (end_of_fn_name);
2983     }
2984   while (*start_formals != '(');
2985 
2986   /* start_of_formals now points to the opening left paren which immediately
2987      follows the name of the function.  */
2988 
2989   /* Note that there may be several formals lists which need to be modified
2990      due to the possibility that the return type of this function is a
2991      pointer-to-function type.  If there are several formals lists, we
2992      convert them in left-to-right order here.  */
2993 
2994 #ifndef UNPROTOIZE
2995   this_f_list_chain_item = definition->f_list_chain;
2996 #endif /* !defined (UNPROTOIZE) */
2997 
2998   for (;;)
2999     {
3000       {
3001 	int depth;
3002 
3003 	end_formals = start_formals + 1;
3004 	depth = 1;
3005 	for (; depth; check_source (++end_formals < clean_text_limit, 0))
3006 	  {
3007 	    switch (*end_formals)
3008 	      {
3009 	      case '(':
3010 		depth++;
3011 		break;
3012 	      case ')':
3013 		depth--;
3014 		break;
3015 	      }
3016 	  }
3017 	end_formals--;
3018       }
3019 
3020       /* end_formals now points to the closing right paren of the formals
3021 	 list whose left paren is pointed to by start_formals.  */
3022 
3023       /* Now, if we are protoizing, we insert the new ANSI-style formals list
3024 	 attached to the associated definition of this function.  If however
3025 	 we are unprotoizing, then we simply delete any formals list which
3026 	 may be present.  */
3027 
3028       output_up_to (start_formals);
3029 #ifndef UNPROTOIZE
3030       if (this_f_list_chain_item)
3031 	{
3032 	  output_string (this_f_list_chain_item->formals_list);
3033 	  this_f_list_chain_item = this_f_list_chain_item->chain_next;
3034 	}
3035       else
3036 	{
3037 	  if (!quiet_flag)
3038 	    notice ("%s: warning: too many parameter lists in declaration of '%s'\n",
3039 		    pname, def_dec_p->hash_entry->symbol);
3040 	  check_source (0, end_formals);  /* leave the declaration intact */
3041 	}
3042 #endif /* !defined (UNPROTOIZE) */
3043       clean_read_ptr = end_formals - 1;
3044 
3045       /* Now see if it looks like there may be another formals list associated
3046 	 with the function declaration that we are converting (following the
3047 	 formals list that we just converted.  */
3048 
3049       {
3050 	const char *another_r_paren = forward_to_next_token_char (end_formals);
3051 
3052 	if ((*another_r_paren != ')')
3053 	    || (*(start_formals = forward_to_next_token_char (another_r_paren)) != '('))
3054 	  {
3055 #ifndef UNPROTOIZE
3056 	    if (this_f_list_chain_item)
3057 	      {
3058 		if (!quiet_flag)
3059 		  notice ("\n%s: warning: too few parameter lists in declaration of '%s'\n",
3060 			  pname, def_dec_p->hash_entry->symbol);
3061 		check_source (0, start_formals); /* leave the decl intact */
3062 	      }
3063 #endif /* !defined (UNPROTOIZE) */
3064 	    break;
3065 
3066 	  }
3067       }
3068 
3069       /* There does appear to be yet another formals list, so loop around
3070 	 again, and convert it also.  */
3071     }
3072 }
3073 
3074 /* Edit a whole group of formals lists, starting with the rightmost one
3075    from some set of formals lists.  This routine is called once (from the
3076    outside) for each function declaration which is converted.  It is
3077    recursive however, and it calls itself once for each remaining formal
3078    list that lies to the left of the one it was originally called to work
3079    on.  Thus, a whole set gets done in right-to-left order.
3080 
3081    This routine returns nonzero if it thinks that it should not be trying
3082    to convert this particular function definition (because the name of the
3083    function doesn't match the one expected).  */
3084 
3085 static int
edit_formals_lists(const char * end_formals,unsigned int f_list_count,const def_dec_info * def_dec_p)3086 edit_formals_lists (const char *end_formals, unsigned int f_list_count,
3087 		    const def_dec_info *def_dec_p)
3088 {
3089   const char *start_formals;
3090   int depth;
3091 
3092   start_formals = end_formals - 1;
3093   depth = 1;
3094   for (; depth; check_source (--start_formals > clean_read_ptr, 0))
3095     {
3096       switch (*start_formals)
3097 	{
3098 	case '(':
3099 	  depth--;
3100 	  break;
3101 	case ')':
3102 	  depth++;
3103 	  break;
3104 	}
3105     }
3106   start_formals++;
3107 
3108   /* start_formals now points to the opening left paren of the formals list.  */
3109 
3110   f_list_count--;
3111 
3112   if (f_list_count)
3113     {
3114       const char *next_end;
3115 
3116       /* There should be more formal lists to the left of here.  */
3117 
3118       next_end = start_formals - 1;
3119       check_source (next_end > clean_read_ptr, 0);
3120       while (ISSPACE ((const unsigned char)*next_end))
3121 	check_source (--next_end > clean_read_ptr, 0);
3122       check_source (*next_end == ')', next_end);
3123       check_source (--next_end > clean_read_ptr, 0);
3124       check_source (*next_end == ')', next_end);
3125       if (edit_formals_lists (next_end, f_list_count, def_dec_p))
3126 	return 1;
3127     }
3128 
3129   /* Check that the function name in the header we are working on is the same
3130      as the one we would expect to find.  If not, issue a warning and return
3131      nonzero.  */
3132 
3133   if (f_list_count == 0)
3134     {
3135       const char *expected = def_dec_p->hash_entry->symbol;
3136       const char *func_name_start;
3137       const char *func_name_limit;
3138       size_t func_name_len;
3139 
3140       for (func_name_limit = start_formals-1;
3141 	   ISSPACE ((const unsigned char)*func_name_limit); )
3142 	check_source (--func_name_limit > clean_read_ptr, 0);
3143 
3144       for (func_name_start = func_name_limit++;
3145 	   is_id_char (*func_name_start);
3146 	   func_name_start--)
3147 	check_source (func_name_start > clean_read_ptr, 0);
3148       func_name_start++;
3149       func_name_len = func_name_limit - func_name_start;
3150       if (func_name_len == 0)
3151 	check_source (0, func_name_start);
3152       if (func_name_len != strlen (expected)
3153 	  || strncmp (func_name_start, expected, func_name_len))
3154 	{
3155 	  notice ("%s: %d: warning: found '%s' but expected '%s'\n",
3156 		  shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3157 		  identify_lineno (func_name_start),
3158 		  dupnstr (func_name_start, func_name_len),
3159 		  expected);
3160 	  return 1;
3161 	}
3162     }
3163 
3164   output_up_to (start_formals);
3165 
3166 #ifdef UNPROTOIZE
3167   if (f_list_count == 0)
3168     output_string (def_dec_p->formal_names);
3169 #else /* !defined (UNPROTOIZE) */
3170   {
3171     unsigned f_list_depth;
3172     const f_list_chain_item *flci_p = def_dec_p->f_list_chain;
3173 
3174     /* At this point, the current value of f_list count says how many
3175        links we have to follow through the f_list_chain to get to the
3176        particular formals list that we need to output next.  */
3177 
3178     for (f_list_depth = 0; f_list_depth < f_list_count; f_list_depth++)
3179       flci_p = flci_p->chain_next;
3180     output_string (flci_p->formals_list);
3181   }
3182 #endif /* !defined (UNPROTOIZE) */
3183 
3184   clean_read_ptr = end_formals - 1;
3185   return 0;
3186 }
3187 
3188 /* Given a pointer to a byte in the clean text buffer which points to
3189    the beginning of a line that contains a "follower" token for a
3190    function definition header, do whatever is necessary to find the
3191    right closing paren for the rightmost formals list of the function
3192    definition header.  */
3193 
3194 static const char *
find_rightmost_formals_list(const char * clean_text_p)3195 find_rightmost_formals_list (const char *clean_text_p)
3196 {
3197   const char *end_formals;
3198 
3199   /* We are editing a function definition.  The line number we did a seek
3200      to contains the first token which immediately follows the entire set of
3201      formals lists which are part of this particular function definition
3202      header.
3203 
3204      Our job now is to scan leftwards in the clean text looking for the
3205      right-paren which is at the end of the function header's rightmost
3206      formals list.
3207 
3208      If we ignore whitespace, this right paren should be the first one we
3209      see which is (ignoring whitespace) immediately followed either by the
3210      open curly-brace beginning the function body or by an alphabetic
3211      character (in the case where the function definition is in old (K&R)
3212      style and there are some declarations of formal parameters).  */
3213 
3214    /* It is possible that the right paren we are looking for is on the
3215       current line (together with its following token).  Just in case that
3216       might be true, we start out here by skipping down to the right end of
3217       the current line before starting our scan.  */
3218 
3219   for (end_formals = clean_text_p; *end_formals != '\n'; end_formals++)
3220     continue;
3221   end_formals--;
3222 
3223 #ifdef UNPROTOIZE
3224 
3225   /* Now scan backwards while looking for the right end of the rightmost
3226      formals list associated with this function definition.  */
3227 
3228   {
3229     char ch;
3230     const char *l_brace_p;
3231 
3232     /* Look leftward and try to find a right-paren.  */
3233 
3234     while (*end_formals != ')')
3235       {
3236 	if (ISSPACE ((unsigned char)*end_formals))
3237 	  while (ISSPACE ((unsigned char)*end_formals))
3238 	    check_source (--end_formals > clean_read_ptr, 0);
3239 	else
3240 	  check_source (--end_formals > clean_read_ptr, 0);
3241       }
3242 
3243     ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3244     /* Since we are unprotoizing an ANSI-style (prototyped) function
3245        definition, there had better not be anything (except whitespace)
3246        between the end of the ANSI formals list and the beginning of the
3247        function body (i.e. the '{').  */
3248 
3249     check_source (ch == '{', l_brace_p);
3250   }
3251 
3252 #else /* !defined (UNPROTOIZE) */
3253 
3254   /* Now scan backwards while looking for the right end of the rightmost
3255      formals list associated with this function definition.  */
3256 
3257   while (1)
3258     {
3259       char ch;
3260       const char *l_brace_p;
3261 
3262       /* Look leftward and try to find a right-paren.  */
3263 
3264       while (*end_formals != ')')
3265 	{
3266 	  if (ISSPACE ((const unsigned char)*end_formals))
3267 	    while (ISSPACE ((const unsigned char)*end_formals))
3268 	      check_source (--end_formals > clean_read_ptr, 0);
3269 	  else
3270 	    check_source (--end_formals > clean_read_ptr, 0);
3271 	}
3272 
3273       ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3274 
3275       /* Since it is possible that we found a right paren before the starting
3276 	 '{' of the body which IS NOT the one at the end of the real K&R
3277 	 formals list (say for instance, we found one embedded inside one of
3278 	 the old K&R formal parameter declarations) we have to check to be
3279 	 sure that this is in fact the right paren that we were looking for.
3280 
3281 	 The one we were looking for *must* be followed by either a '{' or
3282 	 by an alphabetic character, while others *cannot* validly be followed
3283 	 by such characters.  */
3284 
3285       if ((ch == '{') || ISALPHA ((unsigned char) ch))
3286 	break;
3287 
3288       /* At this point, we have found a right paren, but we know that it is
3289 	 not the one we were looking for, so backup one character and keep
3290 	 looking.  */
3291 
3292       check_source (--end_formals > clean_read_ptr, 0);
3293     }
3294 
3295 #endif /* !defined (UNPROTOIZE) */
3296 
3297   return end_formals;
3298 }
3299 
3300 #ifndef UNPROTOIZE
3301 
3302 /* Insert into the output file a totally new declaration for a function
3303    which (up until now) was being called from within the current block
3304    without having been declared at any point such that the declaration
3305    was visible (i.e. in scope) at the point of the call.
3306 
3307    We need to add in explicit declarations for all such function calls
3308    in order to get the full benefit of prototype-based function call
3309    parameter type checking.  */
3310 
3311 static void
add_local_decl(const def_dec_info * def_dec_p,const char * clean_text_p)3312 add_local_decl (const def_dec_info *def_dec_p, const char *clean_text_p)
3313 {
3314   const char *start_of_block;
3315   const char *function_to_edit = def_dec_p->hash_entry->symbol;
3316 
3317   /* Don't insert new local explicit declarations unless explicitly requested
3318      to do so.  */
3319 
3320   if (!local_flag)
3321     return;
3322 
3323   /* Setup here to recover from confusing source code detected during this
3324      particular "edit".  */
3325 
3326   save_pointers ();
3327   if (setjmp (source_confusion_recovery))
3328     {
3329       restore_pointers ();
3330       notice ("%s: local declaration for function '%s' not inserted\n",
3331 	      pname, function_to_edit);
3332       return;
3333     }
3334 
3335   /* We have already done a seek to the start of the line which should
3336      contain *the* open curly brace which begins the block in which we need
3337      to insert an explicit function declaration (to replace the implicit one).
3338 
3339      Now we scan that line, starting from the left, until we find the
3340      open curly brace we are looking for.  Note that there may actually be
3341      multiple open curly braces on the given line, but we will be happy
3342      with the leftmost one no matter what.  */
3343 
3344   start_of_block = clean_text_p;
3345   while (*start_of_block != '{' && *start_of_block != '\n')
3346     check_source (++start_of_block < clean_text_limit, 0);
3347 
3348   /* Note that the line from the original source could possibly
3349      contain *no* open curly braces!  This happens if the line contains
3350      a macro call which expands into a chunk of text which includes a
3351      block (and that block's associated open and close curly braces).
3352      In cases like this, we give up, issue a warning, and do nothing.  */
3353 
3354   if (*start_of_block != '{')
3355     {
3356       if (!quiet_flag)
3357 	notice ("\n%s: %d: warning: can't add declaration of '%s' into macro call\n",
3358 	  def_dec_p->file->hash_entry->symbol, def_dec_p->line,
3359 	  def_dec_p->hash_entry->symbol);
3360       return;
3361     }
3362 
3363   /* Figure out what a nice (pretty) indentation would be for the new
3364      declaration we are adding.  In order to do this, we must scan forward
3365      from the '{' until we find the first line which starts with some
3366      non-whitespace characters (i.e. real "token" material).  */
3367 
3368   {
3369     const char *ep = forward_to_next_token_char (start_of_block) - 1;
3370     const char *sp;
3371 
3372     /* Now we have ep pointing at the rightmost byte of some existing indent
3373        stuff.  At least that is the hope.
3374 
3375        We can now just scan backwards and find the left end of the existing
3376        indentation string, and then copy it to the output buffer.  */
3377 
3378     for (sp = ep; ISSPACE ((const unsigned char)*sp) && *sp != '\n'; sp--)
3379       continue;
3380 
3381     /* Now write out the open { which began this block, and any following
3382        trash up to and including the last byte of the existing indent that
3383        we just found.  */
3384 
3385     output_up_to (ep);
3386 
3387     /* Now we go ahead and insert the new declaration at this point.
3388 
3389        If the definition of the given function is in the same file that we
3390        are currently editing, and if its full ANSI declaration normally
3391        would start with the keyword `extern', suppress the `extern'.  */
3392 
3393     {
3394       const char *decl = def_dec_p->definition->ansi_decl;
3395 
3396       if ((*decl == 'e') && (def_dec_p->file == def_dec_p->definition->file))
3397 	decl += 7;
3398       output_string (decl);
3399     }
3400 
3401     /* Finally, write out a new indent string, just like the preceding one
3402        that we found.  This will typically include a newline as the first
3403        character of the indent string.  */
3404 
3405     output_bytes (sp, (size_t) (ep - sp) + 1);
3406   }
3407 }
3408 
3409 /* Given a pointer to a file_info record, and a pointer to the beginning
3410    of a line (in the clean text buffer) which is assumed to contain the
3411    first "follower" token for the first function definition header in the
3412    given file, find a good place to insert some new global function
3413    declarations (which will replace scattered and imprecise implicit ones)
3414    and then insert the new explicit declaration at that point in the file.  */
3415 
3416 static void
add_global_decls(const file_info * file_p,const char * clean_text_p)3417 add_global_decls (const file_info *file_p, const char *clean_text_p)
3418 {
3419   const def_dec_info *dd_p;
3420   const char *scan_p;
3421 
3422   /* Setup here to recover from confusing source code detected during this
3423      particular "edit".  */
3424 
3425   save_pointers ();
3426   if (setjmp (source_confusion_recovery))
3427     {
3428       restore_pointers ();
3429       notice ("%s: global declarations for file '%s' not inserted\n",
3430 	      pname, shortpath (NULL, file_p->hash_entry->symbol));
3431       return;
3432     }
3433 
3434   /* Start by finding a good location for adding the new explicit function
3435      declarations.  To do this, we scan backwards, ignoring whitespace
3436      and comments and other junk until we find either a semicolon, or until
3437      we hit the beginning of the file.  */
3438 
3439   scan_p = find_rightmost_formals_list (clean_text_p);
3440   for (;; --scan_p)
3441     {
3442       if (scan_p < clean_text_base)
3443 	break;
3444       check_source (scan_p > clean_read_ptr, 0);
3445       if (*scan_p == ';')
3446 	break;
3447     }
3448 
3449   /* scan_p now points either to a semicolon, or to just before the start
3450      of the whole file.  */
3451 
3452   /* Now scan forward for the first non-whitespace character.  In theory,
3453      this should be the first character of the following function definition
3454      header.  We will put in the added declarations just prior to that.  */
3455 
3456   scan_p++;
3457   while (ISSPACE ((const unsigned char)*scan_p))
3458     scan_p++;
3459   scan_p--;
3460 
3461   output_up_to (scan_p);
3462 
3463   /* Now write out full prototypes for all of the things that had been
3464      implicitly declared in this file (but only those for which we were
3465      actually able to find unique matching definitions).  Avoid duplicates
3466      by marking things that we write out as we go.  */
3467 
3468   {
3469     int some_decls_added = 0;
3470 
3471     for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3472       if (dd_p->is_implicit && dd_p->definition && !dd_p->definition->written)
3473 	{
3474 	  const char *decl = dd_p->definition->ansi_decl;
3475 
3476 	  /* If the function for which we are inserting a declaration is
3477 	     actually defined later in the same file, then suppress the
3478 	     leading `extern' keyword (if there is one).  */
3479 
3480 	  if (*decl == 'e' && (dd_p->file == dd_p->definition->file))
3481 	    decl += 7;
3482 
3483 	  output_string ("\n");
3484 	  output_string (decl);
3485 	  some_decls_added = 1;
3486 	  ((NONCONST def_dec_info *) dd_p->definition)->written = 1;
3487 	}
3488     if (some_decls_added)
3489       output_string ("\n\n");
3490   }
3491 
3492   /* Unmark all of the definitions that we just marked.  */
3493 
3494   for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3495     if (dd_p->definition)
3496       ((NONCONST def_dec_info *) dd_p->definition)->written = 0;
3497 }
3498 
3499 #endif /* !defined (UNPROTOIZE) */
3500 
3501 /* Do the editing operation specifically for a function "definition".  Note
3502    that editing operations for function "declarations" are handled by a
3503    separate routine above.  */
3504 
3505 static void
edit_fn_definition(const def_dec_info * def_dec_p,const char * volatile clean_text_p)3506 edit_fn_definition (const def_dec_info *def_dec_p,
3507 		    const char *volatile clean_text_p)
3508 {
3509   const char *end_formals;
3510   const char *function_to_edit = def_dec_p->hash_entry->symbol;
3511 
3512   /* Setup here to recover from confusing source code detected during this
3513      particular "edit".  */
3514 
3515   save_pointers ();
3516   if (setjmp (source_confusion_recovery))
3517     {
3518       restore_pointers ();
3519       notice ("%s: definition of function '%s' not converted\n",
3520 	      pname, function_to_edit);
3521       return;
3522     }
3523 
3524   end_formals = find_rightmost_formals_list (clean_text_p);
3525 
3526   /* end_of_formals now points to the closing right paren of the rightmost
3527      formals list which is actually part of the `header' of the function
3528      definition that we are converting.  */
3529 
3530   /* If the header of this function definition looks like it declares a
3531      function with a variable number of arguments, and if the way it does
3532      that is different from that way we would like it (i.e. varargs vs.
3533      stdarg) then issue a warning and leave the header unconverted.  */
3534 
3535   if (other_variable_style_function (def_dec_p->ansi_decl))
3536     {
3537       if (!quiet_flag)
3538 	notice ("%s: %d: warning: definition of %s not converted\n",
3539 		shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3540 		identify_lineno (end_formals),
3541 		other_var_style);
3542       output_up_to (end_formals);
3543       return;
3544     }
3545 
3546   if (edit_formals_lists (end_formals, def_dec_p->f_list_count, def_dec_p))
3547     {
3548       restore_pointers ();
3549       notice ("%s: definition of function '%s' not converted\n",
3550 	      pname, function_to_edit);
3551       return;
3552     }
3553 
3554   /* Have to output the last right paren because this never gets flushed by
3555      edit_formals_list.  */
3556 
3557   output_up_to (end_formals);
3558 
3559 #ifdef UNPROTOIZE
3560   {
3561     const char *decl_p;
3562     const char *semicolon_p;
3563     const char *limit_p;
3564     const char *scan_p;
3565     int had_newlines = 0;
3566 
3567     /* Now write out the K&R style formal declarations, one per line.  */
3568 
3569     decl_p = def_dec_p->formal_decls;
3570     limit_p = decl_p + strlen (decl_p);
3571     for (;decl_p < limit_p; decl_p = semicolon_p + 2)
3572       {
3573 	for (semicolon_p = decl_p; *semicolon_p != ';'; semicolon_p++)
3574 	  continue;
3575 	output_string ("\n");
3576 	output_string (indent_string);
3577 	output_bytes (decl_p, (size_t) ((semicolon_p + 1) - decl_p));
3578       }
3579 
3580     /* If there are no newlines between the end of the formals list and the
3581        start of the body, we should insert one now.  */
3582 
3583     for (scan_p = end_formals+1; *scan_p != '{'; )
3584       {
3585 	if (*scan_p == '\n')
3586 	  {
3587 	    had_newlines = 1;
3588 	    break;
3589 	  }
3590 	check_source (++scan_p < clean_text_limit, 0);
3591       }
3592     if (!had_newlines)
3593       output_string ("\n");
3594   }
3595 #else /* !defined (UNPROTOIZE) */
3596   /* If we are protoizing, there may be some flotsam & jetsam (like comments
3597      and preprocessing directives) after the old formals list but before
3598      the following { and we would like to preserve that stuff while effectively
3599      deleting the existing K&R formal parameter declarations.  We do so here
3600      in a rather tricky way.  Basically, we white out any stuff *except*
3601      the comments/pp-directives in the original text buffer, then, if there
3602      is anything in this area *other* than whitespace, we output it.  */
3603   {
3604     const char *end_formals_orig;
3605     const char *start_body;
3606     const char *start_body_orig;
3607     const char *scan;
3608     const char *scan_orig;
3609     int have_flotsam = 0;
3610     int have_newlines = 0;
3611 
3612     for (start_body = end_formals + 1; *start_body != '{';)
3613       check_source (++start_body < clean_text_limit, 0);
3614 
3615     end_formals_orig = orig_text_base + (end_formals - clean_text_base);
3616     start_body_orig = orig_text_base + (start_body - clean_text_base);
3617     scan = end_formals + 1;
3618     scan_orig = end_formals_orig + 1;
3619     for (; scan < start_body; scan++, scan_orig++)
3620       {
3621 	if (*scan == *scan_orig)
3622 	  {
3623 	    have_newlines |= (*scan_orig == '\n');
3624 	    /* Leave identical whitespace alone.  */
3625 	    if (!ISSPACE ((const unsigned char)*scan_orig))
3626 	      *((NONCONST char *) scan_orig) = ' '; /* identical - so whiteout */
3627 	  }
3628 	else
3629 	  have_flotsam = 1;
3630       }
3631     if (have_flotsam)
3632       output_bytes (end_formals_orig + 1,
3633 		    (size_t) (start_body_orig - end_formals_orig) - 1);
3634     else
3635       if (have_newlines)
3636 	output_string ("\n");
3637       else
3638 	output_string (" ");
3639     clean_read_ptr = start_body - 1;
3640   }
3641 #endif /* !defined (UNPROTOIZE) */
3642 }
3643 
3644 /* Clean up the clean text buffer.  Do this by converting comments and
3645    preprocessing directives into spaces.   Also convert line continuations
3646    into whitespace.  Also, whiteout string and character literals.  */
3647 
3648 static void
do_cleaning(char * new_clean_text_base,const char * new_clean_text_limit)3649 do_cleaning (char *new_clean_text_base, const char *new_clean_text_limit)
3650 {
3651   char *scan_p;
3652   int non_whitespace_since_newline = 0;
3653 
3654   for (scan_p = new_clean_text_base; scan_p < new_clean_text_limit; scan_p++)
3655     {
3656       switch (*scan_p)
3657 	{
3658 	case '/':			/* Handle comments.  */
3659 	  if (scan_p[1] != '*')
3660 	    goto regular;
3661 	  non_whitespace_since_newline = 1;
3662 	  scan_p[0] = ' ';
3663 	  scan_p[1] = ' ';
3664 	  scan_p += 2;
3665 	  while (scan_p[1] != '/' || scan_p[0] != '*')
3666 	    {
3667 	      if (!ISSPACE ((const unsigned char)*scan_p))
3668 		*scan_p = ' ';
3669 	      ++scan_p;
3670 	      gcc_assert (scan_p < new_clean_text_limit);
3671 	    }
3672 	  *scan_p++ = ' ';
3673 	  *scan_p = ' ';
3674 	  break;
3675 
3676 	case '#':			/* Handle pp directives.  */
3677 	  if (non_whitespace_since_newline)
3678 	    goto regular;
3679 	  *scan_p = ' ';
3680 	  while (scan_p[1] != '\n' || scan_p[0] == '\\')
3681 	    {
3682 	      if (!ISSPACE ((const unsigned char)*scan_p))
3683 		*scan_p = ' ';
3684 	      ++scan_p;
3685 	      gcc_assert (scan_p < new_clean_text_limit);
3686 	    }
3687 	  *scan_p++ = ' ';
3688 	  break;
3689 
3690 	case '\'':			/* Handle character literals.  */
3691 	  non_whitespace_since_newline = 1;
3692 	  while (scan_p[1] != '\'' || scan_p[0] == '\\')
3693 	    {
3694 	      if (scan_p[0] == '\\'
3695 		  && !ISSPACE ((const unsigned char) scan_p[1]))
3696 		scan_p[1] = ' ';
3697 	      if (!ISSPACE ((const unsigned char)*scan_p))
3698 		*scan_p = ' ';
3699 	      ++scan_p;
3700 	      gcc_assert (scan_p < new_clean_text_limit);
3701 	    }
3702 	  *scan_p++ = ' ';
3703 	  break;
3704 
3705 	case '"':			/* Handle string literals.  */
3706 	  non_whitespace_since_newline = 1;
3707 	  while (scan_p[1] != '"' || scan_p[0] == '\\')
3708 	    {
3709 	      if (scan_p[0] == '\\'
3710 		  && !ISSPACE ((const unsigned char) scan_p[1]))
3711 		scan_p[1] = ' ';
3712 	      if (!ISSPACE ((const unsigned char)*scan_p))
3713 		*scan_p = ' ';
3714 	      ++scan_p;
3715 	      gcc_assert (scan_p < new_clean_text_limit);
3716 	    }
3717 	  if (!ISSPACE ((const unsigned char)*scan_p))
3718 	    *scan_p = ' ';
3719 	  scan_p++;
3720 	  break;
3721 
3722 	case '\\':			/* Handle line continuations.  */
3723 	  if (scan_p[1] != '\n')
3724 	    goto regular;
3725 	  *scan_p = ' ';
3726 	  break;
3727 
3728 	case '\n':
3729 	  non_whitespace_since_newline = 0;	/* Reset.  */
3730 	  break;
3731 
3732 	case ' ':
3733 	case '\v':
3734 	case '\t':
3735 	case '\r':
3736 	case '\f':
3737 	case '\b':
3738 	  break;		/* Whitespace characters.  */
3739 
3740 	default:
3741 regular:
3742 	  non_whitespace_since_newline = 1;
3743 	  break;
3744 	}
3745     }
3746 }
3747 
3748 /* Given a pointer to the closing right parenthesis for a particular formals
3749    list (in the clean text buffer) find the corresponding left parenthesis
3750    and return a pointer to it.  */
3751 
3752 static const char *
careful_find_l_paren(const char * p)3753 careful_find_l_paren (const char *p)
3754 {
3755   const char *q;
3756   int paren_depth;
3757 
3758   for (paren_depth = 1, q = p-1; paren_depth; check_source (--q >= clean_text_base, 0))
3759     {
3760       switch (*q)
3761 	{
3762 	case ')':
3763 	  paren_depth++;
3764 	  break;
3765 	case '(':
3766 	  paren_depth--;
3767 	  break;
3768 	}
3769     }
3770   return ++q;
3771 }
3772 
3773 /* Scan the clean text buffer for cases of function definitions that we
3774    don't really know about because they were preprocessed out when the
3775    aux info files were created.
3776 
3777    In this version of protoize/unprotoize we just give a warning for each
3778    one found.  A later version may be able to at least unprotoize such
3779    missed items.
3780 
3781    Note that we may easily find all function definitions simply by
3782    looking for places where there is a left paren which is (ignoring
3783    whitespace) immediately followed by either a left-brace or by an
3784    upper or lower case letter.  Whenever we find this combination, we
3785    have also found a function definition header.
3786 
3787    Finding function *declarations* using syntactic clues is much harder.
3788    I will probably try to do this in a later version though.  */
3789 
3790 static void
scan_for_missed_items(const file_info * file_p)3791 scan_for_missed_items (const file_info *file_p)
3792 {
3793   static const char *scan_p;
3794   const char *limit = clean_text_limit - 3;
3795   static const char *backup_limit;
3796 
3797   backup_limit = clean_text_base - 1;
3798 
3799   for (scan_p = clean_text_base; scan_p < limit; scan_p++)
3800     {
3801       if (*scan_p == ')')
3802 	{
3803 	  static const char *last_r_paren;
3804 	  const char *ahead_p;
3805 
3806 	  last_r_paren = scan_p;
3807 
3808 	  for (ahead_p = scan_p + 1; ISSPACE ((const unsigned char)*ahead_p); )
3809 	    check_source (++ahead_p < limit, limit);
3810 
3811 	  scan_p = ahead_p - 1;
3812 
3813 	  if (ISALPHA ((const unsigned char)*ahead_p) || *ahead_p == '{')
3814 	    {
3815 	      const char *last_l_paren;
3816 	      const int lineno = identify_lineno (ahead_p);
3817 
3818 	      if (setjmp (source_confusion_recovery))
3819 		continue;
3820 
3821 	      /* We know we have a function definition header.  Now skip
3822 	         leftwards over all of its associated formals lists.  */
3823 
3824 	      do
3825 		{
3826 		  last_l_paren = careful_find_l_paren (last_r_paren);
3827 		  for (last_r_paren = last_l_paren-1;
3828 		       ISSPACE ((const unsigned char)*last_r_paren); )
3829 		    check_source (--last_r_paren >= backup_limit, backup_limit);
3830 		}
3831 	      while (*last_r_paren == ')');
3832 
3833 	      if (is_id_char (*last_r_paren))
3834 		{
3835 		  const char *id_limit = last_r_paren + 1;
3836 		  const char *id_start;
3837 		  size_t id_length;
3838 		  const def_dec_info *dd_p;
3839 
3840 		  for (id_start = id_limit-1; is_id_char (*id_start); )
3841 		    check_source (--id_start >= backup_limit, backup_limit);
3842 		  id_start++;
3843 		  backup_limit = id_start;
3844 		  if ((id_length = (size_t) (id_limit - id_start)) == 0)
3845 		    goto not_missed;
3846 
3847 		  {
3848 		    char *func_name = alloca (id_length + 1);
3849 		    static const char * const stmt_keywords[]
3850 		      = { "if", "else", "do", "while", "for", "switch", "case", "return", 0 };
3851 		    const char * const *stmt_keyword;
3852 
3853 		    strncpy (func_name, id_start, id_length);
3854 		    func_name[id_length] = '\0';
3855 
3856 		    /* We must check here to see if we are actually looking at
3857 		       a statement rather than an actual function call.  */
3858 
3859 		    for (stmt_keyword = stmt_keywords; *stmt_keyword; stmt_keyword++)
3860 		      if (!strcmp (func_name, *stmt_keyword))
3861 			goto not_missed;
3862 
3863 #if 0
3864 		    notice ("%s: found definition of '%s' at %s(%d)\n",
3865 			    pname,
3866 			    func_name,
3867 			    shortpath (NULL, file_p->hash_entry->symbol),
3868 			    identify_lineno (id_start));
3869 #endif				/* 0 */
3870 		    /* We really should check for a match of the function name
3871 		       here also, but why bother.  */
3872 
3873 		    for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3874 		      if (dd_p->is_func_def && dd_p->line == lineno)
3875 			goto not_missed;
3876 
3877 		    /* If we make it here, then we did not know about this
3878 		       function definition.  */
3879 
3880 		    notice ("%s: %d: warning: '%s' excluded by preprocessing\n",
3881 			    shortpath (NULL, file_p->hash_entry->symbol),
3882 			    identify_lineno (id_start), func_name);
3883 		    notice ("%s: function definition not converted\n",
3884 			    pname);
3885 		  }
3886 		not_missed: ;
3887 	        }
3888 	    }
3889 	}
3890     }
3891 }
3892 
3893 /* Do all editing operations for a single source file (either a "base" file
3894    or an "include" file).  To do this we read the file into memory, keep a
3895    virgin copy there, make another cleaned in-core copy of the original file
3896    (i.e. one in which all of the comments and preprocessing directives have
3897    been replaced with whitespace), then use these two in-core copies of the
3898    file to make a new edited in-core copy of the file.  Finally, rename the
3899    original file (as a way of saving it), and then write the edited version
3900    of the file from core to a disk file of the same name as the original.
3901 
3902    Note that the trick of making a copy of the original sans comments &
3903    preprocessing directives make the editing a whole lot easier.  */
3904 
3905 static void
edit_file(const hash_table_entry * hp)3906 edit_file (const hash_table_entry *hp)
3907 {
3908   struct stat stat_buf;
3909   const file_info *file_p = hp->fip;
3910   char *new_orig_text_base;
3911   char *new_orig_text_limit;
3912   char *new_clean_text_base;
3913   char *new_clean_text_limit;
3914   size_t orig_size;
3915   size_t repl_size;
3916   int first_definition_in_file;
3917 
3918   /* If we are not supposed to be converting this file, or if there is
3919      nothing in there which needs converting, just skip this file.  */
3920 
3921   if (!needs_to_be_converted (file_p))
3922     return;
3923 
3924   convert_filename = file_p->hash_entry->symbol;
3925 
3926   /* Convert a file if it is in a directory where we want conversion
3927      and the file is not excluded.  */
3928 
3929   if (!directory_specified_p (convert_filename)
3930       || file_excluded_p (convert_filename))
3931     {
3932       if (!quiet_flag
3933 #ifdef UNPROTOIZE
3934 	  /* Don't even mention "system" include files unless we are
3935 	     protoizing.  If we are protoizing, we mention these as a
3936 	     gentle way of prodding the user to convert his "system"
3937 	     include files to prototype format.  */
3938 	  && !in_system_include_dir (convert_filename)
3939 #endif /* defined (UNPROTOIZE) */
3940 	  )
3941 	notice ("%s: '%s' not converted\n",
3942 		pname, shortpath (NULL, convert_filename));
3943       return;
3944     }
3945 
3946   /* Let the user know what we are up to.  */
3947 
3948   if (nochange_flag)
3949     notice ("%s: would convert file '%s'\n",
3950 	    pname, shortpath (NULL, convert_filename));
3951   else
3952     notice ("%s: converting file '%s'\n",
3953 	    pname, shortpath (NULL, convert_filename));
3954   fflush (stderr);
3955 
3956   /* Find out the size (in bytes) of the original file.  */
3957 
3958   /* The cast avoids an erroneous warning on AIX.  */
3959   if (stat (convert_filename, &stat_buf) == -1)
3960     {
3961       int errno_val = errno;
3962       notice ("%s: can't get status for file '%s': %s\n",
3963 	      pname, shortpath (NULL, convert_filename),
3964 	      xstrerror (errno_val));
3965       return;
3966     }
3967   orig_size = stat_buf.st_size;
3968 
3969   /* Allocate a buffer to hold the original text.  */
3970 
3971   orig_text_base = new_orig_text_base = xmalloc (orig_size + 2);
3972   orig_text_limit = new_orig_text_limit = new_orig_text_base + orig_size;
3973 
3974   /* Allocate a buffer to hold the cleaned-up version of the original text.  */
3975 
3976   clean_text_base = new_clean_text_base = xmalloc (orig_size + 2);
3977   clean_text_limit = new_clean_text_limit = new_clean_text_base + orig_size;
3978   clean_read_ptr = clean_text_base - 1;
3979 
3980   /* Allocate a buffer that will hopefully be large enough to hold the entire
3981      converted output text.  As an initial guess for the maximum size of the
3982      output buffer, use 125% of the size of the original + some extra.  This
3983      buffer can be expanded later as needed.  */
3984 
3985   repl_size = orig_size + (orig_size >> 2) + 4096;
3986   repl_text_base = xmalloc (repl_size + 2);
3987   repl_text_limit = repl_text_base + repl_size - 1;
3988   repl_write_ptr = repl_text_base - 1;
3989 
3990   {
3991     int input_file;
3992     int fd_flags;
3993 
3994     /* Open the file to be converted in READ ONLY mode.  */
3995 
3996     fd_flags = O_RDONLY;
3997 #ifdef O_BINARY
3998     /* Use binary mode to avoid having to deal with different EOL characters.  */
3999     fd_flags |= O_BINARY;
4000 #endif
4001     if ((input_file = open (convert_filename, fd_flags, 0444)) == -1)
4002       {
4003 	int errno_val = errno;
4004 	notice ("%s: can't open file '%s' for reading: %s\n",
4005 		pname, shortpath (NULL, convert_filename),
4006 		xstrerror (errno_val));
4007 	return;
4008       }
4009 
4010     /* Read the entire original source text file into the original text buffer
4011        in one swell fwoop.  Then figure out where the end of the text is and
4012        make sure that it ends with a newline followed by a null.  */
4013 
4014     if (safe_read (input_file, new_orig_text_base, orig_size) !=
4015 	(int) orig_size)
4016       {
4017 	int errno_val = errno;
4018 	close (input_file);
4019 	notice ("\n%s: error reading input file '%s': %s\n",
4020 		pname, shortpath (NULL, convert_filename),
4021 		xstrerror (errno_val));
4022 	return;
4023       }
4024 
4025     close (input_file);
4026   }
4027 
4028   if (orig_size == 0 || orig_text_limit[-1] != '\n')
4029     {
4030       *new_orig_text_limit++ = '\n';
4031       orig_text_limit++;
4032     }
4033 
4034   /* Create the cleaned up copy of the original text.  */
4035 
4036   memcpy (new_clean_text_base, orig_text_base,
4037 	  (size_t) (orig_text_limit - orig_text_base));
4038   do_cleaning (new_clean_text_base, new_clean_text_limit);
4039 
4040 #if 0
4041   {
4042     int clean_file;
4043     size_t clean_size = orig_text_limit - orig_text_base;
4044     char *const clean_filename = alloca (strlen (convert_filename) + 6 + 1);
4045 
4046     /* Open (and create) the clean file.  */
4047 
4048     strcpy (clean_filename, convert_filename);
4049     strcat (clean_filename, ".clean");
4050     if ((clean_file = creat (clean_filename, 0666)) == -1)
4051       {
4052 	int errno_val = errno;
4053 	notice ("%s: can't create/open clean file '%s': %s\n",
4054 		pname, shortpath (NULL, clean_filename),
4055 		xstrerror (errno_val));
4056 	return;
4057       }
4058 
4059     /* Write the clean file.  */
4060 
4061     safe_write (clean_file, new_clean_text_base, clean_size, clean_filename);
4062 
4063     close (clean_file);
4064   }
4065 #endif /* 0 */
4066 
4067   /* Do a simplified scan of the input looking for things that were not
4068      mentioned in the aux info files because of the fact that they were
4069      in a region of the source which was preprocessed-out (via #if or
4070      via #ifdef).  */
4071 
4072   scan_for_missed_items (file_p);
4073 
4074   /* Setup to do line-oriented forward seeking in the clean text buffer.  */
4075 
4076   last_known_line_number = 1;
4077   last_known_line_start = clean_text_base;
4078 
4079   /* Now get down to business and make all of the necessary edits.  */
4080 
4081   {
4082     const def_dec_info *def_dec_p;
4083 
4084     first_definition_in_file = 1;
4085     def_dec_p = file_p->defs_decs;
4086     for (; def_dec_p; def_dec_p = def_dec_p->next_in_file)
4087       {
4088 	const char *clean_text_p = seek_to_line (def_dec_p->line);
4089 
4090 	/* clean_text_p now points to the first character of the line which
4091 	   contains the `terminator' for the declaration or definition that
4092 	   we are about to process.  */
4093 
4094 #ifndef UNPROTOIZE
4095 
4096 	if (global_flag && def_dec_p->is_func_def && first_definition_in_file)
4097 	  {
4098 	    add_global_decls (def_dec_p->file, clean_text_p);
4099 	    first_definition_in_file = 0;
4100 	  }
4101 
4102 	/* Don't edit this item if it is already in prototype format or if it
4103 	   is a function declaration and we have found no corresponding
4104 	   definition.  */
4105 
4106 	if (def_dec_p->prototyped
4107 	    || (!def_dec_p->is_func_def && !def_dec_p->definition))
4108 	  continue;
4109 
4110 #endif /* !defined (UNPROTOIZE) */
4111 
4112 	if (def_dec_p->is_func_def)
4113 	  edit_fn_definition (def_dec_p, clean_text_p);
4114 	else
4115 #ifndef UNPROTOIZE
4116 	if (def_dec_p->is_implicit)
4117 	  add_local_decl (def_dec_p, clean_text_p);
4118 	else
4119 #endif /* !defined (UNPROTOIZE) */
4120 	  edit_fn_declaration (def_dec_p, clean_text_p);
4121       }
4122   }
4123 
4124   /* Finalize things.  Output the last trailing part of the original text.  */
4125 
4126   output_up_to (clean_text_limit - 1);
4127 
4128   /* If this is just a test run, stop now and just deallocate the buffers.  */
4129 
4130   if (nochange_flag)
4131     {
4132       free (new_orig_text_base);
4133       free (new_clean_text_base);
4134       free (repl_text_base);
4135       return;
4136     }
4137 
4138   /* Change the name of the original input file.  This is just a quick way of
4139      saving the original file.  */
4140 
4141   if (!nosave_flag)
4142     {
4143       char *new_filename
4144 	= xmalloc (strlen (convert_filename) + strlen (save_suffix) + 2);
4145 
4146       strcpy (new_filename, convert_filename);
4147 #ifdef __MSDOS__
4148       /* MSDOS filenames are restricted to 8.3 format, so we save `foo.c'
4149 	 as `foo.<save_suffix>'.  */
4150       new_filename[(strlen (convert_filename) - 1] = '\0';
4151 #endif
4152       strcat (new_filename, save_suffix);
4153 
4154       /* Don't overwrite existing file.  */
4155       if (access (new_filename, F_OK) == 0)
4156 	{
4157 	  if (!quiet_flag)
4158 	    notice ("%s: warning: file '%s' already saved in '%s'\n",
4159 		    pname,
4160 		    shortpath (NULL, convert_filename),
4161 		    shortpath (NULL, new_filename));
4162 	}
4163       else if (rename (convert_filename, new_filename) == -1)
4164 	{
4165 	  int errno_val = errno;
4166 	  notice ("%s: can't link file '%s' to '%s': %s\n",
4167 		  pname,
4168 		  shortpath (NULL, convert_filename),
4169 		  shortpath (NULL, new_filename),
4170 		  xstrerror (errno_val));
4171 	  return;
4172 	}
4173     }
4174 
4175   if (unlink (convert_filename) == -1)
4176     {
4177       int errno_val = errno;
4178       /* The file may have already been renamed.  */
4179       if (errno_val != ENOENT)
4180 	{
4181 	  notice ("%s: can't delete file '%s': %s\n",
4182 		  pname, shortpath (NULL, convert_filename),
4183 		  xstrerror (errno_val));
4184 	  return;
4185 	}
4186     }
4187 
4188   {
4189     int output_file;
4190 
4191     /* Open (and create) the output file.  */
4192 
4193     if ((output_file = creat (convert_filename, 0666)) == -1)
4194       {
4195 	int errno_val = errno;
4196 	notice ("%s: can't create/open output file '%s': %s\n",
4197 		pname, shortpath (NULL, convert_filename),
4198 		xstrerror (errno_val));
4199 	return;
4200       }
4201 #ifdef O_BINARY
4202     /* Use binary mode to avoid changing the existing EOL character.  */
4203     setmode (output_file, O_BINARY);
4204 #endif
4205 
4206     /* Write the output file.  */
4207 
4208     {
4209       unsigned int out_size = (repl_write_ptr + 1) - repl_text_base;
4210 
4211       safe_write (output_file, repl_text_base, out_size, convert_filename);
4212     }
4213 
4214     close (output_file);
4215   }
4216 
4217   /* Deallocate the conversion buffers.  */
4218 
4219   free (new_orig_text_base);
4220   free (new_clean_text_base);
4221   free (repl_text_base);
4222 
4223   /* Change the mode of the output file to match the original file.  */
4224 
4225   /* The cast avoids an erroneous warning on AIX.  */
4226   if (chmod (convert_filename, stat_buf.st_mode) == -1)
4227     {
4228       int errno_val = errno;
4229       notice ("%s: can't change mode of file '%s': %s\n",
4230 	      pname, shortpath (NULL, convert_filename),
4231 	      xstrerror (errno_val));
4232     }
4233 
4234   /* Note:  We would try to change the owner and group of the output file
4235      to match those of the input file here, except that may not be a good
4236      thing to do because it might be misleading.  Also, it might not even
4237      be possible to do that (on BSD systems with quotas for instance).  */
4238 }
4239 
4240 /* Do all of the individual steps needed to do the protoization (or
4241    unprotoization) of the files referenced in the aux_info files given
4242    in the command line.  */
4243 
4244 static void
4245 do_processing (void)
4246 {
4247   const char * const *base_pp;
4248   const char * const * const end_pps
4249     = &base_source_filenames[n_base_source_files];
4250 
4251 #ifndef UNPROTOIZE
4252   int syscalls_len;
4253 #endif /* !defined (UNPROTOIZE) */
4254 
4255   /* One-by-one, check (and create if necessary), open, and read all of the
4256      stuff in each aux_info file.  After reading each aux_info file, the
4257      aux_info_file just read will be automatically deleted unless the
4258      keep_flag is set.  */
4259 
4260   for (base_pp = base_source_filenames; base_pp < end_pps; base_pp++)
4261     process_aux_info_file (*base_pp, keep_flag, 0);
4262 
4263 #ifndef UNPROTOIZE
4264 
4265   /* Also open and read the special SYSCALLS.c aux_info file which gives us
4266      the prototypes for all of the standard system-supplied functions.  */
4267 
4268   if (nondefault_syscalls_dir)
4269     {
4270       syscalls_absolute_filename
4271 	= xmalloc (strlen (nondefault_syscalls_dir) + 1
4272 		   + sizeof (syscalls_filename));
4273       strcpy (syscalls_absolute_filename, nondefault_syscalls_dir);
4274     }
4275   else
4276     {
4277       GET_ENVIRONMENT (default_syscalls_dir, "GCC_EXEC_PREFIX");
4278       if (!default_syscalls_dir)
4279 	{
4280 	  default_syscalls_dir = standard_exec_prefix;
4281 	}
4282       syscalls_absolute_filename
4283 	= xmalloc (strlen (default_syscalls_dir) + 0
4284 		   + strlen (target_machine) + 1
4285 		   + strlen (target_version) + 1
4286 		   + sizeof (syscalls_filename));
4287       strcpy (syscalls_absolute_filename, default_syscalls_dir);
4288       strcat (syscalls_absolute_filename, target_machine);
4289       strcat (syscalls_absolute_filename, "/");
4290       strcat (syscalls_absolute_filename, target_version);
4291       strcat (syscalls_absolute_filename, "/");
4292     }
4293 
4294   syscalls_len = strlen (syscalls_absolute_filename);
4295   if (! IS_DIR_SEPARATOR (*(syscalls_absolute_filename + syscalls_len - 1)))
4296     {
4297       *(syscalls_absolute_filename + syscalls_len++) = DIR_SEPARATOR;
4298       *(syscalls_absolute_filename + syscalls_len) = '\0';
4299     }
4300   strcat (syscalls_absolute_filename, syscalls_filename);
4301 
4302   /* Call process_aux_info_file in such a way that it does not try to
4303      delete the SYSCALLS aux_info file.  */
4304 
4305   process_aux_info_file (syscalls_absolute_filename, 1, 1);
4306 
4307 #endif /* !defined (UNPROTOIZE) */
4308 
4309   /* When we first read in all of the information from the aux_info files
4310      we saved in it descending line number order, because that was likely to
4311      be faster.  Now however, we want the chains of def & dec records to
4312      appear in ascending line number order as we get further away from the
4313      file_info record that they hang from.  The following line causes all of
4314      these lists to be rearranged into ascending line number order.  */
4315 
4316   visit_each_hash_node (filename_primary, reverse_def_dec_list);
4317 
4318 #ifndef UNPROTOIZE
4319 
4320   /* Now do the "real" work.  The following line causes each declaration record
4321      to be "visited".  For each of these nodes, an attempt is made to match
4322      up the function declaration with a corresponding function definition,
4323      which should have a full prototype-format formals list with it.  Once
4324      these match-ups are made, the conversion of the function declarations
4325      to prototype format can be made.  */
4326 
4327   visit_each_hash_node (function_name_primary, connect_defs_and_decs);
4328 
4329 #endif /* !defined (UNPROTOIZE) */
4330 
4331   /* Now convert each file that can be converted (and needs to be).  */
4332 
4333   visit_each_hash_node (filename_primary, edit_file);
4334 
4335 #ifndef UNPROTOIZE
4336 
4337   /* If we are working in cplusplus mode, try to rename all .c files to .C
4338      files.  Don't panic if some of the renames don't work.  */
4339 
4340   if (cplusplus_flag && !nochange_flag)
4341     visit_each_hash_node (filename_primary, rename_c_file);
4342 
4343 #endif /* !defined (UNPROTOIZE) */
4344 }
4345 
4346 static const struct option longopts[] =
4347 {
4348   {"version", 0, 0, 'V'},
4349   {"file_name", 0, 0, 'p'},
4350   {"quiet", 0, 0, 'q'},
4351   {"silent", 0, 0, 'q'},
4352   {"force", 0, 0, 'f'},
4353   {"keep", 0, 0, 'k'},
4354   {"nosave", 0, 0, 'N'},
4355   {"nochange", 0, 0, 'n'},
4356   {"compiler-options", 1, 0, 'c'},
4357   {"exclude", 1, 0, 'x'},
4358   {"directory", 1, 0, 'd'},
4359 #ifdef UNPROTOIZE
4360   {"indent", 1, 0, 'i'},
4361 #else
4362   {"local", 0, 0, 'l'},
4363   {"global", 0, 0, 'g'},
4364   {"c++", 0, 0, 'C'},
4365   {"syscalls-dir", 1, 0, 'B'},
4366 #endif
4367   {0, 0, 0, 0}
4368 };
4369 
4370 extern int main (int, char **const);
4371 
4372 int
4373 main (int argc, char **const argv)
4374 {
4375   int longind;
4376   int c;
4377   const char *params = "";
4378 
4379   pname = strrchr (argv[0], DIR_SEPARATOR);
4380 #ifdef DIR_SEPARATOR_2
4381   {
4382     char *slash;
4383 
4384     slash = strrchr (pname ? pname : argv[0], DIR_SEPARATOR_2);
4385     if (slash)
4386       pname = slash;
4387   }
4388 #endif
4389   pname = pname ? pname+1 : argv[0];
4390 
4391 #ifdef SIGCHLD
4392   /* We *MUST* set SIGCHLD to SIG_DFL so that the wait4() call will
4393      receive the signal.  A different setting is inheritable */
4394   signal (SIGCHLD, SIG_DFL);
4395 #endif
4396 
4397   /* Unlock the stdio streams.  */
4398   unlock_std_streams ();
4399 
4400   gcc_init_libintl ();
4401 
4402   cwd_buffer = getpwd ();
4403   if (!cwd_buffer)
4404     {
4405       notice ("%s: cannot get working directory: %s\n",
4406 	      pname, xstrerror(errno));
4407       return (FATAL_EXIT_CODE);
4408     }
4409 
4410   /* By default, convert the files in the current directory.  */
4411   directory_list = string_list_cons (cwd_buffer, NULL);
4412 
4413   while ((c = getopt_long (argc, argv,
4414 #ifdef UNPROTOIZE
4415 			   "c:d:i:knNp:qvVx:",
4416 #else
4417 			   "B:c:Cd:gklnNp:qvVx:",
4418 #endif
4419 			   longopts, &longind)) != EOF)
4420     {
4421       if (c == 0)		/* Long option.  */
4422 	c = longopts[longind].val;
4423       switch (c)
4424 	{
4425 	case 'p':
4426 	  compiler_file_name = optarg;
4427 	  break;
4428 	case 'd':
4429 	  directory_list
4430 	    = string_list_cons (abspath (NULL, optarg), directory_list);
4431 	  break;
4432 	case 'x':
4433 	  exclude_list = string_list_cons (optarg, exclude_list);
4434 	  break;
4435 
4436 	case 'v':
4437 	case 'V':
4438 	  version_flag = 1;
4439 	  break;
4440 	case 'q':
4441 	  quiet_flag = 1;
4442 	  break;
4443 #if 0
4444 	case 'f':
4445 	  force_flag = 1;
4446 	  break;
4447 #endif
4448 	case 'n':
4449 	  nochange_flag = 1;
4450 	  keep_flag = 1;
4451 	  break;
4452 	case 'N':
4453 	  nosave_flag = 1;
4454 	  break;
4455 	case 'k':
4456 	  keep_flag = 1;
4457 	  break;
4458 	case 'c':
4459 	  params = optarg;
4460 	  break;
4461 #ifdef UNPROTOIZE
4462 	case 'i':
4463 	  indent_string = optarg;
4464 	  break;
4465 #else				/* !defined (UNPROTOIZE) */
4466 	case 'l':
4467 	  local_flag = 1;
4468 	  break;
4469 	case 'g':
4470 	  global_flag = 1;
4471 	  break;
4472 	case 'C':
4473 	  cplusplus_flag = 1;
4474 	  break;
4475 	case 'B':
4476 	  nondefault_syscalls_dir = optarg;
4477 	  break;
4478 #endif				/* !defined (UNPROTOIZE) */
4479 	default:
4480 	  usage ();
4481 	}
4482     }
4483 
4484   /* Set up compile_params based on -p and -c options.  */
4485   munge_compile_params (params);
4486 
4487   n_base_source_files = argc - optind;
4488 
4489   /* Now actually make a list of the base source filenames.  */
4490 
4491   base_source_filenames
4492     = xmalloc ((n_base_source_files + 1) * sizeof (char *));
4493   n_base_source_files = 0;
4494   for (; optind < argc; optind++)
4495     {
4496       const char *path = abspath (NULL, argv[optind]);
4497       int len = strlen (path);
4498 
4499       if (path[len-1] == 'c' && path[len-2] == '.')
4500 	base_source_filenames[n_base_source_files++] = path;
4501       else
4502 	{
4503 	  notice ("%s: input file names must have .c suffixes: %s\n",
4504 		  pname, shortpath (NULL, path));
4505 	  errors++;
4506 	}
4507     }
4508 
4509 #ifndef UNPROTOIZE
4510   /* We are only interested in the very first identifier token in the
4511      definition of `va_list', so if there is more junk after that first
4512      identifier token, delete it from the `varargs_style_indicator'.  */
4513   {
4514     const char *cp;
4515 
4516     for (cp = varargs_style_indicator; ISIDNUM (*cp); cp++)
4517       continue;
4518     if (*cp != 0)
4519       varargs_style_indicator = savestring (varargs_style_indicator,
4520 					    cp - varargs_style_indicator);
4521   }
4522 #endif /* !defined (UNPROTOIZE) */
4523 
4524   if (errors)
4525     usage ();
4526   else
4527     {
4528       if (version_flag)
4529 	fprintf (stderr, "%s: %s\n", pname, version_string);
4530       do_processing ();
4531     }
4532 
4533   return (errors ? FATAL_EXIT_CODE : SUCCESS_EXIT_CODE);
4534 }
4535