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 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, 59 Temple Place - Suite 330, Boston, MA
20 02111-1307, 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 #undef abort
38 #include "version.h"
39 
40 /* Include getopt.h for the sake of getopt_long.  */
41 #include "getopt.h"
42 
43 /* Macro to see if the path elements match.  */
44 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
45 #define IS_SAME_PATH_CHAR(a,b) (TOUPPER (a) == TOUPPER (b))
46 #else
47 #define IS_SAME_PATH_CHAR(a,b) ((a) == (b))
48 #endif
49 
50 /* Macro to see if the paths match.  */
51 #define IS_SAME_PATH(a,b) (FILENAME_CMP (a, b) == 0)
52 
53 /* Suffix for aux-info files.  */
54 #ifdef __MSDOS__
55 #define AUX_INFO_SUFFIX "X"
56 #else
57 #define AUX_INFO_SUFFIX ".X"
58 #endif
59 
60 /* Suffix for saved files.  */
61 #ifdef __MSDOS__
62 #define SAVE_SUFFIX "sav"
63 #else
64 #define SAVE_SUFFIX ".save"
65 #endif
66 
67 /* Suffix for renamed C++ files.  */
68 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
69 #define CPLUS_FILE_SUFFIX "cc"
70 #else
71 #define CPLUS_FILE_SUFFIX "C"
72 #endif
73 
74 static void usage (void) ATTRIBUTE_NORETURN;
75 static void aux_info_corrupted (void) ATTRIBUTE_NORETURN;
76 static void declare_source_confusing (const char *) ATTRIBUTE_NORETURN;
77 static const char *shortpath (const char *, const char *);
78 extern void fancy_abort  (void) ATTRIBUTE_NORETURN;
79 static void notice (const char *, ...) ATTRIBUTE_PRINTF_1;
80 static char *savestring (const char *, unsigned int);
81 static char *dupnstr (const char *, size_t);
82 static int safe_read (int, void *, int);
83 static void safe_write (int, void *, int, const char *);
84 static void save_pointers (void);
85 static void restore_pointers (void);
86 static int is_id_char (int);
87 static int in_system_include_dir (const char *);
88 static int directory_specified_p (const char *);
89 static int file_excluded_p (const char *);
90 static char *unexpand_if_needed (const char *);
91 static char *abspath (const char *, const char *);
92 static void check_aux_info (int);
93 static const char *find_corresponding_lparen (const char *);
94 static int referenced_file_is_newer (const char *, time_t);
95 static void save_def_or_dec (const char *, int);
96 static void munge_compile_params (const char *);
97 static int gen_aux_info_file (const char *);
98 static void process_aux_info_file (const char *, int, int);
99 static int identify_lineno (const char *);
100 static void check_source (int, const char *);
101 static const char *seek_to_line (int);
102 static const char *forward_to_next_token_char (const char *);
103 static void output_bytes (const char *, size_t);
104 static void output_string (const char *);
105 static void output_up_to (const char *);
106 static int other_variable_style_function (const char *);
107 static const char *find_rightmost_formals_list (const char *);
108 static void do_cleaning (char *, const char *);
109 static const char *careful_find_l_paren (const char *);
110 static void do_processing (void);
111 
112 /* Look for these where the `const' qualifier is intentionally cast aside.  */
113 #define NONCONST
114 
115 /* Define a default place to find the SYSCALLS.X file.  */
116 
117 #ifndef UNPROTOIZE
118 
119 #ifndef STANDARD_EXEC_PREFIX
120 #define STANDARD_EXEC_PREFIX "/usr/local/lib/gcc-lib/"
121 #endif /* !defined STANDARD_EXEC_PREFIX */
122 
123 static const char * const standard_exec_prefix = STANDARD_EXEC_PREFIX;
124 static const char * const target_machine = DEFAULT_TARGET_MACHINE;
125 static const char * const target_version = DEFAULT_TARGET_VERSION;
126 
127 #endif /* !defined (UNPROTOIZE) */
128 
129 /* Suffix of aux_info files.  */
130 
131 static const char * const aux_info_suffix = AUX_INFO_SUFFIX;
132 
133 /* String to attach to filenames for saved versions of original files.  */
134 
135 static const char * const save_suffix = SAVE_SUFFIX;
136 
137 #ifndef UNPROTOIZE
138 
139 /* String to attach to C filenames renamed to C++.  */
140 
141 static const char * const cplus_suffix = CPLUS_FILE_SUFFIX;
142 
143 /* File name of the file which contains descriptions of standard system
144    routines.  Note that we never actually do anything with this file per se,
145    but we do read in its corresponding aux_info file.  */
146 
147 static const char syscalls_filename[] = "SYSCALLS.c";
148 
149 /* Default place to find the above file.  */
150 
151 static const char * default_syscalls_dir;
152 
153 /* Variable to hold the complete absolutized filename of the SYSCALLS.c.X
154    file.  */
155 
156 static char * syscalls_absolute_filename;
157 
158 #endif /* !defined (UNPROTOIZE) */
159 
160 /* Type of the structure that holds information about macro unexpansions.  */
161 
162 struct unexpansion_struct {
163   const char *const expanded;
164   const char *const contracted;
165 };
166 typedef struct unexpansion_struct unexpansion;
167 
168 /* A table of conversions that may need to be made for some (stupid) older
169    operating systems where these types are preprocessor macros rather than
170    typedefs (as they really ought to be).
171 
172    WARNING: The contracted forms must be as small (or smaller) as the
173    expanded forms, or else havoc will ensue.  */
174 
175 static const unexpansion unexpansions[] = {
176   { "struct _iobuf", "FILE" },
177   { 0, 0 }
178 };
179 
180 /* The number of "primary" slots in the hash tables for filenames and for
181    function names.  This can be as big or as small as you like, except that
182    it must be a power of two.  */
183 
184 #define HASH_TABLE_SIZE		(1 << 9)
185 
186 /* Bit mask to use when computing hash values.  */
187 
188 static const int hash_mask = (HASH_TABLE_SIZE - 1);
189 
190 
191 /* Datatype for lists of directories or filenames.  */
192 struct string_list
193 {
194   const char *name;
195   struct string_list *next;
196 };
197 
198 static struct string_list *string_list_cons (const char *,
199 					     struct string_list *);
200 
201 /* List of directories in which files should be converted.  */
202 
203 struct string_list *directory_list;
204 
205 /* List of file names which should not be converted.
206    A file is excluded if the end of its name, following a /,
207    matches one of the names in this list.  */
208 
209 struct string_list *exclude_list;
210 
211 /* The name of the other style of variable-number-of-parameters functions
212    (i.e. the style that we want to leave unconverted because we don't yet
213    know how to convert them to this style.  This string is used in warning
214    messages.  */
215 
216 /* Also define here the string that we can search for in the parameter lists
217    taken from the .X files which will unambiguously indicate that we have
218    found a varargs style function.  */
219 
220 #ifdef UNPROTOIZE
221 static const char * const other_var_style = "stdarg";
222 #else /* !defined (UNPROTOIZE) */
223 static const char * const other_var_style = "varargs";
224 static const char *varargs_style_indicator = "va_alist";
225 #endif /* !defined (UNPROTOIZE) */
226 
227 /* The following two types are used to create hash tables.  In this program,
228    there are two hash tables which are used to store and quickly lookup two
229    different classes of strings.  The first type of strings stored in the
230    first hash table are absolute filenames of files which protoize needs to
231    know about.  The second type of strings (stored in the second hash table)
232    are function names.  It is this second class of strings which really
233    inspired the use of the hash tables, because there may be a lot of them.  */
234 
235 typedef struct hash_table_entry_struct hash_table_entry;
236 
237 /* Do some typedefs so that we don't have to write "struct" so often.  */
238 
239 typedef struct def_dec_info_struct def_dec_info;
240 typedef struct file_info_struct file_info;
241 typedef struct f_list_chain_item_struct f_list_chain_item;
242 
243 #ifndef UNPROTOIZE
244 static int is_syscalls_file (const file_info *);
245 static void rename_c_file (const hash_table_entry *);
246 static const def_dec_info *find_extern_def (const def_dec_info *,
247 					    const def_dec_info *);
248 static const def_dec_info *find_static_definition (const def_dec_info *);
249 static void connect_defs_and_decs (const hash_table_entry *);
250 static void add_local_decl (const def_dec_info *, const char *);
251 static void add_global_decls (const file_info *, const char *);
252 #endif /* ! UNPROTOIZE */
253 static int needs_to_be_converted (const file_info *);
254 static void visit_each_hash_node (const hash_table_entry *,
255 				  void (*)(const hash_table_entry *));
256 static hash_table_entry *add_symbol (hash_table_entry *, const char *);
257 static hash_table_entry *lookup (hash_table_entry *, const char *);
258 static void free_def_dec (def_dec_info *);
259 static file_info *find_file (const char *, int);
260 static void reverse_def_dec_list (const hash_table_entry *);
261 static void edit_fn_declaration (const def_dec_info *, const char *);
262 static int edit_formals_lists (const char *, unsigned int,
263 			       const def_dec_info *);
264 static void edit_fn_definition (const def_dec_info *, const char *);
265 static void scan_for_missed_items (const file_info *);
266 static void edit_file (const hash_table_entry *);
267 
268 /* In the struct below, note that the "_info" field has two different uses
269    depending on the type of hash table we are in (i.e. either the filenames
270    hash table or the function names hash table).  In the filenames hash table
271    the info fields of the entries point to the file_info struct which is
272    associated with each filename (1 per filename).  In the function names
273    hash table, the info field points to the head of a singly linked list of
274    def_dec_info entries which are all defs or decs of the function whose
275    name is pointed to by the "symbol" field.  Keeping all of the defs/decs
276    for a given function name on a special list specifically for that function
277    name makes it quick and easy to find out all of the important information
278    about a given (named) function.  */
279 
280 struct hash_table_entry_struct {
281   hash_table_entry *		hash_next;	/* -> to secondary entries */
282   const char *			symbol;		/* -> to the hashed string */
283   union {
284     const def_dec_info *	_ddip;
285     file_info *			_fip;
286   } _info;
287 };
288 #define ddip _info._ddip
289 #define fip _info._fip
290 
291 /* Define a type specifically for our two hash tables.  */
292 
293 typedef hash_table_entry hash_table[HASH_TABLE_SIZE];
294 
295 /* The following struct holds all of the important information about any
296    single filename (e.g. file) which we need to know about.  */
297 
298 struct file_info_struct {
299   const hash_table_entry *	hash_entry; /* -> to associated hash entry */
300   const def_dec_info *		defs_decs;  /* -> to chain of defs/decs */
301   time_t			mtime;      /* Time of last modification.  */
302 };
303 
304 /* Due to the possibility that functions may return pointers to functions,
305    (which may themselves have their own parameter lists) and due to the
306    fact that returned pointers-to-functions may be of type "pointer-to-
307    function-returning-pointer-to-function" (ad nauseum) we have to keep
308    an entire chain of ANSI style formal parameter lists for each function.
309 
310    Normally, for any given function, there will only be one formals list
311    on the chain, but you never know.
312 
313    Note that the head of each chain of formals lists is pointed to by the
314    `f_list_chain' field of the corresponding def_dec_info record.
315 
316    For any given chain, the item at the head of the chain is the *leftmost*
317    parameter list seen in the actual C language function declaration.  If
318    there are other members of the chain, then these are linked in left-to-right
319    order from the head of the chain.  */
320 
321 struct f_list_chain_item_struct {
322   const f_list_chain_item *	chain_next;	/* -> to next item on chain */
323   const char *			formals_list;	/* -> to formals list string */
324 };
325 
326 /* The following struct holds all of the important information about any
327    single function definition or declaration which we need to know about.
328    Note that for unprotoize we don't need to know very much because we
329    never even create records for stuff that we don't intend to convert
330    (like for instance defs and decs which are already in old K&R format
331    and "implicit" function declarations).  */
332 
333 struct def_dec_info_struct {
334   const def_dec_info *	next_in_file;	/* -> to rest of chain for file */
335   file_info *        	file;		/* -> file_info for containing file */
336   int        		line;		/* source line number of def/dec */
337   const char *		ansi_decl;	/* -> left end of ansi decl */
338   hash_table_entry *	hash_entry;	/* -> hash entry for function name */
339   unsigned int        	is_func_def;	/* = 0 means this is a declaration */
340   const def_dec_info *	next_for_func;	/* -> to rest of chain for func name */
341   unsigned int		f_list_count;	/* count of formals lists we expect */
342   char			prototyped;	/* = 0 means already prototyped */
343 #ifndef UNPROTOIZE
344   const f_list_chain_item * f_list_chain;	/* -> chain of formals lists */
345   const def_dec_info *	definition;	/* -> def/dec containing related def */
346   char	        	is_static;	/* = 0 means visibility is "extern"  */
347   char			is_implicit;	/* != 0 for implicit func decl's */
348   char			written;	/* != 0 means written for implicit */
349 #else /* !defined (UNPROTOIZE) */
350   const char *		formal_names;	/* -> to list of names of formals */
351   const char *		formal_decls;	/* -> to string of formal declarations */
352 #endif /* !defined (UNPROTOIZE) */
353 };
354 
355 /* Pointer to the tail component of the filename by which this program was
356    invoked.  Used everywhere in error and warning messages.  */
357 
358 static const char *pname;
359 
360 /* Error counter.  Will be nonzero if we should give up at the next convenient
361    stopping point.  */
362 
363 static int errors = 0;
364 
365 /* Option flags.  */
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 static int version_flag = 0;		/* Print our version number.  */
374 static int quiet_flag = 0;		/* Don't print messages normally.  */
375 static int nochange_flag = 0;		/* Don't convert, just say what files
376 					   we would have converted.  */
377 static int nosave_flag = 0;		/* Don't save the old version.  */
378 static 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 static int local_flag = 0;		/* Insert new local decls (when?).  */
385 static int global_flag = 0;		/* set by -g option */
386 static 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 * msgid,...)508 notice (const char *msgid, ...)
509 {
510   va_list ap;
511 
512   va_start (ap, msgid);
513   vfprintf (stderr, _(msgid), 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 /* More 'friendly' abort that prints the line and file.
529    config.h can #define abort fancy_abort if you like that sort of thing.  */
530 
531 void
fancy_abort(void)532 fancy_abort (void)
533 {
534   notice ("%s: internal abort\n", pname);
535   exit (FATAL_EXIT_CODE);
536 }
537 
538 /* Make a duplicate of the first N bytes of a given string in a newly
539    allocated area.  */
540 
541 static char *
dupnstr(const char * s,size_t n)542 dupnstr (const char *s, size_t n)
543 {
544   char *ret_val = xmalloc (n + 1);
545 
546   strncpy (ret_val, s, n);
547   ret_val[n] = '\0';
548   return ret_val;
549 }
550 
551 /* Read LEN bytes at PTR from descriptor DESC, for file FILENAME,
552    retrying if necessary.  Return the actual number of bytes read.  */
553 
554 static int
safe_read(int desc,void * ptr,int len)555 safe_read (int desc, void *ptr, int len)
556 {
557   int left = len;
558   while (left > 0) {
559     int nchars = read (desc, ptr, left);
560     if (nchars < 0)
561       {
562 #ifdef EINTR
563 	if (errno == EINTR)
564 	  continue;
565 #endif
566 	return nchars;
567       }
568     if (nchars == 0)
569       break;
570     /* Arithmetic on void pointers is a gcc extension.  */
571     ptr = (char *) ptr + nchars;
572     left -= nchars;
573   }
574   return len - left;
575 }
576 
577 /* Write LEN bytes at PTR to descriptor DESC,
578    retrying if necessary, and treating any real error as fatal.  */
579 
580 static void
safe_write(int desc,void * ptr,int len,const char * out_fname)581 safe_write (int desc, void *ptr, int len, const char *out_fname)
582 {
583   while (len > 0) {
584     int written = write (desc, ptr, len);
585     if (written < 0)
586       {
587 	int errno_val = errno;
588 #ifdef EINTR
589 	if (errno_val == EINTR)
590 	  continue;
591 #endif
592 	notice ("%s: error writing file `%s': %s\n",
593 		pname, shortpath (NULL, out_fname), xstrerror (errno_val));
594 	return;
595       }
596     /* Arithmetic on void pointers is a gcc extension.  */
597     ptr = (char *) ptr + written;
598     len -= written;
599   }
600 }
601 
602 /* Get setup to recover in case the edit we are about to do goes awry.  */
603 
604 static void
save_pointers(void)605 save_pointers (void)
606 {
607   saved_clean_read_ptr = clean_read_ptr;
608   saved_repl_write_ptr = repl_write_ptr;
609 }
610 
611 /* Call this routine to recover our previous state whenever something looks
612    too confusing in the source code we are trying to edit.  */
613 
614 static void
restore_pointers(void)615 restore_pointers (void)
616 {
617   clean_read_ptr = saved_clean_read_ptr;
618   repl_write_ptr = saved_repl_write_ptr;
619 }
620 
621 /* Return true if the given character is a valid identifier character.  */
622 
623 static int
is_id_char(int ch)624 is_id_char (int ch)
625 {
626   return (ISIDNUM (ch) || (ch == '$'));
627 }
628 
629 /* Give a message indicating the proper way to invoke this program and then
630    exit with nonzero status.  */
631 
632 static void
usage(void)633 usage (void)
634 {
635 #ifdef UNPROTOIZE
636   notice ("%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ filename ... ]'\n",
637 	  pname, pname);
638 #else /* !defined (UNPROTOIZE) */
639   notice ("%s: usage '%s [ -VqfnkNlgC ] [ -B <dirname> ] [ filename ... ]'\n",
640 	  pname, pname);
641 #endif /* !defined (UNPROTOIZE) */
642   exit (FATAL_EXIT_CODE);
643 }
644 
645 /* Return true if the given filename (assumed to be an absolute filename)
646    designates a file residing anywhere beneath any one of the "system"
647    include directories.  */
648 
649 static int
in_system_include_dir(const char * path)650 in_system_include_dir (const char *path)
651 {
652   const struct default_include *p;
653 
654   if (! IS_ABSOLUTE_PATH (path))
655     abort ();		/* Must be an absolutized filename.  */
656 
657   for (p = cpp_include_defaults; p->fname; p++)
658     if (!strncmp (path, p->fname, strlen (p->fname))
659 	&& IS_DIR_SEPARATOR (path[strlen (p->fname)]))
660       return 1;
661   return 0;
662 }
663 
664 #if 0
665 /* Return true if the given filename designates a file that the user has
666    read access to and for which the user has write access to the containing
667    directory.  */
668 
669 static int
file_could_be_converted(const char * path)670 file_could_be_converted (const char *path)
671 {
672   char *const dir_name = alloca (strlen (path) + 1);
673 
674   if (access (path, R_OK))
675     return 0;
676 
677   {
678     char *dir_last_slash;
679 
680     strcpy (dir_name, path);
681     dir_last_slash = strrchr (dir_name, DIR_SEPARATOR);
682 #ifdef DIR_SEPARATOR_2
683     {
684       char *slash;
685 
686       slash = strrchr (dir_last_slash ? dir_last_slash : dir_name,
687 		       DIR_SEPARATOR_2);
688       if (slash)
689 	dir_last_slash = slash;
690     }
691 #endif
692     if (dir_last_slash)
693       *dir_last_slash = '\0';
694     else
695       abort ();  /* Should have been an absolutized filename.  */
696   }
697 
698   if (access (path, W_OK))
699     return 0;
700 
701   return 1;
702 }
703 
704 /* Return true if the given filename designates a file that we are allowed
705    to modify.  Files which we should not attempt to modify are (a) "system"
706    include files, and (b) files which the user doesn't have write access to,
707    and (c) files which reside in directories which the user doesn't have
708    write access to.  Unless requested to be quiet, give warnings about
709    files that we will not try to convert for one reason or another.  An
710    exception is made for "system" include files, which we never try to
711    convert and for which we don't issue the usual warnings.  */
712 
713 static int
file_normally_convertible(const char * path)714 file_normally_convertible (const char *path)
715 {
716   char *const dir_name = alloca (strlen (path) + 1);
717 
718   if (in_system_include_dir (path))
719     return 0;
720 
721   {
722     char *dir_last_slash;
723 
724     strcpy (dir_name, path);
725     dir_last_slash = strrchr (dir_name, DIR_SEPARATOR);
726 #ifdef DIR_SEPARATOR_2
727     {
728       char *slash;
729 
730       slash = strrchr (dir_last_slash ? dir_last_slash : dir_name,
731 		       DIR_SEPARATOR_2);
732       if (slash)
733 	dir_last_slash = slash;
734     }
735 #endif
736     if (dir_last_slash)
737       *dir_last_slash = '\0';
738     else
739       abort ();  /* Should have been an absolutized filename.  */
740   }
741 
742   if (access (path, R_OK))
743     {
744       if (!quiet_flag)
745 	notice ("%s: warning: no read access for file `%s'\n",
746 		pname, shortpath (NULL, path));
747       return 0;
748     }
749 
750   if (access (path, W_OK))
751     {
752       if (!quiet_flag)
753 	notice ("%s: warning: no write access for file `%s'\n",
754 		pname, shortpath (NULL, path));
755       return 0;
756     }
757 
758   if (access (dir_name, W_OK))
759     {
760       if (!quiet_flag)
761 	notice ("%s: warning: no write access for dir containing `%s'\n",
762 		pname, shortpath (NULL, path));
763       return 0;
764     }
765 
766   return 1;
767 }
768 #endif /* 0 */
769 
770 #ifndef UNPROTOIZE
771 
772 /* Return true if the given file_info struct refers to the special SYSCALLS.c.X
773    file.  Return false otherwise.  */
774 
775 static int
is_syscalls_file(const file_info * fi_p)776 is_syscalls_file (const file_info *fi_p)
777 {
778   char const *f = fi_p->hash_entry->symbol;
779   size_t fl = strlen (f), sysl = sizeof (syscalls_filename) - 1;
780   return sysl <= fl  &&  strcmp (f + fl - sysl, syscalls_filename) == 0;
781 }
782 
783 #endif /* !defined (UNPROTOIZE) */
784 
785 /* Check to see if this file will need to have anything done to it on this
786    run.  If there is nothing in the given file which both needs conversion
787    and for which we have the necessary stuff to do the conversion, return
788    false.  Otherwise, return true.
789 
790    Note that (for protoize) it is only valid to call this function *after*
791    the connections between declarations and definitions have all been made
792    by connect_defs_and_decs.  */
793 
794 static int
needs_to_be_converted(const file_info * file_p)795 needs_to_be_converted (const file_info *file_p)
796 {
797   const def_dec_info *ddp;
798 
799 #ifndef UNPROTOIZE
800 
801   if (is_syscalls_file (file_p))
802     return 0;
803 
804 #endif /* !defined (UNPROTOIZE) */
805 
806   for (ddp = file_p->defs_decs; ddp; ddp = ddp->next_in_file)
807 
808     if (
809 
810 #ifndef UNPROTOIZE
811 
812       /* ... and if we a protoizing and this function is in old style ...  */
813       !ddp->prototyped
814       /* ... and if this a definition or is a decl with an associated def ...  */
815       && (ddp->is_func_def || (!ddp->is_func_def && ddp->definition))
816 
817 #else /* defined (UNPROTOIZE) */
818 
819       /* ... and if we are unprotoizing and this function is in new style ...  */
820       ddp->prototyped
821 
822 #endif /* defined (UNPROTOIZE) */
823       )
824 	  /* ... then the containing file needs converting.  */
825 	  return -1;
826   return 0;
827 }
828 
829 /* Return 1 if the file name NAME is in a directory
830    that should be converted.  */
831 
832 static int
directory_specified_p(const char * name)833 directory_specified_p (const char *name)
834 {
835   struct string_list *p;
836 
837   for (p = directory_list; p; p = p->next)
838     if (!strncmp (name, p->name, strlen (p->name))
839 	&& IS_DIR_SEPARATOR (name[strlen (p->name)]))
840       {
841 	const char *q = name + strlen (p->name) + 1;
842 
843 	/* If there are more slashes, it's in a subdir, so
844 	   this match doesn't count.  */
845 	while (*q++)
846 	  if (IS_DIR_SEPARATOR (*(q-1)))
847 	    goto lose;
848 	return 1;
849 
850       lose: ;
851       }
852 
853   return 0;
854 }
855 
856 /* Return 1 if the file named NAME should be excluded from conversion.  */
857 
858 static int
file_excluded_p(const char * name)859 file_excluded_p (const char *name)
860 {
861   struct string_list *p;
862   int len = strlen (name);
863 
864   for (p = exclude_list; p; p = p->next)
865     if (!strcmp (name + len - strlen (p->name), p->name)
866 	&& IS_DIR_SEPARATOR (name[len - strlen (p->name) - 1]))
867       return 1;
868 
869   return 0;
870 }
871 
872 /* Construct a new element of a string_list.
873    STRING is the new element value, and REST holds the remaining elements.  */
874 
875 static struct string_list *
string_list_cons(const char * string,struct string_list * rest)876 string_list_cons (const char *string, struct string_list *rest)
877 {
878   struct string_list *temp = xmalloc (sizeof (struct string_list));
879 
880   temp->next = rest;
881   temp->name = string;
882   return temp;
883 }
884 
885 /* ??? The GNU convention for mentioning function args in its comments
886    is to capitalize them.  So change "hash_tab_p" to HASH_TAB_P below.
887    Likewise for all the other functions.  */
888 
889 /* Given a hash table, apply some function to each node in the table. The
890    table to traverse is given as the "hash_tab_p" argument, and the
891    function to be applied to each node in the table is given as "func"
892    argument.  */
893 
894 static void
visit_each_hash_node(const hash_table_entry * hash_tab_p,void (* func)(const hash_table_entry *))895 visit_each_hash_node (const hash_table_entry *hash_tab_p,
896 		      void (*func) (const hash_table_entry *))
897 {
898   const hash_table_entry *primary;
899 
900   for (primary = hash_tab_p; primary < &hash_tab_p[HASH_TABLE_SIZE]; primary++)
901     if (primary->symbol)
902       {
903 	hash_table_entry *second;
904 
905 	(*func)(primary);
906 	for (second = primary->hash_next; second; second = second->hash_next)
907 	  (*func) (second);
908       }
909 }
910 
911 /* Initialize all of the fields of a new hash table entry, pointed
912    to by the "p" parameter.  Note that the space to hold the entry
913    is assumed to have already been allocated before this routine is
914    called.  */
915 
916 static hash_table_entry *
add_symbol(hash_table_entry * p,const char * s)917 add_symbol (hash_table_entry *p, const char *s)
918 {
919   p->hash_next = NULL;
920   p->symbol = xstrdup (s);
921   p->ddip = NULL;
922   p->fip = NULL;
923   return p;
924 }
925 
926 /* Look for a particular function name or filename in the particular
927    hash table indicated by "hash_tab_p".  If the name is not in the
928    given hash table, add it.  Either way, return a pointer to the
929    hash table entry for the given name.  */
930 
931 static hash_table_entry *
lookup(hash_table_entry * hash_tab_p,const char * search_symbol)932 lookup (hash_table_entry *hash_tab_p, const char *search_symbol)
933 {
934   int hash_value = 0;
935   const char *search_symbol_char_p = search_symbol;
936   hash_table_entry *p;
937 
938   while (*search_symbol_char_p)
939     hash_value += *search_symbol_char_p++;
940   hash_value &= hash_mask;
941   p = &hash_tab_p[hash_value];
942   if (! p->symbol)
943       return add_symbol (p, search_symbol);
944   if (!strcmp (p->symbol, search_symbol))
945     return p;
946   while (p->hash_next)
947     {
948       p = p->hash_next;
949       if (!strcmp (p->symbol, search_symbol))
950 	return p;
951     }
952   p->hash_next = xmalloc (sizeof (hash_table_entry));
953   p = p->hash_next;
954   return add_symbol (p, search_symbol);
955 }
956 
957 /* Throw a def/dec record on the junk heap.
958 
959    Also, since we are not using this record anymore, free up all of the
960    stuff it pointed to.  */
961 
962 static void
free_def_dec(def_dec_info * p)963 free_def_dec (def_dec_info *p)
964 {
965   free ((NONCONST void *) p->ansi_decl);
966 
967 #ifndef UNPROTOIZE
968   {
969     const f_list_chain_item * curr;
970     const f_list_chain_item * next;
971 
972     for (curr = p->f_list_chain; curr; curr = next)
973       {
974 	next = curr->chain_next;
975 	free ((NONCONST void *) curr);
976       }
977   }
978 #endif /* !defined (UNPROTOIZE) */
979 
980   free (p);
981 }
982 
983 /* Unexpand as many macro symbol as we can find.
984 
985    If the given line must be unexpanded, make a copy of it in the heap and
986    return a pointer to the unexpanded copy.  Otherwise return NULL.  */
987 
988 static char *
unexpand_if_needed(const char * aux_info_line)989 unexpand_if_needed (const char *aux_info_line)
990 {
991   static char *line_buf = 0;
992   static int line_buf_size = 0;
993   const unexpansion *unexp_p;
994   int got_unexpanded = 0;
995   const char *s;
996   char *copy_p = line_buf;
997 
998   if (line_buf == 0)
999     {
1000       line_buf_size = 1024;
1001       line_buf = xmalloc (line_buf_size);
1002     }
1003 
1004   copy_p = line_buf;
1005 
1006   /* Make a copy of the input string in line_buf, expanding as necessary.  */
1007 
1008   for (s = aux_info_line; *s != '\n'; )
1009     {
1010       for (unexp_p = unexpansions; unexp_p->expanded; unexp_p++)
1011 	{
1012 	  const char *in_p = unexp_p->expanded;
1013 	  size_t len = strlen (in_p);
1014 
1015 	  if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
1016 	    {
1017 	      int size = strlen (unexp_p->contracted);
1018 	      got_unexpanded = 1;
1019 	      if (copy_p + size - line_buf >= line_buf_size)
1020 		{
1021 		  int offset = copy_p - line_buf;
1022 		  line_buf_size *= 2;
1023 		  line_buf_size += size;
1024 		  line_buf = xrealloc (line_buf, line_buf_size);
1025 		  copy_p = line_buf + offset;
1026 		}
1027 	      strcpy (copy_p, unexp_p->contracted);
1028 	      copy_p += size;
1029 
1030 	      /* Assume that there will not be another replacement required
1031 	         within the text just replaced.  */
1032 
1033 	      s += len;
1034 	      goto continue_outer;
1035 	    }
1036 	}
1037       if (copy_p - line_buf == line_buf_size)
1038 	{
1039 	  int offset = copy_p - line_buf;
1040 	  line_buf_size *= 2;
1041 	  line_buf = xrealloc (line_buf, line_buf_size);
1042 	  copy_p = line_buf + offset;
1043 	}
1044       *copy_p++ = *s++;
1045 continue_outer: ;
1046     }
1047   if (copy_p + 2 - line_buf >= line_buf_size)
1048     {
1049       int offset = copy_p - line_buf;
1050       line_buf_size *= 2;
1051       line_buf = xrealloc (line_buf, line_buf_size);
1052       copy_p = line_buf + offset;
1053     }
1054   *copy_p++ = '\n';
1055   *copy_p = '\0';
1056 
1057   return (got_unexpanded ? savestring (line_buf, copy_p - line_buf) : 0);
1058 }
1059 
1060 /* Return the absolutized filename for the given relative
1061    filename.  Note that if that filename is already absolute, it may
1062    still be returned in a modified form because this routine also
1063    eliminates redundant slashes and single dots and eliminates double
1064    dots to get a shortest possible filename from the given input
1065    filename.  The absolutization of relative filenames is made by
1066    assuming that the given filename is to be taken as relative to
1067    the first argument (cwd) or to the current directory if cwd is
1068    NULL.  */
1069 
1070 static char *
abspath(const char * cwd,const char * rel_filename)1071 abspath (const char *cwd, const char *rel_filename)
1072 {
1073   /* Setup the current working directory as needed.  */
1074   const char *const cwd2 = (cwd) ? cwd : cwd_buffer;
1075   char *const abs_buffer = alloca (strlen (cwd2) + strlen (rel_filename) + 2);
1076   char *endp = abs_buffer;
1077   char *outp, *inp;
1078 
1079   /* Copy the  filename (possibly preceded by the current working
1080      directory name) into the absolutization buffer.  */
1081 
1082   {
1083     const char *src_p;
1084 
1085     if (! IS_ABSOLUTE_PATH (rel_filename))
1086       {
1087 	src_p = cwd2;
1088 	while ((*endp++ = *src_p++))
1089 	  continue;
1090 	*(endp-1) = DIR_SEPARATOR;     		/* overwrite null */
1091       }
1092 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1093     else if (IS_DIR_SEPARATOR (rel_filename[0]))
1094       {
1095 	/* A path starting with a directory separator is considered absolute
1096 	   for dos based filesystems, but it's really not -- it's just the
1097 	   convention used throughout GCC and it works. However, in this
1098 	   case, we still need to prepend the drive spec from cwd_buffer.  */
1099 	*endp++ = cwd2[0];
1100 	*endp++ = cwd2[1];
1101       }
1102 #endif
1103     src_p = rel_filename;
1104     while ((*endp++ = *src_p++))
1105       continue;
1106   }
1107 
1108   /* Now make a copy of abs_buffer into abs_buffer, shortening the
1109      filename (by taking out slashes and dots) as we go.  */
1110 
1111   outp = inp = abs_buffer;
1112   *outp++ = *inp++;        	/* copy first slash */
1113 #if defined (apollo) || defined (_WIN32) || defined (__INTERIX)
1114   if (IS_DIR_SEPARATOR (inp[0]))
1115     *outp++ = *inp++;        	/* copy second slash */
1116 #endif
1117   for (;;)
1118     {
1119       if (!inp[0])
1120 	break;
1121       else if (IS_DIR_SEPARATOR (inp[0]) && IS_DIR_SEPARATOR (outp[-1]))
1122 	{
1123 	  inp++;
1124 	  continue;
1125 	}
1126       else if (inp[0] == '.' && IS_DIR_SEPARATOR (outp[-1]))
1127 	{
1128 	  if (!inp[1])
1129 	    break;
1130 	  else if (IS_DIR_SEPARATOR (inp[1]))
1131 	    {
1132 	      inp += 2;
1133 	      continue;
1134 	    }
1135 	  else if ((inp[1] == '.') && (inp[2] == 0
1136 	                               || IS_DIR_SEPARATOR (inp[2])))
1137 	    {
1138 	      inp += (IS_DIR_SEPARATOR (inp[2])) ? 3 : 2;
1139 	      outp -= 2;
1140 	      while (outp >= abs_buffer && ! IS_DIR_SEPARATOR (*outp))
1141 	      	outp--;
1142 	      if (outp < abs_buffer)
1143 		{
1144 		  /* Catch cases like /.. where we try to backup to a
1145 		     point above the absolute root of the logical file
1146 		     system.  */
1147 
1148 		  notice ("%s: invalid file name: %s\n",
1149 			  pname, rel_filename);
1150 		  exit (FATAL_EXIT_CODE);
1151 		}
1152 	      *++outp = '\0';
1153 	      continue;
1154 	    }
1155 	}
1156       *outp++ = *inp++;
1157     }
1158 
1159   /* On exit, make sure that there is a trailing null, and make sure that
1160      the last character of the returned string is *not* a slash.  */
1161 
1162   *outp = '\0';
1163   if (IS_DIR_SEPARATOR (outp[-1]))
1164     *--outp  = '\0';
1165 
1166   /* Make a copy (in the heap) of the stuff left in the absolutization
1167      buffer and return a pointer to the copy.  */
1168 
1169   return savestring (abs_buffer, outp - abs_buffer);
1170 }
1171 
1172 /* Given a filename (and possibly a directory name from which the filename
1173    is relative) return a string which is the shortest possible
1174    equivalent for the corresponding full (absolutized) filename.  The
1175    shortest possible equivalent may be constructed by converting the
1176    absolutized filename to be a relative filename (i.e. relative to
1177    the actual current working directory).  However if a relative filename
1178    is longer, then the full absolute filename is returned.
1179 
1180    KNOWN BUG:
1181 
1182    Note that "simple-minded" conversion of any given type of filename (either
1183    relative or absolute) may not result in a valid equivalent filename if any
1184    subpart of the original filename is actually a symbolic link.  */
1185 
1186 static const char *
shortpath(const char * cwd,const char * filename)1187 shortpath (const char *cwd, const char *filename)
1188 {
1189   char *rel_buffer;
1190   char *rel_buf_p;
1191   char *cwd_p = cwd_buffer;
1192   char *path_p;
1193   int unmatched_slash_count = 0;
1194   size_t filename_len = strlen (filename);
1195 
1196   path_p = abspath (cwd, filename);
1197   rel_buf_p = rel_buffer = xmalloc (filename_len);
1198 
1199   while (*cwd_p && IS_SAME_PATH_CHAR (*cwd_p, *path_p))
1200     {
1201       cwd_p++;
1202       path_p++;
1203     }
1204   if (!*cwd_p && (!*path_p || IS_DIR_SEPARATOR (*path_p)))
1205     {
1206       /* whole pwd matched */
1207       if (!*path_p)        	/* input *is* the current path! */
1208 	return ".";
1209       else
1210 	return ++path_p;
1211     }
1212   else
1213     {
1214       if (*path_p)
1215 	{
1216 	  --cwd_p;
1217 	  --path_p;
1218 	  while (! IS_DIR_SEPARATOR (*cwd_p))     /* backup to last slash */
1219 	    {
1220 	      --cwd_p;
1221 	      --path_p;
1222 	    }
1223 	  cwd_p++;
1224 	  path_p++;
1225 	  unmatched_slash_count++;
1226 	}
1227 
1228       /* Find out how many directory levels in cwd were *not* matched.  */
1229       while (*cwd_p++)
1230 	if (IS_DIR_SEPARATOR (*(cwd_p-1)))
1231 	  unmatched_slash_count++;
1232 
1233       /* Now we know how long the "short name" will be.
1234 	 Reject it if longer than the input.  */
1235       if (unmatched_slash_count * 3 + strlen (path_p) >= filename_len)
1236 	return filename;
1237 
1238       /* For each of them, put a `../' at the beginning of the short name.  */
1239       while (unmatched_slash_count--)
1240 	{
1241 	  /* Give up if the result gets to be longer
1242 	     than the absolute path name.  */
1243 	  if (rel_buffer + filename_len <= rel_buf_p + 3)
1244 	    return filename;
1245 	  *rel_buf_p++ = '.';
1246 	  *rel_buf_p++ = '.';
1247 	  *rel_buf_p++ = DIR_SEPARATOR;
1248 	}
1249 
1250       /* Then tack on the unmatched part of the desired file's name.  */
1251       do
1252 	{
1253 	  if (rel_buffer + filename_len <= rel_buf_p)
1254 	    return filename;
1255 	}
1256       while ((*rel_buf_p++ = *path_p++));
1257 
1258       --rel_buf_p;
1259       if (IS_DIR_SEPARATOR (*(rel_buf_p-1)))
1260 	*--rel_buf_p = '\0';
1261       return rel_buffer;
1262     }
1263 }
1264 
1265 /* Lookup the given filename in the hash table for filenames.  If it is a
1266    new one, then the hash table info pointer will be null.  In this case,
1267    we create a new file_info record to go with the filename, and we initialize
1268    that record with some reasonable values.  */
1269 
1270 /* FILENAME was const, but that causes a warning on AIX when calling stat.
1271    That is probably a bug in AIX, but might as well avoid the warning.  */
1272 
1273 static file_info *
find_file(const char * filename,int do_not_stat)1274 find_file (const char *filename, int do_not_stat)
1275 {
1276   hash_table_entry *hash_entry_p;
1277 
1278   hash_entry_p = lookup (filename_primary, filename);
1279   if (hash_entry_p->fip)
1280     return hash_entry_p->fip;
1281   else
1282     {
1283       struct stat stat_buf;
1284       file_info *file_p = xmalloc (sizeof (file_info));
1285 
1286       /* If we cannot get status on any given source file, give a warning
1287 	 and then just set its time of last modification to infinity.  */
1288 
1289       if (do_not_stat)
1290 	stat_buf.st_mtime = (time_t) 0;
1291       else
1292 	{
1293 	  if (stat (filename, &stat_buf) == -1)
1294 	    {
1295 	      int errno_val = errno;
1296 	      notice ("%s: %s: can't get status: %s\n",
1297 		      pname, shortpath (NULL, filename),
1298 		      xstrerror (errno_val));
1299 	      stat_buf.st_mtime = (time_t) -1;
1300 	    }
1301 	}
1302 
1303       hash_entry_p->fip = file_p;
1304       file_p->hash_entry = hash_entry_p;
1305       file_p->defs_decs = NULL;
1306       file_p->mtime = stat_buf.st_mtime;
1307       return file_p;
1308     }
1309 }
1310 
1311 /* Generate a fatal error because some part of the aux_info file is
1312    messed up.  */
1313 
1314 static void
aux_info_corrupted(void)1315 aux_info_corrupted (void)
1316 {
1317   notice ("\n%s: fatal error: aux info file corrupted at line %d\n",
1318 	  pname, current_aux_info_lineno);
1319   exit (FATAL_EXIT_CODE);
1320 }
1321 
1322 /* ??? This comment is vague.  Say what the condition is for.  */
1323 /* Check to see that a condition is true.  This is kind of like an assert.  */
1324 
1325 static void
check_aux_info(int cond)1326 check_aux_info (int cond)
1327 {
1328   if (! cond)
1329     aux_info_corrupted ();
1330 }
1331 
1332 /* Given a pointer to the closing right parenthesis for a particular formals
1333    list (in an aux_info file) find the corresponding left parenthesis and
1334    return a pointer to it.  */
1335 
1336 static const char *
find_corresponding_lparen(const char * p)1337 find_corresponding_lparen (const char *p)
1338 {
1339   const char *q;
1340   int paren_depth;
1341 
1342   for (paren_depth = 1, q = p-1; paren_depth; q--)
1343     {
1344       switch (*q)
1345 	{
1346 	case ')':
1347 	  paren_depth++;
1348 	  break;
1349 	case '(':
1350 	  paren_depth--;
1351 	  break;
1352 	}
1353     }
1354   return ++q;
1355 }
1356 
1357 /* Given a line from  an aux info file, and a time at which the aux info
1358    file it came from was created, check to see if the item described in
1359    the line comes from a file which has been modified since the aux info
1360    file was created.  If so, return nonzero, else return zero.  */
1361 
1362 static int
referenced_file_is_newer(const char * l,time_t aux_info_mtime)1363 referenced_file_is_newer (const char *l, time_t aux_info_mtime)
1364 {
1365   const char *p;
1366   file_info *fi_p;
1367   char *filename;
1368 
1369   check_aux_info (l[0] == '/');
1370   check_aux_info (l[1] == '*');
1371   check_aux_info (l[2] == ' ');
1372 
1373   {
1374     const char *filename_start = p = l + 3;
1375 
1376     while (*p != ':'
1377 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1378 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1379 #endif
1380 	   )
1381       p++;
1382     filename = alloca ((size_t) (p - filename_start) + 1);
1383     strncpy (filename, filename_start, (size_t) (p - filename_start));
1384     filename[p-filename_start] = '\0';
1385   }
1386 
1387   /* Call find_file to find the file_info record associated with the file
1388      which contained this particular def or dec item.  Note that this call
1389      may cause a new file_info record to be created if this is the first time
1390      that we have ever known about this particular file.  */
1391 
1392   fi_p = find_file (abspath (invocation_filename, filename), 0);
1393 
1394   return (fi_p->mtime > aux_info_mtime);
1395 }
1396 
1397 /* Given a line of info from the aux_info file, create a new
1398    def_dec_info record to remember all of the important information about
1399    a function definition or declaration.
1400 
1401    Link this record onto the list of such records for the particular file in
1402    which it occurred in proper (descending) line number order (for now).
1403 
1404    If there is an identical record already on the list for the file, throw
1405    this one away.  Doing so takes care of the (useless and troublesome)
1406    duplicates which are bound to crop up due to multiple inclusions of any
1407    given individual header file.
1408 
1409    Finally, link the new def_dec record onto the list of such records
1410    pertaining to this particular function name.  */
1411 
1412 static void
save_def_or_dec(const char * l,int is_syscalls)1413 save_def_or_dec (const char *l, int is_syscalls)
1414 {
1415   const char *p;
1416   const char *semicolon_p;
1417   def_dec_info *def_dec_p = xmalloc (sizeof (def_dec_info));
1418 
1419 #ifndef UNPROTOIZE
1420   def_dec_p->written = 0;
1421 #endif /* !defined (UNPROTOIZE) */
1422 
1423   /* Start processing the line by picking off 5 pieces of information from
1424      the left hand end of the line.  These are filename, line number,
1425      new/old/implicit flag (new = ANSI prototype format), definition or
1426      declaration flag, and extern/static flag).  */
1427 
1428   check_aux_info (l[0] == '/');
1429   check_aux_info (l[1] == '*');
1430   check_aux_info (l[2] == ' ');
1431 
1432   {
1433     const char *filename_start = p = l + 3;
1434     char *filename;
1435 
1436     while (*p != ':'
1437 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1438 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1439 #endif
1440 	   )
1441       p++;
1442     filename = alloca ((size_t) (p - filename_start) + 1);
1443     strncpy (filename, filename_start, (size_t) (p - filename_start));
1444     filename[p-filename_start] = '\0';
1445 
1446     /* Call find_file to find the file_info record associated with the file
1447        which contained this particular def or dec item.  Note that this call
1448        may cause a new file_info record to be created if this is the first time
1449        that we have ever known about this particular file.
1450 
1451        Note that we started out by forcing all of the base source file names
1452        (i.e. the names of the aux_info files with the .X stripped off) into the
1453        filenames hash table, and we simultaneously setup file_info records for
1454        all of these base file names (even if they may be useless later).
1455        The file_info records for all of these "base" file names (properly)
1456        act as file_info records for the "original" (i.e. un-included) files
1457        which were submitted to gcc for compilation (when the -aux-info
1458        option was used).  */
1459 
1460     def_dec_p->file = find_file (abspath (invocation_filename, filename), is_syscalls);
1461   }
1462 
1463   {
1464     const char *line_number_start = ++p;
1465     char line_number[10];
1466 
1467     while (*p != ':'
1468 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1469 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1470 #endif
1471 	   )
1472       p++;
1473     strncpy (line_number, line_number_start, (size_t) (p - line_number_start));
1474     line_number[p-line_number_start] = '\0';
1475     def_dec_p->line = atoi (line_number);
1476   }
1477 
1478   /* Check that this record describes a new-style, old-style, or implicit
1479      definition or declaration.  */
1480 
1481   p++;	/* Skip over the `:'.  */
1482   check_aux_info ((*p == 'N') || (*p == 'O') || (*p == 'I'));
1483 
1484   /* Is this a new style (ANSI prototyped) definition or declaration? */
1485 
1486   def_dec_p->prototyped = (*p == 'N');
1487 
1488 #ifndef UNPROTOIZE
1489 
1490   /* Is this an implicit declaration? */
1491 
1492   def_dec_p->is_implicit = (*p == 'I');
1493 
1494 #endif /* !defined (UNPROTOIZE) */
1495 
1496   p++;
1497 
1498   check_aux_info ((*p == 'C') || (*p == 'F'));
1499 
1500   /* Is this item a function definition (F) or a declaration (C).  Note that
1501      we treat item taken from the syscalls file as though they were function
1502      definitions regardless of what the stuff in the file says.  */
1503 
1504   def_dec_p->is_func_def = ((*p++ == 'F') || is_syscalls);
1505 
1506 #ifndef UNPROTOIZE
1507   def_dec_p->definition = 0;	/* Fill this in later if protoizing.  */
1508 #endif /* !defined (UNPROTOIZE) */
1509 
1510   check_aux_info (*p++ == ' ');
1511   check_aux_info (*p++ == '*');
1512   check_aux_info (*p++ == '/');
1513   check_aux_info (*p++ == ' ');
1514 
1515 #ifdef UNPROTOIZE
1516   check_aux_info ((!strncmp (p, "static", 6)) || (!strncmp (p, "extern", 6)));
1517 #else /* !defined (UNPROTOIZE) */
1518   if (!strncmp (p, "static", 6))
1519     def_dec_p->is_static = -1;
1520   else if (!strncmp (p, "extern", 6))
1521     def_dec_p->is_static = 0;
1522   else
1523     check_aux_info (0);	/* Didn't find either `extern' or `static'.  */
1524 #endif /* !defined (UNPROTOIZE) */
1525 
1526   {
1527     const char *ansi_start = p;
1528 
1529     p += 6;	/* Pass over the "static" or "extern".  */
1530 
1531     /* We are now past the initial stuff.  Search forward from here to find
1532        the terminating semicolon that should immediately follow the entire
1533        ANSI format function declaration.  */
1534 
1535     while (*++p != ';')
1536       continue;
1537 
1538     semicolon_p = p;
1539 
1540     /* Make a copy of the ansi declaration part of the line from the aux_info
1541        file.  */
1542 
1543     def_dec_p->ansi_decl
1544       = dupnstr (ansi_start, (size_t) ((semicolon_p+1) - ansi_start));
1545 
1546     /* Backup and point at the final right paren of the final argument list.  */
1547 
1548     p--;
1549 
1550 #ifndef UNPROTOIZE
1551     def_dec_p->f_list_chain = NULL;
1552 #endif /* !defined (UNPROTOIZE) */
1553 
1554     while (p != ansi_start && (p[-1] == ' ' || p[-1] == '\t')) p--;
1555     if (*p != ')')
1556       {
1557 	free_def_dec (def_dec_p);
1558 	return;
1559       }
1560   }
1561 
1562   /* Now isolate a whole set of formal argument lists, one-by-one.  Normally,
1563      there will only be one list to isolate, but there could be more.  */
1564 
1565   def_dec_p->f_list_count = 0;
1566 
1567   for (;;)
1568     {
1569       const char *left_paren_p = find_corresponding_lparen (p);
1570 #ifndef UNPROTOIZE
1571       {
1572 	f_list_chain_item *cip = xmalloc (sizeof (f_list_chain_item));
1573 
1574 	cip->formals_list
1575 	  = dupnstr (left_paren_p + 1, (size_t) (p - (left_paren_p+1)));
1576 
1577 	/* Add the new chain item at the head of the current list.  */
1578 
1579 	cip->chain_next = def_dec_p->f_list_chain;
1580 	def_dec_p->f_list_chain = cip;
1581       }
1582 #endif /* !defined (UNPROTOIZE) */
1583       def_dec_p->f_list_count++;
1584 
1585       p = left_paren_p - 2;
1586 
1587       /* p must now point either to another right paren, or to the last
1588 	 character of the name of the function that was declared/defined.
1589 	 If p points to another right paren, then this indicates that we
1590 	 are dealing with multiple formals lists.  In that case, there
1591 	 really should be another right paren preceding this right paren.  */
1592 
1593       if (*p != ')')
1594 	break;
1595       else
1596 	check_aux_info (*--p == ')');
1597     }
1598 
1599 
1600   {
1601     const char *past_fn = p + 1;
1602 
1603     check_aux_info (*past_fn == ' ');
1604 
1605     /* Scan leftwards over the identifier that names the function.  */
1606 
1607     while (is_id_char (*p))
1608       p--;
1609     p++;
1610 
1611     /* p now points to the leftmost character of the function name.  */
1612 
1613     {
1614       char *fn_string = alloca (past_fn - p + 1);
1615 
1616       strncpy (fn_string, p, (size_t) (past_fn - p));
1617       fn_string[past_fn-p] = '\0';
1618       def_dec_p->hash_entry = lookup (function_name_primary, fn_string);
1619     }
1620   }
1621 
1622   /* Look at all of the defs and decs for this function name that we have
1623      collected so far.  If there is already one which is at the same
1624      line number in the same file, then we can discard this new def_dec_info
1625      record.
1626 
1627      As an extra assurance that any such pair of (nominally) identical
1628      function declarations are in fact identical, we also compare the
1629      ansi_decl parts of the lines from the aux_info files just to be on
1630      the safe side.
1631 
1632      This comparison will fail if (for instance) the user was playing
1633      messy games with the preprocessor which ultimately causes one
1634      function declaration in one header file to look differently when
1635      that file is included by two (or more) other files.  */
1636 
1637   {
1638     const def_dec_info *other;
1639 
1640     for (other = def_dec_p->hash_entry->ddip; other; other = other->next_for_func)
1641       {
1642 	if (def_dec_p->line == other->line && def_dec_p->file == other->file)
1643 	  {
1644 	    if (strcmp (def_dec_p->ansi_decl, other->ansi_decl))
1645 	      {
1646 	        notice ("%s:%d: declaration of function `%s' takes different forms\n",
1647 			def_dec_p->file->hash_entry->symbol,
1648 			def_dec_p->line,
1649 			def_dec_p->hash_entry->symbol);
1650 	        exit (FATAL_EXIT_CODE);
1651 	      }
1652 	    free_def_dec (def_dec_p);
1653 	    return;
1654 	  }
1655       }
1656   }
1657 
1658 #ifdef UNPROTOIZE
1659 
1660   /* If we are doing unprotoizing, we must now setup the pointers that will
1661      point to the K&R name list and to the K&R argument declarations list.
1662 
1663      Note that if this is only a function declaration, then we should not
1664      expect to find any K&R style formals list following the ANSI-style
1665      formals list.  This is because GCC knows that such information is
1666      useless in the case of function declarations (function definitions
1667      are a different story however).
1668 
1669      Since we are unprotoizing, we don't need any such lists anyway.
1670      All we plan to do is to delete all characters between ()'s in any
1671      case.  */
1672 
1673   def_dec_p->formal_names = NULL;
1674   def_dec_p->formal_decls = NULL;
1675 
1676   if (def_dec_p->is_func_def)
1677     {
1678       p = semicolon_p;
1679       check_aux_info (*++p == ' ');
1680       check_aux_info (*++p == '/');
1681       check_aux_info (*++p == '*');
1682       check_aux_info (*++p == ' ');
1683       check_aux_info (*++p == '(');
1684 
1685       {
1686 	const char *kr_names_start = ++p;   /* Point just inside '('.  */
1687 
1688 	while (*p++ != ')')
1689 	  continue;
1690 	p--;		/* point to closing right paren */
1691 
1692 	/* Make a copy of the K&R parameter names list.  */
1693 
1694 	def_dec_p->formal_names
1695 	  = dupnstr (kr_names_start, (size_t) (p - kr_names_start));
1696       }
1697 
1698       check_aux_info (*++p == ' ');
1699       p++;
1700 
1701       /* p now points to the first character of the K&R style declarations
1702 	 list (if there is one) or to the star-slash combination that ends
1703 	 the comment in which such lists get embedded.  */
1704 
1705       /* Make a copy of the K&R formal decls list and set the def_dec record
1706 	 to point to it.  */
1707 
1708       if (*p == '*')		/* Are there no K&R declarations? */
1709 	{
1710 	  check_aux_info (*++p == '/');
1711 	  def_dec_p->formal_decls = "";
1712 	}
1713       else
1714 	{
1715 	  const char *kr_decls_start = p;
1716 
1717 	  while (p[0] != '*' || p[1] != '/')
1718 	    p++;
1719 	  p--;
1720 
1721 	  check_aux_info (*p == ' ');
1722 
1723 	  def_dec_p->formal_decls
1724 	    = dupnstr (kr_decls_start, (size_t) (p - kr_decls_start));
1725 	}
1726 
1727       /* Handle a special case.  If we have a function definition marked as
1728 	 being in "old" style, and if its formal names list is empty, then
1729 	 it may actually have the string "void" in its real formals list
1730 	 in the original source code.  Just to make sure, we will get setup
1731 	 to convert such things anyway.
1732 
1733 	 This kludge only needs to be here because of an insurmountable
1734 	 problem with generating .X files.  */
1735 
1736       if (!def_dec_p->prototyped && !*def_dec_p->formal_names)
1737 	def_dec_p->prototyped = 1;
1738     }
1739 
1740   /* Since we are unprotoizing, if this item is already in old (K&R) style,
1741      we can just ignore it.  If that is true, throw away the itme now.  */
1742 
1743   if (!def_dec_p->prototyped)
1744     {
1745       free_def_dec (def_dec_p);
1746       return;
1747     }
1748 
1749 #endif /* defined (UNPROTOIZE) */
1750 
1751   /* Add this record to the head of the list of records pertaining to this
1752      particular function name.  */
1753 
1754   def_dec_p->next_for_func = def_dec_p->hash_entry->ddip;
1755   def_dec_p->hash_entry->ddip = def_dec_p;
1756 
1757   /* Add this new def_dec_info record to the sorted list of def_dec_info
1758      records for this file.  Note that we don't have to worry about duplicates
1759      (caused by multiple inclusions of header files) here because we have
1760      already eliminated duplicates above.  */
1761 
1762   if (!def_dec_p->file->defs_decs)
1763     {
1764       def_dec_p->file->defs_decs = def_dec_p;
1765       def_dec_p->next_in_file = NULL;
1766     }
1767   else
1768     {
1769       int line = def_dec_p->line;
1770       const def_dec_info *prev = NULL;
1771       const def_dec_info *curr = def_dec_p->file->defs_decs;
1772       const def_dec_info *next = curr->next_in_file;
1773 
1774       while (next && (line < curr->line))
1775 	{
1776 	  prev = curr;
1777 	  curr = next;
1778 	  next = next->next_in_file;
1779 	}
1780       if (line >= curr->line)
1781 	{
1782 	  def_dec_p->next_in_file = curr;
1783 	  if (prev)
1784 	    ((NONCONST def_dec_info *) prev)->next_in_file = def_dec_p;
1785 	  else
1786 	    def_dec_p->file->defs_decs = def_dec_p;
1787 	}
1788       else	/* assert (next == NULL); */
1789 	{
1790 	  ((NONCONST def_dec_info *) curr)->next_in_file = def_dec_p;
1791 	  /* assert (next == NULL); */
1792 	  def_dec_p->next_in_file = next;
1793 	}
1794     }
1795 }
1796 
1797 /* Set up the vector COMPILE_PARAMS which is the argument list for running GCC.
1798    Also set input_file_name_index and aux_info_file_name_index
1799    to the indices of the slots where the file names should go.  */
1800 
1801 /* We initialize the vector by  removing -g, -O, -S, -c, and -o options,
1802    and adding '-aux-info AUXFILE -S  -o /dev/null INFILE' at the end.  */
1803 
1804 static void
munge_compile_params(const char * params_list)1805 munge_compile_params (const char *params_list)
1806 {
1807   /* Build up the contents in a temporary vector
1808      that is so big that to has to be big enough.  */
1809   const char **temp_params
1810     = alloca ((strlen (params_list) + 8) * sizeof (char *));
1811   int param_count = 0;
1812   const char *param;
1813   struct stat st;
1814 
1815   temp_params[param_count++] = compiler_file_name;
1816   for (;;)
1817     {
1818       while (ISSPACE ((const unsigned char)*params_list))
1819 	params_list++;
1820       if (!*params_list)
1821 	break;
1822       param = params_list;
1823       while (*params_list && !ISSPACE ((const unsigned char)*params_list))
1824 	params_list++;
1825       if (param[0] != '-')
1826 	temp_params[param_count++]
1827 	  = dupnstr (param, (size_t) (params_list - param));
1828       else
1829 	{
1830 	  switch (param[1])
1831 	    {
1832 	    case 'g':
1833 	    case 'O':
1834 	    case 'S':
1835 	    case 'c':
1836 	      break;		/* Don't copy these.  */
1837 	    case 'o':
1838 	      while (ISSPACE ((const unsigned char)*params_list))
1839 		params_list++;
1840 	      while (*params_list
1841 		     && !ISSPACE ((const unsigned char)*params_list))
1842 		params_list++;
1843 	      break;
1844 	    default:
1845 	      temp_params[param_count++]
1846 		= dupnstr (param, (size_t) (params_list - param));
1847 	    }
1848 	}
1849       if (!*params_list)
1850 	break;
1851     }
1852   temp_params[param_count++] = "-aux-info";
1853 
1854   /* Leave room for the aux-info file name argument.  */
1855   aux_info_file_name_index = param_count;
1856   temp_params[param_count++] = NULL;
1857 
1858   temp_params[param_count++] = "-S";
1859   temp_params[param_count++] = "-o";
1860 
1861   if ((stat (HOST_BIT_BUCKET, &st) == 0)
1862       && (!S_ISDIR (st.st_mode))
1863       && (access (HOST_BIT_BUCKET, W_OK) == 0))
1864     temp_params[param_count++] = HOST_BIT_BUCKET;
1865   else
1866     /* FIXME: This is hardly likely to be right, if HOST_BIT_BUCKET is not
1867        writable.  But until this is rejigged to use make_temp_file(), this
1868        is the best we can do.  */
1869     temp_params[param_count++] = "/dev/null";
1870 
1871   /* Leave room for the input file name argument.  */
1872   input_file_name_index = param_count;
1873   temp_params[param_count++] = NULL;
1874   /* Terminate the list.  */
1875   temp_params[param_count++] = NULL;
1876 
1877   /* Make a copy of the compile_params in heap space.  */
1878 
1879   compile_params = xmalloc (sizeof (char *) * (param_count+1));
1880   memcpy (compile_params, temp_params, sizeof (char *) * param_count);
1881 }
1882 
1883 /* Do a recompilation for the express purpose of generating a new aux_info
1884    file to go with a specific base source file.
1885 
1886    The result is a boolean indicating success.  */
1887 
1888 static int
gen_aux_info_file(const char * base_filename)1889 gen_aux_info_file (const char *base_filename)
1890 {
1891   if (!input_file_name_index)
1892     munge_compile_params ("");
1893 
1894   /* Store the full source file name in the argument vector.  */
1895   compile_params[input_file_name_index] = shortpath (NULL, base_filename);
1896   /* Add .X to source file name to get aux-info file name.  */
1897   compile_params[aux_info_file_name_index] =
1898     concat (compile_params[input_file_name_index], aux_info_suffix, NULL);
1899 
1900   if (!quiet_flag)
1901     notice ("%s: compiling `%s'\n",
1902 	    pname, compile_params[input_file_name_index]);
1903 
1904   {
1905     char *errmsg_fmt, *errmsg_arg;
1906     int wait_status, pid;
1907 
1908     pid = pexecute (compile_params[0], (char * const *) compile_params,
1909 		    pname, NULL, &errmsg_fmt, &errmsg_arg,
1910 		    PEXECUTE_FIRST | PEXECUTE_LAST | PEXECUTE_SEARCH);
1911 
1912     if (pid == -1)
1913       {
1914 	int errno_val = errno;
1915 	fprintf (stderr, "%s: ", pname);
1916 	fprintf (stderr, errmsg_fmt, errmsg_arg);
1917 	fprintf (stderr, ": %s\n", xstrerror (errno_val));
1918 	return 0;
1919       }
1920 
1921     pid = pwait (pid, &wait_status, 0);
1922     if (pid == -1)
1923       {
1924 	notice ("%s: wait: %s\n", pname, xstrerror (errno));
1925 	return 0;
1926       }
1927     if (WIFSIGNALED (wait_status))
1928       {
1929 	notice ("%s: subprocess got fatal signal %d\n",
1930 		pname, WTERMSIG (wait_status));
1931 	return 0;
1932       }
1933     if (WIFEXITED (wait_status))
1934       {
1935 	if (WEXITSTATUS (wait_status) != 0)
1936 	  {
1937 	    notice ("%s: %s exited with status %d\n",
1938 		    pname, compile_params[0], WEXITSTATUS (wait_status));
1939 	    return 0;
1940 	  }
1941 	return 1;
1942       }
1943     abort ();
1944   }
1945 }
1946 
1947 /* Read in all of the information contained in a single aux_info file.
1948    Save all of the important stuff for later.  */
1949 
1950 static void
process_aux_info_file(const char * base_source_filename,int keep_it,int is_syscalls)1951 process_aux_info_file (const char *base_source_filename, int keep_it,
1952 		       int is_syscalls)
1953 {
1954   size_t base_len = strlen (base_source_filename);
1955   char * aux_info_filename = alloca (base_len + strlen (aux_info_suffix) + 1);
1956   char *aux_info_base;
1957   char *aux_info_limit;
1958   char *aux_info_relocated_name;
1959   const char *aux_info_second_line;
1960   time_t aux_info_mtime;
1961   size_t aux_info_size;
1962   int must_create;
1963 
1964   /* Construct the aux_info filename from the base source filename.  */
1965 
1966   strcpy (aux_info_filename, base_source_filename);
1967   strcat (aux_info_filename, aux_info_suffix);
1968 
1969   /* Check that the aux_info file exists and is readable.  If it does not
1970      exist, try to create it (once only).  */
1971 
1972   /* If file doesn't exist, set must_create.
1973      Likewise if it exists and we can read it but it is obsolete.
1974      Otherwise, report an error.  */
1975   must_create = 0;
1976 
1977   /* Come here with must_create set to 1 if file is out of date.  */
1978 start_over: ;
1979 
1980   if (access (aux_info_filename, R_OK) == -1)
1981     {
1982       if (errno == ENOENT)
1983 	{
1984 	  if (is_syscalls)
1985 	    {
1986 	      notice ("%s: warning: missing SYSCALLS file `%s'\n",
1987 		      pname, aux_info_filename);
1988 	      return;
1989 	    }
1990 	  must_create = 1;
1991 	}
1992       else
1993 	{
1994 	  int errno_val = errno;
1995 	  notice ("%s: can't read aux info file `%s': %s\n",
1996 		  pname, shortpath (NULL, aux_info_filename),
1997 		  xstrerror (errno_val));
1998 	  errors++;
1999 	  return;
2000 	}
2001     }
2002 #if 0 /* There is code farther down to take care of this.  */
2003   else
2004     {
2005       struct stat s1, s2;
2006       stat (aux_info_file_name, &s1);
2007       stat (base_source_file_name, &s2);
2008       if (s2.st_mtime > s1.st_mtime)
2009 	must_create = 1;
2010     }
2011 #endif /* 0 */
2012 
2013   /* If we need a .X file, create it, and verify we can read it.  */
2014   if (must_create)
2015     {
2016       if (!gen_aux_info_file (base_source_filename))
2017 	{
2018 	  errors++;
2019 	  return;
2020 	}
2021       if (access (aux_info_filename, R_OK) == -1)
2022 	{
2023 	  int errno_val = errno;
2024 	  notice ("%s: can't read aux info file `%s': %s\n",
2025 		  pname, shortpath (NULL, aux_info_filename),
2026 		  xstrerror (errno_val));
2027 	  errors++;
2028 	  return;
2029 	}
2030     }
2031 
2032   {
2033     struct stat stat_buf;
2034 
2035     /* Get some status information about this aux_info file.  */
2036 
2037     if (stat (aux_info_filename, &stat_buf) == -1)
2038       {
2039 	int errno_val = errno;
2040 	notice ("%s: can't get status of aux info file `%s': %s\n",
2041 		pname, shortpath (NULL, aux_info_filename),
2042 		xstrerror (errno_val));
2043 	errors++;
2044 	return;
2045       }
2046 
2047     /* Check on whether or not this aux_info file is zero length.  If it is,
2048        then just ignore it and return.  */
2049 
2050     if ((aux_info_size = stat_buf.st_size) == 0)
2051       return;
2052 
2053     /* Get the date/time of last modification for this aux_info file and
2054        remember it.  We will have to check that any source files that it
2055        contains information about are at least this old or older.  */
2056 
2057     aux_info_mtime = stat_buf.st_mtime;
2058 
2059     if (!is_syscalls)
2060       {
2061 	/* Compare mod time with the .c file; update .X file if obsolete.
2062 	   The code later on can fail to check the .c file
2063 	   if it did not directly define any functions.  */
2064 
2065 	if (stat (base_source_filename, &stat_buf) == -1)
2066 	  {
2067 	    int errno_val = errno;
2068 	    notice ("%s: can't get status of aux info file `%s': %s\n",
2069 		    pname, shortpath (NULL, base_source_filename),
2070 		    xstrerror (errno_val));
2071 	    errors++;
2072 	    return;
2073 	  }
2074 	if (stat_buf.st_mtime > aux_info_mtime)
2075 	  {
2076 	    must_create = 1;
2077 	    goto start_over;
2078 	  }
2079       }
2080   }
2081 
2082   {
2083     int aux_info_file;
2084     int fd_flags;
2085 
2086     /* Open the aux_info file.  */
2087 
2088     fd_flags = O_RDONLY;
2089 #ifdef O_BINARY
2090     /* Use binary mode to avoid having to deal with different EOL characters.  */
2091     fd_flags |= O_BINARY;
2092 #endif
2093     if ((aux_info_file = open (aux_info_filename, fd_flags, 0444 )) == -1)
2094       {
2095 	int errno_val = errno;
2096 	notice ("%s: can't open aux info file `%s' for reading: %s\n",
2097 		pname, shortpath (NULL, aux_info_filename),
2098 		xstrerror (errno_val));
2099 	return;
2100       }
2101 
2102     /* Allocate space to hold the aux_info file in memory.  */
2103 
2104     aux_info_base = xmalloc (aux_info_size + 1);
2105     aux_info_limit = aux_info_base + aux_info_size;
2106     *aux_info_limit = '\0';
2107 
2108     /* Read the aux_info file into memory.  */
2109 
2110     if (safe_read (aux_info_file, aux_info_base, aux_info_size) !=
2111 	(int) aux_info_size)
2112       {
2113 	int errno_val = errno;
2114 	notice ("%s: error reading aux info file `%s': %s\n",
2115 		pname, shortpath (NULL, aux_info_filename),
2116 		xstrerror (errno_val));
2117 	free (aux_info_base);
2118 	close (aux_info_file);
2119 	return;
2120       }
2121 
2122     /* Close the aux info file.  */
2123 
2124     if (close (aux_info_file))
2125       {
2126 	int errno_val = errno;
2127 	notice ("%s: error closing aux info file `%s': %s\n",
2128 		pname, shortpath (NULL, aux_info_filename),
2129 		xstrerror (errno_val));
2130 	free (aux_info_base);
2131 	close (aux_info_file);
2132 	return;
2133       }
2134   }
2135 
2136   /* Delete the aux_info file (unless requested not to).  If the deletion
2137      fails for some reason, don't even worry about it.  */
2138 
2139   if (must_create && !keep_it)
2140     if (unlink (aux_info_filename) == -1)
2141       {
2142 	int errno_val = errno;
2143 	notice ("%s: can't delete aux info file `%s': %s\n",
2144 		pname, shortpath (NULL, aux_info_filename),
2145 		xstrerror (errno_val));
2146       }
2147 
2148   /* Save a pointer into the first line of the aux_info file which
2149      contains the filename of the directory from which the compiler
2150      was invoked when the associated source file was compiled.
2151      This information is used later to help create complete
2152      filenames out of the (potentially) relative filenames in
2153      the aux_info file.  */
2154 
2155   {
2156     char *p = aux_info_base;
2157 
2158     while (*p != ':'
2159 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
2160 	   || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
2161 #endif
2162 	   )
2163       p++;
2164     p++;
2165     while (*p == ' ')
2166       p++;
2167     invocation_filename = p;	/* Save a pointer to first byte of path.  */
2168     while (*p != ' ')
2169       p++;
2170     *p++ = DIR_SEPARATOR;
2171     *p++ = '\0';
2172     while (*p++ != '\n')
2173       continue;
2174     aux_info_second_line = p;
2175     aux_info_relocated_name = 0;
2176     if (! IS_ABSOLUTE_PATH (invocation_filename))
2177       {
2178 	/* INVOCATION_FILENAME is relative;
2179 	   append it to BASE_SOURCE_FILENAME's dir.  */
2180 	char *dir_end;
2181 	aux_info_relocated_name = xmalloc (base_len + (p-invocation_filename));
2182 	strcpy (aux_info_relocated_name, base_source_filename);
2183 	dir_end = strrchr (aux_info_relocated_name, DIR_SEPARATOR);
2184 #ifdef DIR_SEPARATOR_2
2185 	{
2186 	  char *slash;
2187 
2188 	  slash = strrchr (dir_end ? dir_end : aux_info_relocated_name,
2189 			   DIR_SEPARATOR_2);
2190 	  if (slash)
2191 	    dir_end = slash;
2192 	}
2193 #endif
2194 	if (dir_end)
2195 	  dir_end++;
2196 	else
2197 	  dir_end = aux_info_relocated_name;
2198 	strcpy (dir_end, invocation_filename);
2199 	invocation_filename = aux_info_relocated_name;
2200       }
2201   }
2202 
2203 
2204   {
2205     const char *aux_info_p;
2206 
2207     /* Do a pre-pass on the lines in the aux_info file, making sure that all
2208        of the source files referenced in there are at least as old as this
2209        aux_info file itself.  If not, go back and regenerate the aux_info
2210        file anew.  Don't do any of this for the syscalls file.  */
2211 
2212     if (!is_syscalls)
2213       {
2214 	current_aux_info_lineno = 2;
2215 
2216 	for (aux_info_p = aux_info_second_line; *aux_info_p; )
2217 	  {
2218 	    if (referenced_file_is_newer (aux_info_p, aux_info_mtime))
2219 	      {
2220 		free (aux_info_base);
2221 		free (aux_info_relocated_name);
2222 		if (keep_it && unlink (aux_info_filename) == -1)
2223 		  {
2224 		    int errno_val = errno;
2225 	            notice ("%s: can't delete file `%s': %s\n",
2226 			    pname, shortpath (NULL, aux_info_filename),
2227 			    xstrerror (errno_val));
2228 	            return;
2229 	          }
2230 		must_create = 1;
2231 	        goto start_over;
2232 	      }
2233 
2234 	    /* Skip over the rest of this line to start of next line.  */
2235 
2236 	    while (*aux_info_p != '\n')
2237 	      aux_info_p++;
2238 	    aux_info_p++;
2239 	    current_aux_info_lineno++;
2240 	  }
2241       }
2242 
2243     /* Now do the real pass on the aux_info lines.  Save their information in
2244        the in-core data base.  */
2245 
2246     current_aux_info_lineno = 2;
2247 
2248     for (aux_info_p = aux_info_second_line; *aux_info_p;)
2249       {
2250 	char *unexpanded_line = unexpand_if_needed (aux_info_p);
2251 
2252 	if (unexpanded_line)
2253 	  {
2254 	    save_def_or_dec (unexpanded_line, is_syscalls);
2255 	    free (unexpanded_line);
2256 	  }
2257 	else
2258 	  save_def_or_dec (aux_info_p, is_syscalls);
2259 
2260 	/* Skip over the rest of this line and get to start of next line.  */
2261 
2262 	while (*aux_info_p != '\n')
2263 	  aux_info_p++;
2264 	aux_info_p++;
2265 	current_aux_info_lineno++;
2266       }
2267   }
2268 
2269   free (aux_info_base);
2270   free (aux_info_relocated_name);
2271 }
2272 
2273 #ifndef UNPROTOIZE
2274 
2275 /* Check an individual filename for a .c suffix.  If the filename has this
2276    suffix, rename the file such that its suffix is changed to .C.  This
2277    function implements the -C option.  */
2278 
2279 static void
rename_c_file(const hash_table_entry * hp)2280 rename_c_file (const hash_table_entry *hp)
2281 {
2282   const char *filename = hp->symbol;
2283   int last_char_index = strlen (filename) - 1;
2284   char *const new_filename = alloca (strlen (filename)
2285 				     + strlen (cplus_suffix) + 1);
2286 
2287   /* Note that we don't care here if the given file was converted or not.  It
2288      is possible that the given file was *not* converted, simply because there
2289      was nothing in it which actually required conversion.  Even in this case,
2290      we want to do the renaming.  Note that we only rename files with the .c
2291      suffix (except for the syscalls file, which is left alone).  */
2292 
2293   if (filename[last_char_index] != 'c' || filename[last_char_index-1] != '.'
2294       || IS_SAME_PATH (syscalls_absolute_filename, filename))
2295     return;
2296 
2297   strcpy (new_filename, filename);
2298   strcpy (&new_filename[last_char_index], cplus_suffix);
2299 
2300   if (rename (filename, new_filename) == -1)
2301     {
2302       int errno_val = errno;
2303       notice ("%s: warning: can't rename file `%s' to `%s': %s\n",
2304 	      pname, shortpath (NULL, filename),
2305 	      shortpath (NULL, new_filename), xstrerror (errno_val));
2306       errors++;
2307       return;
2308     }
2309 }
2310 
2311 #endif /* !defined (UNPROTOIZE) */
2312 
2313 /* Take the list of definitions and declarations attached to a particular
2314    file_info node and reverse the order of the list.  This should get the
2315    list into an order such that the item with the lowest associated line
2316    number is nearest the head of the list.  When these lists are originally
2317    built, they are in the opposite order.  We want to traverse them in
2318    normal line number order later (i.e. lowest to highest) so reverse the
2319    order here.  */
2320 
2321 static void
reverse_def_dec_list(const hash_table_entry * hp)2322 reverse_def_dec_list (const hash_table_entry *hp)
2323 {
2324   file_info *file_p = hp->fip;
2325   def_dec_info *prev = NULL;
2326   def_dec_info *current = (def_dec_info *) file_p->defs_decs;
2327 
2328   if (!current)
2329     return;        		/* no list to reverse */
2330 
2331   prev = current;
2332   if (! (current = (def_dec_info *) current->next_in_file))
2333     return;        		/* can't reverse a single list element */
2334 
2335   prev->next_in_file = NULL;
2336 
2337   while (current)
2338     {
2339       def_dec_info *next = (def_dec_info *) current->next_in_file;
2340 
2341       current->next_in_file = prev;
2342       prev = current;
2343       current = next;
2344     }
2345 
2346   file_p->defs_decs = prev;
2347 }
2348 
2349 #ifndef UNPROTOIZE
2350 
2351 /* Find the (only?) extern definition for a particular function name, starting
2352    from the head of the linked list of entries for the given name.  If we
2353    cannot find an extern definition for the given function name, issue a
2354    warning and scrounge around for the next best thing, i.e. an extern
2355    function declaration with a prototype attached to it.  Note that we only
2356    allow such substitutions for extern declarations and never for static
2357    declarations.  That's because the only reason we allow them at all is
2358    to let un-prototyped function declarations for system-supplied library
2359    functions get their prototypes from our own extra SYSCALLS.c.X file which
2360    contains all of the correct prototypes for system functions.  */
2361 
2362 static const def_dec_info *
find_extern_def(const def_dec_info * head,const def_dec_info * user)2363 find_extern_def (const def_dec_info *head, const def_dec_info *user)
2364 {
2365   const def_dec_info *dd_p;
2366   const def_dec_info *extern_def_p = NULL;
2367   int conflict_noted = 0;
2368 
2369   /* Don't act too stupid here.  Somebody may try to convert an entire system
2370      in one swell fwoop (rather than one program at a time, as should be done)
2371      and in that case, we may find that there are multiple extern definitions
2372      of a given function name in the entire set of source files that we are
2373      converting.  If however one of these definitions resides in exactly the
2374      same source file as the reference we are trying to satisfy then in that
2375      case it would be stupid for us to fail to realize that this one definition
2376      *must* be the precise one we are looking for.
2377 
2378      To make sure that we don't miss an opportunity to make this "same file"
2379      leap of faith, we do a prescan of the list of records relating to the
2380      given function name, and we look (on this first scan) *only* for a
2381      definition of the function which is in the same file as the reference
2382      we are currently trying to satisfy.  */
2383 
2384   for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2385     if (dd_p->is_func_def && !dd_p->is_static && dd_p->file == user->file)
2386       return dd_p;
2387 
2388   /* Now, since we have not found a definition in the same file as the
2389      reference, we scan the list again and consider all possibilities from
2390      all files.  Here we may get conflicts with the things listed in the
2391      SYSCALLS.c.X file, but if that happens it only means that the source
2392      code being converted contains its own definition of a function which
2393      could have been supplied by libc.a.  In such cases, we should avoid
2394      issuing the normal warning, and defer to the definition given in the
2395      user's own code.  */
2396 
2397   for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2398     if (dd_p->is_func_def && !dd_p->is_static)
2399       {
2400 	if (!extern_def_p)	/* Previous definition? */
2401 	  extern_def_p = dd_p;	/* Remember the first definition found.  */
2402 	else
2403 	  {
2404 	    /* Ignore definition just found if it came from SYSCALLS.c.X.  */
2405 
2406 	    if (is_syscalls_file (dd_p->file))
2407 	      continue;
2408 
2409 	    /* Quietly replace the definition previously found with the one
2410 	       just found if the previous one was from SYSCALLS.c.X.  */
2411 
2412 	    if (is_syscalls_file (extern_def_p->file))
2413 	      {
2414 	        extern_def_p = dd_p;
2415 	        continue;
2416 	      }
2417 
2418 	    /* If we get here, then there is a conflict between two function
2419 	       declarations for the same function, both of which came from the
2420 	       user's own code.  */
2421 
2422 	    if (!conflict_noted)	/* first time we noticed? */
2423 	      {
2424 		conflict_noted = 1;
2425 		notice ("%s: conflicting extern definitions of '%s'\n",
2426 			pname, head->hash_entry->symbol);
2427 		if (!quiet_flag)
2428 		  {
2429 		    notice ("%s: declarations of '%s' will not be converted\n",
2430 			    pname, head->hash_entry->symbol);
2431 		    notice ("%s: conflict list for '%s' follows:\n",
2432 			    pname, head->hash_entry->symbol);
2433 		    fprintf (stderr, "%s:     %s(%d): %s\n",
2434 			     pname,
2435 			     shortpath (NULL, extern_def_p->file->hash_entry->symbol),
2436 			     extern_def_p->line, extern_def_p->ansi_decl);
2437 		  }
2438 	      }
2439 	    if (!quiet_flag)
2440 	      fprintf (stderr, "%s:     %s(%d): %s\n",
2441 		       pname,
2442 		       shortpath (NULL, dd_p->file->hash_entry->symbol),
2443 		       dd_p->line, dd_p->ansi_decl);
2444 	  }
2445       }
2446 
2447   /* We want to err on the side of caution, so if we found multiple conflicting
2448      definitions for the same function, treat this as being that same as if we
2449      had found no definitions (i.e. return NULL).  */
2450 
2451   if (conflict_noted)
2452     return NULL;
2453 
2454   if (!extern_def_p)
2455     {
2456       /* We have no definitions for this function so do the next best thing.
2457 	 Search for an extern declaration already in prototype form.  */
2458 
2459       for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2460 	if (!dd_p->is_func_def && !dd_p->is_static && dd_p->prototyped)
2461 	  {
2462 	    extern_def_p = dd_p;	/* save a pointer to the definition */
2463 	    if (!quiet_flag)
2464 	      notice ("%s: warning: using formals list from %s(%d) for function `%s'\n",
2465 		      pname,
2466 		      shortpath (NULL, dd_p->file->hash_entry->symbol),
2467 		      dd_p->line, dd_p->hash_entry->symbol);
2468 	    break;
2469 	  }
2470 
2471       /* Gripe about unprototyped function declarations that we found no
2472 	 corresponding definition (or other source of prototype information)
2473 	 for.
2474 
2475 	 Gripe even if the unprototyped declaration we are worried about
2476 	 exists in a file in one of the "system" include directories.  We
2477 	 can gripe about these because we should have at least found a
2478 	 corresponding (pseudo) definition in the SYSCALLS.c.X file.  If we
2479 	 didn't, then that means that the SYSCALLS.c.X file is missing some
2480 	 needed prototypes for this particular system.  That is worth telling
2481 	 the user about!  */
2482 
2483       if (!extern_def_p)
2484 	{
2485 	  const char *file = user->file->hash_entry->symbol;
2486 
2487 	  if (!quiet_flag)
2488 	    if (in_system_include_dir (file))
2489 	      {
2490 		/* Why copy this string into `needed' at all?
2491 		   Why not just use user->ansi_decl without copying?  */
2492 		char *needed = alloca (strlen (user->ansi_decl) + 1);
2493 	        char *p;
2494 
2495 	        strcpy (needed, user->ansi_decl);
2496 	        p = strstr (needed, user->hash_entry->symbol)
2497 	            + strlen (user->hash_entry->symbol) + 2;
2498 		/* Avoid having ??? in the string.  */
2499 		*p++ = '?';
2500 		*p++ = '?';
2501 		*p++ = '?';
2502 	        strcpy (p, ");");
2503 
2504 	        notice ("%s: %d: `%s' used but missing from SYSCALLS\n",
2505 			shortpath (NULL, file), user->line,
2506 			needed+7);	/* Don't print "extern " */
2507 	      }
2508 #if 0
2509 	    else
2510 	      notice ("%s: %d: warning: no extern definition for `%s'\n",
2511 		      shortpath (NULL, file), user->line,
2512 		      user->hash_entry->symbol);
2513 #endif
2514 	}
2515     }
2516   return extern_def_p;
2517 }
2518 
2519 /* Find the (only?) static definition for a particular function name in a
2520    given file.  Here we get the function-name and the file info indirectly
2521    from the def_dec_info record pointer which is passed in.  */
2522 
2523 static const def_dec_info *
find_static_definition(const def_dec_info * user)2524 find_static_definition (const def_dec_info *user)
2525 {
2526   const def_dec_info *head = user->hash_entry->ddip;
2527   const def_dec_info *dd_p;
2528   int num_static_defs = 0;
2529   const def_dec_info *static_def_p = NULL;
2530 
2531   for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2532     if (dd_p->is_func_def && dd_p->is_static && (dd_p->file == user->file))
2533       {
2534 	static_def_p = dd_p;	/* save a pointer to the definition */
2535 	num_static_defs++;
2536       }
2537   if (num_static_defs == 0)
2538     {
2539       if (!quiet_flag)
2540 	notice ("%s: warning: no static definition for `%s' in file `%s'\n",
2541 		pname, head->hash_entry->symbol,
2542 		shortpath (NULL, user->file->hash_entry->symbol));
2543     }
2544   else if (num_static_defs > 1)
2545     {
2546       notice ("%s: multiple static defs of `%s' in file `%s'\n",
2547 	      pname, head->hash_entry->symbol,
2548 	      shortpath (NULL, user->file->hash_entry->symbol));
2549       return NULL;
2550     }
2551   return static_def_p;
2552 }
2553 
2554 /* Find good prototype style formal argument lists for all of the function
2555    declarations which didn't have them before now.
2556 
2557    To do this we consider each function name one at a time.  For each function
2558    name, we look at the items on the linked list of def_dec_info records for
2559    that particular name.
2560 
2561    Somewhere on this list we should find one (and only one) def_dec_info
2562    record which represents the actual function definition, and this record
2563    should have a nice formal argument list already associated with it.
2564 
2565    Thus, all we have to do is to connect up all of the other def_dec_info
2566    records for this particular function name to the special one which has
2567    the full-blown formals list.
2568 
2569    Of course it is a little more complicated than just that.  See below for
2570    more details.  */
2571 
2572 static void
connect_defs_and_decs(const hash_table_entry * hp)2573 connect_defs_and_decs (const hash_table_entry *hp)
2574 {
2575   const def_dec_info *dd_p;
2576   const def_dec_info *extern_def_p = NULL;
2577   int first_extern_reference = 1;
2578 
2579   /* Traverse the list of definitions and declarations for this particular
2580      function name.  For each item on the list, if it is a function
2581      definition (either old style or new style) then GCC has already been
2582      kind enough to produce a prototype for us, and it is associated with
2583      the item already, so declare the item as its own associated "definition".
2584 
2585      Also, for each item which is only a function declaration, but which
2586      nonetheless has its own prototype already (obviously supplied by the user)
2587      declare the item as its own definition.
2588 
2589      Note that when/if there are multiple user-supplied prototypes already
2590      present for multiple declarations of any given function, these multiple
2591      prototypes *should* all match exactly with one another and with the
2592      prototype for the actual function definition.  We don't check for this
2593      here however, since we assume that the compiler must have already done
2594      this consistency checking when it was creating the .X files.  */
2595 
2596   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2597     if (dd_p->prototyped)
2598       ((NONCONST def_dec_info *) dd_p)->definition = dd_p;
2599 
2600   /* Traverse the list of definitions and declarations for this particular
2601      function name.  For each item on the list, if it is an extern function
2602      declaration and if it has no associated definition yet, go try to find
2603      the matching extern definition for the declaration.
2604 
2605      When looking for the matching function definition, warn the user if we
2606      fail to find one.
2607 
2608      If we find more that one function definition also issue a warning.
2609 
2610      Do the search for the matching definition only once per unique function
2611      name (and only when absolutely needed) so that we can avoid putting out
2612      redundant warning messages, and so that we will only put out warning
2613      messages when there is actually a reference (i.e. a declaration) for
2614      which we need to find a matching definition.  */
2615 
2616   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2617     if (!dd_p->is_func_def && !dd_p->is_static && !dd_p->definition)
2618       {
2619 	if (first_extern_reference)
2620 	  {
2621 	    extern_def_p = find_extern_def (hp->ddip, dd_p);
2622 	    first_extern_reference = 0;
2623 	  }
2624 	((NONCONST def_dec_info *) dd_p)->definition = extern_def_p;
2625       }
2626 
2627   /* Traverse the list of definitions and declarations for this particular
2628      function name.  For each item on the list, if it is a static function
2629      declaration and if it has no associated definition yet, go try to find
2630      the matching static definition for the declaration within the same file.
2631 
2632      When looking for the matching function definition, warn the user if we
2633      fail to find one in the same file with the declaration, and refuse to
2634      convert this kind of cross-file static function declaration.  After all,
2635      this is stupid practice and should be discouraged.
2636 
2637      We don't have to worry about the possibility that there is more than one
2638      matching function definition in the given file because that would have
2639      been flagged as an error by the compiler.
2640 
2641      Do the search for the matching definition only once per unique
2642      function-name/source-file pair (and only when absolutely needed) so that
2643      we can avoid putting out redundant warning messages, and so that we will
2644      only put out warning messages when there is actually a reference (i.e. a
2645      declaration) for which we actually need to find a matching definition.  */
2646 
2647   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2648     if (!dd_p->is_func_def && dd_p->is_static && !dd_p->definition)
2649       {
2650 	const def_dec_info *dd_p2;
2651 	const def_dec_info *static_def;
2652 
2653 	/* We have now found a single static declaration for which we need to
2654 	   find a matching definition.  We want to minimize the work (and the
2655 	   number of warnings), so we will find an appropriate (matching)
2656 	   static definition for this declaration, and then distribute it
2657 	   (as the definition for) any and all other static declarations
2658 	   for this function name which occur within the same file, and which
2659 	   do not already have definitions.
2660 
2661 	   Note that a trick is used here to prevent subsequent attempts to
2662 	   call find_static_definition for a given function-name & file
2663 	   if the first such call returns NULL.  Essentially, we convert
2664 	   these NULL return values to -1, and put the -1 into the definition
2665 	   field for each other static declaration from the same file which
2666 	   does not already have an associated definition.
2667 	   This makes these other static declarations look like they are
2668 	   actually defined already when the outer loop here revisits them
2669 	   later on.  Thus, the outer loop will skip over them.  Later, we
2670 	   turn the -1's back to NULL's.  */
2671 
2672 	((NONCONST def_dec_info *) dd_p)->definition =
2673 	  (static_def = find_static_definition (dd_p))
2674 	  ? static_def
2675 	  : (const def_dec_info *) -1;
2676 
2677 	for (dd_p2 = dd_p->next_for_func; dd_p2; dd_p2 = dd_p2->next_for_func)
2678 	  if (!dd_p2->is_func_def && dd_p2->is_static
2679 	      && !dd_p2->definition && (dd_p2->file == dd_p->file))
2680 	    ((NONCONST def_dec_info *) dd_p2)->definition = dd_p->definition;
2681       }
2682 
2683   /* Convert any dummy (-1) definitions we created in the step above back to
2684      NULL's (as they should be).  */
2685 
2686   for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2687     if (dd_p->definition == (def_dec_info *) -1)
2688       ((NONCONST def_dec_info *) dd_p)->definition = NULL;
2689 }
2690 
2691 #endif /* !defined (UNPROTOIZE) */
2692 
2693 /* Give a pointer into the clean text buffer, return a number which is the
2694    original source line number that the given pointer points into.  */
2695 
2696 static int
identify_lineno(const char * clean_p)2697 identify_lineno (const char *clean_p)
2698 {
2699   int line_num = 1;
2700   const char *scan_p;
2701 
2702   for (scan_p = clean_text_base; scan_p <= clean_p; scan_p++)
2703     if (*scan_p == '\n')
2704       line_num++;
2705   return line_num;
2706 }
2707 
2708 /* Issue an error message and give up on doing this particular edit.  */
2709 
2710 static void
declare_source_confusing(const char * clean_p)2711 declare_source_confusing (const char *clean_p)
2712 {
2713   if (!quiet_flag)
2714     {
2715       if (clean_p == 0)
2716 	notice ("%s: %d: warning: source too confusing\n",
2717 		shortpath (NULL, convert_filename), last_known_line_number);
2718       else
2719 	notice ("%s: %d: warning: source too confusing\n",
2720 		shortpath (NULL, convert_filename),
2721 		identify_lineno (clean_p));
2722     }
2723   longjmp (source_confusion_recovery, 1);
2724 }
2725 
2726 /* Check that a condition which is expected to be true in the original source
2727    code is in fact true.  If not, issue an error message and give up on
2728    converting this particular source file.  */
2729 
2730 static void
check_source(int cond,const char * clean_p)2731 check_source (int cond, const char *clean_p)
2732 {
2733   if (!cond)
2734     declare_source_confusing (clean_p);
2735 }
2736 
2737 /* If we think of the in-core cleaned text buffer as a memory mapped
2738    file (with the variable last_known_line_start acting as sort of a
2739    file pointer) then we can imagine doing "seeks" on the buffer.  The
2740    following routine implements a kind of "seek" operation for the in-core
2741    (cleaned) copy of the source file.  When finished, it returns a pointer to
2742    the start of a given (numbered) line in the cleaned text buffer.
2743 
2744    Note that protoize only has to "seek" in the forward direction on the
2745    in-core cleaned text file buffers, and it never needs to back up.
2746 
2747    This routine is made a little bit faster by remembering the line number
2748    (and pointer value) supplied (and returned) from the previous "seek".
2749    This prevents us from always having to start all over back at the top
2750    of the in-core cleaned buffer again.  */
2751 
2752 static const char *
seek_to_line(int n)2753 seek_to_line (int n)
2754 {
2755   if (n < last_known_line_number)
2756     abort ();
2757 
2758   while (n > last_known_line_number)
2759     {
2760       while (*last_known_line_start != '\n')
2761 	check_source (++last_known_line_start < clean_text_limit, 0);
2762       last_known_line_start++;
2763       last_known_line_number++;
2764     }
2765   return last_known_line_start;
2766 }
2767 
2768 /* Given a pointer to a character in the cleaned text buffer, return a pointer
2769    to the next non-whitespace character which follows it.  */
2770 
2771 static const char *
forward_to_next_token_char(const char * ptr)2772 forward_to_next_token_char (const char *ptr)
2773 {
2774   for (++ptr; ISSPACE ((const unsigned char)*ptr);
2775        check_source (++ptr < clean_text_limit, 0))
2776     continue;
2777   return ptr;
2778 }
2779 
2780 /* Copy a chunk of text of length `len' and starting at `str' to the current
2781    output buffer.  Note that all attempts to add stuff to the current output
2782    buffer ultimately go through here.  */
2783 
2784 static void
output_bytes(const char * str,size_t len)2785 output_bytes (const char *str, size_t len)
2786 {
2787   if ((repl_write_ptr + 1) + len >= repl_text_limit)
2788     {
2789       size_t new_size = (repl_text_limit - repl_text_base) << 1;
2790       char *new_buf = xrealloc (repl_text_base, new_size);
2791 
2792       repl_write_ptr = new_buf + (repl_write_ptr - repl_text_base);
2793       repl_text_base = new_buf;
2794       repl_text_limit = new_buf + new_size;
2795     }
2796   memcpy (repl_write_ptr + 1, str, len);
2797   repl_write_ptr += len;
2798 }
2799 
2800 /* Copy all bytes (except the trailing null) of a null terminated string to
2801    the current output buffer.  */
2802 
2803 static void
output_string(const char * str)2804 output_string (const char *str)
2805 {
2806   output_bytes (str, strlen (str));
2807 }
2808 
2809 /* Copy some characters from the original text buffer to the current output
2810    buffer.
2811 
2812    This routine takes a pointer argument `p' which is assumed to be a pointer
2813    into the cleaned text buffer.  The bytes which are copied are the `original'
2814    equivalents for the set of bytes between the last value of `clean_read_ptr'
2815    and the argument value `p'.
2816 
2817    The set of bytes copied however, comes *not* from the cleaned text buffer,
2818    but rather from the direct counterparts of these bytes within the original
2819    text buffer.
2820 
2821    Thus, when this function is called, some bytes from the original text
2822    buffer (which may include original comments and preprocessing directives)
2823    will be copied into the  output buffer.
2824 
2825    Note that the request implied when this routine is called includes the
2826    byte pointed to by the argument pointer `p'.  */
2827 
2828 static void
output_up_to(const char * p)2829 output_up_to (const char *p)
2830 {
2831   size_t copy_length = (size_t) (p - clean_read_ptr);
2832   const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
2833 
2834   if (copy_length == 0)
2835     return;
2836 
2837   output_bytes (copy_start, copy_length);
2838   clean_read_ptr = p;
2839 }
2840 
2841 /* Given a pointer to a def_dec_info record which represents some form of
2842    definition of a function (perhaps a real definition, or in lieu of that
2843    perhaps just a declaration with a full prototype) return true if this
2844    function is one which we should avoid converting.  Return false
2845    otherwise.  */
2846 
2847 static int
other_variable_style_function(const char * ansi_header)2848 other_variable_style_function (const char *ansi_header)
2849 {
2850 #ifdef UNPROTOIZE
2851 
2852   /* See if we have a stdarg function, or a function which has stdarg style
2853      parameters or a stdarg style return type.  */
2854 
2855   return strstr (ansi_header, "...") != 0;
2856 
2857 #else /* !defined (UNPROTOIZE) */
2858 
2859   /* See if we have a varargs function, or a function which has varargs style
2860      parameters or a varargs style return type.  */
2861 
2862   const char *p;
2863   int len = strlen (varargs_style_indicator);
2864 
2865   for (p = ansi_header; p; )
2866     {
2867       const char *candidate;
2868 
2869       if ((candidate = strstr (p, varargs_style_indicator)) == 0)
2870 	return 0;
2871       else
2872 	if (!is_id_char (candidate[-1]) && !is_id_char (candidate[len]))
2873 	  return 1;
2874 	else
2875 	  p = candidate + 1;
2876     }
2877   return 0;
2878 #endif /* !defined (UNPROTOIZE) */
2879 }
2880 
2881 /* Do the editing operation specifically for a function "declaration".  Note
2882    that editing for function "definitions" are handled in a separate routine
2883    below.  */
2884 
2885 static void
edit_fn_declaration(const def_dec_info * def_dec_p,const char * volatile clean_text_p)2886 edit_fn_declaration (const def_dec_info *def_dec_p,
2887 		     const char *volatile clean_text_p)
2888 {
2889   const char *start_formals;
2890   const char *end_formals;
2891   const char *function_to_edit = def_dec_p->hash_entry->symbol;
2892   size_t func_name_len = strlen (function_to_edit);
2893   const char *end_of_fn_name;
2894 
2895 #ifndef UNPROTOIZE
2896 
2897   const f_list_chain_item *this_f_list_chain_item;
2898   const def_dec_info *definition = def_dec_p->definition;
2899 
2900   /* If we are protoizing, and if we found no corresponding definition for
2901      this particular function declaration, then just leave this declaration
2902      exactly as it is.  */
2903 
2904   if (!definition)
2905     return;
2906 
2907   /* If we are protoizing, and if the corresponding definition that we found
2908      for this particular function declaration defined an old style varargs
2909      function, then we want to issue a warning and just leave this function
2910      declaration unconverted.  */
2911 
2912   if (other_variable_style_function (definition->ansi_decl))
2913     {
2914       if (!quiet_flag)
2915 	notice ("%s: %d: warning: varargs function declaration not converted\n",
2916 		shortpath (NULL, def_dec_p->file->hash_entry->symbol),
2917 		def_dec_p->line);
2918       return;
2919     }
2920 
2921 #endif /* !defined (UNPROTOIZE) */
2922 
2923   /* Setup here to recover from confusing source code detected during this
2924      particular "edit".  */
2925 
2926   save_pointers ();
2927   if (setjmp (source_confusion_recovery))
2928     {
2929       restore_pointers ();
2930       notice ("%s: declaration of function `%s' not converted\n",
2931 	      pname, function_to_edit);
2932       return;
2933     }
2934 
2935   /* We are editing a function declaration.  The line number we did a seek to
2936      contains the comma or semicolon which follows the declaration.  Our job
2937      now is to scan backwards looking for the function name.  This name *must*
2938      be followed by open paren (ignoring whitespace, of course).  We need to
2939      replace everything between that open paren and the corresponding closing
2940      paren.  If we are protoizing, we need to insert the prototype-style
2941      formals lists.  If we are unprotoizing, we need to just delete everything
2942      between the pairs of opening and closing parens.  */
2943 
2944   /* First move up to the end of the line.  */
2945 
2946   while (*clean_text_p != '\n')
2947     check_source (++clean_text_p < clean_text_limit, 0);
2948   clean_text_p--;  /* Point to just before the newline character.  */
2949 
2950   /* Now we can scan backwards for the function name.  */
2951 
2952   do
2953     {
2954       for (;;)
2955 	{
2956 	  /* Scan leftwards until we find some character which can be
2957 	     part of an identifier.  */
2958 
2959 	  while (!is_id_char (*clean_text_p))
2960 	    check_source (--clean_text_p > clean_read_ptr, 0);
2961 
2962 	  /* Scan backwards until we find a char that cannot be part of an
2963 	     identifier.  */
2964 
2965 	  while (is_id_char (*clean_text_p))
2966 	    check_source (--clean_text_p > clean_read_ptr, 0);
2967 
2968 	  /* Having found an "id break", see if the following id is the one
2969 	     that we are looking for.  If so, then exit from this loop.  */
2970 
2971 	  if (!strncmp (clean_text_p+1, function_to_edit, func_name_len))
2972 	    {
2973 	      char ch = *(clean_text_p + 1 + func_name_len);
2974 
2975 	      /* Must also check to see that the name in the source text
2976 	         ends where it should (in order to prevent bogus matches
2977 	         on similar but longer identifiers.  */
2978 
2979 	      if (! is_id_char (ch))
2980 	        break;			/* exit from loop */
2981 	    }
2982 	}
2983 
2984       /* We have now found the first perfect match for the function name in
2985 	 our backward search.  This may or may not be the actual function
2986 	 name at the start of the actual function declaration (i.e. we could
2987 	 have easily been mislead).  We will try to avoid getting fooled too
2988 	 often by looking forward for the open paren which should follow the
2989 	 identifier we just found.  We ignore whitespace while hunting.  If
2990 	 the next non-whitespace byte we see is *not* an open left paren,
2991 	 then we must assume that we have been fooled and we start over
2992 	 again accordingly.  Note that there is no guarantee, that even if
2993 	 we do see the open paren, that we are in the right place.
2994 	 Programmers do the strangest things sometimes!  */
2995 
2996       end_of_fn_name = clean_text_p + strlen (def_dec_p->hash_entry->symbol);
2997       start_formals = forward_to_next_token_char (end_of_fn_name);
2998     }
2999   while (*start_formals != '(');
3000 
3001   /* start_of_formals now points to the opening left paren which immediately
3002      follows the name of the function.  */
3003 
3004   /* Note that there may be several formals lists which need to be modified
3005      due to the possibility that the return type of this function is a
3006      pointer-to-function type.  If there are several formals lists, we
3007      convert them in left-to-right order here.  */
3008 
3009 #ifndef UNPROTOIZE
3010   this_f_list_chain_item = definition->f_list_chain;
3011 #endif /* !defined (UNPROTOIZE) */
3012 
3013   for (;;)
3014     {
3015       {
3016 	int depth;
3017 
3018 	end_formals = start_formals + 1;
3019 	depth = 1;
3020 	for (; depth; check_source (++end_formals < clean_text_limit, 0))
3021 	  {
3022 	    switch (*end_formals)
3023 	      {
3024 	      case '(':
3025 		depth++;
3026 		break;
3027 	      case ')':
3028 		depth--;
3029 		break;
3030 	      }
3031 	  }
3032 	end_formals--;
3033       }
3034 
3035       /* end_formals now points to the closing right paren of the formals
3036 	 list whose left paren is pointed to by start_formals.  */
3037 
3038       /* Now, if we are protoizing, we insert the new ANSI-style formals list
3039 	 attached to the associated definition of this function.  If however
3040 	 we are unprotoizing, then we simply delete any formals list which
3041 	 may be present.  */
3042 
3043       output_up_to (start_formals);
3044 #ifndef UNPROTOIZE
3045       if (this_f_list_chain_item)
3046 	{
3047 	  output_string (this_f_list_chain_item->formals_list);
3048 	  this_f_list_chain_item = this_f_list_chain_item->chain_next;
3049 	}
3050       else
3051 	{
3052 	  if (!quiet_flag)
3053 	    notice ("%s: warning: too many parameter lists in declaration of `%s'\n",
3054 		    pname, def_dec_p->hash_entry->symbol);
3055 	  check_source (0, end_formals);  /* leave the declaration intact */
3056 	}
3057 #endif /* !defined (UNPROTOIZE) */
3058       clean_read_ptr = end_formals - 1;
3059 
3060       /* Now see if it looks like there may be another formals list associated
3061 	 with the function declaration that we are converting (following the
3062 	 formals list that we just converted.  */
3063 
3064       {
3065 	const char *another_r_paren = forward_to_next_token_char (end_formals);
3066 
3067 	if ((*another_r_paren != ')')
3068 	    || (*(start_formals = forward_to_next_token_char (another_r_paren)) != '('))
3069 	  {
3070 #ifndef UNPROTOIZE
3071 	    if (this_f_list_chain_item)
3072 	      {
3073 		if (!quiet_flag)
3074 		  notice ("\n%s: warning: too few parameter lists in declaration of `%s'\n",
3075 			  pname, def_dec_p->hash_entry->symbol);
3076 		check_source (0, start_formals); /* leave the decl intact */
3077 	      }
3078 #endif /* !defined (UNPROTOIZE) */
3079 	    break;
3080 
3081 	  }
3082       }
3083 
3084       /* There does appear to be yet another formals list, so loop around
3085 	 again, and convert it also.  */
3086     }
3087 }
3088 
3089 /* Edit a whole group of formals lists, starting with the rightmost one
3090    from some set of formals lists.  This routine is called once (from the
3091    outside) for each function declaration which is converted.  It is
3092    recursive however, and it calls itself once for each remaining formal
3093    list that lies to the left of the one it was originally called to work
3094    on.  Thus, a whole set gets done in right-to-left order.
3095 
3096    This routine returns nonzero if it thinks that it should not be trying
3097    to convert this particular function definition (because the name of the
3098    function doesn't match the one expected).  */
3099 
3100 static int
edit_formals_lists(const char * end_formals,unsigned int f_list_count,const def_dec_info * def_dec_p)3101 edit_formals_lists (const char *end_formals, unsigned int f_list_count,
3102 		    const def_dec_info *def_dec_p)
3103 {
3104   const char *start_formals;
3105   int depth;
3106 
3107   start_formals = end_formals - 1;
3108   depth = 1;
3109   for (; depth; check_source (--start_formals > clean_read_ptr, 0))
3110     {
3111       switch (*start_formals)
3112 	{
3113 	case '(':
3114 	  depth--;
3115 	  break;
3116 	case ')':
3117 	  depth++;
3118 	  break;
3119 	}
3120     }
3121   start_formals++;
3122 
3123   /* start_formals now points to the opening left paren of the formals list.  */
3124 
3125   f_list_count--;
3126 
3127   if (f_list_count)
3128     {
3129       const char *next_end;
3130 
3131       /* There should be more formal lists to the left of here.  */
3132 
3133       next_end = start_formals - 1;
3134       check_source (next_end > clean_read_ptr, 0);
3135       while (ISSPACE ((const unsigned char)*next_end))
3136 	check_source (--next_end > clean_read_ptr, 0);
3137       check_source (*next_end == ')', next_end);
3138       check_source (--next_end > clean_read_ptr, 0);
3139       check_source (*next_end == ')', next_end);
3140       if (edit_formals_lists (next_end, f_list_count, def_dec_p))
3141 	return 1;
3142     }
3143 
3144   /* Check that the function name in the header we are working on is the same
3145      as the one we would expect to find.  If not, issue a warning and return
3146      nonzero.  */
3147 
3148   if (f_list_count == 0)
3149     {
3150       const char *expected = def_dec_p->hash_entry->symbol;
3151       const char *func_name_start;
3152       const char *func_name_limit;
3153       size_t func_name_len;
3154 
3155       for (func_name_limit = start_formals-1;
3156 	   ISSPACE ((const unsigned char)*func_name_limit); )
3157 	check_source (--func_name_limit > clean_read_ptr, 0);
3158 
3159       for (func_name_start = func_name_limit++;
3160 	   is_id_char (*func_name_start);
3161 	   func_name_start--)
3162 	check_source (func_name_start > clean_read_ptr, 0);
3163       func_name_start++;
3164       func_name_len = func_name_limit - func_name_start;
3165       if (func_name_len == 0)
3166 	check_source (0, func_name_start);
3167       if (func_name_len != strlen (expected)
3168 	  || strncmp (func_name_start, expected, func_name_len))
3169 	{
3170 	  notice ("%s: %d: warning: found `%s' but expected `%s'\n",
3171 		  shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3172 		  identify_lineno (func_name_start),
3173 		  dupnstr (func_name_start, func_name_len),
3174 		  expected);
3175 	  return 1;
3176 	}
3177     }
3178 
3179   output_up_to (start_formals);
3180 
3181 #ifdef UNPROTOIZE
3182   if (f_list_count == 0)
3183     output_string (def_dec_p->formal_names);
3184 #else /* !defined (UNPROTOIZE) */
3185   {
3186     unsigned f_list_depth;
3187     const f_list_chain_item *flci_p = def_dec_p->f_list_chain;
3188 
3189     /* At this point, the current value of f_list count says how many
3190        links we have to follow through the f_list_chain to get to the
3191        particular formals list that we need to output next.  */
3192 
3193     for (f_list_depth = 0; f_list_depth < f_list_count; f_list_depth++)
3194       flci_p = flci_p->chain_next;
3195     output_string (flci_p->formals_list);
3196   }
3197 #endif /* !defined (UNPROTOIZE) */
3198 
3199   clean_read_ptr = end_formals - 1;
3200   return 0;
3201 }
3202 
3203 /* Given a pointer to a byte in the clean text buffer which points to
3204    the beginning of a line that contains a "follower" token for a
3205    function definition header, do whatever is necessary to find the
3206    right closing paren for the rightmost formals list of the function
3207    definition header.  */
3208 
3209 static const char *
find_rightmost_formals_list(const char * clean_text_p)3210 find_rightmost_formals_list (const char *clean_text_p)
3211 {
3212   const char *end_formals;
3213 
3214   /* We are editing a function definition.  The line number we did a seek
3215      to contains the first token which immediately follows the entire set of
3216      formals lists which are part of this particular function definition
3217      header.
3218 
3219      Our job now is to scan leftwards in the clean text looking for the
3220      right-paren which is at the end of the function header's rightmost
3221      formals list.
3222 
3223      If we ignore whitespace, this right paren should be the first one we
3224      see which is (ignoring whitespace) immediately followed either by the
3225      open curly-brace beginning the function body or by an alphabetic
3226      character (in the case where the function definition is in old (K&R)
3227      style and there are some declarations of formal parameters).  */
3228 
3229    /* It is possible that the right paren we are looking for is on the
3230       current line (together with its following token).  Just in case that
3231       might be true, we start out here by skipping down to the right end of
3232       the current line before starting our scan.  */
3233 
3234   for (end_formals = clean_text_p; *end_formals != '\n'; end_formals++)
3235     continue;
3236   end_formals--;
3237 
3238 #ifdef UNPROTOIZE
3239 
3240   /* Now scan backwards while looking for the right end of the rightmost
3241      formals list associated with this function definition.  */
3242 
3243   {
3244     char ch;
3245     const char *l_brace_p;
3246 
3247     /* Look leftward and try to find a right-paren.  */
3248 
3249     while (*end_formals != ')')
3250       {
3251 	if (ISSPACE ((unsigned char)*end_formals))
3252 	  while (ISSPACE ((unsigned char)*end_formals))
3253 	    check_source (--end_formals > clean_read_ptr, 0);
3254 	else
3255 	  check_source (--end_formals > clean_read_ptr, 0);
3256       }
3257 
3258     ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3259     /* Since we are unprotoizing an ANSI-style (prototyped) function
3260        definition, there had better not be anything (except whitespace)
3261        between the end of the ANSI formals list and the beginning of the
3262        function body (i.e. the '{').  */
3263 
3264     check_source (ch == '{', l_brace_p);
3265   }
3266 
3267 #else /* !defined (UNPROTOIZE) */
3268 
3269   /* Now scan backwards while looking for the right end of the rightmost
3270      formals list associated with this function definition.  */
3271 
3272   while (1)
3273     {
3274       char ch;
3275       const char *l_brace_p;
3276 
3277       /* Look leftward and try to find a right-paren.  */
3278 
3279       while (*end_formals != ')')
3280 	{
3281 	  if (ISSPACE ((const unsigned char)*end_formals))
3282 	    while (ISSPACE ((const unsigned char)*end_formals))
3283 	      check_source (--end_formals > clean_read_ptr, 0);
3284 	  else
3285 	    check_source (--end_formals > clean_read_ptr, 0);
3286 	}
3287 
3288       ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3289 
3290       /* Since it is possible that we found a right paren before the starting
3291 	 '{' of the body which IS NOT the one at the end of the real K&R
3292 	 formals list (say for instance, we found one embedded inside one of
3293 	 the old K&R formal parameter declarations) we have to check to be
3294 	 sure that this is in fact the right paren that we were looking for.
3295 
3296 	 The one we were looking for *must* be followed by either a '{' or
3297 	 by an alphabetic character, while others *cannot* validly be followed
3298 	 by such characters.  */
3299 
3300       if ((ch == '{') || ISALPHA ((unsigned char) ch))
3301 	break;
3302 
3303       /* At this point, we have found a right paren, but we know that it is
3304 	 not the one we were looking for, so backup one character and keep
3305 	 looking.  */
3306 
3307       check_source (--end_formals > clean_read_ptr, 0);
3308     }
3309 
3310 #endif /* !defined (UNPROTOIZE) */
3311 
3312   return end_formals;
3313 }
3314 
3315 #ifndef UNPROTOIZE
3316 
3317 /* Insert into the output file a totally new declaration for a function
3318    which (up until now) was being called from within the current block
3319    without having been declared at any point such that the declaration
3320    was visible (i.e. in scope) at the point of the call.
3321 
3322    We need to add in explicit declarations for all such function calls
3323    in order to get the full benefit of prototype-based function call
3324    parameter type checking.  */
3325 
3326 static void
add_local_decl(const def_dec_info * def_dec_p,const char * clean_text_p)3327 add_local_decl (const def_dec_info *def_dec_p, const char *clean_text_p)
3328 {
3329   const char *start_of_block;
3330   const char *function_to_edit = def_dec_p->hash_entry->symbol;
3331 
3332   /* Don't insert new local explicit declarations unless explicitly requested
3333      to do so.  */
3334 
3335   if (!local_flag)
3336     return;
3337 
3338   /* Setup here to recover from confusing source code detected during this
3339      particular "edit".  */
3340 
3341   save_pointers ();
3342   if (setjmp (source_confusion_recovery))
3343     {
3344       restore_pointers ();
3345       notice ("%s: local declaration for function `%s' not inserted\n",
3346 	      pname, function_to_edit);
3347       return;
3348     }
3349 
3350   /* We have already done a seek to the start of the line which should
3351      contain *the* open curly brace which begins the block in which we need
3352      to insert an explicit function declaration (to replace the implicit one).
3353 
3354      Now we scan that line, starting from the left, until we find the
3355      open curly brace we are looking for.  Note that there may actually be
3356      multiple open curly braces on the given line, but we will be happy
3357      with the leftmost one no matter what.  */
3358 
3359   start_of_block = clean_text_p;
3360   while (*start_of_block != '{' && *start_of_block != '\n')
3361     check_source (++start_of_block < clean_text_limit, 0);
3362 
3363   /* Note that the line from the original source could possibly
3364      contain *no* open curly braces!  This happens if the line contains
3365      a macro call which expands into a chunk of text which includes a
3366      block (and that block's associated open and close curly braces).
3367      In cases like this, we give up, issue a warning, and do nothing.  */
3368 
3369   if (*start_of_block != '{')
3370     {
3371       if (!quiet_flag)
3372 	notice ("\n%s: %d: warning: can't add declaration of `%s' into macro call\n",
3373 	  def_dec_p->file->hash_entry->symbol, def_dec_p->line,
3374 	  def_dec_p->hash_entry->symbol);
3375       return;
3376     }
3377 
3378   /* Figure out what a nice (pretty) indentation would be for the new
3379      declaration we are adding.  In order to do this, we must scan forward
3380      from the '{' until we find the first line which starts with some
3381      non-whitespace characters (i.e. real "token" material).  */
3382 
3383   {
3384     const char *ep = forward_to_next_token_char (start_of_block) - 1;
3385     const char *sp;
3386 
3387     /* Now we have ep pointing at the rightmost byte of some existing indent
3388        stuff.  At least that is the hope.
3389 
3390        We can now just scan backwards and find the left end of the existing
3391        indentation string, and then copy it to the output buffer.  */
3392 
3393     for (sp = ep; ISSPACE ((const unsigned char)*sp) && *sp != '\n'; sp--)
3394       continue;
3395 
3396     /* Now write out the open { which began this block, and any following
3397        trash up to and including the last byte of the existing indent that
3398        we just found.  */
3399 
3400     output_up_to (ep);
3401 
3402     /* Now we go ahead and insert the new declaration at this point.
3403 
3404        If the definition of the given function is in the same file that we
3405        are currently editing, and if its full ANSI declaration normally
3406        would start with the keyword `extern', suppress the `extern'.  */
3407 
3408     {
3409       const char *decl = def_dec_p->definition->ansi_decl;
3410 
3411       if ((*decl == 'e') && (def_dec_p->file == def_dec_p->definition->file))
3412 	decl += 7;
3413       output_string (decl);
3414     }
3415 
3416     /* Finally, write out a new indent string, just like the preceding one
3417        that we found.  This will typically include a newline as the first
3418        character of the indent string.  */
3419 
3420     output_bytes (sp, (size_t) (ep - sp) + 1);
3421   }
3422 }
3423 
3424 /* Given a pointer to a file_info record, and a pointer to the beginning
3425    of a line (in the clean text buffer) which is assumed to contain the
3426    first "follower" token for the first function definition header in the
3427    given file, find a good place to insert some new global function
3428    declarations (which will replace scattered and imprecise implicit ones)
3429    and then insert the new explicit declaration at that point in the file.  */
3430 
3431 static void
add_global_decls(const file_info * file_p,const char * clean_text_p)3432 add_global_decls (const file_info *file_p, const char *clean_text_p)
3433 {
3434   const def_dec_info *dd_p;
3435   const char *scan_p;
3436 
3437   /* Setup here to recover from confusing source code detected during this
3438      particular "edit".  */
3439 
3440   save_pointers ();
3441   if (setjmp (source_confusion_recovery))
3442     {
3443       restore_pointers ();
3444       notice ("%s: global declarations for file `%s' not inserted\n",
3445 	      pname, shortpath (NULL, file_p->hash_entry->symbol));
3446       return;
3447     }
3448 
3449   /* Start by finding a good location for adding the new explicit function
3450      declarations.  To do this, we scan backwards, ignoring whitespace
3451      and comments and other junk until we find either a semicolon, or until
3452      we hit the beginning of the file.  */
3453 
3454   scan_p = find_rightmost_formals_list (clean_text_p);
3455   for (;; --scan_p)
3456     {
3457       if (scan_p < clean_text_base)
3458 	break;
3459       check_source (scan_p > clean_read_ptr, 0);
3460       if (*scan_p == ';')
3461 	break;
3462     }
3463 
3464   /* scan_p now points either to a semicolon, or to just before the start
3465      of the whole file.  */
3466 
3467   /* Now scan forward for the first non-whitespace character.  In theory,
3468      this should be the first character of the following function definition
3469      header.  We will put in the added declarations just prior to that.  */
3470 
3471   scan_p++;
3472   while (ISSPACE ((const unsigned char)*scan_p))
3473     scan_p++;
3474   scan_p--;
3475 
3476   output_up_to (scan_p);
3477 
3478   /* Now write out full prototypes for all of the things that had been
3479      implicitly declared in this file (but only those for which we were
3480      actually able to find unique matching definitions).  Avoid duplicates
3481      by marking things that we write out as we go.  */
3482 
3483   {
3484     int some_decls_added = 0;
3485 
3486     for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3487       if (dd_p->is_implicit && dd_p->definition && !dd_p->definition->written)
3488 	{
3489 	  const char *decl = dd_p->definition->ansi_decl;
3490 
3491 	  /* If the function for which we are inserting a declaration is
3492 	     actually defined later in the same file, then suppress the
3493 	     leading `extern' keyword (if there is one).  */
3494 
3495 	  if (*decl == 'e' && (dd_p->file == dd_p->definition->file))
3496 	    decl += 7;
3497 
3498 	  output_string ("\n");
3499 	  output_string (decl);
3500 	  some_decls_added = 1;
3501 	  ((NONCONST def_dec_info *) dd_p->definition)->written = 1;
3502 	}
3503     if (some_decls_added)
3504       output_string ("\n\n");
3505   }
3506 
3507   /* Unmark all of the definitions that we just marked.  */
3508 
3509   for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3510     if (dd_p->definition)
3511       ((NONCONST def_dec_info *) dd_p->definition)->written = 0;
3512 }
3513 
3514 #endif /* !defined (UNPROTOIZE) */
3515 
3516 /* Do the editing operation specifically for a function "definition".  Note
3517    that editing operations for function "declarations" are handled by a
3518    separate routine above.  */
3519 
3520 static void
edit_fn_definition(const def_dec_info * def_dec_p,const char * clean_text_p)3521 edit_fn_definition (const def_dec_info *def_dec_p, const char *clean_text_p)
3522 {
3523   const char *end_formals;
3524   const char *function_to_edit = def_dec_p->hash_entry->symbol;
3525 
3526   /* Setup here to recover from confusing source code detected during this
3527      particular "edit".  */
3528 
3529   save_pointers ();
3530   if (setjmp (source_confusion_recovery))
3531     {
3532       restore_pointers ();
3533       notice ("%s: definition of function `%s' not converted\n",
3534 	      pname, function_to_edit);
3535       return;
3536     }
3537 
3538   end_formals = find_rightmost_formals_list (clean_text_p);
3539 
3540   /* end_of_formals now points to the closing right paren of the rightmost
3541      formals list which is actually part of the `header' of the function
3542      definition that we are converting.  */
3543 
3544   /* If the header of this function definition looks like it declares a
3545      function with a variable number of arguments, and if the way it does
3546      that is different from that way we would like it (i.e. varargs vs.
3547      stdarg) then issue a warning and leave the header unconverted.  */
3548 
3549   if (other_variable_style_function (def_dec_p->ansi_decl))
3550     {
3551       if (!quiet_flag)
3552 	notice ("%s: %d: warning: definition of %s not converted\n",
3553 		shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3554 		identify_lineno (end_formals),
3555 		other_var_style);
3556       output_up_to (end_formals);
3557       return;
3558     }
3559 
3560   if (edit_formals_lists (end_formals, def_dec_p->f_list_count, def_dec_p))
3561     {
3562       restore_pointers ();
3563       notice ("%s: definition of function `%s' not converted\n",
3564 	      pname, function_to_edit);
3565       return;
3566     }
3567 
3568   /* Have to output the last right paren because this never gets flushed by
3569      edit_formals_list.  */
3570 
3571   output_up_to (end_formals);
3572 
3573 #ifdef UNPROTOIZE
3574   {
3575     const char *decl_p;
3576     const char *semicolon_p;
3577     const char *limit_p;
3578     const char *scan_p;
3579     int had_newlines = 0;
3580 
3581     /* Now write out the K&R style formal declarations, one per line.  */
3582 
3583     decl_p = def_dec_p->formal_decls;
3584     limit_p = decl_p + strlen (decl_p);
3585     for (;decl_p < limit_p; decl_p = semicolon_p + 2)
3586       {
3587 	for (semicolon_p = decl_p; *semicolon_p != ';'; semicolon_p++)
3588 	  continue;
3589 	output_string ("\n");
3590 	output_string (indent_string);
3591 	output_bytes (decl_p, (size_t) ((semicolon_p + 1) - decl_p));
3592       }
3593 
3594     /* If there are no newlines between the end of the formals list and the
3595        start of the body, we should insert one now.  */
3596 
3597     for (scan_p = end_formals+1; *scan_p != '{'; )
3598       {
3599 	if (*scan_p == '\n')
3600 	  {
3601 	    had_newlines = 1;
3602 	    break;
3603 	  }
3604 	check_source (++scan_p < clean_text_limit, 0);
3605       }
3606     if (!had_newlines)
3607       output_string ("\n");
3608   }
3609 #else /* !defined (UNPROTOIZE) */
3610   /* If we are protoizing, there may be some flotsam & jetsam (like comments
3611      and preprocessing directives) after the old formals list but before
3612      the following { and we would like to preserve that stuff while effectively
3613      deleting the existing K&R formal parameter declarations.  We do so here
3614      in a rather tricky way.  Basically, we white out any stuff *except*
3615      the comments/pp-directives in the original text buffer, then, if there
3616      is anything in this area *other* than whitespace, we output it.  */
3617   {
3618     const char *end_formals_orig;
3619     const char *start_body;
3620     const char *start_body_orig;
3621     const char *scan;
3622     const char *scan_orig;
3623     int have_flotsam = 0;
3624     int have_newlines = 0;
3625 
3626     for (start_body = end_formals + 1; *start_body != '{';)
3627       check_source (++start_body < clean_text_limit, 0);
3628 
3629     end_formals_orig = orig_text_base + (end_formals - clean_text_base);
3630     start_body_orig = orig_text_base + (start_body - clean_text_base);
3631     scan = end_formals + 1;
3632     scan_orig = end_formals_orig + 1;
3633     for (; scan < start_body; scan++, scan_orig++)
3634       {
3635 	if (*scan == *scan_orig)
3636 	  {
3637 	    have_newlines |= (*scan_orig == '\n');
3638 	    /* Leave identical whitespace alone.  */
3639 	    if (!ISSPACE ((const unsigned char)*scan_orig))
3640 	      *((NONCONST char *) scan_orig) = ' '; /* identical - so whiteout */
3641 	  }
3642 	else
3643 	  have_flotsam = 1;
3644       }
3645     if (have_flotsam)
3646       output_bytes (end_formals_orig + 1,
3647 		    (size_t) (start_body_orig - end_formals_orig) - 1);
3648     else
3649       if (have_newlines)
3650 	output_string ("\n");
3651       else
3652 	output_string (" ");
3653     clean_read_ptr = start_body - 1;
3654   }
3655 #endif /* !defined (UNPROTOIZE) */
3656 }
3657 
3658 /* Clean up the clean text buffer.  Do this by converting comments and
3659    preprocessing directives into spaces.   Also convert line continuations
3660    into whitespace.  Also, whiteout string and character literals.  */
3661 
3662 static void
do_cleaning(char * new_clean_text_base,const char * new_clean_text_limit)3663 do_cleaning (char *new_clean_text_base, const char *new_clean_text_limit)
3664 {
3665   char *scan_p;
3666   int non_whitespace_since_newline = 0;
3667 
3668   for (scan_p = new_clean_text_base; scan_p < new_clean_text_limit; scan_p++)
3669     {
3670       switch (*scan_p)
3671 	{
3672 	case '/':			/* Handle comments.  */
3673 	  if (scan_p[1] != '*')
3674 	    goto regular;
3675 	  non_whitespace_since_newline = 1;
3676 	  scan_p[0] = ' ';
3677 	  scan_p[1] = ' ';
3678 	  scan_p += 2;
3679 	  while (scan_p[1] != '/' || scan_p[0] != '*')
3680 	    {
3681 	      if (!ISSPACE ((const unsigned char)*scan_p))
3682 		*scan_p = ' ';
3683 	      if (++scan_p >= new_clean_text_limit)
3684 		abort ();
3685 	    }
3686 	  *scan_p++ = ' ';
3687 	  *scan_p = ' ';
3688 	  break;
3689 
3690 	case '#':			/* Handle pp directives.  */
3691 	  if (non_whitespace_since_newline)
3692 	    goto regular;
3693 	  *scan_p = ' ';
3694 	  while (scan_p[1] != '\n' || scan_p[0] == '\\')
3695 	    {
3696 	      if (!ISSPACE ((const unsigned char)*scan_p))
3697 		*scan_p = ' ';
3698 	      if (++scan_p >= new_clean_text_limit)
3699 		abort ();
3700 	    }
3701 	  *scan_p++ = ' ';
3702 	  break;
3703 
3704 	case '\'':			/* Handle character literals.  */
3705 	  non_whitespace_since_newline = 1;
3706 	  while (scan_p[1] != '\'' || scan_p[0] == '\\')
3707 	    {
3708 	      if (scan_p[0] == '\\'
3709 		  && !ISSPACE ((const unsigned char) scan_p[1]))
3710 		scan_p[1] = ' ';
3711 	      if (!ISSPACE ((const unsigned char)*scan_p))
3712 		*scan_p = ' ';
3713 	      if (++scan_p >= new_clean_text_limit)
3714 		abort ();
3715 	    }
3716 	  *scan_p++ = ' ';
3717 	  break;
3718 
3719 	case '"':			/* Handle string literals.  */
3720 	  non_whitespace_since_newline = 1;
3721 	  while (scan_p[1] != '"' || scan_p[0] == '\\')
3722 	    {
3723 	      if (scan_p[0] == '\\'
3724 		  && !ISSPACE ((const unsigned char) scan_p[1]))
3725 		scan_p[1] = ' ';
3726 	      if (!ISSPACE ((const unsigned char)*scan_p))
3727 		*scan_p = ' ';
3728 	      if (++scan_p >= new_clean_text_limit)
3729 		abort ();
3730 	    }
3731 	  if (!ISSPACE ((const unsigned char)*scan_p))
3732 	    *scan_p = ' ';
3733 	  scan_p++;
3734 	  break;
3735 
3736 	case '\\':			/* Handle line continuations.  */
3737 	  if (scan_p[1] != '\n')
3738 	    goto regular;
3739 	  *scan_p = ' ';
3740 	  break;
3741 
3742 	case '\n':
3743 	  non_whitespace_since_newline = 0;	/* Reset.  */
3744 	  break;
3745 
3746 	case ' ':
3747 	case '\v':
3748 	case '\t':
3749 	case '\r':
3750 	case '\f':
3751 	case '\b':
3752 	  break;		/* Whitespace characters.  */
3753 
3754 	default:
3755 regular:
3756 	  non_whitespace_since_newline = 1;
3757 	  break;
3758 	}
3759     }
3760 }
3761 
3762 /* Given a pointer to the closing right parenthesis for a particular formals
3763    list (in the clean text buffer) find the corresponding left parenthesis
3764    and return a pointer to it.  */
3765 
3766 static const char *
careful_find_l_paren(const char * p)3767 careful_find_l_paren (const char *p)
3768 {
3769   const char *q;
3770   int paren_depth;
3771 
3772   for (paren_depth = 1, q = p-1; paren_depth; check_source (--q >= clean_text_base, 0))
3773     {
3774       switch (*q)
3775 	{
3776 	case ')':
3777 	  paren_depth++;
3778 	  break;
3779 	case '(':
3780 	  paren_depth--;
3781 	  break;
3782 	}
3783     }
3784   return ++q;
3785 }
3786 
3787 /* Scan the clean text buffer for cases of function definitions that we
3788    don't really know about because they were preprocessed out when the
3789    aux info files were created.
3790 
3791    In this version of protoize/unprotoize we just give a warning for each
3792    one found.  A later version may be able to at least unprotoize such
3793    missed items.
3794 
3795    Note that we may easily find all function definitions simply by
3796    looking for places where there is a left paren which is (ignoring
3797    whitespace) immediately followed by either a left-brace or by an
3798    upper or lower case letter.  Whenever we find this combination, we
3799    have also found a function definition header.
3800 
3801    Finding function *declarations* using syntactic clues is much harder.
3802    I will probably try to do this in a later version though.  */
3803 
3804 static void
scan_for_missed_items(const file_info * file_p)3805 scan_for_missed_items (const file_info *file_p)
3806 {
3807   static const char *scan_p;
3808   const char *limit = clean_text_limit - 3;
3809   static const char *backup_limit;
3810 
3811   backup_limit = clean_text_base - 1;
3812 
3813   for (scan_p = clean_text_base; scan_p < limit; scan_p++)
3814     {
3815       if (*scan_p == ')')
3816 	{
3817 	  static const char *last_r_paren;
3818 	  const char *ahead_p;
3819 
3820 	  last_r_paren = scan_p;
3821 
3822 	  for (ahead_p = scan_p + 1; ISSPACE ((const unsigned char)*ahead_p); )
3823 	    check_source (++ahead_p < limit, limit);
3824 
3825 	  scan_p = ahead_p - 1;
3826 
3827 	  if (ISALPHA ((const unsigned char)*ahead_p) || *ahead_p == '{')
3828 	    {
3829 	      const char *last_l_paren;
3830 	      const int lineno = identify_lineno (ahead_p);
3831 
3832 	      if (setjmp (source_confusion_recovery))
3833 		continue;
3834 
3835 	      /* We know we have a function definition header.  Now skip
3836 	         leftwards over all of its associated formals lists.  */
3837 
3838 	      do
3839 		{
3840 		  last_l_paren = careful_find_l_paren (last_r_paren);
3841 		  for (last_r_paren = last_l_paren-1;
3842 		       ISSPACE ((const unsigned char)*last_r_paren); )
3843 		    check_source (--last_r_paren >= backup_limit, backup_limit);
3844 		}
3845 	      while (*last_r_paren == ')');
3846 
3847 	      if (is_id_char (*last_r_paren))
3848 		{
3849 		  const char *id_limit = last_r_paren + 1;
3850 		  const char *id_start;
3851 		  size_t id_length;
3852 		  const def_dec_info *dd_p;
3853 
3854 		  for (id_start = id_limit-1; is_id_char (*id_start); )
3855 		    check_source (--id_start >= backup_limit, backup_limit);
3856 		  id_start++;
3857 		  backup_limit = id_start;
3858 		  if ((id_length = (size_t) (id_limit - id_start)) == 0)
3859 		    goto not_missed;
3860 
3861 		  {
3862 		    char *func_name = alloca (id_length + 1);
3863 		    static const char * const stmt_keywords[]
3864 		      = { "if", "else", "do", "while", "for", "switch", "case", "return", 0 };
3865 		    const char * const *stmt_keyword;
3866 
3867 		    strncpy (func_name, id_start, id_length);
3868 		    func_name[id_length] = '\0';
3869 
3870 		    /* We must check here to see if we are actually looking at
3871 		       a statement rather than an actual function call.  */
3872 
3873 		    for (stmt_keyword = stmt_keywords; *stmt_keyword; stmt_keyword++)
3874 		      if (!strcmp (func_name, *stmt_keyword))
3875 			goto not_missed;
3876 
3877 #if 0
3878 		    notice ("%s: found definition of `%s' at %s(%d)\n",
3879 			    pname,
3880 			    func_name,
3881 			    shortpath (NULL, file_p->hash_entry->symbol),
3882 			    identify_lineno (id_start));
3883 #endif				/* 0 */
3884 		    /* We really should check for a match of the function name
3885 		       here also, but why bother.  */
3886 
3887 		    for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3888 		      if (dd_p->is_func_def && dd_p->line == lineno)
3889 			goto not_missed;
3890 
3891 		    /* If we make it here, then we did not know about this
3892 		       function definition.  */
3893 
3894 		    notice ("%s: %d: warning: `%s' excluded by preprocessing\n",
3895 			    shortpath (NULL, file_p->hash_entry->symbol),
3896 			    identify_lineno (id_start), func_name);
3897 		    notice ("%s: function definition not converted\n",
3898 			    pname);
3899 		  }
3900 		not_missed: ;
3901 	        }
3902 	    }
3903 	}
3904     }
3905 }
3906 
3907 /* Do all editing operations for a single source file (either a "base" file
3908    or an "include" file).  To do this we read the file into memory, keep a
3909    virgin copy there, make another cleaned in-core copy of the original file
3910    (i.e. one in which all of the comments and preprocessing directives have
3911    been replaced with whitespace), then use these two in-core copies of the
3912    file to make a new edited in-core copy of the file.  Finally, rename the
3913    original file (as a way of saving it), and then write the edited version
3914    of the file from core to a disk file of the same name as the original.
3915 
3916    Note that the trick of making a copy of the original sans comments &
3917    preprocessing directives make the editing a whole lot easier.  */
3918 
3919 static void
edit_file(const hash_table_entry * hp)3920 edit_file (const hash_table_entry *hp)
3921 {
3922   struct stat stat_buf;
3923   const file_info *file_p = hp->fip;
3924   char *new_orig_text_base;
3925   char *new_orig_text_limit;
3926   char *new_clean_text_base;
3927   char *new_clean_text_limit;
3928   size_t orig_size;
3929   size_t repl_size;
3930   int first_definition_in_file;
3931 
3932   /* If we are not supposed to be converting this file, or if there is
3933      nothing in there which needs converting, just skip this file.  */
3934 
3935   if (!needs_to_be_converted (file_p))
3936     return;
3937 
3938   convert_filename = file_p->hash_entry->symbol;
3939 
3940   /* Convert a file if it is in a directory where we want conversion
3941      and the file is not excluded.  */
3942 
3943   if (!directory_specified_p (convert_filename)
3944       || file_excluded_p (convert_filename))
3945     {
3946       if (!quiet_flag
3947 #ifdef UNPROTOIZE
3948 	  /* Don't even mention "system" include files unless we are
3949 	     protoizing.  If we are protoizing, we mention these as a
3950 	     gentle way of prodding the user to convert his "system"
3951 	     include files to prototype format.  */
3952 	  && !in_system_include_dir (convert_filename)
3953 #endif /* defined (UNPROTOIZE) */
3954 	  )
3955 	notice ("%s: `%s' not converted\n",
3956 		pname, shortpath (NULL, convert_filename));
3957       return;
3958     }
3959 
3960   /* Let the user know what we are up to.  */
3961 
3962   if (nochange_flag)
3963     notice ("%s: would convert file `%s'\n",
3964 	    pname, shortpath (NULL, convert_filename));
3965   else
3966     notice ("%s: converting file `%s'\n",
3967 	    pname, shortpath (NULL, convert_filename));
3968   fflush (stderr);
3969 
3970   /* Find out the size (in bytes) of the original file.  */
3971 
3972   /* The cast avoids an erroneous warning on AIX.  */
3973   if (stat (convert_filename, &stat_buf) == -1)
3974     {
3975       int errno_val = errno;
3976       notice ("%s: can't get status for file `%s': %s\n",
3977 	      pname, shortpath (NULL, convert_filename),
3978 	      xstrerror (errno_val));
3979       return;
3980     }
3981   orig_size = stat_buf.st_size;
3982 
3983   /* Allocate a buffer to hold the original text.  */
3984 
3985   orig_text_base = new_orig_text_base = xmalloc (orig_size + 2);
3986   orig_text_limit = new_orig_text_limit = new_orig_text_base + orig_size;
3987 
3988   /* Allocate a buffer to hold the cleaned-up version of the original text.  */
3989 
3990   clean_text_base = new_clean_text_base = xmalloc (orig_size + 2);
3991   clean_text_limit = new_clean_text_limit = new_clean_text_base + orig_size;
3992   clean_read_ptr = clean_text_base - 1;
3993 
3994   /* Allocate a buffer that will hopefully be large enough to hold the entire
3995      converted output text.  As an initial guess for the maximum size of the
3996      output buffer, use 125% of the size of the original + some extra.  This
3997      buffer can be expanded later as needed.  */
3998 
3999   repl_size = orig_size + (orig_size >> 2) + 4096;
4000   repl_text_base = xmalloc (repl_size + 2);
4001   repl_text_limit = repl_text_base + repl_size - 1;
4002   repl_write_ptr = repl_text_base - 1;
4003 
4004   {
4005     int input_file;
4006     int fd_flags;
4007 
4008     /* Open the file to be converted in READ ONLY mode.  */
4009 
4010     fd_flags = O_RDONLY;
4011 #ifdef O_BINARY
4012     /* Use binary mode to avoid having to deal with different EOL characters.  */
4013     fd_flags |= O_BINARY;
4014 #endif
4015     if ((input_file = open (convert_filename, fd_flags, 0444)) == -1)
4016       {
4017 	int errno_val = errno;
4018 	notice ("%s: can't open file `%s' for reading: %s\n",
4019 		pname, shortpath (NULL, convert_filename),
4020 		xstrerror (errno_val));
4021 	return;
4022       }
4023 
4024     /* Read the entire original source text file into the original text buffer
4025        in one swell fwoop.  Then figure out where the end of the text is and
4026        make sure that it ends with a newline followed by a null.  */
4027 
4028     if (safe_read (input_file, new_orig_text_base, orig_size) !=
4029 	(int) orig_size)
4030       {
4031 	int errno_val = errno;
4032 	close (input_file);
4033 	notice ("\n%s: error reading input file `%s': %s\n",
4034 		pname, shortpath (NULL, convert_filename),
4035 		xstrerror (errno_val));
4036 	return;
4037       }
4038 
4039     close (input_file);
4040   }
4041 
4042   if (orig_size == 0 || orig_text_limit[-1] != '\n')
4043     {
4044       *new_orig_text_limit++ = '\n';
4045       orig_text_limit++;
4046     }
4047 
4048   /* Create the cleaned up copy of the original text.  */
4049 
4050   memcpy (new_clean_text_base, orig_text_base,
4051 	  (size_t) (orig_text_limit - orig_text_base));
4052   do_cleaning (new_clean_text_base, new_clean_text_limit);
4053 
4054 #if 0
4055   {
4056     int clean_file;
4057     size_t clean_size = orig_text_limit - orig_text_base;
4058     char *const clean_filename = alloca (strlen (convert_filename) + 6 + 1);
4059 
4060     /* Open (and create) the clean file.  */
4061 
4062     strcpy (clean_filename, convert_filename);
4063     strcat (clean_filename, ".clean");
4064     if ((clean_file = creat (clean_filename, 0666)) == -1)
4065       {
4066 	int errno_val = errno;
4067 	notice ("%s: can't create/open clean file `%s': %s\n",
4068 		pname, shortpath (NULL, clean_filename),
4069 		xstrerror (errno_val));
4070 	return;
4071       }
4072 
4073     /* Write the clean file.  */
4074 
4075     safe_write (clean_file, new_clean_text_base, clean_size, clean_filename);
4076 
4077     close (clean_file);
4078   }
4079 #endif /* 0 */
4080 
4081   /* Do a simplified scan of the input looking for things that were not
4082      mentioned in the aux info files because of the fact that they were
4083      in a region of the source which was preprocessed-out (via #if or
4084      via #ifdef).  */
4085 
4086   scan_for_missed_items (file_p);
4087 
4088   /* Setup to do line-oriented forward seeking in the clean text buffer.  */
4089 
4090   last_known_line_number = 1;
4091   last_known_line_start = clean_text_base;
4092 
4093   /* Now get down to business and make all of the necessary edits.  */
4094 
4095   {
4096     const def_dec_info *def_dec_p;
4097 
4098     first_definition_in_file = 1;
4099     def_dec_p = file_p->defs_decs;
4100     for (; def_dec_p; def_dec_p = def_dec_p->next_in_file)
4101       {
4102 	const char *clean_text_p = seek_to_line (def_dec_p->line);
4103 
4104 	/* clean_text_p now points to the first character of the line which
4105 	   contains the `terminator' for the declaration or definition that
4106 	   we are about to process.  */
4107 
4108 #ifndef UNPROTOIZE
4109 
4110 	if (global_flag && def_dec_p->is_func_def && first_definition_in_file)
4111 	  {
4112 	    add_global_decls (def_dec_p->file, clean_text_p);
4113 	    first_definition_in_file = 0;
4114 	  }
4115 
4116 	/* Don't edit this item if it is already in prototype format or if it
4117 	   is a function declaration and we have found no corresponding
4118 	   definition.  */
4119 
4120 	if (def_dec_p->prototyped
4121 	    || (!def_dec_p->is_func_def && !def_dec_p->definition))
4122 	  continue;
4123 
4124 #endif /* !defined (UNPROTOIZE) */
4125 
4126 	if (def_dec_p->is_func_def)
4127 	  edit_fn_definition (def_dec_p, clean_text_p);
4128 	else
4129 #ifndef UNPROTOIZE
4130 	if (def_dec_p->is_implicit)
4131 	  add_local_decl (def_dec_p, clean_text_p);
4132 	else
4133 #endif /* !defined (UNPROTOIZE) */
4134 	  edit_fn_declaration (def_dec_p, clean_text_p);
4135       }
4136   }
4137 
4138   /* Finalize things.  Output the last trailing part of the original text.  */
4139 
4140   output_up_to (clean_text_limit - 1);
4141 
4142   /* If this is just a test run, stop now and just deallocate the buffers.  */
4143 
4144   if (nochange_flag)
4145     {
4146       free (new_orig_text_base);
4147       free (new_clean_text_base);
4148       free (repl_text_base);
4149       return;
4150     }
4151 
4152   /* Change the name of the original input file.  This is just a quick way of
4153      saving the original file.  */
4154 
4155   if (!nosave_flag)
4156     {
4157       char *new_filename
4158 	= xmalloc (strlen (convert_filename) + strlen (save_suffix) + 2);
4159 
4160       strcpy (new_filename, convert_filename);
4161 #ifdef __MSDOS__
4162       /* MSDOS filenames are restricted to 8.3 format, so we save `foo.c'
4163 	 as `foo.<save_suffix>'.  */
4164       new_filename[(strlen (convert_filename) - 1] = '\0';
4165 #endif
4166       strcat (new_filename, save_suffix);
4167 
4168       /* Don't overwrite existing file.  */
4169       if (access (new_filename, F_OK) == 0)
4170 	{
4171 	  if (!quiet_flag)
4172 	    notice ("%s: warning: file `%s' already saved in `%s'\n",
4173 		    pname,
4174 		    shortpath (NULL, convert_filename),
4175 		    shortpath (NULL, new_filename));
4176 	}
4177       else if (rename (convert_filename, new_filename) == -1)
4178 	{
4179 	  int errno_val = errno;
4180 	  notice ("%s: can't link file `%s' to `%s': %s\n",
4181 		  pname,
4182 		  shortpath (NULL, convert_filename),
4183 		  shortpath (NULL, new_filename),
4184 		  xstrerror (errno_val));
4185 	  return;
4186 	}
4187     }
4188 
4189   if (unlink (convert_filename) == -1)
4190     {
4191       int errno_val = errno;
4192       /* The file may have already been renamed.  */
4193       if (errno_val != ENOENT)
4194 	{
4195 	  notice ("%s: can't delete file `%s': %s\n",
4196 		  pname, shortpath (NULL, convert_filename),
4197 		  xstrerror (errno_val));
4198 	  return;
4199 	}
4200     }
4201 
4202   {
4203     int output_file;
4204 
4205     /* Open (and create) the output file.  */
4206 
4207     if ((output_file = creat (convert_filename, 0666)) == -1)
4208       {
4209 	int errno_val = errno;
4210 	notice ("%s: can't create/open output file `%s': %s\n",
4211 		pname, shortpath (NULL, convert_filename),
4212 		xstrerror (errno_val));
4213 	return;
4214       }
4215 #ifdef O_BINARY
4216     /* Use binary mode to avoid changing the existing EOL character.  */
4217     setmode (output_file, O_BINARY);
4218 #endif
4219 
4220     /* Write the output file.  */
4221 
4222     {
4223       unsigned int out_size = (repl_write_ptr + 1) - repl_text_base;
4224 
4225       safe_write (output_file, repl_text_base, out_size, convert_filename);
4226     }
4227 
4228     close (output_file);
4229   }
4230 
4231   /* Deallocate the conversion buffers.  */
4232 
4233   free (new_orig_text_base);
4234   free (new_clean_text_base);
4235   free (repl_text_base);
4236 
4237   /* Change the mode of the output file to match the original file.  */
4238 
4239   /* The cast avoids an erroneous warning on AIX.  */
4240   if (chmod (convert_filename, stat_buf.st_mode) == -1)
4241     {
4242       int errno_val = errno;
4243       notice ("%s: can't change mode of file `%s': %s\n",
4244 	      pname, shortpath (NULL, convert_filename),
4245 	      xstrerror (errno_val));
4246     }
4247 
4248   /* Note:  We would try to change the owner and group of the output file
4249      to match those of the input file here, except that may not be a good
4250      thing to do because it might be misleading.  Also, it might not even
4251      be possible to do that (on BSD systems with quotas for instance).  */
4252 }
4253 
4254 /* Do all of the individual steps needed to do the protoization (or
4255    unprotoization) of the files referenced in the aux_info files given
4256    in the command line.  */
4257 
4258 static void
4259 do_processing (void)
4260 {
4261   const char * const *base_pp;
4262   const char * const * const end_pps
4263     = &base_source_filenames[n_base_source_files];
4264 
4265 #ifndef UNPROTOIZE
4266   int syscalls_len;
4267 #endif /* !defined (UNPROTOIZE) */
4268 
4269   /* One-by-one, check (and create if necessary), open, and read all of the
4270      stuff in each aux_info file.  After reading each aux_info file, the
4271      aux_info_file just read will be automatically deleted unless the
4272      keep_flag is set.  */
4273 
4274   for (base_pp = base_source_filenames; base_pp < end_pps; base_pp++)
4275     process_aux_info_file (*base_pp, keep_flag, 0);
4276 
4277 #ifndef UNPROTOIZE
4278 
4279   /* Also open and read the special SYSCALLS.c aux_info file which gives us
4280      the prototypes for all of the standard system-supplied functions.  */
4281 
4282   if (nondefault_syscalls_dir)
4283     {
4284       syscalls_absolute_filename
4285 	= xmalloc (strlen (nondefault_syscalls_dir) + 1
4286 		   + sizeof (syscalls_filename));
4287       strcpy (syscalls_absolute_filename, nondefault_syscalls_dir);
4288     }
4289   else
4290     {
4291       GET_ENVIRONMENT (default_syscalls_dir, "GCC_EXEC_PREFIX");
4292       if (!default_syscalls_dir)
4293 	{
4294 	  default_syscalls_dir = standard_exec_prefix;
4295 	}
4296       syscalls_absolute_filename
4297 	= xmalloc (strlen (default_syscalls_dir) + 0
4298 		   + strlen (target_machine) + 1
4299 		   + strlen (target_version) + 1
4300 		   + sizeof (syscalls_filename));
4301       strcpy (syscalls_absolute_filename, default_syscalls_dir);
4302       strcat (syscalls_absolute_filename, target_machine);
4303       strcat (syscalls_absolute_filename, "/");
4304       strcat (syscalls_absolute_filename, target_version);
4305       strcat (syscalls_absolute_filename, "/");
4306     }
4307 
4308   syscalls_len = strlen (syscalls_absolute_filename);
4309   if (! IS_DIR_SEPARATOR (*(syscalls_absolute_filename + syscalls_len - 1)))
4310     {
4311       *(syscalls_absolute_filename + syscalls_len++) = DIR_SEPARATOR;
4312       *(syscalls_absolute_filename + syscalls_len) = '\0';
4313     }
4314   strcat (syscalls_absolute_filename, syscalls_filename);
4315 
4316   /* Call process_aux_info_file in such a way that it does not try to
4317      delete the SYSCALLS aux_info file.  */
4318 
4319   process_aux_info_file (syscalls_absolute_filename, 1, 1);
4320 
4321 #endif /* !defined (UNPROTOIZE) */
4322 
4323   /* When we first read in all of the information from the aux_info files
4324      we saved in it descending line number order, because that was likely to
4325      be faster.  Now however, we want the chains of def & dec records to
4326      appear in ascending line number order as we get further away from the
4327      file_info record that they hang from.  The following line causes all of
4328      these lists to be rearranged into ascending line number order.  */
4329 
4330   visit_each_hash_node (filename_primary, reverse_def_dec_list);
4331 
4332 #ifndef UNPROTOIZE
4333 
4334   /* Now do the "real" work.  The following line causes each declaration record
4335      to be "visited".  For each of these nodes, an attempt is made to match
4336      up the function declaration with a corresponding function definition,
4337      which should have a full prototype-format formals list with it.  Once
4338      these match-ups are made, the conversion of the function declarations
4339      to prototype format can be made.  */
4340 
4341   visit_each_hash_node (function_name_primary, connect_defs_and_decs);
4342 
4343 #endif /* !defined (UNPROTOIZE) */
4344 
4345   /* Now convert each file that can be converted (and needs to be).  */
4346 
4347   visit_each_hash_node (filename_primary, edit_file);
4348 
4349 #ifndef UNPROTOIZE
4350 
4351   /* If we are working in cplusplus mode, try to rename all .c files to .C
4352      files.  Don't panic if some of the renames don't work.  */
4353 
4354   if (cplusplus_flag && !nochange_flag)
4355     visit_each_hash_node (filename_primary, rename_c_file);
4356 
4357 #endif /* !defined (UNPROTOIZE) */
4358 }
4359 
4360 static const struct option longopts[] =
4361 {
4362   {"version", 0, 0, 'V'},
4363   {"file_name", 0, 0, 'p'},
4364   {"quiet", 0, 0, 'q'},
4365   {"silent", 0, 0, 'q'},
4366   {"force", 0, 0, 'f'},
4367   {"keep", 0, 0, 'k'},
4368   {"nosave", 0, 0, 'N'},
4369   {"nochange", 0, 0, 'n'},
4370   {"compiler-options", 1, 0, 'c'},
4371   {"exclude", 1, 0, 'x'},
4372   {"directory", 1, 0, 'd'},
4373 #ifdef UNPROTOIZE
4374   {"indent", 1, 0, 'i'},
4375 #else
4376   {"local", 0, 0, 'l'},
4377   {"global", 0, 0, 'g'},
4378   {"c++", 0, 0, 'C'},
4379   {"syscalls-dir", 1, 0, 'B'},
4380 #endif
4381   {0, 0, 0, 0}
4382 };
4383 
4384 extern int main (int, char **const);
4385 
4386 int
4387 main (int argc, char **const argv)
4388 {
4389   int longind;
4390   int c;
4391   const char *params = "";
4392 
4393   pname = strrchr (argv[0], DIR_SEPARATOR);
4394 #ifdef DIR_SEPARATOR_2
4395   {
4396     char *slash;
4397 
4398     slash = strrchr (pname ? pname : argv[0], DIR_SEPARATOR_2);
4399     if (slash)
4400       pname = slash;
4401   }
4402 #endif
4403   pname = pname ? pname+1 : argv[0];
4404 
4405 #ifdef SIGCHLD
4406   /* We *MUST* set SIGCHLD to SIG_DFL so that the wait4() call will
4407      receive the signal.  A different setting is inheritable */
4408   signal (SIGCHLD, SIG_DFL);
4409 #endif
4410 
4411   gcc_init_libintl ();
4412 
4413   cwd_buffer = getpwd ();
4414   if (!cwd_buffer)
4415     {
4416       notice ("%s: cannot get working directory: %s\n",
4417 	      pname, xstrerror(errno));
4418       return (FATAL_EXIT_CODE);
4419     }
4420 
4421   /* By default, convert the files in the current directory.  */
4422   directory_list = string_list_cons (cwd_buffer, NULL);
4423 
4424   while ((c = getopt_long (argc, argv,
4425 #ifdef UNPROTOIZE
4426 			   "c:d:i:knNp:qvVx:",
4427 #else
4428 			   "B:c:Cd:gklnNp:qvVx:",
4429 #endif
4430 			   longopts, &longind)) != EOF)
4431     {
4432       if (c == 0)		/* Long option.  */
4433 	c = longopts[longind].val;
4434       switch (c)
4435 	{
4436 	case 'p':
4437 	  compiler_file_name = optarg;
4438 	  break;
4439 	case 'd':
4440 	  directory_list
4441 	    = string_list_cons (abspath (NULL, optarg), directory_list);
4442 	  break;
4443 	case 'x':
4444 	  exclude_list = string_list_cons (optarg, exclude_list);
4445 	  break;
4446 
4447 	case 'v':
4448 	case 'V':
4449 	  version_flag = 1;
4450 	  break;
4451 	case 'q':
4452 	  quiet_flag = 1;
4453 	  break;
4454 #if 0
4455 	case 'f':
4456 	  force_flag = 1;
4457 	  break;
4458 #endif
4459 	case 'n':
4460 	  nochange_flag = 1;
4461 	  keep_flag = 1;
4462 	  break;
4463 	case 'N':
4464 	  nosave_flag = 1;
4465 	  break;
4466 	case 'k':
4467 	  keep_flag = 1;
4468 	  break;
4469 	case 'c':
4470 	  params = optarg;
4471 	  break;
4472 #ifdef UNPROTOIZE
4473 	case 'i':
4474 	  indent_string = optarg;
4475 	  break;
4476 #else				/* !defined (UNPROTOIZE) */
4477 	case 'l':
4478 	  local_flag = 1;
4479 	  break;
4480 	case 'g':
4481 	  global_flag = 1;
4482 	  break;
4483 	case 'C':
4484 	  cplusplus_flag = 1;
4485 	  break;
4486 	case 'B':
4487 	  nondefault_syscalls_dir = optarg;
4488 	  break;
4489 #endif				/* !defined (UNPROTOIZE) */
4490 	default:
4491 	  usage ();
4492 	}
4493     }
4494 
4495   /* Set up compile_params based on -p and -c options.  */
4496   munge_compile_params (params);
4497 
4498   n_base_source_files = argc - optind;
4499 
4500   /* Now actually make a list of the base source filenames.  */
4501 
4502   base_source_filenames
4503     = xmalloc ((n_base_source_files + 1) * sizeof (char *));
4504   n_base_source_files = 0;
4505   for (; optind < argc; optind++)
4506     {
4507       const char *path = abspath (NULL, argv[optind]);
4508       int len = strlen (path);
4509 
4510       if (path[len-1] == 'c' && path[len-2] == '.')
4511 	base_source_filenames[n_base_source_files++] = path;
4512       else
4513 	{
4514 	  notice ("%s: input file names must have .c suffixes: %s\n",
4515 		  pname, shortpath (NULL, path));
4516 	  errors++;
4517 	}
4518     }
4519 
4520 #ifndef UNPROTOIZE
4521   /* We are only interested in the very first identifier token in the
4522      definition of `va_list', so if there is more junk after that first
4523      identifier token, delete it from the `varargs_style_indicator'.  */
4524   {
4525     const char *cp;
4526 
4527     for (cp = varargs_style_indicator; ISIDNUM (*cp); cp++)
4528       continue;
4529     if (*cp != 0)
4530       varargs_style_indicator = savestring (varargs_style_indicator,
4531 					    cp - varargs_style_indicator);
4532   }
4533 #endif /* !defined (UNPROTOIZE) */
4534 
4535   if (errors)
4536     usage ();
4537   else
4538     {
4539       if (version_flag)
4540 	fprintf (stderr, "%s: %s\n", pname, version_string);
4541       do_processing ();
4542     }
4543 
4544   return (errors ? FATAL_EXIT_CODE : SUCCESS_EXIT_CODE);
4545 }
4546