Merge from mainline (gomp-merge-2005-02-26).
[official-gcc.git] / gcc / protoize.c
blob093e4400817223498375b58a3982f710b5e78649
1 /* Protoize program - Original version by Ron Guilmette (rfg@segfault.us.com).
2 Copyright (C) 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3 1999, 2000, 2001, 2002, 2003, 2004 Free Software Foundation, Inc.
5 This file is part of GCC.
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.
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.
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. */
22 #include "config.h"
23 #include "system.h"
24 #include "coretypes.h"
25 #include "tm.h"
26 #include "intl.h"
27 #include "cppdefault.h"
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"
40 /* Include getopt.h for the sake of getopt_long. */
41 #include "getopt.h"
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
50 /* Macro to see if the paths match. */
51 #define IS_SAME_PATH(a,b) (FILENAME_CMP (a, b) == 0)
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
60 /* Suffix for saved files. */
61 #ifdef __MSDOS__
62 #define SAVE_SUFFIX "sav"
63 #else
64 #define SAVE_SUFFIX ".save"
65 #endif
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
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 static void notice (const char *, ...) ATTRIBUTE_PRINTF_1;
79 static char *savestring (const char *, unsigned int);
80 static char *dupnstr (const char *, size_t);
81 static int safe_read (int, void *, int);
82 static void safe_write (int, void *, int, const char *);
83 static void save_pointers (void);
84 static void restore_pointers (void);
85 static int is_id_char (int);
86 static int in_system_include_dir (const char *);
87 static int directory_specified_p (const char *);
88 static int file_excluded_p (const char *);
89 static char *unexpand_if_needed (const char *);
90 static char *abspath (const char *, const char *);
91 static void check_aux_info (int);
92 static const char *find_corresponding_lparen (const char *);
93 static int referenced_file_is_newer (const char *, time_t);
94 static void save_def_or_dec (const char *, int);
95 static void munge_compile_params (const char *);
96 static int gen_aux_info_file (const char *);
97 static void process_aux_info_file (const char *, int, int);
98 static int identify_lineno (const char *);
99 static void check_source (int, const char *);
100 static const char *seek_to_line (int);
101 static const char *forward_to_next_token_char (const char *);
102 static void output_bytes (const char *, size_t);
103 static void output_string (const char *);
104 static void output_up_to (const char *);
105 static int other_variable_style_function (const char *);
106 static const char *find_rightmost_formals_list (const char *);
107 static void do_cleaning (char *, const char *);
108 static const char *careful_find_l_paren (const char *);
109 static void do_processing (void);
111 /* Look for these where the `const' qualifier is intentionally cast aside. */
112 #define NONCONST
114 /* Define a default place to find the SYSCALLS.X file. */
116 #ifndef UNPROTOIZE
118 #ifndef STANDARD_EXEC_PREFIX
119 #define STANDARD_EXEC_PREFIX "/usr/local/lib/gcc-lib/"
120 #endif /* !defined STANDARD_EXEC_PREFIX */
122 static const char * const standard_exec_prefix = STANDARD_EXEC_PREFIX;
123 static const char * const target_machine = DEFAULT_TARGET_MACHINE;
124 static const char * const target_version = DEFAULT_TARGET_VERSION;
126 #endif /* !defined (UNPROTOIZE) */
128 /* Suffix of aux_info files. */
130 static const char * const aux_info_suffix = AUX_INFO_SUFFIX;
132 /* String to attach to filenames for saved versions of original files. */
134 static const char * const save_suffix = SAVE_SUFFIX;
136 #ifndef UNPROTOIZE
138 /* String to attach to C filenames renamed to C++. */
140 static const char * const cplus_suffix = CPLUS_FILE_SUFFIX;
142 /* File name of the file which contains descriptions of standard system
143 routines. Note that we never actually do anything with this file per se,
144 but we do read in its corresponding aux_info file. */
146 static const char syscalls_filename[] = "SYSCALLS.c";
148 /* Default place to find the above file. */
150 static const char * default_syscalls_dir;
152 /* Variable to hold the complete absolutized filename of the SYSCALLS.c.X
153 file. */
155 static char * syscalls_absolute_filename;
157 #endif /* !defined (UNPROTOIZE) */
159 /* Type of the structure that holds information about macro unexpansions. */
161 struct unexpansion_struct {
162 const char *const expanded;
163 const char *const contracted;
165 typedef struct unexpansion_struct unexpansion;
167 /* A table of conversions that may need to be made for some (stupid) older
168 operating systems where these types are preprocessor macros rather than
169 typedefs (as they really ought to be).
171 WARNING: The contracted forms must be as small (or smaller) as the
172 expanded forms, or else havoc will ensue. */
174 static const unexpansion unexpansions[] = {
175 { "struct _iobuf", "FILE" },
176 { 0, 0 }
179 /* The number of "primary" slots in the hash tables for filenames and for
180 function names. This can be as big or as small as you like, except that
181 it must be a power of two. */
183 #define HASH_TABLE_SIZE (1 << 9)
185 /* Bit mask to use when computing hash values. */
187 static const int hash_mask = (HASH_TABLE_SIZE - 1);
190 /* Datatype for lists of directories or filenames. */
191 struct string_list
193 const char *name;
194 struct string_list *next;
197 static struct string_list *string_list_cons (const char *,
198 struct string_list *);
200 /* List of directories in which files should be converted. */
202 struct string_list *directory_list;
204 /* List of file names which should not be converted.
205 A file is excluded if the end of its name, following a /,
206 matches one of the names in this list. */
208 struct string_list *exclude_list;
210 /* The name of the other style of variable-number-of-parameters functions
211 (i.e. the style that we want to leave unconverted because we don't yet
212 know how to convert them to this style. This string is used in warning
213 messages. */
215 /* Also define here the string that we can search for in the parameter lists
216 taken from the .X files which will unambiguously indicate that we have
217 found a varargs style function. */
219 #ifdef UNPROTOIZE
220 static const char * const other_var_style = "stdarg";
221 #else /* !defined (UNPROTOIZE) */
222 static const char * const other_var_style = "varargs";
223 static const char *varargs_style_indicator = "va_alist";
224 #endif /* !defined (UNPROTOIZE) */
226 /* The following two types are used to create hash tables. In this program,
227 there are two hash tables which are used to store and quickly lookup two
228 different classes of strings. The first type of strings stored in the
229 first hash table are absolute filenames of files which protoize needs to
230 know about. The second type of strings (stored in the second hash table)
231 are function names. It is this second class of strings which really
232 inspired the use of the hash tables, because there may be a lot of them. */
234 typedef struct hash_table_entry_struct hash_table_entry;
236 /* Do some typedefs so that we don't have to write "struct" so often. */
238 typedef struct def_dec_info_struct def_dec_info;
239 typedef struct file_info_struct file_info;
240 typedef struct f_list_chain_item_struct f_list_chain_item;
242 #ifndef UNPROTOIZE
243 static int is_syscalls_file (const file_info *);
244 static void rename_c_file (const hash_table_entry *);
245 static const def_dec_info *find_extern_def (const def_dec_info *,
246 const def_dec_info *);
247 static const def_dec_info *find_static_definition (const def_dec_info *);
248 static void connect_defs_and_decs (const hash_table_entry *);
249 static void add_local_decl (const def_dec_info *, const char *);
250 static void add_global_decls (const file_info *, const char *);
251 #endif /* ! UNPROTOIZE */
252 static int needs_to_be_converted (const file_info *);
253 static void visit_each_hash_node (const hash_table_entry *,
254 void (*)(const hash_table_entry *));
255 static hash_table_entry *add_symbol (hash_table_entry *, const char *);
256 static hash_table_entry *lookup (hash_table_entry *, const char *);
257 static void free_def_dec (def_dec_info *);
258 static file_info *find_file (const char *, int);
259 static void reverse_def_dec_list (const hash_table_entry *);
260 static void edit_fn_declaration (const def_dec_info *, const char *);
261 static int edit_formals_lists (const char *, unsigned int,
262 const def_dec_info *);
263 static void edit_fn_definition (const def_dec_info *, const char *);
264 static void scan_for_missed_items (const file_info *);
265 static void edit_file (const hash_table_entry *);
267 /* In the struct below, note that the "_info" field has two different uses
268 depending on the type of hash table we are in (i.e. either the filenames
269 hash table or the function names hash table). In the filenames hash table
270 the info fields of the entries point to the file_info struct which is
271 associated with each filename (1 per filename). In the function names
272 hash table, the info field points to the head of a singly linked list of
273 def_dec_info entries which are all defs or decs of the function whose
274 name is pointed to by the "symbol" field. Keeping all of the defs/decs
275 for a given function name on a special list specifically for that function
276 name makes it quick and easy to find out all of the important information
277 about a given (named) function. */
279 struct hash_table_entry_struct {
280 hash_table_entry * hash_next; /* -> to secondary entries */
281 const char * symbol; /* -> to the hashed string */
282 union {
283 const def_dec_info * _ddip;
284 file_info * _fip;
285 } _info;
287 #define ddip _info._ddip
288 #define fip _info._fip
290 /* Define a type specifically for our two hash tables. */
292 typedef hash_table_entry hash_table[HASH_TABLE_SIZE];
294 /* The following struct holds all of the important information about any
295 single filename (e.g. file) which we need to know about. */
297 struct file_info_struct {
298 const hash_table_entry * hash_entry; /* -> to associated hash entry */
299 const def_dec_info * defs_decs; /* -> to chain of defs/decs */
300 time_t mtime; /* Time of last modification. */
303 /* Due to the possibility that functions may return pointers to functions,
304 (which may themselves have their own parameter lists) and due to the
305 fact that returned pointers-to-functions may be of type "pointer-to-
306 function-returning-pointer-to-function" (ad nauseum) we have to keep
307 an entire chain of ANSI style formal parameter lists for each function.
309 Normally, for any given function, there will only be one formals list
310 on the chain, but you never know.
312 Note that the head of each chain of formals lists is pointed to by the
313 `f_list_chain' field of the corresponding def_dec_info record.
315 For any given chain, the item at the head of the chain is the *leftmost*
316 parameter list seen in the actual C language function declaration. If
317 there are other members of the chain, then these are linked in left-to-right
318 order from the head of the chain. */
320 struct f_list_chain_item_struct {
321 const f_list_chain_item * chain_next; /* -> to next item on chain */
322 const char * formals_list; /* -> to formals list string */
325 /* The following struct holds all of the important information about any
326 single function definition or declaration which we need to know about.
327 Note that for unprotoize we don't need to know very much because we
328 never even create records for stuff that we don't intend to convert
329 (like for instance defs and decs which are already in old K&R format
330 and "implicit" function declarations). */
332 struct def_dec_info_struct {
333 const def_dec_info * next_in_file; /* -> to rest of chain for file */
334 file_info * file; /* -> file_info for containing file */
335 int line; /* source line number of def/dec */
336 const char * ansi_decl; /* -> left end of ansi decl */
337 hash_table_entry * hash_entry; /* -> hash entry for function name */
338 unsigned int is_func_def; /* = 0 means this is a declaration */
339 const def_dec_info * next_for_func; /* -> to rest of chain for func name */
340 unsigned int f_list_count; /* count of formals lists we expect */
341 char prototyped; /* = 0 means already prototyped */
342 #ifndef UNPROTOIZE
343 const f_list_chain_item * f_list_chain; /* -> chain of formals lists */
344 const def_dec_info * definition; /* -> def/dec containing related def */
345 char is_static; /* = 0 means visibility is "extern" */
346 char is_implicit; /* != 0 for implicit func decl's */
347 char written; /* != 0 means written for implicit */
348 #else /* !defined (UNPROTOIZE) */
349 const char * formal_names; /* -> to list of names of formals */
350 const char * formal_decls; /* -> to string of formal declarations */
351 #endif /* !defined (UNPROTOIZE) */
354 /* Pointer to the tail component of the filename by which this program was
355 invoked. Used everywhere in error and warning messages. */
357 static const char *pname;
359 /* Error counter. Will be nonzero if we should give up at the next convenient
360 stopping point. */
362 static int errors = 0;
364 /* Option flags. */
365 /* ??? These comments should say what the flag mean as well as the options
366 that set them. */
368 /* File name to use for running gcc. Allows GCC 2 to be named
369 something other than gcc. */
370 static const char *compiler_file_name = "gcc";
372 static int version_flag = 0; /* Print our version number. */
373 static int quiet_flag = 0; /* Don't print messages normally. */
374 static int nochange_flag = 0; /* Don't convert, just say what files
375 we would have converted. */
376 static int nosave_flag = 0; /* Don't save the old version. */
377 static int keep_flag = 0; /* Don't delete the .X files. */
378 static const char ** compile_params = 0; /* Option string for gcc. */
379 #ifdef UNPROTOIZE
380 static const char *indent_string = " "; /* Indentation for newly
381 inserted parm decls. */
382 #else /* !defined (UNPROTOIZE) */
383 static int local_flag = 0; /* Insert new local decls (when?). */
384 static int global_flag = 0; /* set by -g option */
385 static int cplusplus_flag = 0; /* Rename converted files to *.C. */
386 static const char *nondefault_syscalls_dir = 0; /* Dir to look for
387 SYSCALLS.c.X in. */
388 #endif /* !defined (UNPROTOIZE) */
390 /* An index into the compile_params array where we should insert the source
391 file name when we are ready to exec the C compiler. A zero value indicates
392 that we have not yet called munge_compile_params. */
394 static int input_file_name_index = 0;
396 /* An index into the compile_params array where we should insert the filename
397 for the aux info file, when we run the C compiler. */
398 static int aux_info_file_name_index = 0;
400 /* Count of command line arguments which were "filename" arguments. */
402 static int n_base_source_files = 0;
404 /* Points to a malloc'ed list of pointers to all of the filenames of base
405 source files which were specified on the command line. */
407 static const char **base_source_filenames;
409 /* Line number of the line within the current aux_info file that we
410 are currently processing. Used for error messages in case the prototypes
411 info file is corrupted somehow. */
413 static int current_aux_info_lineno;
415 /* Pointer to the name of the source file currently being converted. */
417 static const char *convert_filename;
419 /* Pointer to relative root string (taken from aux_info file) which indicates
420 where directory the user was in when he did the compilation step that
421 produced the containing aux_info file. */
423 static const char *invocation_filename;
425 /* Pointer to the base of the input buffer that holds the original text for the
426 source file currently being converted. */
428 static const char *orig_text_base;
430 /* Pointer to the byte just beyond the end of the input buffer that holds the
431 original text for the source file currently being converted. */
433 static const char *orig_text_limit;
435 /* Pointer to the base of the input buffer that holds the cleaned text for the
436 source file currently being converted. */
438 static const char *clean_text_base;
440 /* Pointer to the byte just beyond the end of the input buffer that holds the
441 cleaned text for the source file currently being converted. */
443 static const char *clean_text_limit;
445 /* Pointer to the last byte in the cleaned text buffer that we have already
446 (virtually) copied to the output buffer (or decided to ignore). */
448 static const char * clean_read_ptr;
450 /* Pointer to the base of the output buffer that holds the replacement text
451 for the source file currently being converted. */
453 static char *repl_text_base;
455 /* Pointer to the byte just beyond the end of the output buffer that holds the
456 replacement text for the source file currently being converted. */
458 static char *repl_text_limit;
460 /* Pointer to the last byte which has been stored into the output buffer.
461 The next byte to be stored should be stored just past where this points
462 to. */
464 static char * repl_write_ptr;
466 /* Pointer into the cleaned text buffer for the source file we are currently
467 converting. This points to the first character of the line that we last
468 did a "seek_to_line" to (see below). */
470 static const char *last_known_line_start;
472 /* Number of the line (in the cleaned text buffer) that we last did a
473 "seek_to_line" to. Will be one if we just read a new source file
474 into the cleaned text buffer. */
476 static int last_known_line_number;
478 /* The filenames hash table. */
480 static hash_table filename_primary;
482 /* The function names hash table. */
484 static hash_table function_name_primary;
486 /* The place to keep the recovery address which is used only in cases where
487 we get hopelessly confused by something in the cleaned original text. */
489 static jmp_buf source_confusion_recovery;
491 /* A pointer to the current directory filename (used by abspath). */
493 static char *cwd_buffer;
495 /* A place to save the read pointer until we are sure that an individual
496 attempt at editing will succeed. */
498 static const char * saved_clean_read_ptr;
500 /* A place to save the write pointer until we are sure that an individual
501 attempt at editing will succeed. */
503 static char * saved_repl_write_ptr;
505 /* Translate and output an error message. */
506 static void
507 notice (const char *msgid, ...)
509 va_list ap;
511 va_start (ap, msgid);
512 vfprintf (stderr, _(msgid), ap);
513 va_end (ap);
517 /* Make a copy of a string INPUT with size SIZE. */
519 static char *
520 savestring (const char *input, unsigned int size)
522 char *output = xmalloc (size + 1);
523 strcpy (output, input);
524 return output;
528 /* Make a duplicate of the first N bytes of a given string in a newly
529 allocated area. */
531 static char *
532 dupnstr (const char *s, size_t n)
534 char *ret_val = xmalloc (n + 1);
536 strncpy (ret_val, s, n);
537 ret_val[n] = '\0';
538 return ret_val;
541 /* Read LEN bytes at PTR from descriptor DESC, for file FILENAME,
542 retrying if necessary. Return the actual number of bytes read. */
544 static int
545 safe_read (int desc, void *ptr, int len)
547 int left = len;
548 while (left > 0) {
549 int nchars = read (desc, ptr, left);
550 if (nchars < 0)
552 #ifdef EINTR
553 if (errno == EINTR)
554 continue;
555 #endif
556 return nchars;
558 if (nchars == 0)
559 break;
560 /* Arithmetic on void pointers is a gcc extension. */
561 ptr = (char *) ptr + nchars;
562 left -= nchars;
564 return len - left;
567 /* Write LEN bytes at PTR to descriptor DESC,
568 retrying if necessary, and treating any real error as fatal. */
570 static void
571 safe_write (int desc, void *ptr, int len, const char *out_fname)
573 while (len > 0) {
574 int written = write (desc, ptr, len);
575 if (written < 0)
577 int errno_val = errno;
578 #ifdef EINTR
579 if (errno_val == EINTR)
580 continue;
581 #endif
582 notice ("%s: error writing file '%s': %s\n",
583 pname, shortpath (NULL, out_fname), xstrerror (errno_val));
584 return;
586 /* Arithmetic on void pointers is a gcc extension. */
587 ptr = (char *) ptr + written;
588 len -= written;
592 /* Get setup to recover in case the edit we are about to do goes awry. */
594 static void
595 save_pointers (void)
597 saved_clean_read_ptr = clean_read_ptr;
598 saved_repl_write_ptr = repl_write_ptr;
601 /* Call this routine to recover our previous state whenever something looks
602 too confusing in the source code we are trying to edit. */
604 static void
605 restore_pointers (void)
607 clean_read_ptr = saved_clean_read_ptr;
608 repl_write_ptr = saved_repl_write_ptr;
611 /* Return true if the given character is a valid identifier character. */
613 static int
614 is_id_char (int ch)
616 return (ISIDNUM (ch) || (ch == '$'));
619 /* Give a message indicating the proper way to invoke this program and then
620 exit with nonzero status. */
622 static void
623 usage (void)
625 #ifdef UNPROTOIZE
626 notice ("%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ filename ... ]'\n",
627 pname, pname);
628 #else /* !defined (UNPROTOIZE) */
629 notice ("%s: usage '%s [ -VqfnkNlgC ] [ -B <dirname> ] [ filename ... ]'\n",
630 pname, pname);
631 #endif /* !defined (UNPROTOIZE) */
632 exit (FATAL_EXIT_CODE);
635 /* Return true if the given filename (assumed to be an absolute filename)
636 designates a file residing anywhere beneath any one of the "system"
637 include directories. */
639 static int
640 in_system_include_dir (const char *path)
642 const struct default_include *p;
644 if (! IS_ABSOLUTE_PATH (path))
645 abort (); /* Must be an absolutized filename. */
647 for (p = cpp_include_defaults; p->fname; p++)
648 if (!strncmp (path, p->fname, strlen (p->fname))
649 && IS_DIR_SEPARATOR (path[strlen (p->fname)]))
650 return 1;
651 return 0;
654 #if 0
655 /* Return true if the given filename designates a file that the user has
656 read access to and for which the user has write access to the containing
657 directory. */
659 static int
660 file_could_be_converted (const char *path)
662 char *const dir_name = alloca (strlen (path) + 1);
664 if (access (path, R_OK))
665 return 0;
668 char *dir_last_slash;
670 strcpy (dir_name, path);
671 dir_last_slash = strrchr (dir_name, DIR_SEPARATOR);
672 #ifdef DIR_SEPARATOR_2
674 char *slash;
676 slash = strrchr (dir_last_slash ? dir_last_slash : dir_name,
677 DIR_SEPARATOR_2);
678 if (slash)
679 dir_last_slash = slash;
681 #endif
682 if (dir_last_slash)
683 *dir_last_slash = '\0';
684 else
685 abort (); /* Should have been an absolutized filename. */
688 if (access (path, W_OK))
689 return 0;
691 return 1;
694 /* Return true if the given filename designates a file that we are allowed
695 to modify. Files which we should not attempt to modify are (a) "system"
696 include files, and (b) files which the user doesn't have write access to,
697 and (c) files which reside in directories which the user doesn't have
698 write access to. Unless requested to be quiet, give warnings about
699 files that we will not try to convert for one reason or another. An
700 exception is made for "system" include files, which we never try to
701 convert and for which we don't issue the usual warnings. */
703 static int
704 file_normally_convertible (const char *path)
706 char *const dir_name = alloca (strlen (path) + 1);
708 if (in_system_include_dir (path))
709 return 0;
712 char *dir_last_slash;
714 strcpy (dir_name, path);
715 dir_last_slash = strrchr (dir_name, DIR_SEPARATOR);
716 #ifdef DIR_SEPARATOR_2
718 char *slash;
720 slash = strrchr (dir_last_slash ? dir_last_slash : dir_name,
721 DIR_SEPARATOR_2);
722 if (slash)
723 dir_last_slash = slash;
725 #endif
726 if (dir_last_slash)
727 *dir_last_slash = '\0';
728 else
729 abort (); /* Should have been an absolutized filename. */
732 if (access (path, R_OK))
734 if (!quiet_flag)
735 notice ("%s: warning: no read access for file '%s'\n",
736 pname, shortpath (NULL, path));
737 return 0;
740 if (access (path, W_OK))
742 if (!quiet_flag)
743 notice ("%s: warning: no write access for file '%s'\n",
744 pname, shortpath (NULL, path));
745 return 0;
748 if (access (dir_name, W_OK))
750 if (!quiet_flag)
751 notice ("%s: warning: no write access for dir containing '%s'\n",
752 pname, shortpath (NULL, path));
753 return 0;
756 return 1;
758 #endif /* 0 */
760 #ifndef UNPROTOIZE
762 /* Return true if the given file_info struct refers to the special SYSCALLS.c.X
763 file. Return false otherwise. */
765 static int
766 is_syscalls_file (const file_info *fi_p)
768 char const *f = fi_p->hash_entry->symbol;
769 size_t fl = strlen (f), sysl = sizeof (syscalls_filename) - 1;
770 return sysl <= fl && strcmp (f + fl - sysl, syscalls_filename) == 0;
773 #endif /* !defined (UNPROTOIZE) */
775 /* Check to see if this file will need to have anything done to it on this
776 run. If there is nothing in the given file which both needs conversion
777 and for which we have the necessary stuff to do the conversion, return
778 false. Otherwise, return true.
780 Note that (for protoize) it is only valid to call this function *after*
781 the connections between declarations and definitions have all been made
782 by connect_defs_and_decs. */
784 static int
785 needs_to_be_converted (const file_info *file_p)
787 const def_dec_info *ddp;
789 #ifndef UNPROTOIZE
791 if (is_syscalls_file (file_p))
792 return 0;
794 #endif /* !defined (UNPROTOIZE) */
796 for (ddp = file_p->defs_decs; ddp; ddp = ddp->next_in_file)
798 if (
800 #ifndef UNPROTOIZE
802 /* ... and if we a protoizing and this function is in old style ... */
803 !ddp->prototyped
804 /* ... and if this a definition or is a decl with an associated def ... */
805 && (ddp->is_func_def || (!ddp->is_func_def && ddp->definition))
807 #else /* defined (UNPROTOIZE) */
809 /* ... and if we are unprotoizing and this function is in new style ... */
810 ddp->prototyped
812 #endif /* defined (UNPROTOIZE) */
814 /* ... then the containing file needs converting. */
815 return -1;
816 return 0;
819 /* Return 1 if the file name NAME is in a directory
820 that should be converted. */
822 static int
823 directory_specified_p (const char *name)
825 struct string_list *p;
827 for (p = directory_list; p; p = p->next)
828 if (!strncmp (name, p->name, strlen (p->name))
829 && IS_DIR_SEPARATOR (name[strlen (p->name)]))
831 const char *q = name + strlen (p->name) + 1;
833 /* If there are more slashes, it's in a subdir, so
834 this match doesn't count. */
835 while (*q++)
836 if (IS_DIR_SEPARATOR (*(q-1)))
837 goto lose;
838 return 1;
840 lose: ;
843 return 0;
846 /* Return 1 if the file named NAME should be excluded from conversion. */
848 static int
849 file_excluded_p (const char *name)
851 struct string_list *p;
852 int len = strlen (name);
854 for (p = exclude_list; p; p = p->next)
855 if (!strcmp (name + len - strlen (p->name), p->name)
856 && IS_DIR_SEPARATOR (name[len - strlen (p->name) - 1]))
857 return 1;
859 return 0;
862 /* Construct a new element of a string_list.
863 STRING is the new element value, and REST holds the remaining elements. */
865 static struct string_list *
866 string_list_cons (const char *string, struct string_list *rest)
868 struct string_list *temp = xmalloc (sizeof (struct string_list));
870 temp->next = rest;
871 temp->name = string;
872 return temp;
875 /* ??? The GNU convention for mentioning function args in its comments
876 is to capitalize them. So change "hash_tab_p" to HASH_TAB_P below.
877 Likewise for all the other functions. */
879 /* Given a hash table, apply some function to each node in the table. The
880 table to traverse is given as the "hash_tab_p" argument, and the
881 function to be applied to each node in the table is given as "func"
882 argument. */
884 static void
885 visit_each_hash_node (const hash_table_entry *hash_tab_p,
886 void (*func) (const hash_table_entry *))
888 const hash_table_entry *primary;
890 for (primary = hash_tab_p; primary < &hash_tab_p[HASH_TABLE_SIZE]; primary++)
891 if (primary->symbol)
893 hash_table_entry *second;
895 (*func)(primary);
896 for (second = primary->hash_next; second; second = second->hash_next)
897 (*func) (second);
901 /* Initialize all of the fields of a new hash table entry, pointed
902 to by the "p" parameter. Note that the space to hold the entry
903 is assumed to have already been allocated before this routine is
904 called. */
906 static hash_table_entry *
907 add_symbol (hash_table_entry *p, const char *s)
909 p->hash_next = NULL;
910 p->symbol = xstrdup (s);
911 p->ddip = NULL;
912 p->fip = NULL;
913 return p;
916 /* Look for a particular function name or filename in the particular
917 hash table indicated by "hash_tab_p". If the name is not in the
918 given hash table, add it. Either way, return a pointer to the
919 hash table entry for the given name. */
921 static hash_table_entry *
922 lookup (hash_table_entry *hash_tab_p, const char *search_symbol)
924 int hash_value = 0;
925 const char *search_symbol_char_p = search_symbol;
926 hash_table_entry *p;
928 while (*search_symbol_char_p)
929 hash_value += *search_symbol_char_p++;
930 hash_value &= hash_mask;
931 p = &hash_tab_p[hash_value];
932 if (! p->symbol)
933 return add_symbol (p, search_symbol);
934 if (!strcmp (p->symbol, search_symbol))
935 return p;
936 while (p->hash_next)
938 p = p->hash_next;
939 if (!strcmp (p->symbol, search_symbol))
940 return p;
942 p->hash_next = xmalloc (sizeof (hash_table_entry));
943 p = p->hash_next;
944 return add_symbol (p, search_symbol);
947 /* Throw a def/dec record on the junk heap.
949 Also, since we are not using this record anymore, free up all of the
950 stuff it pointed to. */
952 static void
953 free_def_dec (def_dec_info *p)
955 free ((NONCONST void *) p->ansi_decl);
957 #ifndef UNPROTOIZE
959 const f_list_chain_item * curr;
960 const f_list_chain_item * next;
962 for (curr = p->f_list_chain; curr; curr = next)
964 next = curr->chain_next;
965 free ((NONCONST void *) curr);
968 #endif /* !defined (UNPROTOIZE) */
970 free (p);
973 /* Unexpand as many macro symbols as we can find.
975 If the given line must be unexpanded, make a copy of it in the heap and
976 return a pointer to the unexpanded copy. Otherwise return NULL. */
978 static char *
979 unexpand_if_needed (const char *aux_info_line)
981 static char *line_buf = 0;
982 static int line_buf_size = 0;
983 const unexpansion *unexp_p;
984 int got_unexpanded = 0;
985 const char *s;
986 char *copy_p = line_buf;
988 if (line_buf == 0)
990 line_buf_size = 1024;
991 line_buf = xmalloc (line_buf_size);
994 copy_p = line_buf;
996 /* Make a copy of the input string in line_buf, expanding as necessary. */
998 for (s = aux_info_line; *s != '\n'; )
1000 for (unexp_p = unexpansions; unexp_p->expanded; unexp_p++)
1002 const char *in_p = unexp_p->expanded;
1003 size_t len = strlen (in_p);
1005 if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
1007 int size = strlen (unexp_p->contracted);
1008 got_unexpanded = 1;
1009 if (copy_p + size - line_buf >= line_buf_size)
1011 int offset = copy_p - line_buf;
1012 line_buf_size *= 2;
1013 line_buf_size += size;
1014 line_buf = xrealloc (line_buf, line_buf_size);
1015 copy_p = line_buf + offset;
1017 strcpy (copy_p, unexp_p->contracted);
1018 copy_p += size;
1020 /* Assume that there will not be another replacement required
1021 within the text just replaced. */
1023 s += len;
1024 goto continue_outer;
1027 if (copy_p - line_buf == line_buf_size)
1029 int offset = copy_p - line_buf;
1030 line_buf_size *= 2;
1031 line_buf = xrealloc (line_buf, line_buf_size);
1032 copy_p = line_buf + offset;
1034 *copy_p++ = *s++;
1035 continue_outer: ;
1037 if (copy_p + 2 - line_buf >= line_buf_size)
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;
1044 *copy_p++ = '\n';
1045 *copy_p = '\0';
1047 return (got_unexpanded ? savestring (line_buf, copy_p - line_buf) : 0);
1050 /* Return the absolutized filename for the given relative
1051 filename. Note that if that filename is already absolute, it may
1052 still be returned in a modified form because this routine also
1053 eliminates redundant slashes and single dots and eliminates double
1054 dots to get a shortest possible filename from the given input
1055 filename. The absolutization of relative filenames is made by
1056 assuming that the given filename is to be taken as relative to
1057 the first argument (cwd) or to the current directory if cwd is
1058 NULL. */
1060 static char *
1061 abspath (const char *cwd, const char *rel_filename)
1063 /* Setup the current working directory as needed. */
1064 const char *const cwd2 = (cwd) ? cwd : cwd_buffer;
1065 char *const abs_buffer = alloca (strlen (cwd2) + strlen (rel_filename) + 2);
1066 char *endp = abs_buffer;
1067 char *outp, *inp;
1069 /* Copy the filename (possibly preceded by the current working
1070 directory name) into the absolutization buffer. */
1073 const char *src_p;
1075 if (! IS_ABSOLUTE_PATH (rel_filename))
1077 src_p = cwd2;
1078 while ((*endp++ = *src_p++))
1079 continue;
1080 *(endp-1) = DIR_SEPARATOR; /* overwrite null */
1082 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1083 else if (IS_DIR_SEPARATOR (rel_filename[0]))
1085 /* A path starting with a directory separator is considered absolute
1086 for dos based filesystems, but it's really not -- it's just the
1087 convention used throughout GCC and it works. However, in this
1088 case, we still need to prepend the drive spec from cwd_buffer. */
1089 *endp++ = cwd2[0];
1090 *endp++ = cwd2[1];
1092 #endif
1093 src_p = rel_filename;
1094 while ((*endp++ = *src_p++))
1095 continue;
1098 /* Now make a copy of abs_buffer into abs_buffer, shortening the
1099 filename (by taking out slashes and dots) as we go. */
1101 outp = inp = abs_buffer;
1102 *outp++ = *inp++; /* copy first slash */
1103 #if defined (apollo) || defined (_WIN32) || defined (__INTERIX)
1104 if (IS_DIR_SEPARATOR (inp[0]))
1105 *outp++ = *inp++; /* copy second slash */
1106 #endif
1107 for (;;)
1109 if (!inp[0])
1110 break;
1111 else if (IS_DIR_SEPARATOR (inp[0]) && IS_DIR_SEPARATOR (outp[-1]))
1113 inp++;
1114 continue;
1116 else if (inp[0] == '.' && IS_DIR_SEPARATOR (outp[-1]))
1118 if (!inp[1])
1119 break;
1120 else if (IS_DIR_SEPARATOR (inp[1]))
1122 inp += 2;
1123 continue;
1125 else if ((inp[1] == '.') && (inp[2] == 0
1126 || IS_DIR_SEPARATOR (inp[2])))
1128 inp += (IS_DIR_SEPARATOR (inp[2])) ? 3 : 2;
1129 outp -= 2;
1130 while (outp >= abs_buffer && ! IS_DIR_SEPARATOR (*outp))
1131 outp--;
1132 if (outp < abs_buffer)
1134 /* Catch cases like /.. where we try to backup to a
1135 point above the absolute root of the logical file
1136 system. */
1138 notice ("%s: invalid file name: %s\n",
1139 pname, rel_filename);
1140 exit (FATAL_EXIT_CODE);
1142 *++outp = '\0';
1143 continue;
1146 *outp++ = *inp++;
1149 /* On exit, make sure that there is a trailing null, and make sure that
1150 the last character of the returned string is *not* a slash. */
1152 *outp = '\0';
1153 if (IS_DIR_SEPARATOR (outp[-1]))
1154 *--outp = '\0';
1156 /* Make a copy (in the heap) of the stuff left in the absolutization
1157 buffer and return a pointer to the copy. */
1159 return savestring (abs_buffer, outp - abs_buffer);
1162 /* Given a filename (and possibly a directory name from which the filename
1163 is relative) return a string which is the shortest possible
1164 equivalent for the corresponding full (absolutized) filename. The
1165 shortest possible equivalent may be constructed by converting the
1166 absolutized filename to be a relative filename (i.e. relative to
1167 the actual current working directory). However if a relative filename
1168 is longer, then the full absolute filename is returned.
1170 KNOWN BUG:
1172 Note that "simple-minded" conversion of any given type of filename (either
1173 relative or absolute) may not result in a valid equivalent filename if any
1174 subpart of the original filename is actually a symbolic link. */
1176 static const char *
1177 shortpath (const char *cwd, const char *filename)
1179 char *rel_buffer;
1180 char *rel_buf_p;
1181 char *cwd_p = cwd_buffer;
1182 char *path_p;
1183 int unmatched_slash_count = 0;
1184 size_t filename_len = strlen (filename);
1186 path_p = abspath (cwd, filename);
1187 rel_buf_p = rel_buffer = xmalloc (filename_len);
1189 while (*cwd_p && IS_SAME_PATH_CHAR (*cwd_p, *path_p))
1191 cwd_p++;
1192 path_p++;
1194 if (!*cwd_p && (!*path_p || IS_DIR_SEPARATOR (*path_p)))
1196 /* whole pwd matched */
1197 if (!*path_p) /* input *is* the current path! */
1198 return ".";
1199 else
1200 return ++path_p;
1202 else
1204 if (*path_p)
1206 --cwd_p;
1207 --path_p;
1208 while (! IS_DIR_SEPARATOR (*cwd_p)) /* backup to last slash */
1210 --cwd_p;
1211 --path_p;
1213 cwd_p++;
1214 path_p++;
1215 unmatched_slash_count++;
1218 /* Find out how many directory levels in cwd were *not* matched. */
1219 while (*cwd_p++)
1220 if (IS_DIR_SEPARATOR (*(cwd_p-1)))
1221 unmatched_slash_count++;
1223 /* Now we know how long the "short name" will be.
1224 Reject it if longer than the input. */
1225 if (unmatched_slash_count * 3 + strlen (path_p) >= filename_len)
1226 return filename;
1228 /* For each of them, put a `../' at the beginning of the short name. */
1229 while (unmatched_slash_count--)
1231 /* Give up if the result gets to be longer
1232 than the absolute path name. */
1233 if (rel_buffer + filename_len <= rel_buf_p + 3)
1234 return filename;
1235 *rel_buf_p++ = '.';
1236 *rel_buf_p++ = '.';
1237 *rel_buf_p++ = DIR_SEPARATOR;
1240 /* Then tack on the unmatched part of the desired file's name. */
1243 if (rel_buffer + filename_len <= rel_buf_p)
1244 return filename;
1246 while ((*rel_buf_p++ = *path_p++));
1248 --rel_buf_p;
1249 if (IS_DIR_SEPARATOR (*(rel_buf_p-1)))
1250 *--rel_buf_p = '\0';
1251 return rel_buffer;
1255 /* Lookup the given filename in the hash table for filenames. If it is a
1256 new one, then the hash table info pointer will be null. In this case,
1257 we create a new file_info record to go with the filename, and we initialize
1258 that record with some reasonable values. */
1260 /* FILENAME was const, but that causes a warning on AIX when calling stat.
1261 That is probably a bug in AIX, but might as well avoid the warning. */
1263 static file_info *
1264 find_file (const char *filename, int do_not_stat)
1266 hash_table_entry *hash_entry_p;
1268 hash_entry_p = lookup (filename_primary, filename);
1269 if (hash_entry_p->fip)
1270 return hash_entry_p->fip;
1271 else
1273 struct stat stat_buf;
1274 file_info *file_p = xmalloc (sizeof (file_info));
1276 /* If we cannot get status on any given source file, give a warning
1277 and then just set its time of last modification to infinity. */
1279 if (do_not_stat)
1280 stat_buf.st_mtime = (time_t) 0;
1281 else
1283 if (stat (filename, &stat_buf) == -1)
1285 int errno_val = errno;
1286 notice ("%s: %s: can't get status: %s\n",
1287 pname, shortpath (NULL, filename),
1288 xstrerror (errno_val));
1289 stat_buf.st_mtime = (time_t) -1;
1293 hash_entry_p->fip = file_p;
1294 file_p->hash_entry = hash_entry_p;
1295 file_p->defs_decs = NULL;
1296 file_p->mtime = stat_buf.st_mtime;
1297 return file_p;
1301 /* Generate a fatal error because some part of the aux_info file is
1302 messed up. */
1304 static void
1305 aux_info_corrupted (void)
1307 notice ("\n%s: fatal error: aux info file corrupted at line %d\n",
1308 pname, current_aux_info_lineno);
1309 exit (FATAL_EXIT_CODE);
1312 /* ??? This comment is vague. Say what the condition is for. */
1313 /* Check to see that a condition is true. This is kind of like an assert. */
1315 static void
1316 check_aux_info (int cond)
1318 if (! cond)
1319 aux_info_corrupted ();
1322 /* Given a pointer to the closing right parenthesis for a particular formals
1323 list (in an aux_info file) find the corresponding left parenthesis and
1324 return a pointer to it. */
1326 static const char *
1327 find_corresponding_lparen (const char *p)
1329 const char *q;
1330 int paren_depth;
1332 for (paren_depth = 1, q = p-1; paren_depth; q--)
1334 switch (*q)
1336 case ')':
1337 paren_depth++;
1338 break;
1339 case '(':
1340 paren_depth--;
1341 break;
1344 return ++q;
1347 /* Given a line from an aux info file, and a time at which the aux info
1348 file it came from was created, check to see if the item described in
1349 the line comes from a file which has been modified since the aux info
1350 file was created. If so, return nonzero, else return zero. */
1352 static int
1353 referenced_file_is_newer (const char *l, time_t aux_info_mtime)
1355 const char *p;
1356 file_info *fi_p;
1357 char *filename;
1359 check_aux_info (l[0] == '/');
1360 check_aux_info (l[1] == '*');
1361 check_aux_info (l[2] == ' ');
1364 const char *filename_start = p = l + 3;
1366 while (*p != ':'
1367 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1368 || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1369 #endif
1371 p++;
1372 filename = alloca ((size_t) (p - filename_start) + 1);
1373 strncpy (filename, filename_start, (size_t) (p - filename_start));
1374 filename[p-filename_start] = '\0';
1377 /* Call find_file to find the file_info record associated with the file
1378 which contained this particular def or dec item. Note that this call
1379 may cause a new file_info record to be created if this is the first time
1380 that we have ever known about this particular file. */
1382 fi_p = find_file (abspath (invocation_filename, filename), 0);
1384 return (fi_p->mtime > aux_info_mtime);
1387 /* Given a line of info from the aux_info file, create a new
1388 def_dec_info record to remember all of the important information about
1389 a function definition or declaration.
1391 Link this record onto the list of such records for the particular file in
1392 which it occurred in proper (descending) line number order (for now).
1394 If there is an identical record already on the list for the file, throw
1395 this one away. Doing so takes care of the (useless and troublesome)
1396 duplicates which are bound to crop up due to multiple inclusions of any
1397 given individual header file.
1399 Finally, link the new def_dec record onto the list of such records
1400 pertaining to this particular function name. */
1402 static void
1403 save_def_or_dec (const char *l, int is_syscalls)
1405 const char *p;
1406 const char *semicolon_p;
1407 def_dec_info *def_dec_p = xmalloc (sizeof (def_dec_info));
1409 #ifndef UNPROTOIZE
1410 def_dec_p->written = 0;
1411 #endif /* !defined (UNPROTOIZE) */
1413 /* Start processing the line by picking off 5 pieces of information from
1414 the left hand end of the line. These are filename, line number,
1415 new/old/implicit flag (new = ANSI prototype format), definition or
1416 declaration flag, and extern/static flag). */
1418 check_aux_info (l[0] == '/');
1419 check_aux_info (l[1] == '*');
1420 check_aux_info (l[2] == ' ');
1423 const char *filename_start = p = l + 3;
1424 char *filename;
1426 while (*p != ':'
1427 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1428 || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1429 #endif
1431 p++;
1432 filename = alloca ((size_t) (p - filename_start) + 1);
1433 strncpy (filename, filename_start, (size_t) (p - filename_start));
1434 filename[p-filename_start] = '\0';
1436 /* Call find_file to find the file_info record associated with the file
1437 which contained this particular def or dec item. Note that this call
1438 may cause a new file_info record to be created if this is the first time
1439 that we have ever known about this particular file.
1441 Note that we started out by forcing all of the base source file names
1442 (i.e. the names of the aux_info files with the .X stripped off) into the
1443 filenames hash table, and we simultaneously setup file_info records for
1444 all of these base file names (even if they may be useless later).
1445 The file_info records for all of these "base" file names (properly)
1446 act as file_info records for the "original" (i.e. un-included) files
1447 which were submitted to gcc for compilation (when the -aux-info
1448 option was used). */
1450 def_dec_p->file = find_file (abspath (invocation_filename, filename), is_syscalls);
1454 const char *line_number_start = ++p;
1455 char line_number[10];
1457 while (*p != ':'
1458 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
1459 || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
1460 #endif
1462 p++;
1463 strncpy (line_number, line_number_start, (size_t) (p - line_number_start));
1464 line_number[p-line_number_start] = '\0';
1465 def_dec_p->line = atoi (line_number);
1468 /* Check that this record describes a new-style, old-style, or implicit
1469 definition or declaration. */
1471 p++; /* Skip over the `:'. */
1472 check_aux_info ((*p == 'N') || (*p == 'O') || (*p == 'I'));
1474 /* Is this a new style (ANSI prototyped) definition or declaration? */
1476 def_dec_p->prototyped = (*p == 'N');
1478 #ifndef UNPROTOIZE
1480 /* Is this an implicit declaration? */
1482 def_dec_p->is_implicit = (*p == 'I');
1484 #endif /* !defined (UNPROTOIZE) */
1486 p++;
1488 check_aux_info ((*p == 'C') || (*p == 'F'));
1490 /* Is this item a function definition (F) or a declaration (C). Note that
1491 we treat item taken from the syscalls file as though they were function
1492 definitions regardless of what the stuff in the file says. */
1494 def_dec_p->is_func_def = ((*p++ == 'F') || is_syscalls);
1496 #ifndef UNPROTOIZE
1497 def_dec_p->definition = 0; /* Fill this in later if protoizing. */
1498 #endif /* !defined (UNPROTOIZE) */
1500 check_aux_info (*p++ == ' ');
1501 check_aux_info (*p++ == '*');
1502 check_aux_info (*p++ == '/');
1503 check_aux_info (*p++ == ' ');
1505 #ifdef UNPROTOIZE
1506 check_aux_info ((!strncmp (p, "static", 6)) || (!strncmp (p, "extern", 6)));
1507 #else /* !defined (UNPROTOIZE) */
1508 if (!strncmp (p, "static", 6))
1509 def_dec_p->is_static = -1;
1510 else if (!strncmp (p, "extern", 6))
1511 def_dec_p->is_static = 0;
1512 else
1513 check_aux_info (0); /* Didn't find either `extern' or `static'. */
1514 #endif /* !defined (UNPROTOIZE) */
1517 const char *ansi_start = p;
1519 p += 6; /* Pass over the "static" or "extern". */
1521 /* We are now past the initial stuff. Search forward from here to find
1522 the terminating semicolon that should immediately follow the entire
1523 ANSI format function declaration. */
1525 while (*++p != ';')
1526 continue;
1528 semicolon_p = p;
1530 /* Make a copy of the ansi declaration part of the line from the aux_info
1531 file. */
1533 def_dec_p->ansi_decl
1534 = dupnstr (ansi_start, (size_t) ((semicolon_p+1) - ansi_start));
1536 /* Backup and point at the final right paren of the final argument list. */
1538 p--;
1540 #ifndef UNPROTOIZE
1541 def_dec_p->f_list_chain = NULL;
1542 #endif /* !defined (UNPROTOIZE) */
1544 while (p != ansi_start && (p[-1] == ' ' || p[-1] == '\t')) p--;
1545 if (*p != ')')
1547 free_def_dec (def_dec_p);
1548 return;
1552 /* Now isolate a whole set of formal argument lists, one-by-one. Normally,
1553 there will only be one list to isolate, but there could be more. */
1555 def_dec_p->f_list_count = 0;
1557 for (;;)
1559 const char *left_paren_p = find_corresponding_lparen (p);
1560 #ifndef UNPROTOIZE
1562 f_list_chain_item *cip = xmalloc (sizeof (f_list_chain_item));
1564 cip->formals_list
1565 = dupnstr (left_paren_p + 1, (size_t) (p - (left_paren_p+1)));
1567 /* Add the new chain item at the head of the current list. */
1569 cip->chain_next = def_dec_p->f_list_chain;
1570 def_dec_p->f_list_chain = cip;
1572 #endif /* !defined (UNPROTOIZE) */
1573 def_dec_p->f_list_count++;
1575 p = left_paren_p - 2;
1577 /* p must now point either to another right paren, or to the last
1578 character of the name of the function that was declared/defined.
1579 If p points to another right paren, then this indicates that we
1580 are dealing with multiple formals lists. In that case, there
1581 really should be another right paren preceding this right paren. */
1583 if (*p != ')')
1584 break;
1585 else
1586 check_aux_info (*--p == ')');
1591 const char *past_fn = p + 1;
1593 check_aux_info (*past_fn == ' ');
1595 /* Scan leftwards over the identifier that names the function. */
1597 while (is_id_char (*p))
1598 p--;
1599 p++;
1601 /* p now points to the leftmost character of the function name. */
1604 char *fn_string = alloca (past_fn - p + 1);
1606 strncpy (fn_string, p, (size_t) (past_fn - p));
1607 fn_string[past_fn-p] = '\0';
1608 def_dec_p->hash_entry = lookup (function_name_primary, fn_string);
1612 /* Look at all of the defs and decs for this function name that we have
1613 collected so far. If there is already one which is at the same
1614 line number in the same file, then we can discard this new def_dec_info
1615 record.
1617 As an extra assurance that any such pair of (nominally) identical
1618 function declarations are in fact identical, we also compare the
1619 ansi_decl parts of the lines from the aux_info files just to be on
1620 the safe side.
1622 This comparison will fail if (for instance) the user was playing
1623 messy games with the preprocessor which ultimately causes one
1624 function declaration in one header file to look differently when
1625 that file is included by two (or more) other files. */
1628 const def_dec_info *other;
1630 for (other = def_dec_p->hash_entry->ddip; other; other = other->next_for_func)
1632 if (def_dec_p->line == other->line && def_dec_p->file == other->file)
1634 if (strcmp (def_dec_p->ansi_decl, other->ansi_decl))
1636 notice ("%s:%d: declaration of function '%s' takes different forms\n",
1637 def_dec_p->file->hash_entry->symbol,
1638 def_dec_p->line,
1639 def_dec_p->hash_entry->symbol);
1640 exit (FATAL_EXIT_CODE);
1642 free_def_dec (def_dec_p);
1643 return;
1648 #ifdef UNPROTOIZE
1650 /* If we are doing unprotoizing, we must now setup the pointers that will
1651 point to the K&R name list and to the K&R argument declarations list.
1653 Note that if this is only a function declaration, then we should not
1654 expect to find any K&R style formals list following the ANSI-style
1655 formals list. This is because GCC knows that such information is
1656 useless in the case of function declarations (function definitions
1657 are a different story however).
1659 Since we are unprotoizing, we don't need any such lists anyway.
1660 All we plan to do is to delete all characters between ()'s in any
1661 case. */
1663 def_dec_p->formal_names = NULL;
1664 def_dec_p->formal_decls = NULL;
1666 if (def_dec_p->is_func_def)
1668 p = semicolon_p;
1669 check_aux_info (*++p == ' ');
1670 check_aux_info (*++p == '/');
1671 check_aux_info (*++p == '*');
1672 check_aux_info (*++p == ' ');
1673 check_aux_info (*++p == '(');
1676 const char *kr_names_start = ++p; /* Point just inside '('. */
1678 while (*p++ != ')')
1679 continue;
1680 p--; /* point to closing right paren */
1682 /* Make a copy of the K&R parameter names list. */
1684 def_dec_p->formal_names
1685 = dupnstr (kr_names_start, (size_t) (p - kr_names_start));
1688 check_aux_info (*++p == ' ');
1689 p++;
1691 /* p now points to the first character of the K&R style declarations
1692 list (if there is one) or to the star-slash combination that ends
1693 the comment in which such lists get embedded. */
1695 /* Make a copy of the K&R formal decls list and set the def_dec record
1696 to point to it. */
1698 if (*p == '*') /* Are there no K&R declarations? */
1700 check_aux_info (*++p == '/');
1701 def_dec_p->formal_decls = "";
1703 else
1705 const char *kr_decls_start = p;
1707 while (p[0] != '*' || p[1] != '/')
1708 p++;
1709 p--;
1711 check_aux_info (*p == ' ');
1713 def_dec_p->formal_decls
1714 = dupnstr (kr_decls_start, (size_t) (p - kr_decls_start));
1717 /* Handle a special case. If we have a function definition marked as
1718 being in "old" style, and if its formal names list is empty, then
1719 it may actually have the string "void" in its real formals list
1720 in the original source code. Just to make sure, we will get setup
1721 to convert such things anyway.
1723 This kludge only needs to be here because of an insurmountable
1724 problem with generating .X files. */
1726 if (!def_dec_p->prototyped && !*def_dec_p->formal_names)
1727 def_dec_p->prototyped = 1;
1730 /* Since we are unprotoizing, if this item is already in old (K&R) style,
1731 we can just ignore it. If that is true, throw away the itme now. */
1733 if (!def_dec_p->prototyped)
1735 free_def_dec (def_dec_p);
1736 return;
1739 #endif /* defined (UNPROTOIZE) */
1741 /* Add this record to the head of the list of records pertaining to this
1742 particular function name. */
1744 def_dec_p->next_for_func = def_dec_p->hash_entry->ddip;
1745 def_dec_p->hash_entry->ddip = def_dec_p;
1747 /* Add this new def_dec_info record to the sorted list of def_dec_info
1748 records for this file. Note that we don't have to worry about duplicates
1749 (caused by multiple inclusions of header files) here because we have
1750 already eliminated duplicates above. */
1752 if (!def_dec_p->file->defs_decs)
1754 def_dec_p->file->defs_decs = def_dec_p;
1755 def_dec_p->next_in_file = NULL;
1757 else
1759 int line = def_dec_p->line;
1760 const def_dec_info *prev = NULL;
1761 const def_dec_info *curr = def_dec_p->file->defs_decs;
1762 const def_dec_info *next = curr->next_in_file;
1764 while (next && (line < curr->line))
1766 prev = curr;
1767 curr = next;
1768 next = next->next_in_file;
1770 if (line >= curr->line)
1772 def_dec_p->next_in_file = curr;
1773 if (prev)
1774 ((NONCONST def_dec_info *) prev)->next_in_file = def_dec_p;
1775 else
1776 def_dec_p->file->defs_decs = def_dec_p;
1778 else /* assert (next == NULL); */
1780 ((NONCONST def_dec_info *) curr)->next_in_file = def_dec_p;
1781 /* assert (next == NULL); */
1782 def_dec_p->next_in_file = next;
1787 /* Set up the vector COMPILE_PARAMS which is the argument list for running GCC.
1788 Also set input_file_name_index and aux_info_file_name_index
1789 to the indices of the slots where the file names should go. */
1791 /* We initialize the vector by removing -g, -O, -S, -c, and -o options,
1792 and adding '-aux-info AUXFILE -S -o /dev/null INFILE' at the end. */
1794 static void
1795 munge_compile_params (const char *params_list)
1797 /* Build up the contents in a temporary vector
1798 that is so big that to has to be big enough. */
1799 const char **temp_params
1800 = alloca ((strlen (params_list) + 8) * sizeof (char *));
1801 int param_count = 0;
1802 const char *param;
1803 struct stat st;
1805 temp_params[param_count++] = compiler_file_name;
1806 for (;;)
1808 while (ISSPACE ((const unsigned char)*params_list))
1809 params_list++;
1810 if (!*params_list)
1811 break;
1812 param = params_list;
1813 while (*params_list && !ISSPACE ((const unsigned char)*params_list))
1814 params_list++;
1815 if (param[0] != '-')
1816 temp_params[param_count++]
1817 = dupnstr (param, (size_t) (params_list - param));
1818 else
1820 switch (param[1])
1822 case 'g':
1823 case 'O':
1824 case 'S':
1825 case 'c':
1826 break; /* Don't copy these. */
1827 case 'o':
1828 while (ISSPACE ((const unsigned char)*params_list))
1829 params_list++;
1830 while (*params_list
1831 && !ISSPACE ((const unsigned char)*params_list))
1832 params_list++;
1833 break;
1834 default:
1835 temp_params[param_count++]
1836 = dupnstr (param, (size_t) (params_list - param));
1839 if (!*params_list)
1840 break;
1842 temp_params[param_count++] = "-aux-info";
1844 /* Leave room for the aux-info file name argument. */
1845 aux_info_file_name_index = param_count;
1846 temp_params[param_count++] = NULL;
1848 temp_params[param_count++] = "-S";
1849 temp_params[param_count++] = "-o";
1851 if ((stat (HOST_BIT_BUCKET, &st) == 0)
1852 && (!S_ISDIR (st.st_mode))
1853 && (access (HOST_BIT_BUCKET, W_OK) == 0))
1854 temp_params[param_count++] = HOST_BIT_BUCKET;
1855 else
1856 /* FIXME: This is hardly likely to be right, if HOST_BIT_BUCKET is not
1857 writable. But until this is rejigged to use make_temp_file(), this
1858 is the best we can do. */
1859 temp_params[param_count++] = "/dev/null";
1861 /* Leave room for the input file name argument. */
1862 input_file_name_index = param_count;
1863 temp_params[param_count++] = NULL;
1864 /* Terminate the list. */
1865 temp_params[param_count++] = NULL;
1867 /* Make a copy of the compile_params in heap space. */
1869 compile_params = xmalloc (sizeof (char *) * (param_count+1));
1870 memcpy (compile_params, temp_params, sizeof (char *) * param_count);
1873 /* Do a recompilation for the express purpose of generating a new aux_info
1874 file to go with a specific base source file.
1876 The result is a boolean indicating success. */
1878 static int
1879 gen_aux_info_file (const char *base_filename)
1881 if (!input_file_name_index)
1882 munge_compile_params ("");
1884 /* Store the full source file name in the argument vector. */
1885 compile_params[input_file_name_index] = shortpath (NULL, base_filename);
1886 /* Add .X to source file name to get aux-info file name. */
1887 compile_params[aux_info_file_name_index] =
1888 concat (compile_params[input_file_name_index], aux_info_suffix, NULL);
1890 if (!quiet_flag)
1891 notice ("%s: compiling '%s'\n",
1892 pname, compile_params[input_file_name_index]);
1895 char *errmsg_fmt, *errmsg_arg;
1896 int wait_status, pid;
1898 pid = pexecute (compile_params[0], (char * const *) compile_params,
1899 pname, NULL, &errmsg_fmt, &errmsg_arg,
1900 PEXECUTE_FIRST | PEXECUTE_LAST | PEXECUTE_SEARCH);
1902 if (pid == -1)
1904 int errno_val = errno;
1905 fprintf (stderr, "%s: ", pname);
1906 fprintf (stderr, errmsg_fmt, errmsg_arg);
1907 fprintf (stderr, ": %s\n", xstrerror (errno_val));
1908 return 0;
1911 pid = pwait (pid, &wait_status, 0);
1912 if (pid == -1)
1914 notice ("%s: wait: %s\n", pname, xstrerror (errno));
1915 return 0;
1917 if (WIFSIGNALED (wait_status))
1919 notice ("%s: subprocess got fatal signal %d\n",
1920 pname, WTERMSIG (wait_status));
1921 return 0;
1923 if (WIFEXITED (wait_status))
1925 if (WEXITSTATUS (wait_status) != 0)
1927 notice ("%s: %s exited with status %d\n",
1928 pname, compile_params[0], WEXITSTATUS (wait_status));
1929 return 0;
1931 return 1;
1933 abort ();
1937 /* Read in all of the information contained in a single aux_info file.
1938 Save all of the important stuff for later. */
1940 static void
1941 process_aux_info_file (const char *base_source_filename, int keep_it,
1942 int is_syscalls)
1944 size_t base_len = strlen (base_source_filename);
1945 char * aux_info_filename = alloca (base_len + strlen (aux_info_suffix) + 1);
1946 char *aux_info_base;
1947 char *aux_info_limit;
1948 char *aux_info_relocated_name;
1949 const char *aux_info_second_line;
1950 time_t aux_info_mtime;
1951 size_t aux_info_size;
1952 int must_create;
1954 /* Construct the aux_info filename from the base source filename. */
1956 strcpy (aux_info_filename, base_source_filename);
1957 strcat (aux_info_filename, aux_info_suffix);
1959 /* Check that the aux_info file exists and is readable. If it does not
1960 exist, try to create it (once only). */
1962 /* If file doesn't exist, set must_create.
1963 Likewise if it exists and we can read it but it is obsolete.
1964 Otherwise, report an error. */
1965 must_create = 0;
1967 /* Come here with must_create set to 1 if file is out of date. */
1968 start_over: ;
1970 if (access (aux_info_filename, R_OK) == -1)
1972 if (errno == ENOENT)
1974 if (is_syscalls)
1976 notice ("%s: warning: missing SYSCALLS file '%s'\n",
1977 pname, aux_info_filename);
1978 return;
1980 must_create = 1;
1982 else
1984 int errno_val = errno;
1985 notice ("%s: can't read aux info file '%s': %s\n",
1986 pname, shortpath (NULL, aux_info_filename),
1987 xstrerror (errno_val));
1988 errors++;
1989 return;
1992 #if 0 /* There is code farther down to take care of this. */
1993 else
1995 struct stat s1, s2;
1996 stat (aux_info_file_name, &s1);
1997 stat (base_source_file_name, &s2);
1998 if (s2.st_mtime > s1.st_mtime)
1999 must_create = 1;
2001 #endif /* 0 */
2003 /* If we need a .X file, create it, and verify we can read it. */
2004 if (must_create)
2006 if (!gen_aux_info_file (base_source_filename))
2008 errors++;
2009 return;
2011 if (access (aux_info_filename, R_OK) == -1)
2013 int errno_val = errno;
2014 notice ("%s: can't read aux info file '%s': %s\n",
2015 pname, shortpath (NULL, aux_info_filename),
2016 xstrerror (errno_val));
2017 errors++;
2018 return;
2023 struct stat stat_buf;
2025 /* Get some status information about this aux_info file. */
2027 if (stat (aux_info_filename, &stat_buf) == -1)
2029 int errno_val = errno;
2030 notice ("%s: can't get status of aux info file '%s': %s\n",
2031 pname, shortpath (NULL, aux_info_filename),
2032 xstrerror (errno_val));
2033 errors++;
2034 return;
2037 /* Check on whether or not this aux_info file is zero length. If it is,
2038 then just ignore it and return. */
2040 if ((aux_info_size = stat_buf.st_size) == 0)
2041 return;
2043 /* Get the date/time of last modification for this aux_info file and
2044 remember it. We will have to check that any source files that it
2045 contains information about are at least this old or older. */
2047 aux_info_mtime = stat_buf.st_mtime;
2049 if (!is_syscalls)
2051 /* Compare mod time with the .c file; update .X file if obsolete.
2052 The code later on can fail to check the .c file
2053 if it did not directly define any functions. */
2055 if (stat (base_source_filename, &stat_buf) == -1)
2057 int errno_val = errno;
2058 notice ("%s: can't get status of aux info file '%s': %s\n",
2059 pname, shortpath (NULL, base_source_filename),
2060 xstrerror (errno_val));
2061 errors++;
2062 return;
2064 if (stat_buf.st_mtime > aux_info_mtime)
2066 must_create = 1;
2067 goto start_over;
2073 int aux_info_file;
2074 int fd_flags;
2076 /* Open the aux_info file. */
2078 fd_flags = O_RDONLY;
2079 #ifdef O_BINARY
2080 /* Use binary mode to avoid having to deal with different EOL characters. */
2081 fd_flags |= O_BINARY;
2082 #endif
2083 if ((aux_info_file = open (aux_info_filename, fd_flags, 0444 )) == -1)
2085 int errno_val = errno;
2086 notice ("%s: can't open aux info file '%s' for reading: %s\n",
2087 pname, shortpath (NULL, aux_info_filename),
2088 xstrerror (errno_val));
2089 return;
2092 /* Allocate space to hold the aux_info file in memory. */
2094 aux_info_base = xmalloc (aux_info_size + 1);
2095 aux_info_limit = aux_info_base + aux_info_size;
2096 *aux_info_limit = '\0';
2098 /* Read the aux_info file into memory. */
2100 if (safe_read (aux_info_file, aux_info_base, aux_info_size) !=
2101 (int) aux_info_size)
2103 int errno_val = errno;
2104 notice ("%s: error reading aux info file '%s': %s\n",
2105 pname, shortpath (NULL, aux_info_filename),
2106 xstrerror (errno_val));
2107 free (aux_info_base);
2108 close (aux_info_file);
2109 return;
2112 /* Close the aux info file. */
2114 if (close (aux_info_file))
2116 int errno_val = errno;
2117 notice ("%s: error closing aux info file '%s': %s\n",
2118 pname, shortpath (NULL, aux_info_filename),
2119 xstrerror (errno_val));
2120 free (aux_info_base);
2121 close (aux_info_file);
2122 return;
2126 /* Delete the aux_info file (unless requested not to). If the deletion
2127 fails for some reason, don't even worry about it. */
2129 if (must_create && !keep_it)
2130 if (unlink (aux_info_filename) == -1)
2132 int errno_val = errno;
2133 notice ("%s: can't delete aux info file '%s': %s\n",
2134 pname, shortpath (NULL, aux_info_filename),
2135 xstrerror (errno_val));
2138 /* Save a pointer into the first line of the aux_info file which
2139 contains the filename of the directory from which the compiler
2140 was invoked when the associated source file was compiled.
2141 This information is used later to help create complete
2142 filenames out of the (potentially) relative filenames in
2143 the aux_info file. */
2146 char *p = aux_info_base;
2148 while (*p != ':'
2149 #ifdef HAVE_DOS_BASED_FILE_SYSTEM
2150 || (*p == ':' && *p && *(p+1) && IS_DIR_SEPARATOR (*(p+1)))
2151 #endif
2153 p++;
2154 p++;
2155 while (*p == ' ')
2156 p++;
2157 invocation_filename = p; /* Save a pointer to first byte of path. */
2158 while (*p != ' ')
2159 p++;
2160 *p++ = DIR_SEPARATOR;
2161 *p++ = '\0';
2162 while (*p++ != '\n')
2163 continue;
2164 aux_info_second_line = p;
2165 aux_info_relocated_name = 0;
2166 if (! IS_ABSOLUTE_PATH (invocation_filename))
2168 /* INVOCATION_FILENAME is relative;
2169 append it to BASE_SOURCE_FILENAME's dir. */
2170 char *dir_end;
2171 aux_info_relocated_name = xmalloc (base_len + (p-invocation_filename));
2172 strcpy (aux_info_relocated_name, base_source_filename);
2173 dir_end = strrchr (aux_info_relocated_name, DIR_SEPARATOR);
2174 #ifdef DIR_SEPARATOR_2
2176 char *slash;
2178 slash = strrchr (dir_end ? dir_end : aux_info_relocated_name,
2179 DIR_SEPARATOR_2);
2180 if (slash)
2181 dir_end = slash;
2183 #endif
2184 if (dir_end)
2185 dir_end++;
2186 else
2187 dir_end = aux_info_relocated_name;
2188 strcpy (dir_end, invocation_filename);
2189 invocation_filename = aux_info_relocated_name;
2195 const char *aux_info_p;
2197 /* Do a pre-pass on the lines in the aux_info file, making sure that all
2198 of the source files referenced in there are at least as old as this
2199 aux_info file itself. If not, go back and regenerate the aux_info
2200 file anew. Don't do any of this for the syscalls file. */
2202 if (!is_syscalls)
2204 current_aux_info_lineno = 2;
2206 for (aux_info_p = aux_info_second_line; *aux_info_p; )
2208 if (referenced_file_is_newer (aux_info_p, aux_info_mtime))
2210 free (aux_info_base);
2211 free (aux_info_relocated_name);
2212 if (keep_it && unlink (aux_info_filename) == -1)
2214 int errno_val = errno;
2215 notice ("%s: can't delete file '%s': %s\n",
2216 pname, shortpath (NULL, aux_info_filename),
2217 xstrerror (errno_val));
2218 return;
2220 must_create = 1;
2221 goto start_over;
2224 /* Skip over the rest of this line to start of next line. */
2226 while (*aux_info_p != '\n')
2227 aux_info_p++;
2228 aux_info_p++;
2229 current_aux_info_lineno++;
2233 /* Now do the real pass on the aux_info lines. Save their information in
2234 the in-core data base. */
2236 current_aux_info_lineno = 2;
2238 for (aux_info_p = aux_info_second_line; *aux_info_p;)
2240 char *unexpanded_line = unexpand_if_needed (aux_info_p);
2242 if (unexpanded_line)
2244 save_def_or_dec (unexpanded_line, is_syscalls);
2245 free (unexpanded_line);
2247 else
2248 save_def_or_dec (aux_info_p, is_syscalls);
2250 /* Skip over the rest of this line and get to start of next line. */
2252 while (*aux_info_p != '\n')
2253 aux_info_p++;
2254 aux_info_p++;
2255 current_aux_info_lineno++;
2259 free (aux_info_base);
2260 free (aux_info_relocated_name);
2263 #ifndef UNPROTOIZE
2265 /* Check an individual filename for a .c suffix. If the filename has this
2266 suffix, rename the file such that its suffix is changed to .C. This
2267 function implements the -C option. */
2269 static void
2270 rename_c_file (const hash_table_entry *hp)
2272 const char *filename = hp->symbol;
2273 int last_char_index = strlen (filename) - 1;
2274 char *const new_filename = alloca (strlen (filename)
2275 + strlen (cplus_suffix) + 1);
2277 /* Note that we don't care here if the given file was converted or not. It
2278 is possible that the given file was *not* converted, simply because there
2279 was nothing in it which actually required conversion. Even in this case,
2280 we want to do the renaming. Note that we only rename files with the .c
2281 suffix (except for the syscalls file, which is left alone). */
2283 if (filename[last_char_index] != 'c' || filename[last_char_index-1] != '.'
2284 || IS_SAME_PATH (syscalls_absolute_filename, filename))
2285 return;
2287 strcpy (new_filename, filename);
2288 strcpy (&new_filename[last_char_index], cplus_suffix);
2290 if (rename (filename, new_filename) == -1)
2292 int errno_val = errno;
2293 notice ("%s: warning: can't rename file '%s' to '%s': %s\n",
2294 pname, shortpath (NULL, filename),
2295 shortpath (NULL, new_filename), xstrerror (errno_val));
2296 errors++;
2297 return;
2301 #endif /* !defined (UNPROTOIZE) */
2303 /* Take the list of definitions and declarations attached to a particular
2304 file_info node and reverse the order of the list. This should get the
2305 list into an order such that the item with the lowest associated line
2306 number is nearest the head of the list. When these lists are originally
2307 built, they are in the opposite order. We want to traverse them in
2308 normal line number order later (i.e. lowest to highest) so reverse the
2309 order here. */
2311 static void
2312 reverse_def_dec_list (const hash_table_entry *hp)
2314 file_info *file_p = hp->fip;
2315 def_dec_info *prev = NULL;
2316 def_dec_info *current = (def_dec_info *) file_p->defs_decs;
2318 if (!current)
2319 return; /* no list to reverse */
2321 prev = current;
2322 if (! (current = (def_dec_info *) current->next_in_file))
2323 return; /* can't reverse a single list element */
2325 prev->next_in_file = NULL;
2327 while (current)
2329 def_dec_info *next = (def_dec_info *) current->next_in_file;
2331 current->next_in_file = prev;
2332 prev = current;
2333 current = next;
2336 file_p->defs_decs = prev;
2339 #ifndef UNPROTOIZE
2341 /* Find the (only?) extern definition for a particular function name, starting
2342 from the head of the linked list of entries for the given name. If we
2343 cannot find an extern definition for the given function name, issue a
2344 warning and scrounge around for the next best thing, i.e. an extern
2345 function declaration with a prototype attached to it. Note that we only
2346 allow such substitutions for extern declarations and never for static
2347 declarations. That's because the only reason we allow them at all is
2348 to let un-prototyped function declarations for system-supplied library
2349 functions get their prototypes from our own extra SYSCALLS.c.X file which
2350 contains all of the correct prototypes for system functions. */
2352 static const def_dec_info *
2353 find_extern_def (const def_dec_info *head, const def_dec_info *user)
2355 const def_dec_info *dd_p;
2356 const def_dec_info *extern_def_p = NULL;
2357 int conflict_noted = 0;
2359 /* Don't act too stupid here. Somebody may try to convert an entire system
2360 in one swell fwoop (rather than one program at a time, as should be done)
2361 and in that case, we may find that there are multiple extern definitions
2362 of a given function name in the entire set of source files that we are
2363 converting. If however one of these definitions resides in exactly the
2364 same source file as the reference we are trying to satisfy then in that
2365 case it would be stupid for us to fail to realize that this one definition
2366 *must* be the precise one we are looking for.
2368 To make sure that we don't miss an opportunity to make this "same file"
2369 leap of faith, we do a prescan of the list of records relating to the
2370 given function name, and we look (on this first scan) *only* for a
2371 definition of the function which is in the same file as the reference
2372 we are currently trying to satisfy. */
2374 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2375 if (dd_p->is_func_def && !dd_p->is_static && dd_p->file == user->file)
2376 return dd_p;
2378 /* Now, since we have not found a definition in the same file as the
2379 reference, we scan the list again and consider all possibilities from
2380 all files. Here we may get conflicts with the things listed in the
2381 SYSCALLS.c.X file, but if that happens it only means that the source
2382 code being converted contains its own definition of a function which
2383 could have been supplied by libc.a. In such cases, we should avoid
2384 issuing the normal warning, and defer to the definition given in the
2385 user's own code. */
2387 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2388 if (dd_p->is_func_def && !dd_p->is_static)
2390 if (!extern_def_p) /* Previous definition? */
2391 extern_def_p = dd_p; /* Remember the first definition found. */
2392 else
2394 /* Ignore definition just found if it came from SYSCALLS.c.X. */
2396 if (is_syscalls_file (dd_p->file))
2397 continue;
2399 /* Quietly replace the definition previously found with the one
2400 just found if the previous one was from SYSCALLS.c.X. */
2402 if (is_syscalls_file (extern_def_p->file))
2404 extern_def_p = dd_p;
2405 continue;
2408 /* If we get here, then there is a conflict between two function
2409 declarations for the same function, both of which came from the
2410 user's own code. */
2412 if (!conflict_noted) /* first time we noticed? */
2414 conflict_noted = 1;
2415 notice ("%s: conflicting extern definitions of '%s'\n",
2416 pname, head->hash_entry->symbol);
2417 if (!quiet_flag)
2419 notice ("%s: declarations of '%s' will not be converted\n",
2420 pname, head->hash_entry->symbol);
2421 notice ("%s: conflict list for '%s' follows:\n",
2422 pname, head->hash_entry->symbol);
2423 fprintf (stderr, "%s: %s(%d): %s\n",
2424 pname,
2425 shortpath (NULL, extern_def_p->file->hash_entry->symbol),
2426 extern_def_p->line, extern_def_p->ansi_decl);
2429 if (!quiet_flag)
2430 fprintf (stderr, "%s: %s(%d): %s\n",
2431 pname,
2432 shortpath (NULL, dd_p->file->hash_entry->symbol),
2433 dd_p->line, dd_p->ansi_decl);
2437 /* We want to err on the side of caution, so if we found multiple conflicting
2438 definitions for the same function, treat this as being that same as if we
2439 had found no definitions (i.e. return NULL). */
2441 if (conflict_noted)
2442 return NULL;
2444 if (!extern_def_p)
2446 /* We have no definitions for this function so do the next best thing.
2447 Search for an extern declaration already in prototype form. */
2449 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2450 if (!dd_p->is_func_def && !dd_p->is_static && dd_p->prototyped)
2452 extern_def_p = dd_p; /* save a pointer to the definition */
2453 if (!quiet_flag)
2454 notice ("%s: warning: using formals list from %s(%d) for function '%s'\n",
2455 pname,
2456 shortpath (NULL, dd_p->file->hash_entry->symbol),
2457 dd_p->line, dd_p->hash_entry->symbol);
2458 break;
2461 /* Gripe about unprototyped function declarations that we found no
2462 corresponding definition (or other source of prototype information)
2463 for.
2465 Gripe even if the unprototyped declaration we are worried about
2466 exists in a file in one of the "system" include directories. We
2467 can gripe about these because we should have at least found a
2468 corresponding (pseudo) definition in the SYSCALLS.c.X file. If we
2469 didn't, then that means that the SYSCALLS.c.X file is missing some
2470 needed prototypes for this particular system. That is worth telling
2471 the user about! */
2473 if (!extern_def_p)
2475 const char *file = user->file->hash_entry->symbol;
2477 if (!quiet_flag)
2478 if (in_system_include_dir (file))
2480 /* Why copy this string into `needed' at all?
2481 Why not just use user->ansi_decl without copying? */
2482 char *needed = alloca (strlen (user->ansi_decl) + 1);
2483 char *p;
2485 strcpy (needed, user->ansi_decl);
2486 p = strstr (needed, user->hash_entry->symbol)
2487 + strlen (user->hash_entry->symbol) + 2;
2488 /* Avoid having ??? in the string. */
2489 *p++ = '?';
2490 *p++ = '?';
2491 *p++ = '?';
2492 strcpy (p, ");");
2494 notice ("%s: %d: '%s' used but missing from SYSCALLS\n",
2495 shortpath (NULL, file), user->line,
2496 needed+7); /* Don't print "extern " */
2498 #if 0
2499 else
2500 notice ("%s: %d: warning: no extern definition for '%s'\n",
2501 shortpath (NULL, file), user->line,
2502 user->hash_entry->symbol);
2503 #endif
2506 return extern_def_p;
2509 /* Find the (only?) static definition for a particular function name in a
2510 given file. Here we get the function-name and the file info indirectly
2511 from the def_dec_info record pointer which is passed in. */
2513 static const def_dec_info *
2514 find_static_definition (const def_dec_info *user)
2516 const def_dec_info *head = user->hash_entry->ddip;
2517 const def_dec_info *dd_p;
2518 int num_static_defs = 0;
2519 const def_dec_info *static_def_p = NULL;
2521 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2522 if (dd_p->is_func_def && dd_p->is_static && (dd_p->file == user->file))
2524 static_def_p = dd_p; /* save a pointer to the definition */
2525 num_static_defs++;
2527 if (num_static_defs == 0)
2529 if (!quiet_flag)
2530 notice ("%s: warning: no static definition for '%s' in file '%s'\n",
2531 pname, head->hash_entry->symbol,
2532 shortpath (NULL, user->file->hash_entry->symbol));
2534 else if (num_static_defs > 1)
2536 notice ("%s: multiple static defs of '%s' in file '%s'\n",
2537 pname, head->hash_entry->symbol,
2538 shortpath (NULL, user->file->hash_entry->symbol));
2539 return NULL;
2541 return static_def_p;
2544 /* Find good prototype style formal argument lists for all of the function
2545 declarations which didn't have them before now.
2547 To do this we consider each function name one at a time. For each function
2548 name, we look at the items on the linked list of def_dec_info records for
2549 that particular name.
2551 Somewhere on this list we should find one (and only one) def_dec_info
2552 record which represents the actual function definition, and this record
2553 should have a nice formal argument list already associated with it.
2555 Thus, all we have to do is to connect up all of the other def_dec_info
2556 records for this particular function name to the special one which has
2557 the full-blown formals list.
2559 Of course it is a little more complicated than just that. See below for
2560 more details. */
2562 static void
2563 connect_defs_and_decs (const hash_table_entry *hp)
2565 const def_dec_info *dd_p;
2566 const def_dec_info *extern_def_p = NULL;
2567 int first_extern_reference = 1;
2569 /* Traverse the list of definitions and declarations for this particular
2570 function name. For each item on the list, if it is a function
2571 definition (either old style or new style) then GCC has already been
2572 kind enough to produce a prototype for us, and it is associated with
2573 the item already, so declare the item as its own associated "definition".
2575 Also, for each item which is only a function declaration, but which
2576 nonetheless has its own prototype already (obviously supplied by the user)
2577 declare the item as its own definition.
2579 Note that when/if there are multiple user-supplied prototypes already
2580 present for multiple declarations of any given function, these multiple
2581 prototypes *should* all match exactly with one another and with the
2582 prototype for the actual function definition. We don't check for this
2583 here however, since we assume that the compiler must have already done
2584 this consistency checking when it was creating the .X files. */
2586 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2587 if (dd_p->prototyped)
2588 ((NONCONST def_dec_info *) dd_p)->definition = dd_p;
2590 /* Traverse the list of definitions and declarations for this particular
2591 function name. For each item on the list, if it is an extern function
2592 declaration and if it has no associated definition yet, go try to find
2593 the matching extern definition for the declaration.
2595 When looking for the matching function definition, warn the user if we
2596 fail to find one.
2598 If we find more that one function definition also issue a warning.
2600 Do the search for the matching definition only once per unique function
2601 name (and only when absolutely needed) so that we can avoid putting out
2602 redundant warning messages, and so that we will only put out warning
2603 messages when there is actually a reference (i.e. a declaration) for
2604 which we need to find a matching definition. */
2606 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2607 if (!dd_p->is_func_def && !dd_p->is_static && !dd_p->definition)
2609 if (first_extern_reference)
2611 extern_def_p = find_extern_def (hp->ddip, dd_p);
2612 first_extern_reference = 0;
2614 ((NONCONST def_dec_info *) dd_p)->definition = extern_def_p;
2617 /* Traverse the list of definitions and declarations for this particular
2618 function name. For each item on the list, if it is a static function
2619 declaration and if it has no associated definition yet, go try to find
2620 the matching static definition for the declaration within the same file.
2622 When looking for the matching function definition, warn the user if we
2623 fail to find one in the same file with the declaration, and refuse to
2624 convert this kind of cross-file static function declaration. After all,
2625 this is stupid practice and should be discouraged.
2627 We don't have to worry about the possibility that there is more than one
2628 matching function definition in the given file because that would have
2629 been flagged as an error by the compiler.
2631 Do the search for the matching definition only once per unique
2632 function-name/source-file pair (and only when absolutely needed) so that
2633 we can avoid putting out redundant warning messages, and so that we will
2634 only put out warning messages when there is actually a reference (i.e. a
2635 declaration) for which we actually need to find a matching definition. */
2637 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2638 if (!dd_p->is_func_def && dd_p->is_static && !dd_p->definition)
2640 const def_dec_info *dd_p2;
2641 const def_dec_info *static_def;
2643 /* We have now found a single static declaration for which we need to
2644 find a matching definition. We want to minimize the work (and the
2645 number of warnings), so we will find an appropriate (matching)
2646 static definition for this declaration, and then distribute it
2647 (as the definition for) any and all other static declarations
2648 for this function name which occur within the same file, and which
2649 do not already have definitions.
2651 Note that a trick is used here to prevent subsequent attempts to
2652 call find_static_definition for a given function-name & file
2653 if the first such call returns NULL. Essentially, we convert
2654 these NULL return values to -1, and put the -1 into the definition
2655 field for each other static declaration from the same file which
2656 does not already have an associated definition.
2657 This makes these other static declarations look like they are
2658 actually defined already when the outer loop here revisits them
2659 later on. Thus, the outer loop will skip over them. Later, we
2660 turn the -1's back to NULL's. */
2662 ((NONCONST def_dec_info *) dd_p)->definition =
2663 (static_def = find_static_definition (dd_p))
2664 ? static_def
2665 : (const def_dec_info *) -1;
2667 for (dd_p2 = dd_p->next_for_func; dd_p2; dd_p2 = dd_p2->next_for_func)
2668 if (!dd_p2->is_func_def && dd_p2->is_static
2669 && !dd_p2->definition && (dd_p2->file == dd_p->file))
2670 ((NONCONST def_dec_info *) dd_p2)->definition = dd_p->definition;
2673 /* Convert any dummy (-1) definitions we created in the step above back to
2674 NULL's (as they should be). */
2676 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2677 if (dd_p->definition == (def_dec_info *) -1)
2678 ((NONCONST def_dec_info *) dd_p)->definition = NULL;
2681 #endif /* !defined (UNPROTOIZE) */
2683 /* Give a pointer into the clean text buffer, return a number which is the
2684 original source line number that the given pointer points into. */
2686 static int
2687 identify_lineno (const char *clean_p)
2689 int line_num = 1;
2690 const char *scan_p;
2692 for (scan_p = clean_text_base; scan_p <= clean_p; scan_p++)
2693 if (*scan_p == '\n')
2694 line_num++;
2695 return line_num;
2698 /* Issue an error message and give up on doing this particular edit. */
2700 static void
2701 declare_source_confusing (const char *clean_p)
2703 if (!quiet_flag)
2705 if (clean_p == 0)
2706 notice ("%s: %d: warning: source too confusing\n",
2707 shortpath (NULL, convert_filename), last_known_line_number);
2708 else
2709 notice ("%s: %d: warning: source too confusing\n",
2710 shortpath (NULL, convert_filename),
2711 identify_lineno (clean_p));
2713 longjmp (source_confusion_recovery, 1);
2716 /* Check that a condition which is expected to be true in the original source
2717 code is in fact true. If not, issue an error message and give up on
2718 converting this particular source file. */
2720 static void
2721 check_source (int cond, const char *clean_p)
2723 if (!cond)
2724 declare_source_confusing (clean_p);
2727 /* If we think of the in-core cleaned text buffer as a memory mapped
2728 file (with the variable last_known_line_start acting as sort of a
2729 file pointer) then we can imagine doing "seeks" on the buffer. The
2730 following routine implements a kind of "seek" operation for the in-core
2731 (cleaned) copy of the source file. When finished, it returns a pointer to
2732 the start of a given (numbered) line in the cleaned text buffer.
2734 Note that protoize only has to "seek" in the forward direction on the
2735 in-core cleaned text file buffers, and it never needs to back up.
2737 This routine is made a little bit faster by remembering the line number
2738 (and pointer value) supplied (and returned) from the previous "seek".
2739 This prevents us from always having to start all over back at the top
2740 of the in-core cleaned buffer again. */
2742 static const char *
2743 seek_to_line (int n)
2745 if (n < last_known_line_number)
2746 abort ();
2748 while (n > last_known_line_number)
2750 while (*last_known_line_start != '\n')
2751 check_source (++last_known_line_start < clean_text_limit, 0);
2752 last_known_line_start++;
2753 last_known_line_number++;
2755 return last_known_line_start;
2758 /* Given a pointer to a character in the cleaned text buffer, return a pointer
2759 to the next non-whitespace character which follows it. */
2761 static const char *
2762 forward_to_next_token_char (const char *ptr)
2764 for (++ptr; ISSPACE ((const unsigned char)*ptr);
2765 check_source (++ptr < clean_text_limit, 0))
2766 continue;
2767 return ptr;
2770 /* Copy a chunk of text of length `len' and starting at `str' to the current
2771 output buffer. Note that all attempts to add stuff to the current output
2772 buffer ultimately go through here. */
2774 static void
2775 output_bytes (const char *str, size_t len)
2777 if ((repl_write_ptr + 1) + len >= repl_text_limit)
2779 size_t new_size = (repl_text_limit - repl_text_base) << 1;
2780 char *new_buf = xrealloc (repl_text_base, new_size);
2782 repl_write_ptr = new_buf + (repl_write_ptr - repl_text_base);
2783 repl_text_base = new_buf;
2784 repl_text_limit = new_buf + new_size;
2786 memcpy (repl_write_ptr + 1, str, len);
2787 repl_write_ptr += len;
2790 /* Copy all bytes (except the trailing null) of a null terminated string to
2791 the current output buffer. */
2793 static void
2794 output_string (const char *str)
2796 output_bytes (str, strlen (str));
2799 /* Copy some characters from the original text buffer to the current output
2800 buffer.
2802 This routine takes a pointer argument `p' which is assumed to be a pointer
2803 into the cleaned text buffer. The bytes which are copied are the `original'
2804 equivalents for the set of bytes between the last value of `clean_read_ptr'
2805 and the argument value `p'.
2807 The set of bytes copied however, comes *not* from the cleaned text buffer,
2808 but rather from the direct counterparts of these bytes within the original
2809 text buffer.
2811 Thus, when this function is called, some bytes from the original text
2812 buffer (which may include original comments and preprocessing directives)
2813 will be copied into the output buffer.
2815 Note that the request implied when this routine is called includes the
2816 byte pointed to by the argument pointer `p'. */
2818 static void
2819 output_up_to (const char *p)
2821 size_t copy_length = (size_t) (p - clean_read_ptr);
2822 const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
2824 if (copy_length == 0)
2825 return;
2827 output_bytes (copy_start, copy_length);
2828 clean_read_ptr = p;
2831 /* Given a pointer to a def_dec_info record which represents some form of
2832 definition of a function (perhaps a real definition, or in lieu of that
2833 perhaps just a declaration with a full prototype) return true if this
2834 function is one which we should avoid converting. Return false
2835 otherwise. */
2837 static int
2838 other_variable_style_function (const char *ansi_header)
2840 #ifdef UNPROTOIZE
2842 /* See if we have a stdarg function, or a function which has stdarg style
2843 parameters or a stdarg style return type. */
2845 return strstr (ansi_header, "...") != 0;
2847 #else /* !defined (UNPROTOIZE) */
2849 /* See if we have a varargs function, or a function which has varargs style
2850 parameters or a varargs style return type. */
2852 const char *p;
2853 int len = strlen (varargs_style_indicator);
2855 for (p = ansi_header; p; )
2857 const char *candidate;
2859 if ((candidate = strstr (p, varargs_style_indicator)) == 0)
2860 return 0;
2861 else
2862 if (!is_id_char (candidate[-1]) && !is_id_char (candidate[len]))
2863 return 1;
2864 else
2865 p = candidate + 1;
2867 return 0;
2868 #endif /* !defined (UNPROTOIZE) */
2871 /* Do the editing operation specifically for a function "declaration". Note
2872 that editing for function "definitions" are handled in a separate routine
2873 below. */
2875 static void
2876 edit_fn_declaration (const def_dec_info *def_dec_p,
2877 const char *volatile clean_text_p)
2879 const char *start_formals;
2880 const char *end_formals;
2881 const char *function_to_edit = def_dec_p->hash_entry->symbol;
2882 size_t func_name_len = strlen (function_to_edit);
2883 const char *end_of_fn_name;
2885 #ifndef UNPROTOIZE
2887 const f_list_chain_item *this_f_list_chain_item;
2888 const def_dec_info *definition = def_dec_p->definition;
2890 /* If we are protoizing, and if we found no corresponding definition for
2891 this particular function declaration, then just leave this declaration
2892 exactly as it is. */
2894 if (!definition)
2895 return;
2897 /* If we are protoizing, and if the corresponding definition that we found
2898 for this particular function declaration defined an old style varargs
2899 function, then we want to issue a warning and just leave this function
2900 declaration unconverted. */
2902 if (other_variable_style_function (definition->ansi_decl))
2904 if (!quiet_flag)
2905 notice ("%s: %d: warning: varargs function declaration not converted\n",
2906 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
2907 def_dec_p->line);
2908 return;
2911 #endif /* !defined (UNPROTOIZE) */
2913 /* Setup here to recover from confusing source code detected during this
2914 particular "edit". */
2916 save_pointers ();
2917 if (setjmp (source_confusion_recovery))
2919 restore_pointers ();
2920 notice ("%s: declaration of function '%s' not converted\n",
2921 pname, function_to_edit);
2922 return;
2925 /* We are editing a function declaration. The line number we did a seek to
2926 contains the comma or semicolon which follows the declaration. Our job
2927 now is to scan backwards looking for the function name. This name *must*
2928 be followed by open paren (ignoring whitespace, of course). We need to
2929 replace everything between that open paren and the corresponding closing
2930 paren. If we are protoizing, we need to insert the prototype-style
2931 formals lists. If we are unprotoizing, we need to just delete everything
2932 between the pairs of opening and closing parens. */
2934 /* First move up to the end of the line. */
2936 while (*clean_text_p != '\n')
2937 check_source (++clean_text_p < clean_text_limit, 0);
2938 clean_text_p--; /* Point to just before the newline character. */
2940 /* Now we can scan backwards for the function name. */
2944 for (;;)
2946 /* Scan leftwards until we find some character which can be
2947 part of an identifier. */
2949 while (!is_id_char (*clean_text_p))
2950 check_source (--clean_text_p > clean_read_ptr, 0);
2952 /* Scan backwards until we find a char that cannot be part of an
2953 identifier. */
2955 while (is_id_char (*clean_text_p))
2956 check_source (--clean_text_p > clean_read_ptr, 0);
2958 /* Having found an "id break", see if the following id is the one
2959 that we are looking for. If so, then exit from this loop. */
2961 if (!strncmp (clean_text_p+1, function_to_edit, func_name_len))
2963 char ch = *(clean_text_p + 1 + func_name_len);
2965 /* Must also check to see that the name in the source text
2966 ends where it should (in order to prevent bogus matches
2967 on similar but longer identifiers. */
2969 if (! is_id_char (ch))
2970 break; /* exit from loop */
2974 /* We have now found the first perfect match for the function name in
2975 our backward search. This may or may not be the actual function
2976 name at the start of the actual function declaration (i.e. we could
2977 have easily been mislead). We will try to avoid getting fooled too
2978 often by looking forward for the open paren which should follow the
2979 identifier we just found. We ignore whitespace while hunting. If
2980 the next non-whitespace byte we see is *not* an open left paren,
2981 then we must assume that we have been fooled and we start over
2982 again accordingly. Note that there is no guarantee, that even if
2983 we do see the open paren, that we are in the right place.
2984 Programmers do the strangest things sometimes! */
2986 end_of_fn_name = clean_text_p + strlen (def_dec_p->hash_entry->symbol);
2987 start_formals = forward_to_next_token_char (end_of_fn_name);
2989 while (*start_formals != '(');
2991 /* start_of_formals now points to the opening left paren which immediately
2992 follows the name of the function. */
2994 /* Note that there may be several formals lists which need to be modified
2995 due to the possibility that the return type of this function is a
2996 pointer-to-function type. If there are several formals lists, we
2997 convert them in left-to-right order here. */
2999 #ifndef UNPROTOIZE
3000 this_f_list_chain_item = definition->f_list_chain;
3001 #endif /* !defined (UNPROTOIZE) */
3003 for (;;)
3006 int depth;
3008 end_formals = start_formals + 1;
3009 depth = 1;
3010 for (; depth; check_source (++end_formals < clean_text_limit, 0))
3012 switch (*end_formals)
3014 case '(':
3015 depth++;
3016 break;
3017 case ')':
3018 depth--;
3019 break;
3022 end_formals--;
3025 /* end_formals now points to the closing right paren of the formals
3026 list whose left paren is pointed to by start_formals. */
3028 /* Now, if we are protoizing, we insert the new ANSI-style formals list
3029 attached to the associated definition of this function. If however
3030 we are unprotoizing, then we simply delete any formals list which
3031 may be present. */
3033 output_up_to (start_formals);
3034 #ifndef UNPROTOIZE
3035 if (this_f_list_chain_item)
3037 output_string (this_f_list_chain_item->formals_list);
3038 this_f_list_chain_item = this_f_list_chain_item->chain_next;
3040 else
3042 if (!quiet_flag)
3043 notice ("%s: warning: too many parameter lists in declaration of '%s'\n",
3044 pname, def_dec_p->hash_entry->symbol);
3045 check_source (0, end_formals); /* leave the declaration intact */
3047 #endif /* !defined (UNPROTOIZE) */
3048 clean_read_ptr = end_formals - 1;
3050 /* Now see if it looks like there may be another formals list associated
3051 with the function declaration that we are converting (following the
3052 formals list that we just converted. */
3055 const char *another_r_paren = forward_to_next_token_char (end_formals);
3057 if ((*another_r_paren != ')')
3058 || (*(start_formals = forward_to_next_token_char (another_r_paren)) != '('))
3060 #ifndef UNPROTOIZE
3061 if (this_f_list_chain_item)
3063 if (!quiet_flag)
3064 notice ("\n%s: warning: too few parameter lists in declaration of '%s'\n",
3065 pname, def_dec_p->hash_entry->symbol);
3066 check_source (0, start_formals); /* leave the decl intact */
3068 #endif /* !defined (UNPROTOIZE) */
3069 break;
3074 /* There does appear to be yet another formals list, so loop around
3075 again, and convert it also. */
3079 /* Edit a whole group of formals lists, starting with the rightmost one
3080 from some set of formals lists. This routine is called once (from the
3081 outside) for each function declaration which is converted. It is
3082 recursive however, and it calls itself once for each remaining formal
3083 list that lies to the left of the one it was originally called to work
3084 on. Thus, a whole set gets done in right-to-left order.
3086 This routine returns nonzero if it thinks that it should not be trying
3087 to convert this particular function definition (because the name of the
3088 function doesn't match the one expected). */
3090 static int
3091 edit_formals_lists (const char *end_formals, unsigned int f_list_count,
3092 const def_dec_info *def_dec_p)
3094 const char *start_formals;
3095 int depth;
3097 start_formals = end_formals - 1;
3098 depth = 1;
3099 for (; depth; check_source (--start_formals > clean_read_ptr, 0))
3101 switch (*start_formals)
3103 case '(':
3104 depth--;
3105 break;
3106 case ')':
3107 depth++;
3108 break;
3111 start_formals++;
3113 /* start_formals now points to the opening left paren of the formals list. */
3115 f_list_count--;
3117 if (f_list_count)
3119 const char *next_end;
3121 /* There should be more formal lists to the left of here. */
3123 next_end = start_formals - 1;
3124 check_source (next_end > clean_read_ptr, 0);
3125 while (ISSPACE ((const unsigned char)*next_end))
3126 check_source (--next_end > clean_read_ptr, 0);
3127 check_source (*next_end == ')', next_end);
3128 check_source (--next_end > clean_read_ptr, 0);
3129 check_source (*next_end == ')', next_end);
3130 if (edit_formals_lists (next_end, f_list_count, def_dec_p))
3131 return 1;
3134 /* Check that the function name in the header we are working on is the same
3135 as the one we would expect to find. If not, issue a warning and return
3136 nonzero. */
3138 if (f_list_count == 0)
3140 const char *expected = def_dec_p->hash_entry->symbol;
3141 const char *func_name_start;
3142 const char *func_name_limit;
3143 size_t func_name_len;
3145 for (func_name_limit = start_formals-1;
3146 ISSPACE ((const unsigned char)*func_name_limit); )
3147 check_source (--func_name_limit > clean_read_ptr, 0);
3149 for (func_name_start = func_name_limit++;
3150 is_id_char (*func_name_start);
3151 func_name_start--)
3152 check_source (func_name_start > clean_read_ptr, 0);
3153 func_name_start++;
3154 func_name_len = func_name_limit - func_name_start;
3155 if (func_name_len == 0)
3156 check_source (0, func_name_start);
3157 if (func_name_len != strlen (expected)
3158 || strncmp (func_name_start, expected, func_name_len))
3160 notice ("%s: %d: warning: found '%s' but expected '%s'\n",
3161 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3162 identify_lineno (func_name_start),
3163 dupnstr (func_name_start, func_name_len),
3164 expected);
3165 return 1;
3169 output_up_to (start_formals);
3171 #ifdef UNPROTOIZE
3172 if (f_list_count == 0)
3173 output_string (def_dec_p->formal_names);
3174 #else /* !defined (UNPROTOIZE) */
3176 unsigned f_list_depth;
3177 const f_list_chain_item *flci_p = def_dec_p->f_list_chain;
3179 /* At this point, the current value of f_list count says how many
3180 links we have to follow through the f_list_chain to get to the
3181 particular formals list that we need to output next. */
3183 for (f_list_depth = 0; f_list_depth < f_list_count; f_list_depth++)
3184 flci_p = flci_p->chain_next;
3185 output_string (flci_p->formals_list);
3187 #endif /* !defined (UNPROTOIZE) */
3189 clean_read_ptr = end_formals - 1;
3190 return 0;
3193 /* Given a pointer to a byte in the clean text buffer which points to
3194 the beginning of a line that contains a "follower" token for a
3195 function definition header, do whatever is necessary to find the
3196 right closing paren for the rightmost formals list of the function
3197 definition header. */
3199 static const char *
3200 find_rightmost_formals_list (const char *clean_text_p)
3202 const char *end_formals;
3204 /* We are editing a function definition. The line number we did a seek
3205 to contains the first token which immediately follows the entire set of
3206 formals lists which are part of this particular function definition
3207 header.
3209 Our job now is to scan leftwards in the clean text looking for the
3210 right-paren which is at the end of the function header's rightmost
3211 formals list.
3213 If we ignore whitespace, this right paren should be the first one we
3214 see which is (ignoring whitespace) immediately followed either by the
3215 open curly-brace beginning the function body or by an alphabetic
3216 character (in the case where the function definition is in old (K&R)
3217 style and there are some declarations of formal parameters). */
3219 /* It is possible that the right paren we are looking for is on the
3220 current line (together with its following token). Just in case that
3221 might be true, we start out here by skipping down to the right end of
3222 the current line before starting our scan. */
3224 for (end_formals = clean_text_p; *end_formals != '\n'; end_formals++)
3225 continue;
3226 end_formals--;
3228 #ifdef UNPROTOIZE
3230 /* Now scan backwards while looking for the right end of the rightmost
3231 formals list associated with this function definition. */
3234 char ch;
3235 const char *l_brace_p;
3237 /* Look leftward and try to find a right-paren. */
3239 while (*end_formals != ')')
3241 if (ISSPACE ((unsigned char)*end_formals))
3242 while (ISSPACE ((unsigned char)*end_formals))
3243 check_source (--end_formals > clean_read_ptr, 0);
3244 else
3245 check_source (--end_formals > clean_read_ptr, 0);
3248 ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3249 /* Since we are unprotoizing an ANSI-style (prototyped) function
3250 definition, there had better not be anything (except whitespace)
3251 between the end of the ANSI formals list and the beginning of the
3252 function body (i.e. the '{'). */
3254 check_source (ch == '{', l_brace_p);
3257 #else /* !defined (UNPROTOIZE) */
3259 /* Now scan backwards while looking for the right end of the rightmost
3260 formals list associated with this function definition. */
3262 while (1)
3264 char ch;
3265 const char *l_brace_p;
3267 /* Look leftward and try to find a right-paren. */
3269 while (*end_formals != ')')
3271 if (ISSPACE ((const unsigned char)*end_formals))
3272 while (ISSPACE ((const unsigned char)*end_formals))
3273 check_source (--end_formals > clean_read_ptr, 0);
3274 else
3275 check_source (--end_formals > clean_read_ptr, 0);
3278 ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3280 /* Since it is possible that we found a right paren before the starting
3281 '{' of the body which IS NOT the one at the end of the real K&R
3282 formals list (say for instance, we found one embedded inside one of
3283 the old K&R formal parameter declarations) we have to check to be
3284 sure that this is in fact the right paren that we were looking for.
3286 The one we were looking for *must* be followed by either a '{' or
3287 by an alphabetic character, while others *cannot* validly be followed
3288 by such characters. */
3290 if ((ch == '{') || ISALPHA ((unsigned char) ch))
3291 break;
3293 /* At this point, we have found a right paren, but we know that it is
3294 not the one we were looking for, so backup one character and keep
3295 looking. */
3297 check_source (--end_formals > clean_read_ptr, 0);
3300 #endif /* !defined (UNPROTOIZE) */
3302 return end_formals;
3305 #ifndef UNPROTOIZE
3307 /* Insert into the output file a totally new declaration for a function
3308 which (up until now) was being called from within the current block
3309 without having been declared at any point such that the declaration
3310 was visible (i.e. in scope) at the point of the call.
3312 We need to add in explicit declarations for all such function calls
3313 in order to get the full benefit of prototype-based function call
3314 parameter type checking. */
3316 static void
3317 add_local_decl (const def_dec_info *def_dec_p, const char *clean_text_p)
3319 const char *start_of_block;
3320 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3322 /* Don't insert new local explicit declarations unless explicitly requested
3323 to do so. */
3325 if (!local_flag)
3326 return;
3328 /* Setup here to recover from confusing source code detected during this
3329 particular "edit". */
3331 save_pointers ();
3332 if (setjmp (source_confusion_recovery))
3334 restore_pointers ();
3335 notice ("%s: local declaration for function '%s' not inserted\n",
3336 pname, function_to_edit);
3337 return;
3340 /* We have already done a seek to the start of the line which should
3341 contain *the* open curly brace which begins the block in which we need
3342 to insert an explicit function declaration (to replace the implicit one).
3344 Now we scan that line, starting from the left, until we find the
3345 open curly brace we are looking for. Note that there may actually be
3346 multiple open curly braces on the given line, but we will be happy
3347 with the leftmost one no matter what. */
3349 start_of_block = clean_text_p;
3350 while (*start_of_block != '{' && *start_of_block != '\n')
3351 check_source (++start_of_block < clean_text_limit, 0);
3353 /* Note that the line from the original source could possibly
3354 contain *no* open curly braces! This happens if the line contains
3355 a macro call which expands into a chunk of text which includes a
3356 block (and that block's associated open and close curly braces).
3357 In cases like this, we give up, issue a warning, and do nothing. */
3359 if (*start_of_block != '{')
3361 if (!quiet_flag)
3362 notice ("\n%s: %d: warning: can't add declaration of '%s' into macro call\n",
3363 def_dec_p->file->hash_entry->symbol, def_dec_p->line,
3364 def_dec_p->hash_entry->symbol);
3365 return;
3368 /* Figure out what a nice (pretty) indentation would be for the new
3369 declaration we are adding. In order to do this, we must scan forward
3370 from the '{' until we find the first line which starts with some
3371 non-whitespace characters (i.e. real "token" material). */
3374 const char *ep = forward_to_next_token_char (start_of_block) - 1;
3375 const char *sp;
3377 /* Now we have ep pointing at the rightmost byte of some existing indent
3378 stuff. At least that is the hope.
3380 We can now just scan backwards and find the left end of the existing
3381 indentation string, and then copy it to the output buffer. */
3383 for (sp = ep; ISSPACE ((const unsigned char)*sp) && *sp != '\n'; sp--)
3384 continue;
3386 /* Now write out the open { which began this block, and any following
3387 trash up to and including the last byte of the existing indent that
3388 we just found. */
3390 output_up_to (ep);
3392 /* Now we go ahead and insert the new declaration at this point.
3394 If the definition of the given function is in the same file that we
3395 are currently editing, and if its full ANSI declaration normally
3396 would start with the keyword `extern', suppress the `extern'. */
3399 const char *decl = def_dec_p->definition->ansi_decl;
3401 if ((*decl == 'e') && (def_dec_p->file == def_dec_p->definition->file))
3402 decl += 7;
3403 output_string (decl);
3406 /* Finally, write out a new indent string, just like the preceding one
3407 that we found. This will typically include a newline as the first
3408 character of the indent string. */
3410 output_bytes (sp, (size_t) (ep - sp) + 1);
3414 /* Given a pointer to a file_info record, and a pointer to the beginning
3415 of a line (in the clean text buffer) which is assumed to contain the
3416 first "follower" token for the first function definition header in the
3417 given file, find a good place to insert some new global function
3418 declarations (which will replace scattered and imprecise implicit ones)
3419 and then insert the new explicit declaration at that point in the file. */
3421 static void
3422 add_global_decls (const file_info *file_p, const char *clean_text_p)
3424 const def_dec_info *dd_p;
3425 const char *scan_p;
3427 /* Setup here to recover from confusing source code detected during this
3428 particular "edit". */
3430 save_pointers ();
3431 if (setjmp (source_confusion_recovery))
3433 restore_pointers ();
3434 notice ("%s: global declarations for file '%s' not inserted\n",
3435 pname, shortpath (NULL, file_p->hash_entry->symbol));
3436 return;
3439 /* Start by finding a good location for adding the new explicit function
3440 declarations. To do this, we scan backwards, ignoring whitespace
3441 and comments and other junk until we find either a semicolon, or until
3442 we hit the beginning of the file. */
3444 scan_p = find_rightmost_formals_list (clean_text_p);
3445 for (;; --scan_p)
3447 if (scan_p < clean_text_base)
3448 break;
3449 check_source (scan_p > clean_read_ptr, 0);
3450 if (*scan_p == ';')
3451 break;
3454 /* scan_p now points either to a semicolon, or to just before the start
3455 of the whole file. */
3457 /* Now scan forward for the first non-whitespace character. In theory,
3458 this should be the first character of the following function definition
3459 header. We will put in the added declarations just prior to that. */
3461 scan_p++;
3462 while (ISSPACE ((const unsigned char)*scan_p))
3463 scan_p++;
3464 scan_p--;
3466 output_up_to (scan_p);
3468 /* Now write out full prototypes for all of the things that had been
3469 implicitly declared in this file (but only those for which we were
3470 actually able to find unique matching definitions). Avoid duplicates
3471 by marking things that we write out as we go. */
3474 int some_decls_added = 0;
3476 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3477 if (dd_p->is_implicit && dd_p->definition && !dd_p->definition->written)
3479 const char *decl = dd_p->definition->ansi_decl;
3481 /* If the function for which we are inserting a declaration is
3482 actually defined later in the same file, then suppress the
3483 leading `extern' keyword (if there is one). */
3485 if (*decl == 'e' && (dd_p->file == dd_p->definition->file))
3486 decl += 7;
3488 output_string ("\n");
3489 output_string (decl);
3490 some_decls_added = 1;
3491 ((NONCONST def_dec_info *) dd_p->definition)->written = 1;
3493 if (some_decls_added)
3494 output_string ("\n\n");
3497 /* Unmark all of the definitions that we just marked. */
3499 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3500 if (dd_p->definition)
3501 ((NONCONST def_dec_info *) dd_p->definition)->written = 0;
3504 #endif /* !defined (UNPROTOIZE) */
3506 /* Do the editing operation specifically for a function "definition". Note
3507 that editing operations for function "declarations" are handled by a
3508 separate routine above. */
3510 static void
3511 edit_fn_definition (const def_dec_info *def_dec_p, const char *clean_text_p)
3513 const char *end_formals;
3514 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3516 /* Setup here to recover from confusing source code detected during this
3517 particular "edit". */
3519 save_pointers ();
3520 if (setjmp (source_confusion_recovery))
3522 restore_pointers ();
3523 notice ("%s: definition of function '%s' not converted\n",
3524 pname, function_to_edit);
3525 return;
3528 end_formals = find_rightmost_formals_list (clean_text_p);
3530 /* end_of_formals now points to the closing right paren of the rightmost
3531 formals list which is actually part of the `header' of the function
3532 definition that we are converting. */
3534 /* If the header of this function definition looks like it declares a
3535 function with a variable number of arguments, and if the way it does
3536 that is different from that way we would like it (i.e. varargs vs.
3537 stdarg) then issue a warning and leave the header unconverted. */
3539 if (other_variable_style_function (def_dec_p->ansi_decl))
3541 if (!quiet_flag)
3542 notice ("%s: %d: warning: definition of %s not converted\n",
3543 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3544 identify_lineno (end_formals),
3545 other_var_style);
3546 output_up_to (end_formals);
3547 return;
3550 if (edit_formals_lists (end_formals, def_dec_p->f_list_count, def_dec_p))
3552 restore_pointers ();
3553 notice ("%s: definition of function '%s' not converted\n",
3554 pname, function_to_edit);
3555 return;
3558 /* Have to output the last right paren because this never gets flushed by
3559 edit_formals_list. */
3561 output_up_to (end_formals);
3563 #ifdef UNPROTOIZE
3565 const char *decl_p;
3566 const char *semicolon_p;
3567 const char *limit_p;
3568 const char *scan_p;
3569 int had_newlines = 0;
3571 /* Now write out the K&R style formal declarations, one per line. */
3573 decl_p = def_dec_p->formal_decls;
3574 limit_p = decl_p + strlen (decl_p);
3575 for (;decl_p < limit_p; decl_p = semicolon_p + 2)
3577 for (semicolon_p = decl_p; *semicolon_p != ';'; semicolon_p++)
3578 continue;
3579 output_string ("\n");
3580 output_string (indent_string);
3581 output_bytes (decl_p, (size_t) ((semicolon_p + 1) - decl_p));
3584 /* If there are no newlines between the end of the formals list and the
3585 start of the body, we should insert one now. */
3587 for (scan_p = end_formals+1; *scan_p != '{'; )
3589 if (*scan_p == '\n')
3591 had_newlines = 1;
3592 break;
3594 check_source (++scan_p < clean_text_limit, 0);
3596 if (!had_newlines)
3597 output_string ("\n");
3599 #else /* !defined (UNPROTOIZE) */
3600 /* If we are protoizing, there may be some flotsam & jetsam (like comments
3601 and preprocessing directives) after the old formals list but before
3602 the following { and we would like to preserve that stuff while effectively
3603 deleting the existing K&R formal parameter declarations. We do so here
3604 in a rather tricky way. Basically, we white out any stuff *except*
3605 the comments/pp-directives in the original text buffer, then, if there
3606 is anything in this area *other* than whitespace, we output it. */
3608 const char *end_formals_orig;
3609 const char *start_body;
3610 const char *start_body_orig;
3611 const char *scan;
3612 const char *scan_orig;
3613 int have_flotsam = 0;
3614 int have_newlines = 0;
3616 for (start_body = end_formals + 1; *start_body != '{';)
3617 check_source (++start_body < clean_text_limit, 0);
3619 end_formals_orig = orig_text_base + (end_formals - clean_text_base);
3620 start_body_orig = orig_text_base + (start_body - clean_text_base);
3621 scan = end_formals + 1;
3622 scan_orig = end_formals_orig + 1;
3623 for (; scan < start_body; scan++, scan_orig++)
3625 if (*scan == *scan_orig)
3627 have_newlines |= (*scan_orig == '\n');
3628 /* Leave identical whitespace alone. */
3629 if (!ISSPACE ((const unsigned char)*scan_orig))
3630 *((NONCONST char *) scan_orig) = ' '; /* identical - so whiteout */
3632 else
3633 have_flotsam = 1;
3635 if (have_flotsam)
3636 output_bytes (end_formals_orig + 1,
3637 (size_t) (start_body_orig - end_formals_orig) - 1);
3638 else
3639 if (have_newlines)
3640 output_string ("\n");
3641 else
3642 output_string (" ");
3643 clean_read_ptr = start_body - 1;
3645 #endif /* !defined (UNPROTOIZE) */
3648 /* Clean up the clean text buffer. Do this by converting comments and
3649 preprocessing directives into spaces. Also convert line continuations
3650 into whitespace. Also, whiteout string and character literals. */
3652 static void
3653 do_cleaning (char *new_clean_text_base, const char *new_clean_text_limit)
3655 char *scan_p;
3656 int non_whitespace_since_newline = 0;
3658 for (scan_p = new_clean_text_base; scan_p < new_clean_text_limit; scan_p++)
3660 switch (*scan_p)
3662 case '/': /* Handle comments. */
3663 if (scan_p[1] != '*')
3664 goto regular;
3665 non_whitespace_since_newline = 1;
3666 scan_p[0] = ' ';
3667 scan_p[1] = ' ';
3668 scan_p += 2;
3669 while (scan_p[1] != '/' || scan_p[0] != '*')
3671 if (!ISSPACE ((const unsigned char)*scan_p))
3672 *scan_p = ' ';
3673 if (++scan_p >= new_clean_text_limit)
3674 abort ();
3676 *scan_p++ = ' ';
3677 *scan_p = ' ';
3678 break;
3680 case '#': /* Handle pp directives. */
3681 if (non_whitespace_since_newline)
3682 goto regular;
3683 *scan_p = ' ';
3684 while (scan_p[1] != '\n' || scan_p[0] == '\\')
3686 if (!ISSPACE ((const unsigned char)*scan_p))
3687 *scan_p = ' ';
3688 if (++scan_p >= new_clean_text_limit)
3689 abort ();
3691 *scan_p++ = ' ';
3692 break;
3694 case '\'': /* Handle character literals. */
3695 non_whitespace_since_newline = 1;
3696 while (scan_p[1] != '\'' || scan_p[0] == '\\')
3698 if (scan_p[0] == '\\'
3699 && !ISSPACE ((const unsigned char) scan_p[1]))
3700 scan_p[1] = ' ';
3701 if (!ISSPACE ((const unsigned char)*scan_p))
3702 *scan_p = ' ';
3703 if (++scan_p >= new_clean_text_limit)
3704 abort ();
3706 *scan_p++ = ' ';
3707 break;
3709 case '"': /* Handle string literals. */
3710 non_whitespace_since_newline = 1;
3711 while (scan_p[1] != '"' || scan_p[0] == '\\')
3713 if (scan_p[0] == '\\'
3714 && !ISSPACE ((const unsigned char) scan_p[1]))
3715 scan_p[1] = ' ';
3716 if (!ISSPACE ((const unsigned char)*scan_p))
3717 *scan_p = ' ';
3718 if (++scan_p >= new_clean_text_limit)
3719 abort ();
3721 if (!ISSPACE ((const unsigned char)*scan_p))
3722 *scan_p = ' ';
3723 scan_p++;
3724 break;
3726 case '\\': /* Handle line continuations. */
3727 if (scan_p[1] != '\n')
3728 goto regular;
3729 *scan_p = ' ';
3730 break;
3732 case '\n':
3733 non_whitespace_since_newline = 0; /* Reset. */
3734 break;
3736 case ' ':
3737 case '\v':
3738 case '\t':
3739 case '\r':
3740 case '\f':
3741 case '\b':
3742 break; /* Whitespace characters. */
3744 default:
3745 regular:
3746 non_whitespace_since_newline = 1;
3747 break;
3752 /* Given a pointer to the closing right parenthesis for a particular formals
3753 list (in the clean text buffer) find the corresponding left parenthesis
3754 and return a pointer to it. */
3756 static const char *
3757 careful_find_l_paren (const char *p)
3759 const char *q;
3760 int paren_depth;
3762 for (paren_depth = 1, q = p-1; paren_depth; check_source (--q >= clean_text_base, 0))
3764 switch (*q)
3766 case ')':
3767 paren_depth++;
3768 break;
3769 case '(':
3770 paren_depth--;
3771 break;
3774 return ++q;
3777 /* Scan the clean text buffer for cases of function definitions that we
3778 don't really know about because they were preprocessed out when the
3779 aux info files were created.
3781 In this version of protoize/unprotoize we just give a warning for each
3782 one found. A later version may be able to at least unprotoize such
3783 missed items.
3785 Note that we may easily find all function definitions simply by
3786 looking for places where there is a left paren which is (ignoring
3787 whitespace) immediately followed by either a left-brace or by an
3788 upper or lower case letter. Whenever we find this combination, we
3789 have also found a function definition header.
3791 Finding function *declarations* using syntactic clues is much harder.
3792 I will probably try to do this in a later version though. */
3794 static void
3795 scan_for_missed_items (const file_info *file_p)
3797 static const char *scan_p;
3798 const char *limit = clean_text_limit - 3;
3799 static const char *backup_limit;
3801 backup_limit = clean_text_base - 1;
3803 for (scan_p = clean_text_base; scan_p < limit; scan_p++)
3805 if (*scan_p == ')')
3807 static const char *last_r_paren;
3808 const char *ahead_p;
3810 last_r_paren = scan_p;
3812 for (ahead_p = scan_p + 1; ISSPACE ((const unsigned char)*ahead_p); )
3813 check_source (++ahead_p < limit, limit);
3815 scan_p = ahead_p - 1;
3817 if (ISALPHA ((const unsigned char)*ahead_p) || *ahead_p == '{')
3819 const char *last_l_paren;
3820 const int lineno = identify_lineno (ahead_p);
3822 if (setjmp (source_confusion_recovery))
3823 continue;
3825 /* We know we have a function definition header. Now skip
3826 leftwards over all of its associated formals lists. */
3830 last_l_paren = careful_find_l_paren (last_r_paren);
3831 for (last_r_paren = last_l_paren-1;
3832 ISSPACE ((const unsigned char)*last_r_paren); )
3833 check_source (--last_r_paren >= backup_limit, backup_limit);
3835 while (*last_r_paren == ')');
3837 if (is_id_char (*last_r_paren))
3839 const char *id_limit = last_r_paren + 1;
3840 const char *id_start;
3841 size_t id_length;
3842 const def_dec_info *dd_p;
3844 for (id_start = id_limit-1; is_id_char (*id_start); )
3845 check_source (--id_start >= backup_limit, backup_limit);
3846 id_start++;
3847 backup_limit = id_start;
3848 if ((id_length = (size_t) (id_limit - id_start)) == 0)
3849 goto not_missed;
3852 char *func_name = alloca (id_length + 1);
3853 static const char * const stmt_keywords[]
3854 = { "if", "else", "do", "while", "for", "switch", "case", "return", 0 };
3855 const char * const *stmt_keyword;
3857 strncpy (func_name, id_start, id_length);
3858 func_name[id_length] = '\0';
3860 /* We must check here to see if we are actually looking at
3861 a statement rather than an actual function call. */
3863 for (stmt_keyword = stmt_keywords; *stmt_keyword; stmt_keyword++)
3864 if (!strcmp (func_name, *stmt_keyword))
3865 goto not_missed;
3867 #if 0
3868 notice ("%s: found definition of '%s' at %s(%d)\n",
3869 pname,
3870 func_name,
3871 shortpath (NULL, file_p->hash_entry->symbol),
3872 identify_lineno (id_start));
3873 #endif /* 0 */
3874 /* We really should check for a match of the function name
3875 here also, but why bother. */
3877 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3878 if (dd_p->is_func_def && dd_p->line == lineno)
3879 goto not_missed;
3881 /* If we make it here, then we did not know about this
3882 function definition. */
3884 notice ("%s: %d: warning: '%s' excluded by preprocessing\n",
3885 shortpath (NULL, file_p->hash_entry->symbol),
3886 identify_lineno (id_start), func_name);
3887 notice ("%s: function definition not converted\n",
3888 pname);
3890 not_missed: ;
3897 /* Do all editing operations for a single source file (either a "base" file
3898 or an "include" file). To do this we read the file into memory, keep a
3899 virgin copy there, make another cleaned in-core copy of the original file
3900 (i.e. one in which all of the comments and preprocessing directives have
3901 been replaced with whitespace), then use these two in-core copies of the
3902 file to make a new edited in-core copy of the file. Finally, rename the
3903 original file (as a way of saving it), and then write the edited version
3904 of the file from core to a disk file of the same name as the original.
3906 Note that the trick of making a copy of the original sans comments &
3907 preprocessing directives make the editing a whole lot easier. */
3909 static void
3910 edit_file (const hash_table_entry *hp)
3912 struct stat stat_buf;
3913 const file_info *file_p = hp->fip;
3914 char *new_orig_text_base;
3915 char *new_orig_text_limit;
3916 char *new_clean_text_base;
3917 char *new_clean_text_limit;
3918 size_t orig_size;
3919 size_t repl_size;
3920 int first_definition_in_file;
3922 /* If we are not supposed to be converting this file, or if there is
3923 nothing in there which needs converting, just skip this file. */
3925 if (!needs_to_be_converted (file_p))
3926 return;
3928 convert_filename = file_p->hash_entry->symbol;
3930 /* Convert a file if it is in a directory where we want conversion
3931 and the file is not excluded. */
3933 if (!directory_specified_p (convert_filename)
3934 || file_excluded_p (convert_filename))
3936 if (!quiet_flag
3937 #ifdef UNPROTOIZE
3938 /* Don't even mention "system" include files unless we are
3939 protoizing. If we are protoizing, we mention these as a
3940 gentle way of prodding the user to convert his "system"
3941 include files to prototype format. */
3942 && !in_system_include_dir (convert_filename)
3943 #endif /* defined (UNPROTOIZE) */
3945 notice ("%s: '%s' not converted\n",
3946 pname, shortpath (NULL, convert_filename));
3947 return;
3950 /* Let the user know what we are up to. */
3952 if (nochange_flag)
3953 notice ("%s: would convert file '%s'\n",
3954 pname, shortpath (NULL, convert_filename));
3955 else
3956 notice ("%s: converting file '%s'\n",
3957 pname, shortpath (NULL, convert_filename));
3958 fflush (stderr);
3960 /* Find out the size (in bytes) of the original file. */
3962 /* The cast avoids an erroneous warning on AIX. */
3963 if (stat (convert_filename, &stat_buf) == -1)
3965 int errno_val = errno;
3966 notice ("%s: can't get status for file '%s': %s\n",
3967 pname, shortpath (NULL, convert_filename),
3968 xstrerror (errno_val));
3969 return;
3971 orig_size = stat_buf.st_size;
3973 /* Allocate a buffer to hold the original text. */
3975 orig_text_base = new_orig_text_base = xmalloc (orig_size + 2);
3976 orig_text_limit = new_orig_text_limit = new_orig_text_base + orig_size;
3978 /* Allocate a buffer to hold the cleaned-up version of the original text. */
3980 clean_text_base = new_clean_text_base = xmalloc (orig_size + 2);
3981 clean_text_limit = new_clean_text_limit = new_clean_text_base + orig_size;
3982 clean_read_ptr = clean_text_base - 1;
3984 /* Allocate a buffer that will hopefully be large enough to hold the entire
3985 converted output text. As an initial guess for the maximum size of the
3986 output buffer, use 125% of the size of the original + some extra. This
3987 buffer can be expanded later as needed. */
3989 repl_size = orig_size + (orig_size >> 2) + 4096;
3990 repl_text_base = xmalloc (repl_size + 2);
3991 repl_text_limit = repl_text_base + repl_size - 1;
3992 repl_write_ptr = repl_text_base - 1;
3995 int input_file;
3996 int fd_flags;
3998 /* Open the file to be converted in READ ONLY mode. */
4000 fd_flags = O_RDONLY;
4001 #ifdef O_BINARY
4002 /* Use binary mode to avoid having to deal with different EOL characters. */
4003 fd_flags |= O_BINARY;
4004 #endif
4005 if ((input_file = open (convert_filename, fd_flags, 0444)) == -1)
4007 int errno_val = errno;
4008 notice ("%s: can't open file '%s' for reading: %s\n",
4009 pname, shortpath (NULL, convert_filename),
4010 xstrerror (errno_val));
4011 return;
4014 /* Read the entire original source text file into the original text buffer
4015 in one swell fwoop. Then figure out where the end of the text is and
4016 make sure that it ends with a newline followed by a null. */
4018 if (safe_read (input_file, new_orig_text_base, orig_size) !=
4019 (int) orig_size)
4021 int errno_val = errno;
4022 close (input_file);
4023 notice ("\n%s: error reading input file '%s': %s\n",
4024 pname, shortpath (NULL, convert_filename),
4025 xstrerror (errno_val));
4026 return;
4029 close (input_file);
4032 if (orig_size == 0 || orig_text_limit[-1] != '\n')
4034 *new_orig_text_limit++ = '\n';
4035 orig_text_limit++;
4038 /* Create the cleaned up copy of the original text. */
4040 memcpy (new_clean_text_base, orig_text_base,
4041 (size_t) (orig_text_limit - orig_text_base));
4042 do_cleaning (new_clean_text_base, new_clean_text_limit);
4044 #if 0
4046 int clean_file;
4047 size_t clean_size = orig_text_limit - orig_text_base;
4048 char *const clean_filename = alloca (strlen (convert_filename) + 6 + 1);
4050 /* Open (and create) the clean file. */
4052 strcpy (clean_filename, convert_filename);
4053 strcat (clean_filename, ".clean");
4054 if ((clean_file = creat (clean_filename, 0666)) == -1)
4056 int errno_val = errno;
4057 notice ("%s: can't create/open clean file '%s': %s\n",
4058 pname, shortpath (NULL, clean_filename),
4059 xstrerror (errno_val));
4060 return;
4063 /* Write the clean file. */
4065 safe_write (clean_file, new_clean_text_base, clean_size, clean_filename);
4067 close (clean_file);
4069 #endif /* 0 */
4071 /* Do a simplified scan of the input looking for things that were not
4072 mentioned in the aux info files because of the fact that they were
4073 in a region of the source which was preprocessed-out (via #if or
4074 via #ifdef). */
4076 scan_for_missed_items (file_p);
4078 /* Setup to do line-oriented forward seeking in the clean text buffer. */
4080 last_known_line_number = 1;
4081 last_known_line_start = clean_text_base;
4083 /* Now get down to business and make all of the necessary edits. */
4086 const def_dec_info *def_dec_p;
4088 first_definition_in_file = 1;
4089 def_dec_p = file_p->defs_decs;
4090 for (; def_dec_p; def_dec_p = def_dec_p->next_in_file)
4092 const char *clean_text_p = seek_to_line (def_dec_p->line);
4094 /* clean_text_p now points to the first character of the line which
4095 contains the `terminator' for the declaration or definition that
4096 we are about to process. */
4098 #ifndef UNPROTOIZE
4100 if (global_flag && def_dec_p->is_func_def && first_definition_in_file)
4102 add_global_decls (def_dec_p->file, clean_text_p);
4103 first_definition_in_file = 0;
4106 /* Don't edit this item if it is already in prototype format or if it
4107 is a function declaration and we have found no corresponding
4108 definition. */
4110 if (def_dec_p->prototyped
4111 || (!def_dec_p->is_func_def && !def_dec_p->definition))
4112 continue;
4114 #endif /* !defined (UNPROTOIZE) */
4116 if (def_dec_p->is_func_def)
4117 edit_fn_definition (def_dec_p, clean_text_p);
4118 else
4119 #ifndef UNPROTOIZE
4120 if (def_dec_p->is_implicit)
4121 add_local_decl (def_dec_p, clean_text_p);
4122 else
4123 #endif /* !defined (UNPROTOIZE) */
4124 edit_fn_declaration (def_dec_p, clean_text_p);
4128 /* Finalize things. Output the last trailing part of the original text. */
4130 output_up_to (clean_text_limit - 1);
4132 /* If this is just a test run, stop now and just deallocate the buffers. */
4134 if (nochange_flag)
4136 free (new_orig_text_base);
4137 free (new_clean_text_base);
4138 free (repl_text_base);
4139 return;
4142 /* Change the name of the original input file. This is just a quick way of
4143 saving the original file. */
4145 if (!nosave_flag)
4147 char *new_filename
4148 = xmalloc (strlen (convert_filename) + strlen (save_suffix) + 2);
4150 strcpy (new_filename, convert_filename);
4151 #ifdef __MSDOS__
4152 /* MSDOS filenames are restricted to 8.3 format, so we save `foo.c'
4153 as `foo.<save_suffix>'. */
4154 new_filename[(strlen (convert_filename) - 1] = '\0';
4155 #endif
4156 strcat (new_filename, save_suffix);
4158 /* Don't overwrite existing file. */
4159 if (access (new_filename, F_OK) == 0)
4161 if (!quiet_flag)
4162 notice ("%s: warning: file '%s' already saved in '%s'\n",
4163 pname,
4164 shortpath (NULL, convert_filename),
4165 shortpath (NULL, new_filename));
4167 else if (rename (convert_filename, new_filename) == -1)
4169 int errno_val = errno;
4170 notice ("%s: can't link file '%s' to '%s': %s\n",
4171 pname,
4172 shortpath (NULL, convert_filename),
4173 shortpath (NULL, new_filename),
4174 xstrerror (errno_val));
4175 return;
4179 if (unlink (convert_filename) == -1)
4181 int errno_val = errno;
4182 /* The file may have already been renamed. */
4183 if (errno_val != ENOENT)
4185 notice ("%s: can't delete file '%s': %s\n",
4186 pname, shortpath (NULL, convert_filename),
4187 xstrerror (errno_val));
4188 return;
4193 int output_file;
4195 /* Open (and create) the output file. */
4197 if ((output_file = creat (convert_filename, 0666)) == -1)
4199 int errno_val = errno;
4200 notice ("%s: can't create/open output file '%s': %s\n",
4201 pname, shortpath (NULL, convert_filename),
4202 xstrerror (errno_val));
4203 return;
4205 #ifdef O_BINARY
4206 /* Use binary mode to avoid changing the existing EOL character. */
4207 setmode (output_file, O_BINARY);
4208 #endif
4210 /* Write the output file. */
4213 unsigned int out_size = (repl_write_ptr + 1) - repl_text_base;
4215 safe_write (output_file, repl_text_base, out_size, convert_filename);
4218 close (output_file);
4221 /* Deallocate the conversion buffers. */
4223 free (new_orig_text_base);
4224 free (new_clean_text_base);
4225 free (repl_text_base);
4227 /* Change the mode of the output file to match the original file. */
4229 /* The cast avoids an erroneous warning on AIX. */
4230 if (chmod (convert_filename, stat_buf.st_mode) == -1)
4232 int errno_val = errno;
4233 notice ("%s: can't change mode of file '%s': %s\n",
4234 pname, shortpath (NULL, convert_filename),
4235 xstrerror (errno_val));
4238 /* Note: We would try to change the owner and group of the output file
4239 to match those of the input file here, except that may not be a good
4240 thing to do because it might be misleading. Also, it might not even
4241 be possible to do that (on BSD systems with quotas for instance). */
4244 /* Do all of the individual steps needed to do the protoization (or
4245 unprotoization) of the files referenced in the aux_info files given
4246 in the command line. */
4248 static void
4249 do_processing (void)
4251 const char * const *base_pp;
4252 const char * const * const end_pps
4253 = &base_source_filenames[n_base_source_files];
4255 #ifndef UNPROTOIZE
4256 int syscalls_len;
4257 #endif /* !defined (UNPROTOIZE) */
4259 /* One-by-one, check (and create if necessary), open, and read all of the
4260 stuff in each aux_info file. After reading each aux_info file, the
4261 aux_info_file just read will be automatically deleted unless the
4262 keep_flag is set. */
4264 for (base_pp = base_source_filenames; base_pp < end_pps; base_pp++)
4265 process_aux_info_file (*base_pp, keep_flag, 0);
4267 #ifndef UNPROTOIZE
4269 /* Also open and read the special SYSCALLS.c aux_info file which gives us
4270 the prototypes for all of the standard system-supplied functions. */
4272 if (nondefault_syscalls_dir)
4274 syscalls_absolute_filename
4275 = xmalloc (strlen (nondefault_syscalls_dir) + 1
4276 + sizeof (syscalls_filename));
4277 strcpy (syscalls_absolute_filename, nondefault_syscalls_dir);
4279 else
4281 GET_ENVIRONMENT (default_syscalls_dir, "GCC_EXEC_PREFIX");
4282 if (!default_syscalls_dir)
4284 default_syscalls_dir = standard_exec_prefix;
4286 syscalls_absolute_filename
4287 = xmalloc (strlen (default_syscalls_dir) + 0
4288 + strlen (target_machine) + 1
4289 + strlen (target_version) + 1
4290 + sizeof (syscalls_filename));
4291 strcpy (syscalls_absolute_filename, default_syscalls_dir);
4292 strcat (syscalls_absolute_filename, target_machine);
4293 strcat (syscalls_absolute_filename, "/");
4294 strcat (syscalls_absolute_filename, target_version);
4295 strcat (syscalls_absolute_filename, "/");
4298 syscalls_len = strlen (syscalls_absolute_filename);
4299 if (! IS_DIR_SEPARATOR (*(syscalls_absolute_filename + syscalls_len - 1)))
4301 *(syscalls_absolute_filename + syscalls_len++) = DIR_SEPARATOR;
4302 *(syscalls_absolute_filename + syscalls_len) = '\0';
4304 strcat (syscalls_absolute_filename, syscalls_filename);
4306 /* Call process_aux_info_file in such a way that it does not try to
4307 delete the SYSCALLS aux_info file. */
4309 process_aux_info_file (syscalls_absolute_filename, 1, 1);
4311 #endif /* !defined (UNPROTOIZE) */
4313 /* When we first read in all of the information from the aux_info files
4314 we saved in it descending line number order, because that was likely to
4315 be faster. Now however, we want the chains of def & dec records to
4316 appear in ascending line number order as we get further away from the
4317 file_info record that they hang from. The following line causes all of
4318 these lists to be rearranged into ascending line number order. */
4320 visit_each_hash_node (filename_primary, reverse_def_dec_list);
4322 #ifndef UNPROTOIZE
4324 /* Now do the "real" work. The following line causes each declaration record
4325 to be "visited". For each of these nodes, an attempt is made to match
4326 up the function declaration with a corresponding function definition,
4327 which should have a full prototype-format formals list with it. Once
4328 these match-ups are made, the conversion of the function declarations
4329 to prototype format can be made. */
4331 visit_each_hash_node (function_name_primary, connect_defs_and_decs);
4333 #endif /* !defined (UNPROTOIZE) */
4335 /* Now convert each file that can be converted (and needs to be). */
4337 visit_each_hash_node (filename_primary, edit_file);
4339 #ifndef UNPROTOIZE
4341 /* If we are working in cplusplus mode, try to rename all .c files to .C
4342 files. Don't panic if some of the renames don't work. */
4344 if (cplusplus_flag && !nochange_flag)
4345 visit_each_hash_node (filename_primary, rename_c_file);
4347 #endif /* !defined (UNPROTOIZE) */
4350 static const struct option longopts[] =
4352 {"version", 0, 0, 'V'},
4353 {"file_name", 0, 0, 'p'},
4354 {"quiet", 0, 0, 'q'},
4355 {"silent", 0, 0, 'q'},
4356 {"force", 0, 0, 'f'},
4357 {"keep", 0, 0, 'k'},
4358 {"nosave", 0, 0, 'N'},
4359 {"nochange", 0, 0, 'n'},
4360 {"compiler-options", 1, 0, 'c'},
4361 {"exclude", 1, 0, 'x'},
4362 {"directory", 1, 0, 'd'},
4363 #ifdef UNPROTOIZE
4364 {"indent", 1, 0, 'i'},
4365 #else
4366 {"local", 0, 0, 'l'},
4367 {"global", 0, 0, 'g'},
4368 {"c++", 0, 0, 'C'},
4369 {"syscalls-dir", 1, 0, 'B'},
4370 #endif
4371 {0, 0, 0, 0}
4374 extern int main (int, char **const);
4377 main (int argc, char **const argv)
4379 int longind;
4380 int c;
4381 const char *params = "";
4383 pname = strrchr (argv[0], DIR_SEPARATOR);
4384 #ifdef DIR_SEPARATOR_2
4386 char *slash;
4388 slash = strrchr (pname ? pname : argv[0], DIR_SEPARATOR_2);
4389 if (slash)
4390 pname = slash;
4392 #endif
4393 pname = pname ? pname+1 : argv[0];
4395 #ifdef SIGCHLD
4396 /* We *MUST* set SIGCHLD to SIG_DFL so that the wait4() call will
4397 receive the signal. A different setting is inheritable */
4398 signal (SIGCHLD, SIG_DFL);
4399 #endif
4401 gcc_init_libintl ();
4403 cwd_buffer = getpwd ();
4404 if (!cwd_buffer)
4406 notice ("%s: cannot get working directory: %s\n",
4407 pname, xstrerror(errno));
4408 return (FATAL_EXIT_CODE);
4411 /* By default, convert the files in the current directory. */
4412 directory_list = string_list_cons (cwd_buffer, NULL);
4414 while ((c = getopt_long (argc, argv,
4415 #ifdef UNPROTOIZE
4416 "c:d:i:knNp:qvVx:",
4417 #else
4418 "B:c:Cd:gklnNp:qvVx:",
4419 #endif
4420 longopts, &longind)) != EOF)
4422 if (c == 0) /* Long option. */
4423 c = longopts[longind].val;
4424 switch (c)
4426 case 'p':
4427 compiler_file_name = optarg;
4428 break;
4429 case 'd':
4430 directory_list
4431 = string_list_cons (abspath (NULL, optarg), directory_list);
4432 break;
4433 case 'x':
4434 exclude_list = string_list_cons (optarg, exclude_list);
4435 break;
4437 case 'v':
4438 case 'V':
4439 version_flag = 1;
4440 break;
4441 case 'q':
4442 quiet_flag = 1;
4443 break;
4444 #if 0
4445 case 'f':
4446 force_flag = 1;
4447 break;
4448 #endif
4449 case 'n':
4450 nochange_flag = 1;
4451 keep_flag = 1;
4452 break;
4453 case 'N':
4454 nosave_flag = 1;
4455 break;
4456 case 'k':
4457 keep_flag = 1;
4458 break;
4459 case 'c':
4460 params = optarg;
4461 break;
4462 #ifdef UNPROTOIZE
4463 case 'i':
4464 indent_string = optarg;
4465 break;
4466 #else /* !defined (UNPROTOIZE) */
4467 case 'l':
4468 local_flag = 1;
4469 break;
4470 case 'g':
4471 global_flag = 1;
4472 break;
4473 case 'C':
4474 cplusplus_flag = 1;
4475 break;
4476 case 'B':
4477 nondefault_syscalls_dir = optarg;
4478 break;
4479 #endif /* !defined (UNPROTOIZE) */
4480 default:
4481 usage ();
4485 /* Set up compile_params based on -p and -c options. */
4486 munge_compile_params (params);
4488 n_base_source_files = argc - optind;
4490 /* Now actually make a list of the base source filenames. */
4492 base_source_filenames
4493 = xmalloc ((n_base_source_files + 1) * sizeof (char *));
4494 n_base_source_files = 0;
4495 for (; optind < argc; optind++)
4497 const char *path = abspath (NULL, argv[optind]);
4498 int len = strlen (path);
4500 if (path[len-1] == 'c' && path[len-2] == '.')
4501 base_source_filenames[n_base_source_files++] = path;
4502 else
4504 notice ("%s: input file names must have .c suffixes: %s\n",
4505 pname, shortpath (NULL, path));
4506 errors++;
4510 #ifndef UNPROTOIZE
4511 /* We are only interested in the very first identifier token in the
4512 definition of `va_list', so if there is more junk after that first
4513 identifier token, delete it from the `varargs_style_indicator'. */
4515 const char *cp;
4517 for (cp = varargs_style_indicator; ISIDNUM (*cp); cp++)
4518 continue;
4519 if (*cp != 0)
4520 varargs_style_indicator = savestring (varargs_style_indicator,
4521 cp - varargs_style_indicator);
4523 #endif /* !defined (UNPROTOIZE) */
4525 if (errors)
4526 usage ();
4527 else
4529 if (version_flag)
4530 fprintf (stderr, "%s: %s\n", pname, version_string);
4531 do_processing ();
4534 return (errors ? FATAL_EXIT_CODE : SUCCESS_EXIT_CODE);