1 /* Protoize program - Original version by Ron Guilmette at MCC.
3 Copyright (C) 1989, 1992 Free Software Foundation, Inc.
5 This file is part of GNU CC.
7 GNU CC is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 GNU CC is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GNU CC; see the file COPYING. If not, write to
19 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */
21 /* Any reasonable C++ compiler should have all of the same features
22 as __STDC__ plus more, so make sure that __STDC__ is defined if
23 __cplusplus is defined. */
25 #if defined(__cplusplus) && !defined(__STDC__)
27 #endif /* defined(__cplusplus) && !defined(__STDC__) */
29 #if defined(__GNUC__) || defined (__GNUG__)
30 #define VOLATILE volatile
41 #ifdef POSIX /* We should be able to define _POSIX_SOURCE unconditionally,
42 but some systems respond in buggy ways to it,
43 including Sunos 4.1.1. Which we don't classify as POSIX. */
44 /* In case this is a POSIX system with an ANSI C compiler,
45 ask for definition of all POSIX facilities. */
50 /* Users are not supposed to use _POSIX_SOURCE to say the
51 system is a POSIX system. That is not what _POSIX_SOURCE means! -- rms */
52 /* If the user asked for POSIX via _POSIX_SOURCE, turn on POSIX code. */
53 #if defined(_POSIX_SOURCE) && !defined(POSIX)
61 #include <sys/types.h>
64 #if ! defined (USG) || defined (SVR4)
70 /* Include getopt.h for the sake of getopt_long.
71 We don't need the declaration of getopt, and it could conflict
72 with something from a system header file, so effectively nullify that. */
73 #define getopt getopt_loser
77 extern char *sys_errlist[];
78 extern char *version_string;
80 /* Systems which are compatible only with POSIX 1003.1-1988 (but *not*
81 with POSIX 1003.1-1990), e.g. Ultrix 4.2, might not have
82 const qualifiers in the prototypes in the system include files.
83 Unfortunately, this can lead to GCC issuing lots of warnings for
84 calls to the following functions. To eliminate these warnings we
85 provide the following #defines. */
87 #define my_access(file,flag) access((char *)file, flag)
88 #define my_stat(file,pkt) stat((char *)file, pkt)
89 #define my_execvp(prog,argv) execvp((char *)prog, (char **)argv)
90 #define my_link(file1, file2) link((char *)file1, (char *)file2)
91 #define my_unlink(file) unlink((char *)file)
92 #define my_open(file, mode, flag) open((char *)file, mode, flag)
93 #define my_chmod(file, mode) chmod((char *)file, mode)
97 /* Aliases for pointers to void.
98 These were made to facilitate compilation with other compilers. */
101 typedef void * pointer_type;
102 typedef const void * const_pointer_type;
104 typedef char * pointer_type;
105 typedef char * const_pointer_type;
116 #else /* !defined(POSIX) */
118 #define R_OK 4 /* Test for Read permission */
119 #define W_OK 2 /* Test for Write permission */
120 #define X_OK 1 /* Test for eXecute permission */
121 #define F_OK 0 /* Test for existence of File */
126 /* Declaring stat or __flsbuf with a prototype
127 causes conflicts with system headers on some systems. */
130 extern VOLATILE void abort ();
134 #if 0 /* These conflict with stdio.h on some systems. */
135 extern int fprintf (FILE *, const char *, ...);
136 extern int printf (const char *, ...);
139 extern pointer_type malloc ();
140 extern pointer_type realloc ();
145 extern int fflush ();
150 #if 0 /* Causes trouble on some systems that define setjmp as a macro. */
151 extern int setjmp ();
152 extern void longjmp ();
155 extern char * strcat ();
156 extern int strcmp ();
157 extern char * strcpy ();
158 #if 0 /* size_t from sys/types.h may fail to match GCC.
159 If so, we would get a warning from this.
160 So do without the prototype. */
161 extern size_t strlen (const char *);
163 extern int strncmp ();
164 extern char * strncpy ();
165 extern char * strrchr ();
167 /* Fork is not declared because the declaration caused a conflict
169 #if !(defined (USG) || defined (VMS))
171 #endif /* (defined (USG) || defined (VMS)) */
173 #endif /* !defined (POSIX) */
175 /* Look for these where the `const' qualifier is intentionally cast aside. */
179 /* Define a STRINGIFY macro that's right for ANSI or traditional C. */
182 #define STRINGIFY(STRING) #STRING
184 #define STRINGIFY(STRING) "STRING"
187 /* POSIX systems will not have definitions for WIFEXITED or WEXITSTATUS.
188 Define them correctly and so that they work for all environments. */
191 #define WIFEXITED(status_word) ((*((int *)&status_word) & 0xff) == 0x00)
194 #define WEXITSTATUS(status_word) ((*((int *)&status_word) & 0xff00) >> 8)
196 /* Define a default place to find the SYSCALLS.X file. */
198 #ifndef STD_PROTO_DIR
199 #define STD_PROTO_DIR "/usr/local/lib"
200 #endif /* !defined (STD_PROTO_DIR) */
202 /* Suffix of aux_info files. */
204 static const char * const aux_info_suffix = ".X";
206 /* String to attach to filenames for saved versions of original files. */
208 static const char * const save_suffix = ".save";
212 /* File name of the file which contains descriptions of standard system
213 routines. Note that we never actually do anything with this file per se,
214 but we do read in its corresponding aux_info file. */
216 static const char syscalls_filename[] = "SYSCALLS.c";
218 /* Default place to find the above file. */
220 static const char * const default_syscalls_dir = STD_PROTO_DIR;
222 /* Variable to hold the complete absolutized filename of the SYSCALLS.c.X
225 static char * syscalls_absolute_filename;
227 #endif /* !defined (UNPROTOIZE) */
229 /* Type of the structure that holds information about macro unexpansions. */
231 struct unexpansion_struct {
232 const char *expanded;
233 const char *contracted;
235 typedef struct unexpansion_struct unexpansion;
237 /* A table of conversions that may need to be made for some (stupid) older
238 operating systems where these types are preprocessor macros rather than
239 typedefs (as they really ought to be).
241 WARNING: The contracted forms must be as small (or smaller) as the
242 expanded forms, or else havoc will ensue. */
244 static const unexpansion unexpansions[] = {
245 { "struct _iobuf", "FILE" },
249 /* The number of "primary" slots in the hash tables for filenames and for
250 function names. This can be as big or as small as you like, except that
251 it must be a power of two. */
253 #define HASH_TABLE_SIZE (1 << 9)
255 /* Bit mask to use when computing hash values. */
257 static const int hash_mask = (HASH_TABLE_SIZE - 1);
259 /* Make a table of default system include directories
260 just as it is done in cccp.c. */
262 #ifndef STANDARD_INCLUDE_DIR
263 #define STANDARD_INCLUDE_DIR "/usr/include"
266 #ifndef LOCAL_INCLUDE_DIR
267 #define LOCAL_INCLUDE_DIR "/usr/local/include"
270 struct default_include { char *fname; int cplusplus; } include_defaults[]
271 #ifdef INCLUDE_DEFAULTS
275 /* Pick up GNU C++ specific include files. */
276 { GPLUSPLUS_INCLUDE_DIR, 1},
277 { GCC_INCLUDE_DIR, 0},
279 /* For cross-compilation, this dir name is generated
280 automatically in Makefile.in. */
281 { CROSS_INCLUDE_DIR, 0 },
282 #else /* not CROSS_COMPILE */
283 { LOCAL_INCLUDE_DIR, 0},
284 /* Some systems have an extra dir of include files. */
285 #ifdef SYSTEM_INCLUDE_DIR
286 { SYSTEM_INCLUDE_DIR, 0},
288 { STANDARD_INCLUDE_DIR, 0},
289 #endif /* not CROSS_COMPILE */
292 #endif /* no INCLUDE_DEFAULTS */
294 /* Datatype for lists of directories or filenames. */
298 struct string_list *next;
301 /* List of directories in which files should be converted. */
303 struct string_list *directory_list;
305 /* List of file names which should not be converted.
306 A file is excluded if the end of its name, following a /,
307 matches one of the names in this list. */
309 struct string_list *exclude_list;
311 /* The name of the other style of variable-number-of-parameters functions
312 (i.e. the style that we want to leave unconverted because we don't yet
313 know how to convert them to this style. This string is used in warning
316 /* Also define here the string that we can search for in the parameter lists
317 taken from the .X files which will unambiguously indicate that we have
318 found a varargs style function. */
321 static const char * const other_var_style = "stdarg";
322 #else /* !defined (UNPROTOIZE) */
323 static const char * const other_var_style = "varargs";
324 /* Note that this is a string containing the expansion of va_alist.
325 But in `main' we discard all but the first token. */
326 static const char *varargs_style_indicator = STRINGIFY (va_alist);
327 #endif /* !defined (UNPROTOIZE) */
329 /* The following two types are used to create hash tables. In this program,
330 there are two hash tables which are used to store and quickly lookup two
331 different classes of strings. The first type of strings stored in the
332 first hash table are absolute filenames of files which protoize needs to
333 know about. The second type of strings (stored in the second hash table)
334 are function names. It is this second class of strings which really
335 inspired the use of the hash tables, because there may be a lot of them. */
337 typedef struct hash_table_entry_struct hash_table_entry;
339 /* Do some typedefs so that we don't have to write "struct" so often. */
341 typedef struct def_dec_info_struct def_dec_info;
342 typedef struct file_info_struct file_info;
343 typedef struct f_list_chain_item_struct f_list_chain_item;
345 /* In the struct below, note that the "_info" field has two different uses
346 depending on the type of hash table we are in (i.e. either the filenames
347 hash table or the function names hash table). In the filenames hash table
348 the info fields of the entries point to the file_info struct which is
349 associated with each filename (1 per filename). In the function names
350 hash table, the info field points to the head of a singly linked list of
351 def_dec_info entries which are all defs or decs of the function whose
352 name is pointed to by the "symbol" field. Keeping all of the defs/decs
353 for a given function name on a special list specifically for that function
354 name makes it quick and easy to find out all of the important information
355 about a given (named) function. */
357 struct hash_table_entry_struct {
358 hash_table_entry * hash_next; /* -> to secondary entries */
359 const char * symbol; /* -> to the hashed string */
361 const def_dec_info * _ddip;
365 #define ddip _info._ddip
366 #define fip _info._fip
368 /* Define a type specifically for our two hash tables. */
370 typedef hash_table_entry hash_table[HASH_TABLE_SIZE];
372 /* The following struct holds all of the important information about any
373 single filename (e.g. file) which we need to know about. */
375 struct file_info_struct {
376 const hash_table_entry * hash_entry; /* -> to associated hash entry */
377 const def_dec_info * defs_decs; /* -> to chain of defs/decs */
378 time_t mtime; /* Time of last modification. */
381 /* Due to the possibility that functions may return pointers to functions,
382 (which may themselves have their own parameter lists) and due to the
383 fact that returned pointers-to-functions may be of type "pointer-to-
384 function-returning-pointer-to-function" (ad nauseum) we have to keep
385 an entire chain of ANSI style formal parameter lists for each function.
387 Normally, for any given function, there will only be one formals list
388 on the chain, but you never know.
390 Note that the head of each chain of formals lists is pointed to by the
391 `f_list_chain' field of the corresponding def_dec_info record.
393 For any given chain, the item at the head of the chain is the *leftmost*
394 parameter list seen in the actual C language function declaration. If
395 there are other members of the chain, then these are linked in left-to-right
396 order from the head of the chain. */
398 struct f_list_chain_item_struct {
399 const f_list_chain_item * chain_next; /* -> to next item on chain */
400 const char * formals_list; /* -> to formals list string */
403 /* The following struct holds all of the important information about any
404 single function definition or declaration which we need to know about.
405 Note that for unprotoize we don't need to know very much because we
406 never even create records for stuff that we don't intend to convert
407 (like for instance defs and decs which are already in old K&R format
408 and "implicit" function declarations). */
410 struct def_dec_info_struct {
411 const def_dec_info * next_in_file; /* -> to rest of chain for file */
412 file_info * file; /* -> file_info for containing file */
413 int line; /* source line number of def/dec */
414 const char * ansi_decl; /* -> left end of ansi decl */
415 hash_table_entry * hash_entry; /* -> hash entry for function name */
416 unsigned int is_func_def; /* = 0 means this is a declaration */
417 const def_dec_info * next_for_func; /* -> to rest of chain for func name */
418 unsigned int f_list_count; /* count of formals lists we expect */
419 char prototyped; /* = 0 means already prototyped */
421 const f_list_chain_item * f_list_chain; /* -> chain of formals lists */
422 const def_dec_info * definition; /* -> def/dec containing related def */
423 char is_static; /* = 0 means visiblilty is "extern" */
424 char is_implicit; /* != 0 for implicit func decl's */
425 char written; /* != 0 means written for implicit */
426 #else /* !defined (UNPROTOIZE) */
427 const char * formal_names; /* -> to list of names of formals */
428 const char * formal_decls; /* -> to string of formal declartions */
429 #endif /* !defined (UNPROTOIZE) */
432 /* Pointer to the tail component of the filename by which this program was
433 invoked. Used everywhere in error and warning messages. */
435 static const char *pname;
437 /* Error counter. Will be non-zero if we should give up at the next convenient
440 static int errors = 0;
443 /* ??? These comments should say what the flag mean as well as the options
446 /* File name to use for running gcc. Allows GCC 2 to be named
447 something other than gcc. */
448 static char *compiler_file_name = "gcc";
450 static int version_flag = 0; /* Print our version number. */
451 static int quiet_flag = 0; /* Don't print messages normally. */
452 static int nochange_flag = 0; /* Don't convert, just say what files
453 we would have converted. */
454 static int nosave_flag = 0; /* Don't save the old version. */
455 static int keep_flag = 0; /* Don't delete the .X files. */
456 static const char ** compile_params = 0; /* Option string for gcc. */
458 static const char *indent_string = " "; /* Indentation for newly
459 inserted parm decls. */
460 #else /* !defined (UNPROTOIZE) */
461 static int local_flag = 0; /* Insert new local decls (when?). */
462 static int global_flag = 0; /* set by -g option */
463 static int cplusplus_flag = 0; /* Rename converted files to *.C. */
464 static const char* nondefault_syscalls_dir = 0; /* Dir to look for
466 #endif /* !defined (UNPROTOIZE) */
468 /* An index into the compile_params array where we should insert the source
469 file name when we are ready to exec the C compiler. A zero value indicates
470 that we have not yet called munge_compile_params. */
472 static int input_file_name_index = 0;
474 /* An index into the compile_params array where we should insert the filename
475 for the aux info file, when we run the C compiler. */
476 static int aux_info_file_name_index = 0;
478 /* Count of command line arguments which were "filename" arguments. */
480 static int n_base_source_files = 0;
482 /* Points to a malloc'ed list of pointers to all of the filenames of base
483 source files which were specified on the command line. */
485 static const char **base_source_filenames;
487 /* Line number of the line within the current aux_info file that we
488 are currently processing. Used for error messages in case the prototypes
489 info file is corrupted somehow. */
491 static int current_aux_info_lineno;
493 /* Pointer to the name of the source file currently being converted. */
495 static const char *convert_filename;
497 /* Pointer to relative root string (taken from aux_info file) which indicates
498 where directory the user was in when he did the compilation step that
499 produced the containing aux_info file. */
501 static const char *invocation_filename;
503 /* Pointer to the base of the input buffer that holds the original text for the
504 source file currently being converted. */
506 static const char *orig_text_base;
508 /* Pointer to the byte just beyond the end of the input buffer that holds the
509 original text for the source file currently being converted. */
511 static const char *orig_text_limit;
513 /* Pointer to the base of the input buffer that holds the cleaned text for the
514 source file currently being converted. */
516 static const char *clean_text_base;
518 /* Pointer to the byte just beyond the end of the input buffer that holds the
519 cleaned text for the source file currently being converted. */
521 static const char *clean_text_limit;
523 /* Pointer to the last byte in the cleaned text buffer that we have already
524 (virtually) copied to the output buffer (or decided to ignore). */
526 static const char * clean_read_ptr;
528 /* Pointer to the base of the output buffer that holds the replacement text
529 for the source file currently being converted. */
531 static char *repl_text_base;
533 /* Pointer to the byte just beyond the end of the output buffer that holds the
534 replacement text for the source file currently being converted. */
536 static char *repl_text_limit;
538 /* Pointer to the last byte which has been stored into the output buffer.
539 The next byte to be stored should be stored just past where this points
542 static char * repl_write_ptr;
544 /* Pointer into the cleaned text buffer for the source file we are currently
545 converting. This points to the first character of the line that we last
546 did a "seek_to_line" to (see below). */
548 static const char *last_known_line_start;
550 /* Number of the line (in the cleaned text buffer) that we last did a
551 "seek_to_line" to. Will be one if we just read a new source file
552 into the cleaned text buffer. */
554 static int last_known_line_number;
556 /* The filenames hash table. */
558 static hash_table filename_primary;
560 /* The function names hash table. */
562 static hash_table function_name_primary;
564 /* The place to keep the recovery address which is used only in cases where
565 we get hopelessly confused by something in the cleaned original text. */
567 static jmp_buf source_confusion_recovery;
569 /* A pointer to the current directory filename (used by abspath). */
571 static char *cwd_buffer;
573 /* A place to save the read pointer until we are sure that an individual
574 attempt at editing will succeed. */
576 static const char * saved_clean_read_ptr;
578 /* A place to save the write pointer until we are sure that an individual
579 attempt at editing will succeed. */
581 static char * saved_repl_write_ptr;
583 /* Forward declaration. */
585 static const char *shortpath ();
587 /* Allocate some space, but check that the allocation was successful. */
588 /* alloca.c uses this, so don't make it static. */
596 rv = malloc (byte_count);
599 fprintf (stderr, "\n%s: fatal error: can't allocate %u more bytes of memory\n",
602 return 0; /* avoid warnings */
608 /* Reallocate some space, but check that the reallocation was successful. */
611 xrealloc (old_space, byte_count)
612 pointer_type old_space;
617 rv = realloc (old_space, byte_count);
620 fprintf (stderr, "\n%s: fatal error: can't allocate %u more bytes of memory\n",
623 return 0; /* avoid warnings */
629 /* Deallocate the area pointed to by an arbitrary pointer, but first, strip
630 the `const' qualifier from it and also make sure that the pointer value
635 const_pointer_type p;
638 free ((NONCONST pointer_type) p);
641 /* Make a copy of a string INPUT with size SIZE. */
644 savestring (input, size)
648 char *output = (char *) xmalloc (size + 1);
649 strcpy (output, input);
653 /* More 'friendly' abort that prints the line and file.
654 config.h can #define abort fancy_abort if you like that sort of thing. */
659 fprintf (stderr, "%s: internal abort\n", pname);
663 /* Make a duplicate of a given string in a newly allocated area. */
669 return strcpy ((char *) xmalloc (strlen (s) + 1), s);
672 /* Make a duplicate of the first N bytes of a given string in a newly
680 char *ret_val = strncpy ((char *) xmalloc (n + 1), s, n);
686 /* Return a pointer to the first occurance of s2 within s1 or NULL if s2
687 does not occur within s1. Assume neither s1 nor s2 are null pointers. */
692 const char *const s2;
700 for (p1 = s1, p2 = s2; c = *p2; p1++, p2++)
710 /* Get setup to recover in case the edit we are about to do goes awry. */
715 saved_clean_read_ptr = clean_read_ptr;
716 saved_repl_write_ptr = repl_write_ptr;
719 /* Call this routine to recover our previous state whenever something looks
720 too confusing in the source code we are trying to edit. */
725 clean_read_ptr = saved_clean_read_ptr;
726 repl_write_ptr = saved_repl_write_ptr;
729 /* Return true if the given character is a legal identifier character. */
735 return (isalnum (ch) || (ch == '_') || (ch == '$'));
738 /* Give a message indicating the proper way to invoke this program and then
739 exit with non-zero status. */
745 fprintf (stderr, "%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ filename ... ]'\n",
747 #else /* !defined (UNPROTOIZE) */
748 fprintf (stderr, "%s: usage '%s [ -VqfnkNlgC ] [ -B <diname> ] [ filename ... ]'\n",
750 #endif /* !defined (UNPROTOIZE) */
754 /* Return true if the given filename (assumed to be an absolute filename)
755 designates a file residing anywhere beneath any one of the "system"
756 include directories. */
759 in_system_include_dir (path)
762 struct default_include *p;
765 abort (); /* Must be an absolutized filename. */
767 for (p = include_defaults; p->fname; p++)
768 if (!strncmp (path, p->fname, strlen (p->fname))
769 && path[strlen (p->fname)] == '/')
775 /* Return true if the given filename designates a file that the user has
776 read access to and for which the user has write access to the containing
780 file_could_be_converted (const char *path)
782 char *const dir_name = (char *) alloca (strlen (path) + 1);
784 if (my_access (path, R_OK))
788 char *dir_last_slash;
790 strcpy (dir_name, path);
791 dir_last_slash = strrchr (dir_name, '/');
793 *dir_last_slash = '\0';
795 abort (); /* Should have been an absolutized filename. */
798 if (my_access (path, W_OK))
804 /* Return true if the given filename designates a file that we are allowed
805 to modify. Files which we should not attempt to modify are (a) "system"
806 include files, and (b) files which the user doesn't have write access to,
807 and (c) files which reside in directories which the user doesn't have
808 write access to. Unless requested to be quiet, give warnings about
809 files that we will not try to convert for one reason or another. An
810 exception is made for "system" include files, which we never try to
811 convert and for which we don't issue the usual warnings. */
814 file_normally_convertable (const char *path)
816 char *const dir_name = alloca (strlen (path) + 1);
818 if (in_system_include_dir (path))
822 char *dir_last_slash;
824 strcpy (dir_name, path);
825 dir_last_slash = strrchr (dir_name, '/');
827 *dir_last_slash = '\0';
829 abort (); /* Should have been an absolutized filename. */
832 if (my_access (path, R_OK))
835 fprintf (stderr, "%s: warning: no read access for file `%s'\n",
836 pname, shortpath (NULL, path));
840 if (my_access (path, W_OK))
843 fprintf (stderr, "%s: warning: no write access for file `%s'\n",
844 pname, shortpath (NULL, path));
848 if (my_access (dir_name, W_OK))
851 fprintf (stderr, "%s: warning: no write access for dir containing `%s'\n",
852 pname, shortpath (NULL, path));
862 /* Return true if the given file_info struct refers to the special SYSCALLS.c.X
863 file. Return false otherwise. */
866 is_syscalls_file (fi_p)
867 const file_info *fi_p;
869 char const *f = fi_p->hash_entry->symbol;
870 size_t fl = strlen (f), sysl = sizeof (syscalls_filename) - 1;
871 return sysl <= fl && strcmp (f + fl - sysl, syscalls_filename) == 0;
874 #endif /* !defined (UNPROTOIZE) */
876 /* Check to see if this file will need to have anything done to it on this
877 run. If there is nothing in the given file which both needs conversion
878 and for which we have the necessary stuff to do the conversion, return
879 false. Otherwise, return true.
881 Note that (for protoize) it is only valid to call this function *after*
882 the connections between declarations and definitions have all been made
883 by connect_defs_and_decs. */
886 needs_to_be_converted (file_p)
887 const file_info *file_p;
889 const def_dec_info *ddp;
893 if (is_syscalls_file (file_p))
896 #endif /* !defined (UNPROTOIZE) */
898 for (ddp = file_p->defs_decs; ddp; ddp = ddp->next_in_file)
904 /* ... and if we a protoizing and this function is in old style ... */
906 /* ... and if this a definition or is a decl with an associated def ... */
907 && (ddp->is_func_def || (!ddp->is_func_def && ddp->definition))
909 #else /* defined (UNPROTOIZE) */
911 /* ... and if we are unprotoizing and this function is in new style ... */
914 #endif /* defined (UNPROTOIZE) */
916 /* ... then the containing file needs converting. */
921 /* Return 1 if the file name NAME is in a directory
922 that should be converted. */
925 directory_specified_p (name)
928 struct string_list *p;
930 for (p = directory_list; p; p = p->next)
931 if (!strncmp (name, p->name, strlen (p->name))
932 && name[strlen (p->name)] == '/')
934 const char *q = name + strlen (p->name) + 1;
936 /* If there are more slashes, it's in a subdir, so
937 this match doesn't count. */
949 /* Return 1 if the file named NAME should be excluded from conversion. */
952 file_excluded_p (name)
955 struct string_list *p;
956 int len = strlen (name);
958 for (p = exclude_list; p; p = p->next)
959 if (!strcmp (name + len - strlen (p->name), p->name)
960 && name[len - strlen (p->name) - 1] == '/')
966 /* Construct a new element of a string_list.
967 STRING is the new element value, and REST holds the remaining elements. */
969 static struct string_list *
970 string_list_cons (string, rest)
972 struct string_list *rest;
974 struct string_list *temp
975 = (struct string_list *) xmalloc (sizeof (struct string_list));
982 /* ??? The GNU convention for mentioning function args in its comments
983 is to capitalize them. So change "hash_tab_p" to HASH_TAB_P below.
984 Likewise for all the other functions. */
986 /* Given a hash table, apply some function to each node in the table. The
987 table to traverse is given as the "hash_tab_p" argument, and the
988 function to be applied to each node in the table is given as "func"
992 visit_each_hash_node (hash_tab_p, func)
993 const hash_table_entry *hash_tab_p;
996 const hash_table_entry *primary;
998 for (primary = hash_tab_p; primary < &hash_tab_p[HASH_TABLE_SIZE]; primary++)
1001 hash_table_entry *second;
1004 for (second = primary->hash_next; second; second = second->hash_next)
1009 /* Initialize all of the fields of a new hash table entry, pointed
1010 to by the "p" parameter. Note that the space to hold the entry
1011 is assumed to have already been allocated before this routine is
1014 static hash_table_entry *
1016 hash_table_entry *p;
1019 p->hash_next = NULL;
1020 p->symbol = dupstr (s);
1026 /* Look for a particular function name or filename in the particular
1027 hash table indicated by "hash_tab_p". If the name is not in the
1028 given hash table, add it. Either way, return a pointer to the
1029 hash table entry for the given name. */
1031 static hash_table_entry *
1032 lookup (hash_tab_p, search_symbol)
1033 hash_table_entry *hash_tab_p;
1034 const char *search_symbol;
1037 const char *search_symbol_char_p = search_symbol;
1038 hash_table_entry *p;
1040 while (*search_symbol_char_p)
1041 hash_value += *search_symbol_char_p++;
1042 hash_value &= hash_mask;
1043 p = &hash_tab_p[hash_value];
1045 return add_symbol (p, search_symbol);
1046 if (!strcmp (p->symbol, search_symbol))
1048 while (p->hash_next)
1051 if (!strcmp (p->symbol, search_symbol))
1054 p->hash_next = (hash_table_entry *) xmalloc (sizeof (hash_table_entry));
1056 return add_symbol (p, search_symbol);
1059 /* Throw a def/dec record on the junk heap.
1061 Also, since we are not using this record anymore, free up all of the
1062 stuff it pointed to. */
1068 xfree (p->ansi_decl);
1072 const f_list_chain_item * curr;
1073 const f_list_chain_item * next;
1075 for (curr = p->f_list_chain; curr; curr = next)
1077 next = curr->chain_next;
1081 #endif /* !defined (UNPROTOIZE) */
1086 /* Unexpand as many macro symbol as we can find.
1088 If the given line must be unexpanded, make a copy of it in the heap and
1089 return a pointer to the unexpanded copy. Otherwise return NULL. */
1092 unexpand_if_needed (aux_info_line)
1093 const char *aux_info_line;
1095 static char *line_buf = 0;
1096 static int line_buf_size = 0;
1097 const unexpansion* unexp_p;
1098 int got_unexpanded = 0;
1100 char *copy_p = line_buf;
1104 line_buf_size = 1024;
1105 line_buf = (char *) xmalloc (line_buf_size);
1110 /* Make a copy of the input string in line_buf, expanding as necessary. */
1112 for (s = aux_info_line; *s != '\n'; )
1114 for (unexp_p = unexpansions; unexp_p->expanded; unexp_p++)
1116 const char *in_p = unexp_p->expanded;
1117 size_t len = strlen (in_p);
1119 if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
1121 int size = strlen (unexp_p->contracted);
1123 if (copy_p + size - line_buf >= line_buf_size)
1125 int offset = copy_p - line_buf;
1127 line_buf_size += size;
1128 line_buf = (char *) xrealloc (line_buf, line_buf_size);
1129 copy_p = line_buf + offset;
1131 strcpy (copy_p, unexp_p->contracted);
1134 /* Assume the there will not be another replacement required
1135 within the text just replaced. */
1138 goto continue_outer;
1141 if (copy_p - line_buf == line_buf_size)
1143 int offset = copy_p - line_buf;
1145 line_buf = (char *) xrealloc (line_buf, line_buf_size);
1146 copy_p = line_buf + offset;
1151 if (copy_p + 2 - line_buf >= line_buf_size)
1153 int offset = copy_p - line_buf;
1155 line_buf = (char *) xrealloc (line_buf, line_buf_size);
1156 copy_p = line_buf + offset;
1161 return (got_unexpanded ? dupstr (line_buf) : 0);
1164 /* Return the absolutized filename for the given relative
1165 filename. Note that if that filename is already absolute, it may
1166 still be returned in a modified form because this routine also
1167 eliminates redundant slashes and single dots and eliminates double
1168 dots to get a shortest possible filename from the given input
1169 filename. The absolutization of relative filenames is made by
1170 assuming that the given filename is to be taken as relative to
1171 the first argument (cwd) or to the current directory if cwd is
1175 abspath (cwd, rel_filename)
1177 const char *rel_filename;
1179 /* Setup the current working directory as needed. */
1180 const char *cwd2 = (cwd) ? cwd : cwd_buffer;
1181 char *const abs_buffer
1182 = (char *) alloca (strlen (cwd2) + strlen (rel_filename) + 2);
1183 char *endp = abs_buffer;
1186 /* Copy the filename (possibly preceded by the current working
1187 directory name) into the absolutization buffer. */
1192 if (rel_filename[0] != '/')
1195 while (*endp++ = *src_p++)
1197 *(endp-1) = '/'; /* overwrite null */
1199 src_p = rel_filename;
1200 while (*endp++ = *src_p++)
1204 /* Now make a copy of abs_buffer into abs_buffer, shortening the
1205 filename (by taking out slashes and dots) as we go. */
1207 outp = inp = abs_buffer;
1208 *outp++ = *inp++; /* copy first slash */
1211 *outp++ = *inp++; /* copy second slash */
1217 else if (inp[0] == '/' && outp[-1] == '/')
1222 else if (inp[0] == '.' && outp[-1] == '/')
1226 else if (inp[1] == '/')
1231 else if ((inp[1] == '.') && (inp[2] == 0 || inp[2] == '/'))
1233 inp += (inp[2] == '/') ? 3 : 2;
1235 while (outp >= abs_buffer && *outp != '/')
1237 if (outp < abs_buffer)
1239 /* Catch cases like /.. where we try to backup to a
1240 point above the absolute root of the logical file
1243 fprintf (stderr, "%s: invalid file name: %s\n",
1244 pname, rel_filename);
1254 /* On exit, make sure that there is a trailing null, and make sure that
1255 the last character of the returned string is *not* a slash. */
1258 if (outp[-1] == '/')
1261 /* Make a copy (in the heap) of the stuff left in the absolutization
1262 buffer and return a pointer to the copy. */
1264 return dupstr (abs_buffer);
1267 /* Given a filename (and possibly a directory name from which the filename
1268 is relative) return a string which is the shortest possible
1269 equivalent for the corresponding full (absolutized) filename. The
1270 shortest possible equivalent may be constructed by converting the
1271 absolutized filename to be a relative filename (i.e. relative to
1272 the actual current working directory). However if a relative filename
1273 is longer, then the full absolute filename is returned.
1277 Note that "simple-minded" conversion of any given type of filename (either
1278 relative or absolute) may not result in a valid equivalent filename if any
1279 subpart of the original filename is actually a symbolic link. */
1282 shortpath (cwd, filename)
1284 const char *filename;
1288 char *cwd_p = cwd_buffer;
1290 int unmatched_slash_count = 0;
1291 size_t filename_len = strlen (filename);
1293 path_p = abspath (cwd, filename);
1294 rel_buf_p = rel_buffer = (char *) xmalloc (filename_len);
1296 while (*cwd_p && (*cwd_p == *path_p))
1301 if (!*cwd_p && (!*path_p || *path_p == '/')) /* whole pwd matched */
1303 if (!*path_p) /* input *is* the current path! */
1314 while (*cwd_p != '/') /* backup to last slash */
1321 unmatched_slash_count++;
1324 if (*cwd_p++ == '/')
1325 unmatched_slash_count++;
1326 while (unmatched_slash_count--)
1328 if (rel_buffer + filename_len <= rel_buf_p + 3)
1337 if (rel_buffer + filename_len <= rel_buf_p)
1340 while (*rel_buf_p++ = *path_p++);
1343 if (*(rel_buf_p-1) == '/')
1344 *--rel_buf_p = '\0';
1349 /* Lookup the given filename in the hash table for filenames. If it is a
1350 new one, then the hash table info pointer will be null. In this case,
1351 we create a new file_info record to go with the filename, and we initialize
1352 that record with some reasonable values. */
1355 find_file (filename, do_not_stat)
1358 /* FILENAME was const, but that causes a warning on AIX when calling stat.
1359 That is probably a bug in AIX, but might as well avoid the warning. */
1361 hash_table_entry *hash_entry_p;
1363 hash_entry_p = lookup (filename_primary, filename);
1364 if (hash_entry_p->fip)
1365 return hash_entry_p->fip;
1368 struct stat stat_buf;
1369 file_info *file_p = (file_info *) xmalloc (sizeof (file_info));
1371 /* If we cannot get status on any given source file, give a warning
1372 and then just set its time of last modification to infinity. */
1375 stat_buf.st_mtime = (time_t) 0;
1378 if (my_stat (filename, &stat_buf) == -1)
1380 fprintf (stderr, "%s: error: can't get status of `%s': %s\n",
1381 pname, shortpath (NULL, filename), sys_errlist[errno]);
1382 stat_buf.st_mtime = (time_t) -1;
1386 hash_entry_p->fip = file_p;
1387 file_p->hash_entry = hash_entry_p;
1388 file_p->defs_decs = NULL;
1389 file_p->mtime = stat_buf.st_mtime;
1394 /* Generate a fatal error because some part of the aux_info file is
1398 aux_info_corrupted ()
1400 fprintf (stderr, "\n%s: fatal error: aux info file corrupted at line %d\n",
1401 pname, current_aux_info_lineno);
1405 /* ??? This comment is vague. Say what the condition is for. */
1406 /* Check to see that a condition is true. This is kind of like an assert. */
1409 check_aux_info (cond)
1413 aux_info_corrupted ();
1416 /* Given a pointer to the closing right parenthesis for a particular formals
1417 list (in a aux_info file) find the corresponding left parenthesis and
1418 return a pointer to it. */
1421 find_corresponding_lparen (p)
1427 for (paren_depth = 1, q = p-1; paren_depth; q--)
1442 /* Given a line from an aux info file, and a time at which the aux info
1443 file it came from was created, check to see if the item described in
1444 the line comes from a file which has been modified since the aux info
1445 file was created. If so, return non-zero, else return zero. */
1448 referenced_file_is_newer (l, aux_info_mtime)
1450 time_t aux_info_mtime;
1456 check_aux_info (l[0] == '/');
1457 check_aux_info (l[1] == '*');
1458 check_aux_info (l[2] == ' ');
1461 const char *filename_start = p = l + 3;
1465 filename = (char *) alloca ((size_t) (p - filename_start) + 1);
1466 strncpy (filename, filename_start, (size_t) (p - filename_start));
1467 filename[p-filename_start] = '\0';
1470 /* Call find_file to find the file_info record associated with the file
1471 which contained this particular def or dec item. Note that this call
1472 may cause a new file_info record to be created if this is the first time
1473 that we have ever known about this particular file. */
1475 fi_p = find_file (abspath (invocation_filename, filename), 0);
1477 return (fi_p->mtime > aux_info_mtime);
1480 /* Given a line of info from the aux_info file, create a new
1481 def_dec_info record to remember all of the important information about
1482 a function definition or declaration.
1484 Link this record onto the list of such records for the particular file in
1485 which it occurred in proper (descending) line number order (for now).
1487 If there is an identical record already on the list for the file, throw
1488 this one away. Doing so takes care of the (useless and troublesome)
1489 duplicates which are bound to crop up due to multiple inclusions of any
1490 given individual header file.
1492 Finally, link the new def_dec record onto the list of such records
1493 pertaining to this particular function name. */
1496 save_def_or_dec (l, is_syscalls)
1501 const char *semicolon_p;
1502 def_dec_info *def_dec_p = (def_dec_info *) xmalloc (sizeof (def_dec_info));
1505 def_dec_p->written = 0;
1506 #endif /* !defined (UNPROTOIZE) */
1508 /* Start processing the line by picking off 5 pieces of information from
1509 the left hand end of the line. These are filename, line number,
1510 new/old/implicit flag (new = ANSI prototype format), definition or
1511 declaration flag, and extern/static flag). */
1513 check_aux_info (l[0] == '/');
1514 check_aux_info (l[1] == '*');
1515 check_aux_info (l[2] == ' ');
1518 const char *filename_start = p = l + 3;
1523 filename = (char *) alloca ((size_t) (p - filename_start) + 1);
1524 strncpy (filename, filename_start, (size_t) (p - filename_start));
1525 filename[p-filename_start] = '\0';
1527 /* Call find_file to find the file_info record associated with the file
1528 which contained this particular def or dec item. Note that this call
1529 may cause a new file_info record to be created if this is the first time
1530 that we have ever known about this particular file.
1532 Note that we started out by forcing all of the base source file names
1533 (i.e. the names of the aux_info files with the .X stripped off) into the
1534 filenames hash table, and we simultaneously setup file_info records for
1535 all of these base file names (even if they may be useless later).
1536 The file_info records for all of these "base" file names (properly)
1537 act as file_info records for the "original" (i.e. un-included) files
1538 which were submitted to gcc for compilation (when the -aux-info
1539 option was used). */
1541 def_dec_p->file = find_file (abspath (invocation_filename, filename), is_syscalls);
1545 const char *line_number_start = ++p;
1546 char line_number[10];
1550 strncpy (line_number, line_number_start, (size_t) (p - line_number_start));
1551 line_number[p-line_number_start] = '\0';
1552 def_dec_p->line = atoi (line_number);
1555 /* Check that this record describes a new-style, old-style, or implicit
1556 definition or declaration. */
1558 p++; /* Skip over the `:'. */
1559 check_aux_info ((*p == 'N') || (*p == 'O') || (*p == 'I'));
1561 /* Is this a new style (ANSI prototyped) definition or declaration? */
1563 def_dec_p->prototyped = (*p == 'N');
1567 /* Is this an implicit declaration? */
1569 def_dec_p->is_implicit = (*p == 'I');
1571 #endif /* !defined (UNPROTOIZE) */
1575 check_aux_info ((*p == 'C') || (*p == 'F'));
1577 /* Is this item a function definition (F) or a declaration (C). Note that
1578 we treat item taken from the syscalls file as though they were function
1579 definitions regardless of what the stuff in the file says. */
1581 def_dec_p->is_func_def = ((*p++ == 'F') || is_syscalls);
1584 def_dec_p->definition = 0; /* Fill this in later if protoizing. */
1585 #endif /* !defined (UNPROTOIZE) */
1587 check_aux_info (*p++ == ' ');
1588 check_aux_info (*p++ == '*');
1589 check_aux_info (*p++ == '/');
1590 check_aux_info (*p++ == ' ');
1593 check_aux_info ((!strncmp (p, "static", 6)) || (!strncmp (p, "extern", 6)));
1594 #else /* !defined (UNPROTOIZE) */
1595 if (!strncmp (p, "static", 6))
1596 def_dec_p->is_static = -1;
1597 else if (!strncmp (p, "extern", 6))
1598 def_dec_p->is_static = 0;
1600 check_aux_info (0); /* Didn't find either `extern' or `static'. */
1601 #endif /* !defined (UNPROTOIZE) */
1604 const char *ansi_start = p;
1606 p += 6; /* Pass over the "static" or "extern". */
1608 /* We are now past the initial stuff. Search forward from here to find
1609 the terminating semicolon that should immediately follow the entire
1610 ANSI format function declaration. */
1617 /* Make a copy of the ansi declaration part of the line from the aux_info
1620 def_dec_p->ansi_decl
1621 = dupnstr (ansi_start, (size_t) ((semicolon_p+1) - ansi_start));
1624 /* Backup and point at the final right paren of the final argument list. */
1628 /* Now isolate a whole set of formal argument lists, one-by-one. Normally,
1629 there will only be one list to isolate, but there could be more. */
1631 def_dec_p->f_list_count = 0;
1634 def_dec_p->f_list_chain = NULL;
1635 #endif /* !defined (UNPROTOIZE) */
1639 const char *left_paren_p = find_corresponding_lparen (p);
1642 f_list_chain_item *cip =
1643 (f_list_chain_item *) xmalloc (sizeof (f_list_chain_item));
1646 = dupnstr (left_paren_p + 1, (size_t) (p - (left_paren_p+1)));
1648 /* Add the new chain item at the head of the current list. */
1650 cip->chain_next = def_dec_p->f_list_chain;
1651 def_dec_p->f_list_chain = cip;
1653 #endif /* !defined (UNPROTOIZE) */
1654 def_dec_p->f_list_count++;
1656 p = left_paren_p - 2;
1658 /* p must now point either to another right paren, or to the last
1659 character of the name of the function that was declared/defined.
1660 If p points to another right paren, then this indicates that we
1661 are dealing with multiple formals lists. In that case, there
1662 really should be another right paren preceding this right paren. */
1667 check_aux_info (*--p == ')');
1672 const char *past_fn = p + 1;
1674 check_aux_info (*past_fn == ' ');
1676 /* Scan leftwards over the identifier that names the function. */
1678 while (is_id_char (*p))
1682 /* p now points to the leftmost character of the function name. */
1685 char *fn_string = (char *) alloca (past_fn - p + 1);
1687 strncpy (fn_string, p, (size_t) (past_fn - p));
1688 fn_string[past_fn-p] = '\0';
1689 def_dec_p->hash_entry = lookup (function_name_primary, fn_string);
1693 /* Look at all of the defs and decs for this function name that we have
1694 collected so far. If there is already one which is at the same
1695 line number in the same file, then we can discard this new def_dec_info
1698 As an extra assurance that any such pair of (nominally) identical
1699 function declarations are in fact identical, we also compare the
1700 ansi_decl parts of the lines from the aux_info files just to be on
1703 This comparison will fail if (for instance) the user was playing
1704 messy games with the preprocessor which ultimately causes one
1705 function declaration in one header file to look differently when
1706 that file is included by two (or more) other files. */
1709 const def_dec_info *other;
1711 for (other = def_dec_p->hash_entry->ddip; other; other = other->next_for_func)
1713 if (def_dec_p->line == other->line && def_dec_p->file == other->file)
1715 if (strcmp (def_dec_p->ansi_decl, other->ansi_decl))
1717 fprintf (stderr, "%s: error: declaration of function `%s' at %s(%d) takes different forms\n",
1719 def_dec_p->hash_entry->symbol,
1720 def_dec_p->file->hash_entry->symbol,
1724 free_def_dec (def_dec_p);
1732 /* If we are doing unprotoizing, we must now setup the pointers that will
1733 point to the K&R name list and to the K&R argument declarations list.
1735 Note that if this is only a function declaration, then we should not
1736 expect to find any K&R style formals list following the ANSI-style
1737 formals list. This is because GCC knows that such information is
1738 useless in the case of function declarations (function definitions
1739 are a different story however).
1741 Since we are unprotoizing, we don't need any such lists anyway.
1742 All we plan to do is to delete all characters between ()'s in any
1745 def_dec_p->formal_names = NULL;
1746 def_dec_p->formal_decls = NULL;
1748 if (def_dec_p->is_func_def)
1751 check_aux_info (*++p == ' ');
1752 check_aux_info (*++p == '/');
1753 check_aux_info (*++p == '*');
1754 check_aux_info (*++p == ' ');
1755 check_aux_info (*++p == '(');
1758 const char *kr_names_start = ++p; /* Point just inside '('. */
1762 p--; /* point to closing right paren */
1764 /* Make a copy of the K&R parameter names list. */
1766 def_dec_p->formal_names
1767 = dupnstr (kr_names_start, (size_t) (p - kr_names_start));
1770 check_aux_info (*++p == ' ');
1773 /* p now points to the first character of the K&R style declarations
1774 list (if there is one) or to the star-slash combination that ends
1775 the comment in which such lists get embedded. */
1777 /* Make a copy of the K&R formal decls list and set the def_dec record
1780 if (*p == '*') /* Are there no K&R declarations? */
1782 check_aux_info (*++p == '/');
1783 def_dec_p->formal_decls = "";
1787 const char *kr_decls_start = p;
1789 while (p[0] != '*' || p[1] != '/')
1793 check_aux_info (*p == ' ');
1795 def_dec_p->formal_decls
1796 = dupnstr (kr_decls_start, (size_t) (p - kr_decls_start));
1799 /* Handle a special case. If we have a function definition marked as
1800 being in "old" style, and if it's formal names list is empty, then
1801 it may actually have the string "void" in its real formals list
1802 in the original source code. Just to make sure, we will get setup
1803 to convert such things anyway.
1805 This kludge only needs to be here because of an insurmountable
1806 problem with generating .X files. */
1808 if (!def_dec_p->prototyped && !*def_dec_p->formal_names)
1809 def_dec_p->prototyped = 1;
1812 /* Since we are unprotoizing, if this item is already in old (K&R) style,
1813 we can just ignore it. If that is true, throw away the itme now. */
1815 if (!def_dec_p->prototyped)
1817 free_def_dec (def_dec_p);
1821 #endif /* defined (UNPROTOIZE) */
1823 /* Add this record to the head of the list of records pertaining to this
1824 particular function name. */
1826 def_dec_p->next_for_func = def_dec_p->hash_entry->ddip;
1827 def_dec_p->hash_entry->ddip = def_dec_p;
1829 /* Add this new def_dec_info record to the sorted list of def_dec_info
1830 records for this file. Note that we don't have to worry about duplicates
1831 (caused by multiple inclusions of header files) here because we have
1832 already eliminated duplicates above. */
1834 if (!def_dec_p->file->defs_decs)
1836 def_dec_p->file->defs_decs = def_dec_p;
1837 def_dec_p->next_in_file = NULL;
1841 int line = def_dec_p->line;
1842 const def_dec_info *prev = NULL;
1843 const def_dec_info *curr = def_dec_p->file->defs_decs;
1844 const def_dec_info *next = curr->next_in_file;
1846 while (next && (line < curr->line))
1850 next = next->next_in_file;
1852 if (line >= curr->line)
1854 def_dec_p->next_in_file = curr;
1856 ((NONCONST def_dec_info *) prev)->next_in_file = def_dec_p;
1858 def_dec_p->file->defs_decs = def_dec_p;
1860 else /* assert (next == NULL); */
1862 ((NONCONST def_dec_info *) curr)->next_in_file = def_dec_p;
1863 /* assert (next == NULL); */
1864 def_dec_p->next_in_file = next;
1869 /* Set up the vector COMPILE_PARAMS which is the argument list for running GCC.
1870 Also set input_file_name_index and aux_info_file_name_index
1871 to the indices of the slots where the file names should go. */
1873 /* We initialize the vector by removing -g, -O, -S, -c, and -o options,
1874 and adding '-aux-info AUXFILE -S -o /dev/null INFILE' at the end. */
1877 munge_compile_params (params_list)
1878 const char *params_list;
1880 /* Build up the contents in a temporary vector
1881 that is so big that to has to be big enough. */
1883 = (char **) alloca ((strlen (params_list) + 6) * sizeof (char *));
1884 int param_count = 0;
1887 temp_params[param_count++] = compiler_file_name;
1890 while (isspace (*params_list))
1894 param = params_list;
1895 while (*params_list && !isspace (*params_list))
1897 if (param[0] != '-')
1898 temp_params[param_count++]
1899 = dupnstr (param, (size_t) (params_list - param));
1908 break; /* Don't copy these. */
1910 while (isspace (*params_list))
1912 while (*params_list && !isspace (*params_list))
1916 temp_params[param_count++]
1917 = dupnstr (param, (size_t) (params_list - param));
1923 temp_params[param_count++] = "-aux-info";
1925 /* Leave room for the aux-info file name argument. */
1926 aux_info_file_name_index = param_count;
1927 temp_params[param_count++] = NULL;
1929 temp_params[param_count++] = "-S";
1930 temp_params[param_count++] = "-o";
1931 temp_params[param_count++] = "/dev/null";
1933 /* Leave room for the input file name argument. */
1934 input_file_name_index = param_count;
1935 temp_params[param_count++] = NULL;
1936 /* Terminate the list. */
1937 temp_params[param_count++] = NULL;
1939 /* Make a copy of the compile_params in heap space. */
1942 = (const char **) xmalloc (sizeof (char *) * (param_count+1));
1943 memcpy (compile_params, temp_params, sizeof (char *) * param_count);
1946 /* Do a recompilation for the express purpose of generating a new aux_info
1947 file to go with a specific base source file. */
1950 gen_aux_info_file (base_filename)
1951 const char *base_filename;
1955 if (!input_file_name_index)
1956 munge_compile_params ("");
1958 /* Store the full source file name in the argument vector. */
1959 compile_params[input_file_name_index] = shortpath (NULL, base_filename);
1960 /* Add .X to source file name to get aux-info file name. */
1961 compile_params[aux_info_file_name_index]
1962 = dupnstr (compile_params[input_file_name_index],
1963 (2 + strlen (compile_params[input_file_name_index])));
1964 strcat (compile_params[aux_info_file_name_index], ".X");
1967 fprintf (stderr, "%s: compiling `%s'\n",
1968 pname, compile_params[input_file_name_index]);
1970 if (child_pid = fork ())
1972 if (child_pid == -1)
1974 fprintf (stderr, "%s: error: could not fork process: %s\n",
1975 pname, sys_errlist[errno]);
1980 /* Print out the command line that the other process is now executing. */
1986 fputs ("\t", stderr);
1987 for (arg = compile_params; *arg; arg++)
1989 fputs (*arg, stderr);
1990 fputc (' ', stderr);
1992 fputc ('\n', stderr);
2000 if (wait (&wait_status) == -1)
2002 fprintf (stderr, "%s: error: wait for process failed: %s\n",
2003 pname, sys_errlist[errno]);
2006 if (!WIFEXITED (wait_status))
2008 fprintf (stderr, "%s: error: subprocess %ld did not exit\n",
2009 pname, (long) child_pid);
2010 kill (child_pid, 9);
2013 if (WEXITSTATUS (wait_status) != 0)
2015 fprintf (stderr, "%s: error: %s: compilation failed\n",
2016 pname, base_filename);
2024 if (my_execvp (compile_params[0], (char *const *) compile_params))
2026 int e = errno, f = fileno (stderr);
2027 write (f, pname, strlen (pname));
2029 write (f, compile_params[0], strlen (compile_params[0]));
2031 write (f, sys_errlist[e], strlen (sys_errlist[e]));
2035 return 1; /* Never executed. */
2039 /* Read in all of the information contained in a single aux_info file.
2040 Save all of the important stuff for later. */
2043 process_aux_info_file (base_source_filename, keep_it, is_syscalls)
2044 const char *base_source_filename;
2048 size_t base_len = strlen (base_source_filename);
2049 char * aux_info_filename
2050 = (char *) alloca (base_len + strlen (aux_info_suffix) + 1);
2051 char *aux_info_base;
2052 char *aux_info_limit;
2053 char *aux_info_relocated_name;
2054 const char *aux_info_second_line;
2055 time_t aux_info_mtime;
2056 size_t aux_info_size;
2058 /* Construct the aux_info filename from the base source filename. */
2060 strcpy (aux_info_filename, base_source_filename);
2061 strcat (aux_info_filename, aux_info_suffix);
2063 /* Check that the aux_info file exists and is readable. If it does not
2064 exist, try to create it (once only). */
2072 if (my_access (aux_info_filename, R_OK) == -1)
2074 if (errno == ENOENT && retries == 0)
2078 fprintf (stderr, "%s: warning: missing SYSCALLS file `%s'\n",
2079 pname, aux_info_filename);
2082 if (!gen_aux_info_file (base_source_filename))
2092 fprintf (stderr, "%s: error: can't read aux info file `%s': %s\n",
2093 pname, shortpath (NULL, aux_info_filename),
2094 sys_errlist[errno]);
2102 struct stat stat_buf;
2104 /* Get some status information about this aux_info file. */
2106 if (my_stat (aux_info_filename, &stat_buf) == -1)
2108 fprintf (stderr, "%s: error: can't get status of aux info file `%s': %s\n",
2109 pname, shortpath (NULL, aux_info_filename),
2110 sys_errlist[errno]);
2115 /* Check on whether or not this aux_info file is zero length. If it is,
2116 then just ignore it and return. */
2118 if ((aux_info_size = stat_buf.st_size) == 0)
2121 /* Get the date/time of last modification for this aux_info file and
2122 remember it. We will have to check that any source files that it
2123 contains information about are at least this old or older. */
2125 aux_info_mtime = stat_buf.st_mtime;
2131 /* Open the aux_info file. */
2133 if ((aux_info_file = my_open (aux_info_filename, O_RDONLY, 0444 )) == -1)
2135 fprintf (stderr, "%s: error: can't open aux info file `%s' for reading: %s\n",
2136 pname, shortpath (NULL, aux_info_filename),
2137 sys_errlist[errno]);
2141 /* Allocate space to hold the aux_info file in memory. */
2143 aux_info_base = xmalloc (aux_info_size + 1);
2144 aux_info_limit = aux_info_base + aux_info_size;
2145 *aux_info_limit = '\0';
2147 /* Read the aux_info file into memory. */
2149 if (read (aux_info_file, aux_info_base, aux_info_size) != aux_info_size)
2151 fprintf (stderr, "%s: error: while reading aux info file `%s': %s\n",
2152 pname, shortpath (NULL, aux_info_filename),
2153 sys_errlist[errno]);
2154 free (aux_info_base);
2155 close (aux_info_file);
2159 /* Close the aux info file. */
2161 if (close (aux_info_file))
2163 fprintf (stderr, "%s: error: while closing aux info file `%s': %s\n",
2164 pname, shortpath (NULL, aux_info_filename),
2165 sys_errlist[errno]);
2166 free (aux_info_base);
2167 close (aux_info_file);
2172 /* Delete the aux_info file (unless requested not to). If the deletion
2173 fails for some reason, don't even worry about it. */
2176 if (my_unlink (aux_info_filename) == -1)
2177 fprintf (stderr, "%s: error: can't delete aux info file `%s': %s\n",
2178 pname, shortpath (NULL, aux_info_filename),
2179 sys_errlist[errno]);
2181 /* Save a pointer into the first line of the aux_info file which
2182 contains the filename of the directory from which the compiler
2183 was invoked when the associated source file was compiled.
2184 This information is used later to help create complete
2185 filenames out of the (potentially) relative filenames in
2186 the aux_info file. */
2189 char *p = aux_info_base;
2196 invocation_filename = p; /* Save a pointer to first byte of path. */
2201 while (*p++ != '\n')
2203 aux_info_second_line = p;
2204 aux_info_relocated_name = 0;
2205 if (invocation_filename[0] != '/')
2207 /* INVOCATION_FILENAME is relative;
2208 append it to BASE_SOURCE_FILENAME's dir. */
2210 aux_info_relocated_name = xmalloc (base_len + (p-invocation_filename));
2211 strcpy (aux_info_relocated_name, base_source_filename);
2212 dir_end = strrchr (aux_info_relocated_name, '/');
2216 dir_end = aux_info_relocated_name;
2217 strcpy (dir_end, invocation_filename);
2218 invocation_filename = aux_info_relocated_name;
2224 const char *aux_info_p;
2226 /* Do a pre-pass on the lines in the aux_info file, making sure that all
2227 of the source files referenced in there are at least as old as this
2228 aux_info file itself. If not, go back and regenerate the aux_info
2229 file anew. Don't do any of this for the syscalls file. */
2233 current_aux_info_lineno = 2;
2235 for (aux_info_p = aux_info_second_line; *aux_info_p; )
2237 if (referenced_file_is_newer (aux_info_p, aux_info_mtime))
2239 free (aux_info_base);
2240 xfree (aux_info_relocated_name);
2241 if (keep_it && my_unlink (aux_info_filename) == -1)
2243 fprintf (stderr, "%s: error: can't delete file `%s': %s\n",
2244 pname, shortpath (NULL, aux_info_filename),
2245 sys_errlist[errno]);
2251 /* Skip over the rest of this line to start of next line. */
2253 while (*aux_info_p != '\n')
2256 current_aux_info_lineno++;
2260 /* Now do the real pass on the aux_info lines. Save their information in
2261 the in-core data base. */
2263 current_aux_info_lineno = 2;
2265 for (aux_info_p = aux_info_second_line; *aux_info_p;)
2267 char *unexpanded_line = unexpand_if_needed (aux_info_p);
2269 if (unexpanded_line)
2271 save_def_or_dec (unexpanded_line, is_syscalls);
2272 free (unexpanded_line);
2275 save_def_or_dec (aux_info_p, is_syscalls);
2277 /* Skip over the rest of this line and get to start of next line. */
2279 while (*aux_info_p != '\n')
2282 current_aux_info_lineno++;
2286 free (aux_info_base);
2287 xfree (aux_info_relocated_name);
2292 /* Check an individual filename for a .c suffix. If the filename has this
2293 suffix, rename the file such that its suffix is changed to .C. This
2294 function implements the -C option. */
2298 const hash_table_entry *hp;
2300 const char *filename = hp->symbol;
2301 int last_char_index = strlen (filename) - 1;
2302 char *const new_filename = (char *) alloca (strlen (filename) + 1);
2304 /* Note that we don't care here if the given file was converted or not. It
2305 is possible that the given file was *not* converted, simply because there
2306 was nothing in it which actually required conversion. Even in this case,
2307 we want to do the renaming. Note that we only rename files with the .c
2310 if (filename[last_char_index] != 'c' || filename[last_char_index-1] != '.')
2313 strcpy (new_filename, filename);
2314 new_filename[last_char_index] = 'C';
2316 if (my_link (filename, new_filename) == -1)
2318 fprintf (stderr, "%s: warning: can't link file `%s' to `%s': %s\n",
2319 pname, shortpath (NULL, filename),
2320 shortpath (NULL, new_filename), sys_errlist[errno]);
2325 if (my_unlink (filename) == -1)
2327 fprintf (stderr, "%s: warning: can't delete file `%s': %s\n",
2328 pname, shortpath (NULL, filename), sys_errlist[errno]);
2334 #endif /* !defined (UNPROTOIZE) */
2336 /* Take the list of definitions and declarations attached to a particular
2337 file_info node and reverse the order of the list. This should get the
2338 list into an order such that the item with the lowest associated line
2339 number is nearest the head of the list. When these lists are originally
2340 built, they are in the opposite order. We want to traverse them in
2341 normal line number order later (i.e. lowest to highest) so reverse the
2345 reverse_def_dec_list (hp)
2346 const hash_table_entry *hp;
2348 file_info *file_p = hp->fip;
2349 const def_dec_info *prev = NULL;
2350 const def_dec_info *current = file_p->defs_decs;
2352 if (!( current = file_p->defs_decs))
2353 return; /* no list to reverse */
2356 if (! (current = current->next_in_file))
2357 return; /* can't reverse a single list element */
2359 ((NONCONST def_dec_info *) prev)->next_in_file = NULL;
2363 const def_dec_info *next = current->next_in_file;
2365 ((NONCONST def_dec_info *) current)->next_in_file = prev;
2370 file_p->defs_decs = prev;
2375 /* Find the (only?) extern definition for a particular function name, starting
2376 from the head of the linked list of entries for the given name. If we
2377 cannot find an extern definition for the given function name, issue a
2378 warning and scrounge around for the next best thing, i.e. an extern
2379 function declaration with a prototype attached to it. Note that we only
2380 allow such substitutions for extern declarations and never for static
2381 declarations. That's because the only reason we allow them at all is
2382 to let un-prototyped function declarations for system-supplied library
2383 functions get their prototypes from our own extra SYSCALLS.c.X file which
2384 contains all of the correct prototypes for system functions. */
2386 static const def_dec_info *
2387 find_extern_def (head, user)
2388 const def_dec_info *head;
2389 const def_dec_info *user;
2391 const def_dec_info *dd_p;
2392 const def_dec_info *extern_def_p = NULL;
2393 int conflict_noted = 0;
2395 /* Don't act too stupid here. Somebody may try to convert an entire system
2396 in one swell fwoop (rather than one program at a time, as should be done)
2397 and in that case, we may find that there are multiple extern definitions
2398 of a given function name in the entire set of source files that we are
2399 converting. If however one of these definitions resides in exactly the
2400 same source file as the reference we are trying to satisfy then in that
2401 case it would be stupid for us to fail to realize that this one definition
2402 *must* be the precise one we are looking for.
2404 To make sure that we don't miss an opportunity to make this "same file"
2405 leap of faith, we do a prescan of the list of records relating to the
2406 given function name, and we look (on this first scan) *only* for a
2407 definition of the function which is in the same file as the reference
2408 we are currently trying to satisfy. */
2410 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2411 if (dd_p->is_func_def && !dd_p->is_static && dd_p->file == user->file)
2414 /* Now, since we have not found a definition in the same file as the
2415 reference, we scan the list again and consider all possibilities from
2416 all files. Here we may get conflicts with the things listed in the
2417 SYSCALLS.c.X file, but if that happens it only means that the source
2418 code being converted contains its own definition of a function which
2419 could have been supplied by libc.a. In such cases, we should avoid
2420 issuing the normal warning, and defer to the definition given in the
2423 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2424 if (dd_p->is_func_def && !dd_p->is_static)
2426 if (!extern_def_p) /* Previous definition? */
2427 extern_def_p = dd_p; /* Remember the first definition found. */
2430 /* Ignore definition just found if it came from SYSCALLS.c.X. */
2432 if (is_syscalls_file (dd_p->file))
2435 /* Quietly replace the definition previously found with the one
2436 just found if the previous one was from SYSCALLS.c.X. */
2438 if (is_syscalls_file (extern_def_p->file))
2440 extern_def_p = dd_p;
2444 /* If we get here, then there is a conflict between two function
2445 declarations for the same function, both of which came from the
2448 if (!conflict_noted) /* first time we noticed? */
2451 fprintf (stderr, "%s: error: conflicting extern definitions of '%s'\n",
2452 pname, head->hash_entry->symbol);
2455 fprintf (stderr, "%s: declarations of '%s' will not be converted\n",
2456 pname, head->hash_entry->symbol);
2457 fprintf (stderr, "%s: conflict list for '%s' follows:\n",
2458 pname, head->hash_entry->symbol);
2459 fprintf (stderr, "%s: %s(%d): %s\n",
2461 shortpath (NULL, extern_def_p->file->hash_entry->symbol),
2462 extern_def_p->line, extern_def_p->ansi_decl);
2466 fprintf (stderr, "%s: %s(%d): %s\n",
2468 shortpath (NULL, dd_p->file->hash_entry->symbol),
2469 dd_p->line, dd_p->ansi_decl);
2473 /* We want to err on the side of caution, so if we found multiple conflicting
2474 definitions for the same function, treat this as being that same as if we
2475 had found no definitions (i.e. return NULL). */
2482 /* We have no definitions for this function so do the next best thing.
2483 Search for an extern declaration already in prototype form. */
2485 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2486 if (!dd_p->is_func_def && !dd_p->is_static && dd_p->prototyped)
2488 extern_def_p = dd_p; /* save a pointer to the definition */
2490 fprintf (stderr, "%s: warning: using formals list from %s(%d) for function `%s'\n",
2492 shortpath (NULL, dd_p->file->hash_entry->symbol),
2493 dd_p->line, dd_p->hash_entry->symbol);
2497 /* Gripe about unprototyped function declarations that we found no
2498 corresponding definition (or other source of prototype information)
2501 Gripe even if the unprototyped declaration we are worried about
2502 exists in a file in one of the "system" include directories. We
2503 can gripe about these because we should have at least found a
2504 corresponding (pseudo) definition in the SYSCALLS.c.X file. If we
2505 didn't, then that means that the SYSCALLS.c.X file is missing some
2506 needed prototypes for this particular system. That is worth telling
2511 const char *file = user->file->hash_entry->symbol;
2514 if (in_system_include_dir (file))
2516 /* Why copy this string into `needed' at all?
2517 Why not just use user->ansi_decl without copying? */
2518 char *needed = (char *) alloca (strlen (user->ansi_decl) + 1);
2521 strcpy (needed, user->ansi_decl);
2522 p = (NONCONST char *) substr (needed, user->hash_entry->symbol)
2523 + strlen (user->hash_entry->symbol) + 2;
2524 /* Avoid having ??? in the string. */
2530 fprintf (stderr, "%s: %d: `%s' used but missing from SYSCALLS\n",
2531 shortpath (NULL, file), user->line,
2532 needed+7); /* Don't print "extern " */
2535 fprintf (stderr, "%s: %d: warning: no extern definition for `%s'\n",
2536 shortpath (NULL, file), user->line,
2537 user->hash_entry->symbol);
2540 return extern_def_p;
2543 /* Find the (only?) static definition for a particular function name in a
2544 given file. Here we get the function-name and the file info indirectly
2545 from the def_dec_info record pointer which is passed in. */
2547 static const def_dec_info *
2548 find_static_definition (user)
2549 const def_dec_info *user;
2551 const def_dec_info *head = user->hash_entry->ddip;
2552 const def_dec_info *dd_p;
2553 int num_static_defs = 0;
2554 const def_dec_info *static_def_p = NULL;
2556 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2557 if (dd_p->is_func_def && dd_p->is_static && (dd_p->file == user->file))
2559 static_def_p = dd_p; /* save a pointer to the definition */
2562 if (num_static_defs == 0)
2565 fprintf (stderr, "%s: warning: no static definition for `%s' in file `%s'\n",
2566 pname, head->hash_entry->symbol,
2567 shortpath (NULL, user->file->hash_entry->symbol));
2569 else if (num_static_defs > 1)
2571 fprintf (stderr, "%s: error: multiple static defs of `%s' in file `%s'\n",
2572 pname, head->hash_entry->symbol,
2573 shortpath (NULL, user->file->hash_entry->symbol));
2576 return static_def_p;
2579 /* Find good prototype style formal argument lists for all of the function
2580 declarations which didn't have them before now.
2582 To do this we consider each function name one at a time. For each function
2583 name, we look at the items on the linked list of def_dec_info records for
2584 that particular name.
2586 Somewhere on this list we should find one (and only one) def_dec_info
2587 record which represents the actual function definition, and this record
2588 should have a nice formal argument list already associated with it.
2590 Thus, all we have to do is to connect up all of the other def_dec_info
2591 records for this particular function name to the special one which has
2592 the full-blown formals list.
2594 Of course it is a little more complicated than just that. See below for
2598 connect_defs_and_decs (hp)
2599 const hash_table_entry *hp;
2601 const def_dec_info *dd_p;
2602 const def_dec_info *extern_def_p = NULL;
2603 int first_extern_reference = 1;
2605 /* Traverse the list of definitions and declarations for this particular
2606 function name. For each item on the list, if it is a function
2607 definition (either old style or new style) then GCC has already been
2608 kind enough to produce a prototype for us, and it is associated with
2609 the item already, so declare the item as its own associated "definition".
2611 Also, for each item which is only a function declaration, but which
2612 nonetheless has its own prototype already (obviously supplied by the user)
2613 declare the item as it's own definition.
2615 Note that when/if there are multiple user-supplied prototypes already
2616 present for multiple declarations of any given function, these multiple
2617 prototypes *should* all match exactly with one another and with the
2618 prototype for the actual function definition. We don't check for this
2619 here however, since we assume that the compiler must have already done
2620 this consistency checking when it was creating the .X files. */
2622 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2623 if (dd_p->prototyped)
2624 ((NONCONST def_dec_info *) dd_p)->definition = dd_p;
2626 /* Traverse the list of definitions and declarations for this particular
2627 function name. For each item on the list, if it is an extern function
2628 declaration and if it has no associated definition yet, go try to find
2629 the matching extern definition for the declaration.
2631 When looking for the matching function definition, warn the user if we
2634 If we find more that one function definition also issue a warning.
2636 Do the search for the matching definition only once per unique function
2637 name (and only when absolutely needed) so that we can avoid putting out
2638 redundant warning messages, and so that we will only put out warning
2639 messages when there is actually a reference (i.e. a declaration) for
2640 which we need to find a matching definition. */
2642 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2643 if (!dd_p->is_func_def && !dd_p->is_static && !dd_p->definition)
2645 if (first_extern_reference)
2647 extern_def_p = find_extern_def (hp->ddip, dd_p);
2648 first_extern_reference = 0;
2650 ((NONCONST def_dec_info *) dd_p)->definition = extern_def_p;
2653 /* Traverse the list of definitions and declarations for this particular
2654 function name. For each item on the list, if it is a static function
2655 declaration and if it has no associated definition yet, go try to find
2656 the matching static definition for the declaration within the same file.
2658 When looking for the matching function definition, warn the user if we
2659 fail to find one in the same file with the declaration, and refuse to
2660 convert this kind of cross-file static function declaration. After all,
2661 this is stupid practice and should be discouraged.
2663 We don't have to worry about the possibility that there is more than one
2664 matching function definition in the given file because that would have
2665 been flagged as an error by the compiler.
2667 Do the search for the matching definition only once per unique
2668 function-name/source-file pair (and only when absolutely needed) so that
2669 we can avoid putting out redundant warning messages, and so that we will
2670 only put out warning messages when there is actually a reference (i.e. a
2671 declaration) for which we actually need to find a matching definition. */
2673 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2674 if (!dd_p->is_func_def && dd_p->is_static && !dd_p->definition)
2676 const def_dec_info *dd_p2;
2677 const def_dec_info *static_def;
2679 /* We have now found a single static declaration for which we need to
2680 find a matching definition. We want to minimize the work (and the
2681 number of warnings), so we will find an appropriate (matching)
2682 static definition for this declaration, and then distribute it
2683 (as the definition for) any and all other static declarations
2684 for this function name which occur within the same file, and which
2685 do not already have definitions.
2687 Note that a trick is used here to prevent subsequent attempts to
2688 call find_static_definition for a given function-name & file
2689 if the first such call returns NULL. Essentially, we convert
2690 these NULL return values to -1, and put the -1 into the definition
2691 field for each other static declaration from the same file which
2692 does not already have an associated definition.
2693 This makes these other static declarations look like they are
2694 actually defined already when the outer loop here revisits them
2695 later on. Thus, the outer loop will skip over them. Later, we
2696 turn the -1's back to NULL's. */
2698 ((NONCONST def_dec_info *) dd_p)->definition =
2699 (static_def = find_static_definition (dd_p))
2701 : (const def_dec_info *) -1;
2703 for (dd_p2 = dd_p->next_for_func; dd_p2; dd_p2 = dd_p2->next_for_func)
2704 if (!dd_p2->is_func_def && dd_p2->is_static
2705 && !dd_p2->definition && (dd_p2->file == dd_p->file))
2706 ((NONCONST def_dec_info *)dd_p2)->definition = dd_p->definition;
2709 /* Convert any dummy (-1) definitions we created in the step above back to
2710 NULL's (as they should be). */
2712 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2713 if (dd_p->definition == (def_dec_info *) -1)
2714 ((NONCONST def_dec_info *) dd_p)->definition = NULL;
2717 #endif /* !defined (UNPROTOIZE) */
2719 /* Give a pointer into the clean text buffer, return a number which is the
2720 original source line number that the given pointer points into. */
2723 identify_lineno (clean_p)
2724 const char *clean_p;
2729 for (scan_p = clean_text_base; scan_p <= clean_p; scan_p++)
2730 if (*scan_p == '\n')
2735 /* Issue an error message and give up on doing this particular edit. */
2738 declare_source_confusing (clean_p)
2739 const char *clean_p;
2744 fprintf (stderr, "%s: %d: warning: source too confusing\n",
2745 shortpath (NULL, convert_filename), last_known_line_number);
2747 fprintf (stderr, "%s: %d: warning: source too confusing\n",
2748 shortpath (NULL, convert_filename),
2749 identify_lineno (clean_p));
2751 longjmp (source_confusion_recovery, 1);
2754 /* Check that a condition which is expected to be true in the original source
2755 code is in fact true. If not, issue an error message and give up on
2756 converting this particular source file. */
2759 check_source (cond, clean_p)
2761 const char *clean_p;
2764 declare_source_confusing (clean_p);
2767 /* If we think of the in-core cleaned text buffer as a memory mapped
2768 file (with the variable last_known_line_start acting as sort of a
2769 file pointer) then we can imagine doing "seeks" on the buffer. The
2770 following routine implements a kind of "seek" operation for the in-core
2771 (cleaned) copy of the source file. When finished, it returns a pointer to
2772 the start of a given (numbered) line in the cleaned text buffer.
2774 Note that protoize only has to "seek" in the forward direction on the
2775 in-core cleaned text file buffers, and it never needs to back up.
2777 This routine is made a little bit faster by remembering the line number
2778 (and pointer value) supplied (and returned) from the previous "seek".
2779 This prevents us from always having to start all over back at the top
2780 of the in-core cleaned buffer again. */
2786 if (n < last_known_line_number)
2789 while (n > last_known_line_number)
2791 while (*last_known_line_start != '\n')
2792 check_source (++last_known_line_start < clean_text_limit, 0);
2793 last_known_line_start++;
2794 last_known_line_number++;
2796 return last_known_line_start;
2799 /* Given a pointer to a character in the cleaned text buffer, return a pointer
2800 to the next non-whitepace character which follows it. */
2803 forward_to_next_token_char (ptr)
2806 for (++ptr; isspace (*ptr); check_source (++ptr < clean_text_limit, 0))
2811 /* Copy a chunk of text of length `len' and starting at `str' to the current
2812 output buffer. Note that all attempts to add stuff to the current output
2813 buffer ultimately go through here. */
2816 output_bytes (str, len)
2820 if ((repl_write_ptr + 1) + len >= repl_text_limit)
2822 size_t new_size = (repl_text_limit - repl_text_base) << 1;
2823 char *new_buf = (char *) xrealloc (repl_text_base, new_size);
2825 repl_write_ptr = new_buf + (repl_write_ptr - repl_text_base);
2826 repl_text_base = new_buf;
2827 repl_text_limit = new_buf + new_size;
2829 memcpy (repl_write_ptr + 1, str, len);
2830 repl_write_ptr += len;
2833 /* Copy all bytes (except the trailing null) of a null terminated string to
2834 the current output buffer. */
2840 output_bytes (str, strlen (str));
2843 /* Copy some characters from the original text buffer to the current output
2846 This routine takes a pointer argument `p' which is assumed to be a pointer
2847 into the cleaned text buffer. The bytes which are copied are the `original'
2848 equivalents for the set of bytes between the last value of `clean_read_ptr'
2849 and the argument value `p'.
2851 The set of bytes copied however, comes *not* from the cleaned text buffer,
2852 but rather from the direct counterparts of these bytes within the original
2855 Thus, when this function is called, some bytes from the original text
2856 buffer (which may include original comments and preprocessing directives)
2857 will be copied into the output buffer.
2859 Note that the request implide when this routine is called includes the
2860 byte pointed to by the argument pointer `p'. */
2866 size_t copy_length = (size_t) (p - clean_read_ptr);
2867 const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
2869 if (copy_length == 0)
2872 output_bytes (copy_start, copy_length);
2876 /* Given a pointer to a def_dec_info record which represents some form of
2877 definition of a function (perhaps a real definition, or in lieu of that
2878 perhaps just a declaration with a full prototype) return true if this
2879 function is one which we should avoid converting. Return false
2883 other_variable_style_function (ansi_header)
2884 const char *ansi_header;
2888 /* See if we have a stdarg function, or a function which has stdarg style
2889 parameters or a stdarg style return type. */
2891 return (int) substr (ansi_header, "...");
2893 #else /* !defined (UNPROTOIZE) */
2895 /* See if we have a varargs function, or a function which has varargs style
2896 parameters or a varargs style return type. */
2899 int len = strlen (varargs_style_indicator);
2901 for (p = ansi_header; p; )
2903 const char *candidate;
2905 if ((candidate = substr (p, varargs_style_indicator)) == 0)
2908 if (!is_id_char (candidate[-1]) && !is_id_char (candidate[len]))
2914 #endif /* !defined (UNPROTOIZE) */
2917 /* Do the editing operation specifically for a function "declaration". Note
2918 that editing for function "definitions" are handled in a separate routine
2922 edit_fn_declaration (def_dec_p, clean_text_p)
2923 const def_dec_info *def_dec_p;
2924 const char *VOLATILE clean_text_p;
2926 const char *start_formals;
2927 const char *end_formals;
2928 const char *function_to_edit = def_dec_p->hash_entry->symbol;
2929 size_t func_name_len = strlen (function_to_edit);
2930 const char *end_of_fn_name;
2934 const f_list_chain_item *this_f_list_chain_item;
2935 const def_dec_info *definition = def_dec_p->definition;
2937 /* If we are protoizing, and if we found no corresponding definition for
2938 this particular function declaration, then just leave this declaration
2939 exactly as it is. */
2944 /* If we are protoizing, and if the corresponding definition that we found
2945 for this particular function declaration defined an old style varargs
2946 function, then we want to issue a warning and just leave this function
2947 declaration unconverted. */
2949 if (other_variable_style_function (definition->ansi_decl))
2952 fprintf (stderr, "%s: %d: warning: varargs function declaration not converted\n",
2953 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
2958 #endif /* !defined (UNPROTOIZE) */
2960 /* Setup here to recover from confusing source code detected during this
2961 particular "edit". */
2964 if (setjmp (source_confusion_recovery))
2966 restore_pointers ();
2967 fprintf (stderr, "%s: declaration of function `%s' not converted\n",
2968 pname, function_to_edit);
2972 /* We are editing a function declaration. The line number we did a seek to
2973 contains the comma or semicolon which follows the declaration. Our job
2974 now is to scan backwards looking for the function name. This name *must*
2975 be followed by open paren (ignoring whitespace, of course). We need to
2976 replace everything between that open paren and the corresponding closing
2977 paren. If we are protoizing, we need to insert the prototype-style
2978 formals lists. If we are unprotoizing, we need to just delete everything
2979 between the pairs of opening and closing parens. */
2981 /* First move up to the end of the line. */
2983 while (*clean_text_p != '\n')
2984 check_source (++clean_text_p < clean_text_limit, 0);
2985 clean_text_p--; /* Point to just before the newline character. */
2987 /* Now we can scan backwards for the function name. */
2993 /* Scan leftwards until we find some character which can be
2994 part of an identifier. */
2996 while (!is_id_char (*clean_text_p))
2997 check_source (--clean_text_p > clean_read_ptr, 0);
2999 /* Scan backwards until we find a char that cannot be part of an
3002 while (is_id_char (*clean_text_p))
3003 check_source (--clean_text_p > clean_read_ptr, 0);
3005 /* Having found an "id break", see if the following id is the one
3006 that we are looking for. If so, then exit from this loop. */
3008 if (!strncmp (clean_text_p+1, function_to_edit, func_name_len))
3010 char ch = *(clean_text_p + 1 + func_name_len);
3012 /* Must also check to see that the name in the source text
3013 ends where it should (in order to prevent bogus matches
3014 on similar but longer identifiers. */
3016 if (! is_id_char (ch))
3017 break; /* exit from loop */
3021 /* We have now found the first perfect match for the function name in
3022 our backward search. This may or may not be the actual function
3023 name at the start of the actual function declaration (i.e. we could
3024 have easily been mislead). We will try to avoid getting fooled too
3025 often by looking forward for the open paren which should follow the
3026 identifier we just found. We ignore whitespace while hunting. If
3027 the next non-whitespace byte we see is *not* an open left paren,
3028 then we must assume that we have been fooled and we start over
3029 again accordingly. Note that there is no guarrantee, that even if
3030 we do see the open paren, that we are in the right place.
3031 Programmers do the strangest things sometimes! */
3033 end_of_fn_name = clean_text_p + strlen (def_dec_p->hash_entry->symbol);
3034 start_formals = forward_to_next_token_char (end_of_fn_name);
3036 while (*start_formals != '(');
3038 /* start_of_formals now points to the opening left paren which immediately
3039 follows the name of the function. */
3041 /* Note that there may be several formals lists which need to be modified
3042 due to the possibility that the return type of this function is a
3043 pointer-to-function type. If there are several formals lists, we
3044 convert them in left-to-right order here. */
3047 this_f_list_chain_item = definition->f_list_chain;
3048 #endif /* !defined (UNPROTOIZE) */
3055 end_formals = start_formals + 1;
3057 for (; depth; check_source (++end_formals < clean_text_limit, 0))
3059 switch (*end_formals)
3072 /* end_formals now points to the closing right paren of the formals
3073 list whose left paren is pointed to by start_formals. */
3075 /* Now, if we are protoizing, we insert the new ANSI-style formals list
3076 attached to the associated definition of this function. If however
3077 we are unprotoizing, then we simply delete any formals list which
3080 output_up_to (start_formals);
3082 if (this_f_list_chain_item)
3084 output_string (this_f_list_chain_item->formals_list);
3085 this_f_list_chain_item = this_f_list_chain_item->chain_next;
3090 fprintf (stderr, "%s: warning: too many parameter lists in declaration of `%s'\n",
3091 pname, def_dec_p->hash_entry->symbol);
3092 check_source (0, end_formals); /* leave the declaration intact */
3094 #endif /* !defined (UNPROTOIZE) */
3095 clean_read_ptr = end_formals - 1;
3097 /* Now see if it looks like there may be another formals list associated
3098 with the function declaration that we are converting (following the
3099 formals list that we just converted. */
3102 const char *another_r_paren = forward_to_next_token_char (end_formals);
3104 if ((*another_r_paren != ')')
3105 || (*(start_formals = forward_to_next_token_char (another_r_paren)) != '('))
3108 if (this_f_list_chain_item)
3111 fprintf (stderr, "\n%s: warning: too few parameter lists in declaration of `%s'\n",
3112 pname, def_dec_p->hash_entry->symbol);
3113 check_source (0, start_formals); /* leave the decl intact */
3115 #endif /* !defined (UNPROTOIZE) */
3121 /* There does appear to be yet another formals list, so loop around
3122 again, and convert it also. */
3126 /* Edit a whole group of formals lists, starting with the rightmost one
3127 from some set of formals lists. This routine is called once (from the
3128 outside) for each function declaration which is converted. It is
3129 recursive however, and it calls itself once for each remaining formal
3130 list that lies to the left of the one it was originally called to work
3131 on. Thus, a whole set gets done in right-to-left order.
3133 This routine returns non-zero if it thinks that it should not be trying
3134 to convert this particular function definition (because the name of the
3135 function doesn't match the one expected). */
3138 edit_formals_lists (end_formals, f_list_count, def_dec_p)
3139 const char *end_formals;
3140 unsigned int f_list_count;
3141 const def_dec_info *def_dec_p;
3143 const char *start_formals;
3146 start_formals = end_formals - 1;
3148 for (; depth; check_source (--start_formals > clean_read_ptr, 0))
3150 switch (*start_formals)
3162 /* start_formals now points to the opening left paren of the formals list. */
3168 const char *next_end;
3170 /* There should be more formal lists to the left of here. */
3172 next_end = start_formals - 1;
3173 check_source (next_end > clean_read_ptr, 0);
3174 while (isspace (*next_end))
3175 check_source (--next_end > clean_read_ptr, 0);
3176 check_source (*next_end == ')', next_end);
3177 check_source (--next_end > clean_read_ptr, 0);
3178 check_source (*next_end == ')', next_end);
3179 if (edit_formals_lists (next_end, f_list_count, def_dec_p))
3183 /* Check that the function name in the header we are working on is the same
3184 as the one we would expect to find. If not, issue a warning and return
3187 if (f_list_count == 0)
3189 const char *expected = def_dec_p->hash_entry->symbol;
3190 const char *func_name_start;
3191 const char *func_name_limit;
3192 size_t func_name_len;
3194 for (func_name_limit = start_formals-1; isspace (*func_name_limit); )
3195 check_source (--func_name_limit > clean_read_ptr, 0);
3197 for (func_name_start = func_name_limit++;
3198 is_id_char (*func_name_start);
3200 check_source (func_name_start > clean_read_ptr, 0);
3202 func_name_len = func_name_limit - func_name_start;
3203 if (func_name_len == 0)
3204 check_source (0, func_name_start);
3205 if (func_name_len != strlen (expected)
3206 || strncmp (func_name_start, expected, func_name_len))
3208 fprintf (stderr, "%s: %d: warning: found `%s' but expected `%s'\n",
3209 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3210 identify_lineno (func_name_start),
3211 dupnstr (func_name_start, func_name_len),
3217 output_up_to (start_formals);
3220 if (f_list_count == 0)
3221 output_string (def_dec_p->formal_names);
3222 #else /* !defined (UNPROTOIZE) */
3224 unsigned f_list_depth;
3225 const f_list_chain_item *flci_p = def_dec_p->f_list_chain;
3227 /* At this point, the current value of f_list count says how many
3228 links we have to follow through the f_list_chain to get to the
3229 particular formals list that we need to output next. */
3231 for (f_list_depth = 0; f_list_depth < f_list_count; f_list_depth++)
3232 flci_p = flci_p->chain_next;
3233 output_string (flci_p->formals_list);
3235 #endif /* !defined (UNPROTOIZE) */
3237 clean_read_ptr = end_formals - 1;
3241 /* Given a pointer to a byte in the clean text buffer which points to the
3242 beginning of a line that contains a "follower" token for a function
3243 definition header, do whatever is necessary to find the right closing
3244 paren for the rightmost formals list of the function definition header.
3248 find_rightmost_formals_list (clean_text_p)
3249 const char *clean_text_p;
3251 const char *end_formals;
3253 /* We are editing a function definition. The line number we did a seek
3254 to contains the first token which immediately follows the entire set of
3255 formals lists which are part of this particular function definition
3258 Our job now is to scan leftwards in the clean text looking for the
3259 right-paren which is at the end of the function header's rightmost
3262 If we ignore whitespace, this right paren should be the first one we
3263 see which is (ignoring whitespace) immediately followed either by the
3264 open curly-brace beginning the function body or by an alphabetic
3265 character (in the case where the function definition is in old (K&R)
3266 style and there are some declarations of formal parameters). */
3268 /* It is possible that the right paren we are looking for is on the
3269 current line (together with its following token). Just in case that
3270 might be true, we start out here by skipping down to the right end of
3271 the current line before starting our scan. */
3273 for (end_formals = clean_text_p; *end_formals != '\n'; end_formals++)
3279 /* Now scan backwards while looking for the right end of the rightmost
3280 formals list associated with this function definition. */
3284 const char *l_brace_p;
3286 /* Look leftward and try to find a right-paren. */
3288 while (*end_formals != ')')
3290 if (isspace (*end_formals))
3291 while (isspace (*end_formals))
3292 check_source (--end_formals > clean_read_ptr, 0);
3294 check_source (--end_formals > clean_read_ptr, 0);
3297 ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3298 /* Since we are unprotoizing an ANSI-style (prototyped) function
3299 definition, there had better not be anything (except whitespace)
3300 between the end of the ANSI formals list and the beginning of the
3301 function body (i.e. the '{'). */
3303 check_source (ch == '{', l_brace_p);
3306 #else /* !defined (UNPROTOIZE) */
3308 /* Now scan backwards while looking for the right end of the rightmost
3309 formals list associated with this function definition. */
3314 const char *l_brace_p;
3316 /* Look leftward and try to find a right-paren. */
3318 while (*end_formals != ')')
3320 if (isspace (*end_formals))
3321 while (isspace (*end_formals))
3322 check_source (--end_formals > clean_read_ptr, 0);
3324 check_source (--end_formals > clean_read_ptr, 0);
3327 ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3329 /* Since it is possible that we found a right paren before the starting
3330 '{' of the body which IS NOT the one at the end of the real K&R
3331 formals list (say for instance, we found one embedded inside one of
3332 the old K&R formal parameter declarations) we have to check to be
3333 sure that this is in fact the right paren that we were looking for.
3335 The one we were looking for *must* be followed by either a '{' or
3336 by an alphabetic character, while others *cannot* legally be followed
3337 by such characters. */
3339 if ((ch == '{') || isalpha (ch))
3342 /* At this point, we have found a right paren, but we know that it is
3343 not the one we were looking for, so backup one character and keep
3346 check_source (--end_formals > clean_read_ptr, 0);
3349 #endif /* !defined (UNPROTOIZE) */
3356 /* Insert into the output file a totally new declaration for a function
3357 which (up until now) was being called from within the current block
3358 without having been declared at any point such that the declaration
3359 was visible (i.e. in scope) at the point of the call.
3361 We need to add in explicit declarations for all such function calls
3362 in order to get the full benefit of prototype-based function call
3363 parameter type checking. */
3366 add_local_decl (def_dec_p, clean_text_p)
3367 const def_dec_info *def_dec_p;
3368 const char *clean_text_p;
3370 const char *start_of_block;
3371 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3373 /* Don't insert new local explicit declarations unless explicitly requested
3379 /* Setup here to recover from confusing source code detected during this
3380 particular "edit". */
3383 if (setjmp (source_confusion_recovery))
3385 restore_pointers ();
3386 fprintf (stderr, "%s: local declaration for function `%s' not inserted\n",
3387 pname, function_to_edit);
3391 /* We have already done a seek to the start of the line which should
3392 contain *the* open curly brace which begins the block in which we need
3393 to insert an explicit function declaration (to replace the implicit one).
3395 Now we scan that line, starting from the left, until we find the
3396 open curly brace we are looking for. Note that there may actually be
3397 multiple open curly braces on the given line, but we will be happy
3398 with the leftmost one no matter what. */
3400 start_of_block = clean_text_p;
3401 while (*start_of_block != '{' && *start_of_block != '\n')
3402 check_source (++start_of_block < clean_text_limit, 0);
3404 /* Note that the line from the original source could possibly
3405 contain *no* open curly braces! This happens if the line contains
3406 a macro call which expands into a chunk of text which includes a
3407 block (and that block's associated open and close curly braces).
3408 In cases like this, we give up, issue a warning, and do nothing. */
3410 if (*start_of_block != '{')
3414 "\n%s: %d: warning: can't add declaration of `%s' into macro call\n",
3415 def_dec_p->file->hash_entry->symbol, def_dec_p->line,
3416 def_dec_p->hash_entry->symbol);
3420 /* Figure out what a nice (pretty) indentation would be for the new
3421 declaration we are adding. In order to do this, we must scan forward
3422 from the '{' until we find the first line which starts with some
3423 non-whitespace characters (i.e. real "token" material). */
3426 const char *ep = forward_to_next_token_char (start_of_block) - 1;
3429 /* Now we have ep pointing at the rightmost byte of some existing indent
3430 stuff. At least that is the hope.
3432 We can now just scan backwards and find the left end of the existing
3433 indentation string, and then copy it to the output buffer. */
3435 for (sp = ep; isspace (*sp) && *sp != '\n'; sp--)
3438 /* Now write out the open { which began this block, and any following
3439 trash up to and including the last byte of the existing indent that
3444 /* Now we go ahead and insert the new declaration at this point.
3446 If the definition of the given function is in the same file that we
3447 are currently editing, and if its full ANSI declaration normally
3448 would start with the keyword `extern', suppress the `extern'. */
3451 const char *decl = def_dec_p->definition->ansi_decl;
3453 if ((*decl == 'e') && (def_dec_p->file == def_dec_p->definition->file))
3455 output_string (decl);
3458 /* Finally, write out a new indent string, just like the preceding one
3459 that we found. This will typically include a newline as the first
3460 character of the indent string. */
3462 output_bytes (sp, (size_t) (ep - sp) + 1);
3466 /* Given a pointer to a file_info record, and a pointer to the beginning
3467 of a line (in the clean text buffer) which is assumed to contain the
3468 first "follower" token for the first function definition header in the
3469 given file, find a good place to insert some new global function
3470 declarations (which will replace scattered and imprecise implicit ones)
3471 and then insert the new explicit declaration at that point in the file. */
3474 add_global_decls (file_p, clean_text_p)
3475 const file_info *file_p;
3476 const char *clean_text_p;
3478 const def_dec_info *dd_p;
3481 /* Setup here to recover from confusing source code detected during this
3482 particular "edit". */
3485 if (setjmp (source_confusion_recovery))
3487 restore_pointers ();
3488 fprintf (stderr, "%s: global declarations for file `%s' not inserted\n",
3489 pname, shortpath (NULL, file_p->hash_entry->symbol));
3493 /* Start by finding a good location for adding the new explicit function
3494 declarations. To do this, we scan backwards, ignoring whitespace
3495 and comments and other junk until we find either a semicolon, or until
3496 we hit the beginning of the file. */
3498 scan_p = find_rightmost_formals_list (clean_text_p);
3501 if (scan_p < clean_text_base)
3503 check_source (scan_p > clean_read_ptr, 0);
3508 /* scan_p now points either to a semicolon, or to just before the start
3509 of the whole file. */
3511 /* Now scan forward for the first non-whitespace character. In theory,
3512 this should be the first character of the following function definition
3513 header. We will put in the added declarations just prior to that. */
3516 while (isspace (*scan_p))
3520 output_up_to (scan_p);
3522 /* Now write out full prototypes for all of the things that had been
3523 implicitly declared in this file (but only those for which we were
3524 actually able to find unique matching definitions). Avoid duplicates
3525 by marking things that we write out as we go. */
3528 int some_decls_added = 0;
3530 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3531 if (dd_p->is_implicit && dd_p->definition && !dd_p->definition->written)
3533 const char *decl = dd_p->definition->ansi_decl;
3535 /* If the function for which we are inserting a declaration is
3536 actually defined later in the same file, then suppress the
3537 leading `extern' keyword (if there is one). */
3539 if (*decl == 'e' && (dd_p->file == dd_p->definition->file))
3542 output_string ("\n");
3543 output_string (decl);
3544 some_decls_added = 1;
3545 ((NONCONST def_dec_info *) dd_p->definition)->written = 1;
3547 if (some_decls_added)
3548 output_string ("\n\n");
3551 /* Unmark all of the definitions that we just marked. */
3553 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3554 if (dd_p->definition)
3555 ((NONCONST def_dec_info *) dd_p->definition)->written = 0;
3558 #endif /* !defined (UNPROTOIZE) */
3560 /* Do the editing operation specifically for a function "definition". Note
3561 that editing operations for function "declarations" are handled by a
3562 separate routine above. */
3565 edit_fn_definition (def_dec_p, clean_text_p)
3566 const def_dec_info *def_dec_p;
3567 const char *clean_text_p;
3569 const char *end_formals;
3570 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3572 /* Setup here to recover from confusing source code detected during this
3573 particular "edit". */
3576 if (setjmp (source_confusion_recovery))
3578 restore_pointers ();
3579 fprintf (stderr, "%s: definition of function `%s' not converted\n",
3580 pname, function_to_edit);
3584 end_formals = find_rightmost_formals_list (clean_text_p);
3586 /* end_of_formals now points to the closing right paren of the rightmost
3587 formals list which is actually part of the `header' of the function
3588 definition that we are converting. */
3590 /* If the header of this function definition looks like it declares a
3591 function with a variable number of arguments, and if the way it does
3592 that is different from that way we would like it (i.e. varargs vs.
3593 stdarg) then issue a warning and leave the header unconverted. */
3595 if (other_variable_style_function (def_dec_p->ansi_decl))
3598 fprintf (stderr, "%s: %d: warning: definition of %s not converted\n",
3599 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3600 identify_lineno (end_formals),
3602 output_up_to (end_formals);
3606 if (edit_formals_lists (end_formals, def_dec_p->f_list_count, def_dec_p))
3608 restore_pointers ();
3609 fprintf (stderr, "%s: definition of function `%s' not converted\n",
3610 pname, function_to_edit);
3614 /* Have to output the last right paren because this never gets flushed by
3615 edit_formals_list. */
3617 output_up_to (end_formals);
3622 const char *semicolon_p;
3623 const char *limit_p;
3625 int had_newlines = 0;
3627 /* Now write out the K&R style formal declarations, one per line. */
3629 decl_p = def_dec_p->formal_decls;
3630 limit_p = decl_p + strlen (decl_p);
3631 for (;decl_p < limit_p; decl_p = semicolon_p + 2)
3633 for (semicolon_p = decl_p; *semicolon_p != ';'; semicolon_p++)
3635 output_string ("\n");
3636 output_string (indent_string);
3637 output_bytes (decl_p, (size_t) ((semicolon_p + 1) - decl_p));
3640 /* If there are no newlines between the end of the formals list and the
3641 start of the body, we should insert one now. */
3643 for (scan_p = end_formals+1; *scan_p != '{'; )
3645 if (*scan_p == '\n')
3650 check_source (++scan_p < clean_text_limit, 0);
3653 output_string ("\n");
3655 #else /* !defined (UNPROTOIZE) */
3656 /* If we are protoizing, there may be some flotsum & jetsum (like comments
3657 and preprocessing directives) after the old formals list but before
3658 the following { and we would like to preserve that stuff while effectively
3659 deleting the existing K&R formal parameter declarations. We do so here
3660 in a rather tricky way. Basically, we white out any stuff *except*
3661 the comments/pp-directives in the original text buffer, then, if there
3662 is anything in this area *other* than whitespace, we output it. */
3664 const char *end_formals_orig;
3665 const char *start_body;
3666 const char *start_body_orig;
3668 const char *scan_orig;
3669 int have_flotsum = 0;
3670 int have_newlines = 0;
3672 for (start_body = end_formals + 1; *start_body != '{';)
3673 check_source (++start_body < clean_text_limit, 0);
3675 end_formals_orig = orig_text_base + (end_formals - clean_text_base);
3676 start_body_orig = orig_text_base + (start_body - clean_text_base);
3677 scan = end_formals + 1;
3678 scan_orig = end_formals_orig + 1;
3679 for (; scan < start_body; scan++, scan_orig++)
3681 if (*scan == *scan_orig)
3683 have_newlines |= (*scan_orig == '\n');
3684 /* Leave identical whitespace alone. */
3685 if (!isspace (*scan_orig))
3686 *((NONCONST char *)scan_orig) = ' '; /* identical - so whiteout */
3692 output_bytes (end_formals_orig + 1,
3693 (size_t) (start_body_orig - end_formals_orig) - 1);
3696 output_string ("\n");
3698 output_string (" ");
3699 clean_read_ptr = start_body - 1;
3701 #endif /* !defined (UNPROTOIZE) */
3704 /* Clean up the clean text buffer. Do this by converting comments and
3705 preprocessor directives into spaces. Also convert line continuations
3706 into whitespace. Also, whiteout string and character literals. */
3709 do_cleaning (new_clean_text_base, new_clean_text_limit)
3710 char *new_clean_text_base;
3711 char *new_clean_text_limit;
3714 int non_whitespace_since_newline = 0;
3716 for (scan_p = new_clean_text_base; scan_p < new_clean_text_limit; scan_p++)
3720 case '/': /* Handle comments. */
3721 if (scan_p[1] != '*')
3723 non_whitespace_since_newline = 1;
3727 while (scan_p[1] != '/' || scan_p[0] != '*')
3729 if (!isspace (*scan_p))
3731 if (++scan_p >= new_clean_text_limit)
3738 case '#': /* Handle pp directives. */
3739 if (non_whitespace_since_newline)
3742 while (scan_p[1] != '\n' || scan_p[0] == '\\')
3744 if (!isspace (*scan_p))
3746 if (++scan_p >= new_clean_text_limit)
3752 case '\'': /* Handle character literals. */
3753 non_whitespace_since_newline = 1;
3754 while (scan_p[1] != '\'' || scan_p[0] == '\\')
3756 if (scan_p[0] == '\\' && !isspace (scan_p[1]))
3758 if (!isspace (*scan_p))
3760 if (++scan_p >= new_clean_text_limit)
3766 case '"': /* Handle string literals. */
3767 non_whitespace_since_newline = 1;
3768 while (scan_p[1] != '"' || scan_p[0] == '\\')
3770 if (scan_p[0] == '\\' && !isspace (scan_p[1]))
3772 if (!isspace (*scan_p))
3774 if (++scan_p >= new_clean_text_limit)
3780 case '\\': /* Handle line continuations. */
3781 if (scan_p[1] != '\n')
3787 non_whitespace_since_newline = 0; /* Reset. */
3796 break; /* Whitespace characters. */
3800 non_whitespace_since_newline = 1;
3806 /* Given a pointer to the closing right parenthesis for a particular formals
3807 list (in the clean text buffer) find the corresponding left parenthesis
3808 and return a pointer to it. */
3811 careful_find_l_paren (p)
3817 for (paren_depth = 1, q = p-1; paren_depth; check_source (--q >= clean_text_base, 0))
3832 /* Scan the clean text buffer for cases of function definitions that we
3833 don't really know about because they were preprocessed out when the
3834 aux info files were created.
3836 In this version of protoize/unprotoize we just give a warning for each
3837 one found. A later version may be able to at least unprotoize such
3840 Note that we may easily find all function definitions simply by
3841 looking for places where there is a left paren which is (ignoring
3842 whitespace) immediately followed by either a left-brace or by an
3843 upper or lower case letter. Whenever we find this combination, we
3844 have also found a function definition header.
3846 Finding function *declarations* using syntactic clues is much harder.
3847 I will probably try to do this in a later version though. */
3850 scan_for_missed_items (file_p)
3851 const file_info *file_p;
3853 static const char *scan_p;
3854 const char *limit = clean_text_limit - 3;
3855 static const char *backup_limit;
3857 backup_limit = clean_text_base - 1;
3859 for (scan_p = clean_text_base; scan_p < limit; scan_p++)
3863 static const char *last_r_paren;
3864 const char *ahead_p;
3866 last_r_paren = scan_p;
3868 for (ahead_p = scan_p + 1; isspace (*ahead_p); )
3869 check_source (++ahead_p < limit, limit);
3871 scan_p = ahead_p - 1;
3873 if (isalpha (*ahead_p) || *ahead_p == '{')
3875 const char *last_l_paren;
3876 const int lineno = identify_lineno (ahead_p);
3878 if (setjmp (source_confusion_recovery))
3881 /* We know we have a function definition header. Now skip
3882 leftwards over all of its associated formals lists. */
3886 last_l_paren = careful_find_l_paren (last_r_paren);
3887 for (last_r_paren = last_l_paren-1; isspace (*last_r_paren); )
3888 check_source (--last_r_paren >= backup_limit, backup_limit);
3890 while (*last_r_paren == ')');
3892 if (is_id_char (*last_r_paren))
3894 const char *id_limit = last_r_paren + 1;
3895 const char *id_start;
3897 const def_dec_info *dd_p;
3899 for (id_start = id_limit-1; is_id_char (*id_start); )
3900 check_source (--id_start >= backup_limit, backup_limit);
3902 backup_limit = id_start;
3903 if ((id_length = (size_t) (id_limit - id_start)) == 0)
3907 char *func_name = (char *) alloca (id_length + 1);
3908 static const char * const stmt_keywords[]
3909 = { "if", "while", "for", "switch", "return", 0 };
3910 const char * const *stmt_keyword;
3912 strncpy (func_name, id_start, id_length);
3913 func_name[id_length] = '\0';
3915 /* We must check here to see if we are actually looking at
3916 a statement rather than an actual function call. */
3918 for (stmt_keyword = stmt_keywords; *stmt_keyword; stmt_keyword++)
3919 if (!strcmp (func_name, *stmt_keyword))
3923 fprintf (stderr, "%s: found definition of `%s' at %s(%d)\n",
3926 shortpath (NULL, file_p->hash_entry->symbol),
3927 identify_lineno (id_start));
3929 /* We really should check for a match of the function name
3930 here also, but why bother. */
3932 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3933 if (dd_p->is_func_def && dd_p->line == lineno)
3936 /* If we make it here, then we did not know about this
3937 function definition. */
3939 fprintf (stderr, "%s: %d: warning: `%s' was #if 0\n",
3940 shortpath (NULL, file_p->hash_entry->symbol),
3941 identify_lineno (id_start), func_name);
3942 fprintf (stderr, "%s: function definition not converted\n",
3952 /* Do all editing operations for a single source file (either a "base" file
3953 or an "include" file). To do this we read the file into memory, keep a
3954 virgin copy there, make another cleaned in-core copy of the original file
3955 (i.e. one in which all of the comments and preprocessor directives have
3956 been replaced with whitespace), then use these two in-core copies of the
3957 file to make a new edited in-core copy of the file. Finally, rename the
3958 original file (as a way of saving it), and then write the edited version
3959 of the file from core to a disk file of the same name as the original.
3961 Note that the trick of making a copy of the original sans comments &
3962 preprocessor directives make the editing a whole lot easier. */
3966 const hash_table_entry *hp;
3968 struct stat stat_buf;
3969 const file_info *file_p = hp->fip;
3970 char *new_orig_text_base;
3971 char *new_orig_text_limit;
3972 char *new_clean_text_base;
3973 char *new_clean_text_limit;
3976 int first_definition_in_file;
3978 /* If we are not supposed to be converting this file, or if there is
3979 nothing in there which needs converting, just skip this file. */
3981 if (!needs_to_be_converted (file_p))
3984 convert_filename = file_p->hash_entry->symbol;
3986 /* Convert a file if it is in a directory where we want conversion
3987 and the file is not excluded. */
3989 if (!directory_specified_p (convert_filename)
3990 || file_excluded_p (convert_filename))
3994 /* Don't even mention "system" include files unless we are
3995 protoizing. If we are protoizing, we mention these as a
3996 gentle way of prodding the user to convert his "system"
3997 include files to prototype format. */
3998 && !in_system_include_dir (convert_filename)
3999 #endif /* defined (UNPROTOIZE) */
4001 fprintf (stderr, "%s: file `%s' not converted\n",
4002 pname, shortpath (NULL, convert_filename));
4006 /* Let the user know what we are up to. */
4009 fprintf (stderr, "%s: would convert file `%s'\n",
4010 pname, shortpath (NULL, convert_filename));
4012 fprintf (stderr, "%s: converting file `%s'\n",
4013 pname, shortpath (NULL, convert_filename));
4016 /* Find out the size (in bytes) of the original file. */
4018 /* The cast avoids an erroneous warning on AIX. */
4019 if (my_stat ((char *)convert_filename, &stat_buf) == -1)
4021 fprintf (stderr, "%s: error: can't get status for file `%s': %s\n",
4022 pname, shortpath (NULL, convert_filename), sys_errlist[errno]);
4025 orig_size = stat_buf.st_size;
4027 /* Allocate a buffer to hold the original text. */
4029 orig_text_base = new_orig_text_base = (char *) xmalloc (orig_size + 2);
4030 orig_text_limit = new_orig_text_limit = new_orig_text_base + orig_size;
4032 /* Allocate a buffer to hold the cleaned-up version of the original text. */
4034 clean_text_base = new_clean_text_base = (char *) xmalloc (orig_size + 2);
4035 clean_text_limit = new_clean_text_limit = new_clean_text_base + orig_size;
4036 clean_read_ptr = clean_text_base - 1;
4038 /* Allocate a buffer that will hopefully be large enough to hold the entire
4039 converted output text. As an initial guess for the maximum size of the
4040 output buffer, use 125% of the size of the original + some extra. This
4041 buffer can be expanded later as needed. */
4043 repl_size = orig_size + (orig_size >> 2) + 4096;
4044 repl_text_base = (char *) xmalloc (repl_size + 2);
4045 repl_text_limit = repl_text_base + repl_size - 1;
4046 repl_write_ptr = repl_text_base - 1;
4051 /* Open the file to be converted in READ ONLY mode. */
4053 if ((input_file = my_open (convert_filename, O_RDONLY, 0444)) == -1)
4055 fprintf (stderr, "%s: error: can't open file `%s' for reading: %s\n",
4056 pname, shortpath (NULL, convert_filename),
4057 sys_errlist[errno]);
4061 /* Read the entire original source text file into the original text buffer
4062 in one swell fwoop. Then figure out where the end of the text is and
4063 make sure that it ends with a newline followed by a null. */
4065 if (read (input_file, new_orig_text_base, orig_size) != orig_size)
4068 fprintf (stderr, "\n%s: error: while reading input file `%s': %s\n",
4069 pname, shortpath (NULL, convert_filename),
4070 sys_errlist[errno]);
4077 if (orig_size == 0 || orig_text_limit[-1] != '\n')
4079 *new_orig_text_limit++ = '\n';
4083 /* Create the cleaned up copy of the original text. */
4085 memcpy (new_clean_text_base, orig_text_base,
4086 (size_t) (orig_text_limit - orig_text_base));
4087 do_cleaning (new_clean_text_base, new_clean_text_limit);
4092 size_t clean_size = orig_text_limit - orig_text_base;
4093 char *const clean_filename = (char *) alloca (strlen (convert_filename) + 6 + 1);
4095 /* Open (and create) the clean file. */
4097 strcpy (clean_filename, convert_filename);
4098 strcat (clean_filename, ".clean");
4099 if ((clean_file = creat (clean_filename, 0666)) == -1)
4101 fprintf (stderr, "%s: error: can't create/open clean file `%s': %s\n",
4102 pname, shortpath (NULL, clean_filename),
4103 sys_errlist[errno]);
4107 /* Write the clean file. */
4109 if (write (clean_file, new_clean_text_base, clean_size) != clean_size)
4110 fprintf (stderr, "%s: error: while writing file `%s': %s\n",
4111 pname, shortpath (NULL, clean_filename), sys_errlist[errno]);
4117 /* Do a simplified scan of the input looking for things that were not
4118 mentioned in the aux info files because of the fact that they were
4119 in a region of the source which was preprocessed-out (via #if or
4122 scan_for_missed_items (file_p);
4124 /* Setup to do line-oriented forward seeking in the clean text buffer. */
4126 last_known_line_number = 1;
4127 last_known_line_start = clean_text_base;
4129 /* Now get down to business and make all of the necessary edits. */
4132 const def_dec_info *def_dec_p;
4134 first_definition_in_file = 1;
4135 def_dec_p = file_p->defs_decs;
4136 for (; def_dec_p; def_dec_p = def_dec_p->next_in_file)
4138 const char *clean_text_p = seek_to_line (def_dec_p->line);
4140 /* clean_text_p now points to the first character of the line which
4141 contains the `terminator' for the declaration or definition that
4142 we are about to process. */
4146 if (global_flag && def_dec_p->is_func_def && first_definition_in_file)
4148 add_global_decls (def_dec_p->file, clean_text_p);
4149 first_definition_in_file = 0;
4152 /* Don't edit this item if it is already in prototype format or if it
4153 is a function declaration and we have found no corresponding
4156 if (def_dec_p->prototyped
4157 || (!def_dec_p->is_func_def && !def_dec_p->definition))
4160 #endif /* !defined (UNPROTOIZE) */
4162 if (def_dec_p->is_func_def)
4163 edit_fn_definition (def_dec_p, clean_text_p);
4166 if (def_dec_p->is_implicit)
4167 add_local_decl (def_dec_p, clean_text_p);
4169 #endif /* !defined (UNPROTOIZE) */
4170 edit_fn_declaration (def_dec_p, clean_text_p);
4174 /* Finalize things. Output the last trailing part of the original text. */
4176 output_up_to (clean_text_limit - 1);
4178 /* If this is just a test run, stop now and just deallocate the buffers. */
4182 free (new_orig_text_base);
4183 free (new_clean_text_base);
4184 free (repl_text_base);
4188 /* Change the name of the original input file. This is just a quick way of
4189 saving the original file. */
4193 char *new_filename =
4194 (char *) xmalloc (strlen (convert_filename) + strlen (save_suffix) + 2);
4196 strcpy (new_filename, convert_filename);
4197 strcat (new_filename, save_suffix);
4198 if (my_link (convert_filename, new_filename) == -1)
4200 if (errno == EEXIST)
4203 fprintf (stderr, "%s: warning: file `%s' already saved in `%s'\n",
4205 shortpath (NULL, convert_filename),
4206 shortpath (NULL, new_filename));
4210 fprintf (stderr, "%s: error: can't link file `%s' to `%s': %s\n",
4212 shortpath (NULL, convert_filename),
4213 shortpath (NULL, new_filename),
4214 sys_errlist[errno]);
4220 if (my_unlink (convert_filename) == -1)
4222 fprintf (stderr, "%s: error: can't delete file `%s': %s\n",
4223 pname, shortpath (NULL, convert_filename), sys_errlist[errno]);
4230 /* Open (and create) the output file. */
4232 if ((output_file = creat (convert_filename, 0666)) == -1)
4234 fprintf (stderr, "%s: error: can't create/open output file `%s': %s\n",
4235 pname, shortpath (NULL, convert_filename),
4236 sys_errlist[errno]);
4240 /* Write the output file. */
4243 unsigned int out_size = (repl_write_ptr + 1) - repl_text_base;
4245 if (write (output_file, repl_text_base, out_size) != out_size)
4246 fprintf (stderr, "%s: error: while writing file `%s': %s\n",
4247 pname, shortpath (NULL, convert_filename),
4248 sys_errlist[errno]);
4251 close (output_file);
4254 /* Deallocate the conversion buffers. */
4256 free (new_orig_text_base);
4257 free (new_clean_text_base);
4258 free (repl_text_base);
4260 /* Change the mode of the output file to match the original file. */
4262 /* The cast avoids an erroneous warning on AIX. */
4263 if (my_chmod ((char *)convert_filename, stat_buf.st_mode) == -1)
4264 fprintf (stderr, "%s: error: can't change mode of file `%s': %s\n",
4265 pname, shortpath (NULL, convert_filename), sys_errlist[errno]);
4267 /* Note: We would try to change the owner and group of the output file
4268 to match those of the input file here, except that may not be a good
4269 thing to do because it might be misleading. Also, it might not even
4270 be possible to do that (on BSD systems with quotas for instance). */
4273 /* Do all of the individual steps needed to do the protoization (or
4274 unprotoization) of the files referenced in the aux_info files given
4275 in the command line. */
4280 const char * const *base_pp;
4281 const char * const * const end_pps
4282 = &base_source_filenames[n_base_source_files];
4286 #endif /* !defined (UNPROTOIZE) */
4288 /* One-by-one, check (and create if necessary), open, and read all of the
4289 stuff in each aux_info file. After reading each aux_info file, the
4290 aux_info_file just read will be automatically deleted unless the
4291 keep_flag is set. */
4293 for (base_pp = base_source_filenames; base_pp < end_pps; base_pp++)
4294 process_aux_info_file (*base_pp, keep_flag, 0);
4298 /* Also open and read the special SYSCALLS.c aux_info file which gives us
4299 the prototypes for all of the standard system-supplied functions. */
4301 if (nondefault_syscalls_dir)
4303 syscalls_absolute_filename
4304 = (char *) xmalloc (strlen (nondefault_syscalls_dir)
4305 + sizeof (syscalls_filename) + 1);
4306 strcpy (syscalls_absolute_filename, nondefault_syscalls_dir);
4310 syscalls_absolute_filename
4311 = (char *) xmalloc (strlen (default_syscalls_dir)
4312 + sizeof (syscalls_filename) + 1);
4313 strcpy (syscalls_absolute_filename, default_syscalls_dir);
4316 syscalls_len = strlen (syscalls_absolute_filename);
4317 if (*(syscalls_absolute_filename + syscalls_len - 1) != '/')
4319 *(syscalls_absolute_filename + syscalls_len++) = '/';
4320 *(syscalls_absolute_filename + syscalls_len) = '\0';
4322 strcat (syscalls_absolute_filename, syscalls_filename);
4324 /* Call process_aux_info_file in such a way that it does not try to
4325 delete the SYSCALLS aux_info file. */
4327 process_aux_info_file (syscalls_absolute_filename, 1, 1);
4329 #endif /* !defined (UNPROTOIZE) */
4331 /* When we first read in all of the information from the aux_info files
4332 we saved in it decending line number order, because that was likely to
4333 be faster. Now however, we want the chains of def & dec records to
4334 appear in ascending line number order as we get further away from the
4335 file_info record that they hang from. The following line causes all of
4336 these lists to be rearranged into ascending line number order. */
4338 visit_each_hash_node (filename_primary, reverse_def_dec_list);
4342 /* Now do the "real" work. The following line causes each declaration record
4343 to be "visited". For each of these nodes, an attempt is made to match
4344 up the function declaration with a corresponding function definition,
4345 which should have a full prototype-format formals list with it. Once
4346 these match-ups are made, the conversion of the function declarations
4347 to prototype format can be made. */
4349 visit_each_hash_node (function_name_primary, connect_defs_and_decs);
4351 #endif /* !defined (UNPROTOIZE) */
4353 /* Now convert each file that can be converted (and needs to be). */
4355 visit_each_hash_node (filename_primary, edit_file);
4359 /* If we are working in cplusplus mode, try to rename all .c files to .C
4360 files. Don't panic if some of the renames don't work. */
4362 if (cplusplus_flag && !nochange_flag)
4363 visit_each_hash_node (filename_primary, rename_c_file);
4365 #endif /* !defined (UNPROTOIZE) */
4368 static struct option longopts[] =
4370 {"version", 0, 0, 'V'},
4371 {"file_name", 0, 0, 'p'},
4372 {"quiet", 0, 0, 'q'},
4373 {"silent", 0, 0, 'q'},
4374 {"force", 0, 0, 'f'},
4375 {"keep", 0, 0, 'k'},
4376 {"nosave", 0, 0, 'N'},
4377 {"nochange", 0, 0, 'n'},
4378 {"compiler-options", 1, 0, 'c'},
4379 {"exclude", 1, 0, 'x'},
4380 {"directory", 1, 0, 'd'},
4382 {"indent", 1, 0, 'i'},
4384 {"local", 0, 0, 'l'},
4385 {"global", 0, 0, 'g'},
4387 {"syscalls-dir", 1, 0, 'B'},
4401 pname = strrchr (argv[0], '/');
4402 pname = pname ? pname+1 : argv[0];
4404 cwd_buffer = getpwd ();
4407 fprintf (stderr, "%s: cannot get working directory: %s\n",
4408 pname, sys_errlist[errno]);
4412 /* By default, convert the files in the current directory. */
4413 directory_list = string_list_cons (cwd_buffer, NULL);
4415 while ((c = getopt_long (argc, argv,
4419 "B:c:Cd:gklnNp:qVx:",
4421 longopts, &longind)) != EOF)
4423 if (c == 0) /* Long option. */
4424 c = longopts[longind].val;
4428 compiler_file_name = optarg;
4432 = string_list_cons (abspath (NULL, optarg), directory_list);
4435 exclude_list = string_list_cons (optarg, exclude_list);
4464 indent_string = optarg;
4466 #else /* !defined (UNPROTOIZE) */
4477 nondefault_syscalls_dir = optarg;
4479 #endif /* !defined (UNPROTOIZE) */
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 (const char **) 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;
4504 fprintf (stderr, "%s: input file names must have .c suffixes: %s\n",
4505 pname, shortpath (NULL, path));
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'. */
4517 for (cp = varargs_style_indicator; isalnum (*cp) || *cp == '_'; cp++)
4520 varargs_style_indicator = savestring (varargs_style_indicator,
4521 cp - varargs_style_indicator);
4523 #endif /* !defined (UNPROTOIZE) */
4530 fprintf (stderr, "%s: %s\n", pname, version_string);