* recog.c (preproces_constraints): Zero recog_op_alt before
[official-gcc.git] / gcc / protoize.c
blob7e9205cb1a821c4a305e30db50ed838816f94ab3
1 /* Protoize program - Original version by Ron Guilmette (rfg@segfault.us.com).
2 Copyright (C) 1989, 92-98, 1999 Free Software Foundation, Inc.
4 This file is part of GNU CC.
6 GNU CC is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2, or (at your option)
9 any later version.
11 GNU CC is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with GNU CC; see the file COPYING. If not, write to
18 the Free Software Foundation, 59 Temple Place - Suite 330,
19 Boston, MA 02111-1307, 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__)
26 #define __STDC__ 1
27 #endif /* defined(__cplusplus) && !defined(__STDC__) */
29 #if defined(__GNUC__) || defined (__GNUG__)
30 #define VOLATILE volatile
31 #else
32 #define VOLATILE
33 #endif
35 #ifndef __STDC__
36 #define const
37 #define volatile
38 #endif
40 #include "config.h"
42 #if 0
43 /* Users are not supposed to use _POSIX_SOURCE to say the
44 system is a POSIX system. That is not what _POSIX_SOURCE means! -- rms */
45 /* If the user asked for POSIX via _POSIX_SOURCE, turn on POSIX code. */
46 #if defined(_POSIX_SOURCE) && !defined(POSIX)
47 #define POSIX
48 #endif
49 #endif /* 0 */
51 #ifdef POSIX /* We should be able to define _POSIX_SOURCE unconditionally,
52 but some systems respond in buggy ways to it,
53 including SunOS 4.1.1. Which we don't classify as POSIX. */
54 /* In case this is a POSIX system with an ANSI C compiler,
55 ask for definition of all POSIX facilities. */
56 #undef _POSIX_SOURCE
57 #define _POSIX_SOURCE
58 #endif
60 #include "system.h"
61 #include "intl.h"
62 #undef abort
64 #if ! defined (_WIN32) || defined (__CYGWIN__) || defined (_UWIN)
65 #if defined(POSIX) || defined(CONCURRENT)
66 #include <dirent.h>
67 #else
68 #include <sys/dir.h>
69 #endif
70 #endif
71 #include <setjmp.h>
73 /* Some systems like Linux don't declare rindex if _POSIX_SOURCE is declared,
74 but it normally does declare it. This means that configure thinks we don't
75 need to declare it. Favor using strrchr if it is available. */
77 #ifndef strrchr
78 #ifndef HAVE_STRRCHR
79 #ifdef HAVE_RINDEX
80 #define strrchr rindex
81 #endif
82 #endif
83 #endif
85 /* Include getopt.h for the sake of getopt_long.
86 We don't need the declaration of getopt, and it could conflict
87 with something from a system header file, so effectively nullify that. */
88 #define getopt getopt_loser
89 #include "getopt.h"
90 #undef getopt
92 extern char *version_string;
94 /* Systems which are compatible only with POSIX 1003.1-1988 (but *not*
95 with POSIX 1003.1-1990), e.g. Ultrix 4.2, might not have
96 const qualifiers in the prototypes in the system include files.
97 Unfortunately, this can lead to GCC issuing lots of warnings for
98 calls to the following functions. To eliminate these warnings we
99 provide the following #defines. */
101 #define my_access(file,flag) access((char *)file, flag)
102 #define my_stat(file,pkt) stat((char *)file, pkt)
103 #ifdef __MINGW32__
104 #define my_link(file1, file2) -1
105 #else
106 #define my_link(file1, file2) link((char *)file1, (char *)file2)
107 #endif
108 #define my_unlink(file) unlink((char *)file)
109 #define my_open(file, mode, flag) open((char *)file, mode, flag)
110 #define my_chmod(file, mode) chmod((char *)file, mode)
112 extern char *getpwd ();
114 static void usage PROTO ((void)) ATTRIBUTE_NORETURN;
115 static void aux_info_corrupted PROTO ((void)) ATTRIBUTE_NORETURN;
116 static void declare_source_confusing PROTO ((const char *)) ATTRIBUTE_NORETURN;
118 /* Aliases for pointers to void.
119 These were made to facilitate compilation with old brain-dead DEC C
120 compilers which didn't properly grok `void*' types. */
122 typedef PTR pointer_type;
123 typedef const PTR const_pointer_type;
125 #if defined(POSIX)
127 #include <signal.h>
129 #else /* !defined(POSIX) */
131 /* Declaring stat or __flsbuf with a prototype
132 causes conflicts with system headers on some systems. */
134 #if 0 /* These conflict with stdio.h on some systems. */
135 extern int creat ();
136 extern int fprintf (FILE *, const char *, ...);
137 extern int printf (const char *, ...);
138 extern int open (const char *, int, ...);
139 extern int read ();
140 extern int write ();
141 #endif /* 0 */
142 extern int close ();
143 extern int fflush ();
144 extern int atoi ();
145 extern int puts ();
146 #ifndef fputs /* This may have been #defined by "system.h". */
147 extern int fputs ();
148 #endif
149 #ifndef fputc /* some systems define this as a macro. */
150 extern int fputc ();
151 #endif
152 extern int unlink ();
153 extern int access ();
155 #if 0 /* size_t from sys/types.h may fail to match GCC.
156 If so, we would get a warning from this. */
157 extern size_t strlen ()
158 #endif
160 #endif /* !defined (POSIX) */
162 /* Look for these where the `const' qualifier is intentionally cast aside. */
164 #define NONCONST
166 /* Define a default place to find the SYSCALLS.X file. */
168 #ifndef STD_PROTO_DIR
169 #define STD_PROTO_DIR "/usr/local/lib"
170 #endif /* !defined (STD_PROTO_DIR) */
172 /* Suffix of aux_info files. */
174 static const char * const aux_info_suffix = ".X";
176 /* String to attach to filenames for saved versions of original files. */
178 static const char * const save_suffix = ".save";
180 #ifndef UNPROTOIZE
182 /* File name of the file which contains descriptions of standard system
183 routines. Note that we never actually do anything with this file per se,
184 but we do read in its corresponding aux_info file. */
186 static const char syscalls_filename[] = "SYSCALLS.c";
188 /* Default place to find the above file. */
190 static const char * const default_syscalls_dir = STD_PROTO_DIR;
192 /* Variable to hold the complete absolutized filename of the SYSCALLS.c.X
193 file. */
195 static char * syscalls_absolute_filename;
197 #endif /* !defined (UNPROTOIZE) */
199 /* Type of the structure that holds information about macro unexpansions. */
201 struct unexpansion_struct {
202 const char *expanded;
203 const char *contracted;
205 typedef struct unexpansion_struct unexpansion;
207 /* A table of conversions that may need to be made for some (stupid) older
208 operating systems where these types are preprocessor macros rather than
209 typedefs (as they really ought to be).
211 WARNING: The contracted forms must be as small (or smaller) as the
212 expanded forms, or else havoc will ensue. */
214 static const unexpansion unexpansions[] = {
215 { "struct _iobuf", "FILE" },
216 { 0, 0 }
219 /* The number of "primary" slots in the hash tables for filenames and for
220 function names. This can be as big or as small as you like, except that
221 it must be a power of two. */
223 #define HASH_TABLE_SIZE (1 << 9)
225 /* Bit mask to use when computing hash values. */
227 static const int hash_mask = (HASH_TABLE_SIZE - 1);
229 /* Make a table of default system include directories
230 just as it is done in cccp.c. */
232 #ifndef STANDARD_INCLUDE_DIR
233 #define STANDARD_INCLUDE_DIR "/usr/include"
234 #endif
236 #ifndef LOCAL_INCLUDE_DIR
237 #define LOCAL_INCLUDE_DIR "/usr/local/include"
238 #endif
240 struct default_include { const char *fname;
241 const char *component;
242 int x1, x2; } include_defaults[]
243 #ifdef INCLUDE_DEFAULTS
244 = INCLUDE_DEFAULTS;
245 #else
247 /* Pick up GNU C++ specific include files. */
248 { GPLUSPLUS_INCLUDE_DIR, "G++", 1, 1 },
249 #ifdef CROSS_COMPILE
250 /* This is the dir for fixincludes. Put it just before
251 the files that we fix. */
252 { GCC_INCLUDE_DIR, "GCC", 0, 0 },
253 /* For cross-compilation, this dir name is generated
254 automatically in Makefile.in. */
255 { CROSS_INCLUDE_DIR, 0, 0, 0 },
256 /* This is another place that the target system's headers might be. */
257 { TOOL_INCLUDE_DIR, "BINUTILS", 0, 0 },
258 #else /* not CROSS_COMPILE */
259 /* This should be /use/local/include and should come before
260 the fixincludes-fixed header files. */
261 { LOCAL_INCLUDE_DIR, 0, 0, 1 },
262 /* This is here ahead of GCC_INCLUDE_DIR because assert.h goes here.
263 Likewise, behind LOCAL_INCLUDE_DIR, where glibc puts its assert.h. */
264 { TOOL_INCLUDE_DIR, "BINUTILS", 0, 0 },
265 /* This is the dir for fixincludes. Put it just before
266 the files that we fix. */
267 { GCC_INCLUDE_DIR, "GCC", 0, 0 },
268 /* Some systems have an extra dir of include files. */
269 #ifdef SYSTEM_INCLUDE_DIR
270 { SYSTEM_INCLUDE_DIR, 0, 0, 0 },
271 #endif
272 { STANDARD_INCLUDE_DIR, 0, 0, 0},
273 #endif /* not CROSS_COMPILE */
274 { 0, 0, 0, 0}
276 #endif /* no INCLUDE_DEFAULTS */
278 /* Datatype for lists of directories or filenames. */
279 struct string_list
281 char *name;
282 struct string_list *next;
285 /* List of directories in which files should be converted. */
287 struct string_list *directory_list;
289 /* List of file names which should not be converted.
290 A file is excluded if the end of its name, following a /,
291 matches one of the names in this list. */
293 struct string_list *exclude_list;
295 /* The name of the other style of variable-number-of-parameters functions
296 (i.e. the style that we want to leave unconverted because we don't yet
297 know how to convert them to this style. This string is used in warning
298 messages. */
300 /* Also define here the string that we can search for in the parameter lists
301 taken from the .X files which will unambiguously indicate that we have
302 found a varargs style function. */
304 #ifdef UNPROTOIZE
305 static const char * const other_var_style = "stdarg";
306 #else /* !defined (UNPROTOIZE) */
307 static const char * const other_var_style = "varargs";
308 /* Note that this is a string containing the expansion of va_alist.
309 But in `main' we discard all but the first token. */
310 static const char *varargs_style_indicator = STRINGIFY (va_alist);
311 #endif /* !defined (UNPROTOIZE) */
313 /* The following two types are used to create hash tables. In this program,
314 there are two hash tables which are used to store and quickly lookup two
315 different classes of strings. The first type of strings stored in the
316 first hash table are absolute filenames of files which protoize needs to
317 know about. The second type of strings (stored in the second hash table)
318 are function names. It is this second class of strings which really
319 inspired the use of the hash tables, because there may be a lot of them. */
321 typedef struct hash_table_entry_struct hash_table_entry;
323 /* Do some typedefs so that we don't have to write "struct" so often. */
325 typedef struct def_dec_info_struct def_dec_info;
326 typedef struct file_info_struct file_info;
327 typedef struct f_list_chain_item_struct f_list_chain_item;
329 /* In the struct below, note that the "_info" field has two different uses
330 depending on the type of hash table we are in (i.e. either the filenames
331 hash table or the function names hash table). In the filenames hash table
332 the info fields of the entries point to the file_info struct which is
333 associated with each filename (1 per filename). In the function names
334 hash table, the info field points to the head of a singly linked list of
335 def_dec_info entries which are all defs or decs of the function whose
336 name is pointed to by the "symbol" field. Keeping all of the defs/decs
337 for a given function name on a special list specifically for that function
338 name makes it quick and easy to find out all of the important information
339 about a given (named) function. */
341 struct hash_table_entry_struct {
342 hash_table_entry * hash_next; /* -> to secondary entries */
343 const char * symbol; /* -> to the hashed string */
344 union {
345 const def_dec_info * _ddip;
346 file_info * _fip;
347 } _info;
349 #define ddip _info._ddip
350 #define fip _info._fip
352 /* Define a type specifically for our two hash tables. */
354 typedef hash_table_entry hash_table[HASH_TABLE_SIZE];
356 /* The following struct holds all of the important information about any
357 single filename (e.g. file) which we need to know about. */
359 struct file_info_struct {
360 const hash_table_entry * hash_entry; /* -> to associated hash entry */
361 const def_dec_info * defs_decs; /* -> to chain of defs/decs */
362 time_t mtime; /* Time of last modification. */
365 /* Due to the possibility that functions may return pointers to functions,
366 (which may themselves have their own parameter lists) and due to the
367 fact that returned pointers-to-functions may be of type "pointer-to-
368 function-returning-pointer-to-function" (ad nauseum) we have to keep
369 an entire chain of ANSI style formal parameter lists for each function.
371 Normally, for any given function, there will only be one formals list
372 on the chain, but you never know.
374 Note that the head of each chain of formals lists is pointed to by the
375 `f_list_chain' field of the corresponding def_dec_info record.
377 For any given chain, the item at the head of the chain is the *leftmost*
378 parameter list seen in the actual C language function declaration. If
379 there are other members of the chain, then these are linked in left-to-right
380 order from the head of the chain. */
382 struct f_list_chain_item_struct {
383 const f_list_chain_item * chain_next; /* -> to next item on chain */
384 const char * formals_list; /* -> to formals list string */
387 /* The following struct holds all of the important information about any
388 single function definition or declaration which we need to know about.
389 Note that for unprotoize we don't need to know very much because we
390 never even create records for stuff that we don't intend to convert
391 (like for instance defs and decs which are already in old K&R format
392 and "implicit" function declarations). */
394 struct def_dec_info_struct {
395 const def_dec_info * next_in_file; /* -> to rest of chain for file */
396 file_info * file; /* -> file_info for containing file */
397 int line; /* source line number of def/dec */
398 const char * ansi_decl; /* -> left end of ansi decl */
399 hash_table_entry * hash_entry; /* -> hash entry for function name */
400 unsigned int is_func_def; /* = 0 means this is a declaration */
401 const def_dec_info * next_for_func; /* -> to rest of chain for func name */
402 unsigned int f_list_count; /* count of formals lists we expect */
403 char prototyped; /* = 0 means already prototyped */
404 #ifndef UNPROTOIZE
405 const f_list_chain_item * f_list_chain; /* -> chain of formals lists */
406 const def_dec_info * definition; /* -> def/dec containing related def */
407 char is_static; /* = 0 means visibility is "extern" */
408 char is_implicit; /* != 0 for implicit func decl's */
409 char written; /* != 0 means written for implicit */
410 #else /* !defined (UNPROTOIZE) */
411 const char * formal_names; /* -> to list of names of formals */
412 const char * formal_decls; /* -> to string of formal declarations */
413 #endif /* !defined (UNPROTOIZE) */
416 /* Pointer to the tail component of the filename by which this program was
417 invoked. Used everywhere in error and warning messages. */
419 static const char *pname;
421 /* Error counter. Will be non-zero if we should give up at the next convenient
422 stopping point. */
424 static int errors = 0;
426 /* Option flags. */
427 /* ??? These comments should say what the flag mean as well as the options
428 that set them. */
430 /* File name to use for running gcc. Allows GCC 2 to be named
431 something other than gcc. */
432 static const char *compiler_file_name = "gcc";
434 static int version_flag = 0; /* Print our version number. */
435 static int quiet_flag = 0; /* Don't print messages normally. */
436 static int nochange_flag = 0; /* Don't convert, just say what files
437 we would have converted. */
438 static int nosave_flag = 0; /* Don't save the old version. */
439 static int keep_flag = 0; /* Don't delete the .X files. */
440 static const char ** compile_params = 0; /* Option string for gcc. */
441 #ifdef UNPROTOIZE
442 static const char *indent_string = " "; /* Indentation for newly
443 inserted parm decls. */
444 #else /* !defined (UNPROTOIZE) */
445 static int local_flag = 0; /* Insert new local decls (when?). */
446 static int global_flag = 0; /* set by -g option */
447 static int cplusplus_flag = 0; /* Rename converted files to *.C. */
448 static const char *nondefault_syscalls_dir = 0; /* Dir to look for
449 SYSCALLS.c.X in. */
450 #endif /* !defined (UNPROTOIZE) */
452 /* An index into the compile_params array where we should insert the source
453 file name when we are ready to exec the C compiler. A zero value indicates
454 that we have not yet called munge_compile_params. */
456 static int input_file_name_index = 0;
458 /* An index into the compile_params array where we should insert the filename
459 for the aux info file, when we run the C compiler. */
460 static int aux_info_file_name_index = 0;
462 /* Count of command line arguments which were "filename" arguments. */
464 static int n_base_source_files = 0;
466 /* Points to a malloc'ed list of pointers to all of the filenames of base
467 source files which were specified on the command line. */
469 static const char **base_source_filenames;
471 /* Line number of the line within the current aux_info file that we
472 are currently processing. Used for error messages in case the prototypes
473 info file is corrupted somehow. */
475 static int current_aux_info_lineno;
477 /* Pointer to the name of the source file currently being converted. */
479 static const char *convert_filename;
481 /* Pointer to relative root string (taken from aux_info file) which indicates
482 where directory the user was in when he did the compilation step that
483 produced the containing aux_info file. */
485 static const char *invocation_filename;
487 /* Pointer to the base of the input buffer that holds the original text for the
488 source file currently being converted. */
490 static const char *orig_text_base;
492 /* Pointer to the byte just beyond the end of the input buffer that holds the
493 original text for the source file currently being converted. */
495 static const char *orig_text_limit;
497 /* Pointer to the base of the input buffer that holds the cleaned text for the
498 source file currently being converted. */
500 static const char *clean_text_base;
502 /* Pointer to the byte just beyond the end of the input buffer that holds the
503 cleaned text for the source file currently being converted. */
505 static const char *clean_text_limit;
507 /* Pointer to the last byte in the cleaned text buffer that we have already
508 (virtually) copied to the output buffer (or decided to ignore). */
510 static const char * clean_read_ptr;
512 /* Pointer to the base of the output buffer that holds the replacement text
513 for the source file currently being converted. */
515 static char *repl_text_base;
517 /* Pointer to the byte just beyond the end of the output buffer that holds the
518 replacement text for the source file currently being converted. */
520 static char *repl_text_limit;
522 /* Pointer to the last byte which has been stored into the output buffer.
523 The next byte to be stored should be stored just past where this points
524 to. */
526 static char * repl_write_ptr;
528 /* Pointer into the cleaned text buffer for the source file we are currently
529 converting. This points to the first character of the line that we last
530 did a "seek_to_line" to (see below). */
532 static const char *last_known_line_start;
534 /* Number of the line (in the cleaned text buffer) that we last did a
535 "seek_to_line" to. Will be one if we just read a new source file
536 into the cleaned text buffer. */
538 static int last_known_line_number;
540 /* The filenames hash table. */
542 static hash_table filename_primary;
544 /* The function names hash table. */
546 static hash_table function_name_primary;
548 /* The place to keep the recovery address which is used only in cases where
549 we get hopelessly confused by something in the cleaned original text. */
551 static jmp_buf source_confusion_recovery;
553 /* A pointer to the current directory filename (used by abspath). */
555 static char *cwd_buffer;
557 /* A place to save the read pointer until we are sure that an individual
558 attempt at editing will succeed. */
560 static const char * saved_clean_read_ptr;
562 /* A place to save the write pointer until we are sure that an individual
563 attempt at editing will succeed. */
565 static char * saved_repl_write_ptr;
567 /* Forward declaration. */
569 static const char *shortpath ();
571 /* Translate and output an error message. */
572 static void notice PVPROTO ((const char *, ...))
573 ATTRIBUTE_PRINTF_1;
574 static void
575 notice VPROTO ((const char *msgid, ...))
577 #ifndef ANSI_PROTOTYPES
578 const char *msgid;
579 #endif
580 va_list ap;
582 VA_START (ap, msgid);
584 #ifndef ANSI_PROTOTYPES
585 msgid = va_arg (ap, const char *);
586 #endif
588 vfprintf (stderr, _(msgid), ap);
589 va_end (ap);
593 char *
594 xstrerror(e)
595 int e;
598 #ifdef HAVE_STRERROR
599 return strerror(e);
601 #else
602 if (!e)
603 return "";
605 if (e > 0 && e < sys_nerr)
606 return sys_errlist[e];
608 return "errno = ?";
609 #endif
612 /* Allocate some space, but check that the allocation was successful. */
613 /* alloca.c uses this, so don't make it static. */
615 pointer_type
616 xmalloc (byte_count)
617 size_t byte_count;
619 register pointer_type rv = (pointer_type) malloc (byte_count);
620 if (rv == NULL)
622 notice ("\n%s: virtual memory exceeded\n", pname);
623 exit (FATAL_EXIT_CODE);
625 return rv;
628 /* Reallocate some space, but check that the reallocation was successful. */
630 pointer_type
631 xrealloc (old_space, byte_count)
632 pointer_type old_space;
633 size_t byte_count;
635 register pointer_type rv;
636 if (old_space)
637 rv = (pointer_type) realloc (old_space, byte_count);
638 else
639 rv = (pointer_type) malloc (byte_count);
640 if (rv == NULL)
642 notice ("\n%s: virtual memory exceeded\n", pname);
643 exit (FATAL_EXIT_CODE);
645 return rv;
648 /* Deallocate the area pointed to by an arbitrary pointer, but first, strip
649 the `const' qualifier from it and also make sure that the pointer value
650 is non-null. */
652 void
653 xfree (p)
654 const_pointer_type p;
656 if (p)
657 free ((NONCONST pointer_type) p);
660 /* Make a copy of a string INPUT with size SIZE. */
662 static char *
663 savestring (input, size)
664 const char *input;
665 unsigned int size;
667 char *output = (char *) xmalloc (size + 1);
668 strcpy (output, input);
669 return output;
672 /* Make a copy of the concatenation of INPUT1 and INPUT2. */
674 static char *
675 savestring2 (input1, size1, input2, size2)
676 const char *input1;
677 unsigned int size1;
678 const char *input2;
679 unsigned int size2;
681 char *output = (char *) xmalloc (size1 + size2 + 1);
682 strcpy (output, input1);
683 strcpy (&output[size1], input2);
684 return output;
687 /* More 'friendly' abort that prints the line and file.
688 config.h can #define abort fancy_abort if you like that sort of thing. */
690 void
691 fancy_abort ()
693 notice ("%s: internal abort\n", pname);
694 exit (FATAL_EXIT_CODE);
697 /* Make a duplicate of the first N bytes of a given string in a newly
698 allocated area. */
700 static char *
701 dupnstr (s, n)
702 const char *s;
703 size_t n;
705 char *ret_val = (char *) xmalloc (n + 1);
707 strncpy (ret_val, s, n);
708 ret_val[n] = '\0';
709 return ret_val;
712 /* Return a pointer to the first occurrence of s2 within s1 or NULL if s2
713 does not occur within s1. Assume neither s1 nor s2 are null pointers. */
715 static const char *
716 substr (s1, s2)
717 const char *s1;
718 const char *const s2;
720 for (; *s1 ; s1++)
722 const char *p1;
723 const char *p2;
724 int c;
726 for (p1 = s1, p2 = s2; (c = *p2); p1++, p2++)
727 if (*p1 != c)
728 goto outer;
729 return s1;
730 outer:
733 return 0;
736 /* Read LEN bytes at PTR from descriptor DESC, for file FILENAME,
737 retrying if necessary. Return the actual number of bytes read. */
739 static int
740 safe_read (desc, ptr, len)
741 int desc;
742 char *ptr;
743 int len;
745 int left = len;
746 while (left > 0) {
747 int nchars = read (desc, ptr, left);
748 if (nchars < 0)
750 #ifdef EINTR
751 if (errno == EINTR)
752 continue;
753 #endif
754 return nchars;
756 if (nchars == 0)
757 break;
758 ptr += nchars;
759 left -= nchars;
761 return len - left;
764 /* Write LEN bytes at PTR to descriptor DESC,
765 retrying if necessary, and treating any real error as fatal. */
767 static void
768 safe_write (desc, ptr, len, out_fname)
769 int desc;
770 char *ptr;
771 int len;
772 char *out_fname;
774 while (len > 0) {
775 int written = write (desc, ptr, len);
776 if (written < 0)
778 int errno_val = errno;
779 #ifdef EINTR
780 if (errno_val == EINTR)
781 continue;
782 #endif
783 notice ("%s: error writing file `%s': %s\n",
784 pname, shortpath (NULL, out_fname), xstrerror (errno_val));
785 return;
787 ptr += written;
788 len -= written;
792 /* Get setup to recover in case the edit we are about to do goes awry. */
794 void
795 save_pointers ()
797 saved_clean_read_ptr = clean_read_ptr;
798 saved_repl_write_ptr = repl_write_ptr;
801 /* Call this routine to recover our previous state whenever something looks
802 too confusing in the source code we are trying to edit. */
804 void
805 restore_pointers ()
807 clean_read_ptr = saved_clean_read_ptr;
808 repl_write_ptr = saved_repl_write_ptr;
811 /* Return true if the given character is a valid identifier character. */
813 static int
814 is_id_char (ch)
815 unsigned char ch;
817 return (ISALNUM (ch) || (ch == '_') || (ch == '$'));
820 /* Give a message indicating the proper way to invoke this program and then
821 exit with non-zero status. */
823 static void
824 usage ()
826 #ifdef UNPROTOIZE
827 notice ("%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ filename ... ]'\n",
828 pname, pname);
829 #else /* !defined (UNPROTOIZE) */
830 notice ("%s: usage '%s [ -VqfnkNlgC ] [ -B <dirname> ] [ filename ... ]'\n",
831 pname, pname);
832 #endif /* !defined (UNPROTOIZE) */
833 exit (FATAL_EXIT_CODE);
836 /* Return true if the given filename (assumed to be an absolute filename)
837 designates a file residing anywhere beneath any one of the "system"
838 include directories. */
840 static int
841 in_system_include_dir (path)
842 const char *path;
844 struct default_include *p;
846 if (path[0] != '/')
847 abort (); /* Must be an absolutized filename. */
849 for (p = include_defaults; p->fname; p++)
850 if (!strncmp (path, p->fname, strlen (p->fname))
851 && path[strlen (p->fname)] == '/')
852 return 1;
853 return 0;
856 #if 0
857 /* Return true if the given filename designates a file that the user has
858 read access to and for which the user has write access to the containing
859 directory. */
861 static int
862 file_could_be_converted (const char *path)
864 char *const dir_name = (char *) alloca (strlen (path) + 1);
866 if (my_access (path, R_OK))
867 return 0;
870 char *dir_last_slash;
872 strcpy (dir_name, path);
873 dir_last_slash = strrchr (dir_name, '/');
874 if (dir_last_slash)
875 *dir_last_slash = '\0';
876 else
877 abort (); /* Should have been an absolutized filename. */
880 if (my_access (path, W_OK))
881 return 0;
883 return 1;
886 /* Return true if the given filename designates a file that we are allowed
887 to modify. Files which we should not attempt to modify are (a) "system"
888 include files, and (b) files which the user doesn't have write access to,
889 and (c) files which reside in directories which the user doesn't have
890 write access to. Unless requested to be quiet, give warnings about
891 files that we will not try to convert for one reason or another. An
892 exception is made for "system" include files, which we never try to
893 convert and for which we don't issue the usual warnings. */
895 static int
896 file_normally_convertible (const char *path)
898 char *const dir_name = alloca (strlen (path) + 1);
900 if (in_system_include_dir (path))
901 return 0;
904 char *dir_last_slash;
906 strcpy (dir_name, path);
907 dir_last_slash = strrchr (dir_name, '/');
908 if (dir_last_slash)
909 *dir_last_slash = '\0';
910 else
911 abort (); /* Should have been an absolutized filename. */
914 if (my_access (path, R_OK))
916 if (!quiet_flag)
917 notice ("%s: warning: no read access for file `%s'\n",
918 pname, shortpath (NULL, path));
919 return 0;
922 if (my_access (path, W_OK))
924 if (!quiet_flag)
925 notice ("%s: warning: no write access for file `%s'\n",
926 pname, shortpath (NULL, path));
927 return 0;
930 if (my_access (dir_name, W_OK))
932 if (!quiet_flag)
933 notice ("%s: warning: no write access for dir containing `%s'\n",
934 pname, shortpath (NULL, path));
935 return 0;
938 return 1;
940 #endif /* 0 */
942 #ifndef UNPROTOIZE
944 /* Return true if the given file_info struct refers to the special SYSCALLS.c.X
945 file. Return false otherwise. */
947 static int
948 is_syscalls_file (fi_p)
949 const file_info *fi_p;
951 char const *f = fi_p->hash_entry->symbol;
952 size_t fl = strlen (f), sysl = sizeof (syscalls_filename) - 1;
953 return sysl <= fl && strcmp (f + fl - sysl, syscalls_filename) == 0;
956 #endif /* !defined (UNPROTOIZE) */
958 /* Check to see if this file will need to have anything done to it on this
959 run. If there is nothing in the given file which both needs conversion
960 and for which we have the necessary stuff to do the conversion, return
961 false. Otherwise, return true.
963 Note that (for protoize) it is only valid to call this function *after*
964 the connections between declarations and definitions have all been made
965 by connect_defs_and_decs. */
967 static int
968 needs_to_be_converted (file_p)
969 const file_info *file_p;
971 const def_dec_info *ddp;
973 #ifndef UNPROTOIZE
975 if (is_syscalls_file (file_p))
976 return 0;
978 #endif /* !defined (UNPROTOIZE) */
980 for (ddp = file_p->defs_decs; ddp; ddp = ddp->next_in_file)
982 if (
984 #ifndef UNPROTOIZE
986 /* ... and if we a protoizing and this function is in old style ... */
987 !ddp->prototyped
988 /* ... and if this a definition or is a decl with an associated def ... */
989 && (ddp->is_func_def || (!ddp->is_func_def && ddp->definition))
991 #else /* defined (UNPROTOIZE) */
993 /* ... and if we are unprotoizing and this function is in new style ... */
994 ddp->prototyped
996 #endif /* defined (UNPROTOIZE) */
998 /* ... then the containing file needs converting. */
999 return -1;
1000 return 0;
1003 /* Return 1 if the file name NAME is in a directory
1004 that should be converted. */
1006 static int
1007 directory_specified_p (name)
1008 const char *name;
1010 struct string_list *p;
1012 for (p = directory_list; p; p = p->next)
1013 if (!strncmp (name, p->name, strlen (p->name))
1014 && name[strlen (p->name)] == '/')
1016 const char *q = name + strlen (p->name) + 1;
1018 /* If there are more slashes, it's in a subdir, so
1019 this match doesn't count. */
1020 while (*q)
1021 if (*q++ == '/')
1022 goto lose;
1023 return 1;
1025 lose: ;
1028 return 0;
1031 /* Return 1 if the file named NAME should be excluded from conversion. */
1033 static int
1034 file_excluded_p (name)
1035 const char *name;
1037 struct string_list *p;
1038 int len = strlen (name);
1040 for (p = exclude_list; p; p = p->next)
1041 if (!strcmp (name + len - strlen (p->name), p->name)
1042 && name[len - strlen (p->name) - 1] == '/')
1043 return 1;
1045 return 0;
1048 /* Construct a new element of a string_list.
1049 STRING is the new element value, and REST holds the remaining elements. */
1051 static struct string_list *
1052 string_list_cons (string, rest)
1053 char *string;
1054 struct string_list *rest;
1056 struct string_list *temp
1057 = (struct string_list *) xmalloc (sizeof (struct string_list));
1059 temp->next = rest;
1060 temp->name = string;
1061 return temp;
1064 /* ??? The GNU convention for mentioning function args in its comments
1065 is to capitalize them. So change "hash_tab_p" to HASH_TAB_P below.
1066 Likewise for all the other functions. */
1068 /* Given a hash table, apply some function to each node in the table. The
1069 table to traverse is given as the "hash_tab_p" argument, and the
1070 function to be applied to each node in the table is given as "func"
1071 argument. */
1073 static void
1074 visit_each_hash_node (hash_tab_p, func)
1075 const hash_table_entry *hash_tab_p;
1076 void (*func)();
1078 const hash_table_entry *primary;
1080 for (primary = hash_tab_p; primary < &hash_tab_p[HASH_TABLE_SIZE]; primary++)
1081 if (primary->symbol)
1083 hash_table_entry *second;
1085 (*func)(primary);
1086 for (second = primary->hash_next; second; second = second->hash_next)
1087 (*func) (second);
1091 /* Initialize all of the fields of a new hash table entry, pointed
1092 to by the "p" parameter. Note that the space to hold the entry
1093 is assumed to have already been allocated before this routine is
1094 called. */
1096 static hash_table_entry *
1097 add_symbol (p, s)
1098 hash_table_entry *p;
1099 const char *s;
1101 p->hash_next = NULL;
1102 p->symbol = savestring (s, strlen (s));
1103 p->ddip = NULL;
1104 p->fip = NULL;
1105 return p;
1108 /* Look for a particular function name or filename in the particular
1109 hash table indicated by "hash_tab_p". If the name is not in the
1110 given hash table, add it. Either way, return a pointer to the
1111 hash table entry for the given name. */
1113 static hash_table_entry *
1114 lookup (hash_tab_p, search_symbol)
1115 hash_table_entry *hash_tab_p;
1116 const char *search_symbol;
1118 int hash_value = 0;
1119 const char *search_symbol_char_p = search_symbol;
1120 hash_table_entry *p;
1122 while (*search_symbol_char_p)
1123 hash_value += *search_symbol_char_p++;
1124 hash_value &= hash_mask;
1125 p = &hash_tab_p[hash_value];
1126 if (! p->symbol)
1127 return add_symbol (p, search_symbol);
1128 if (!strcmp (p->symbol, search_symbol))
1129 return p;
1130 while (p->hash_next)
1132 p = p->hash_next;
1133 if (!strcmp (p->symbol, search_symbol))
1134 return p;
1136 p->hash_next = (hash_table_entry *) xmalloc (sizeof (hash_table_entry));
1137 p = p->hash_next;
1138 return add_symbol (p, search_symbol);
1141 /* Throw a def/dec record on the junk heap.
1143 Also, since we are not using this record anymore, free up all of the
1144 stuff it pointed to. */
1146 static void
1147 free_def_dec (p)
1148 def_dec_info *p;
1150 xfree (p->ansi_decl);
1152 #ifndef UNPROTOIZE
1154 const f_list_chain_item * curr;
1155 const f_list_chain_item * next;
1157 for (curr = p->f_list_chain; curr; curr = next)
1159 next = curr->chain_next;
1160 xfree (curr);
1163 #endif /* !defined (UNPROTOIZE) */
1165 xfree (p);
1168 /* Unexpand as many macro symbol as we can find.
1170 If the given line must be unexpanded, make a copy of it in the heap and
1171 return a pointer to the unexpanded copy. Otherwise return NULL. */
1173 static char *
1174 unexpand_if_needed (aux_info_line)
1175 const char *aux_info_line;
1177 static char *line_buf = 0;
1178 static int line_buf_size = 0;
1179 const unexpansion *unexp_p;
1180 int got_unexpanded = 0;
1181 const char *s;
1182 char *copy_p = line_buf;
1184 if (line_buf == 0)
1186 line_buf_size = 1024;
1187 line_buf = (char *) xmalloc (line_buf_size);
1190 copy_p = line_buf;
1192 /* Make a copy of the input string in line_buf, expanding as necessary. */
1194 for (s = aux_info_line; *s != '\n'; )
1196 for (unexp_p = unexpansions; unexp_p->expanded; unexp_p++)
1198 const char *in_p = unexp_p->expanded;
1199 size_t len = strlen (in_p);
1201 if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
1203 int size = strlen (unexp_p->contracted);
1204 got_unexpanded = 1;
1205 if (copy_p + size - line_buf >= line_buf_size)
1207 int offset = copy_p - line_buf;
1208 line_buf_size *= 2;
1209 line_buf_size += size;
1210 line_buf = (char *) xrealloc (line_buf, line_buf_size);
1211 copy_p = line_buf + offset;
1213 strcpy (copy_p, unexp_p->contracted);
1214 copy_p += size;
1216 /* Assume the there will not be another replacement required
1217 within the text just replaced. */
1219 s += len;
1220 goto continue_outer;
1223 if (copy_p - line_buf == line_buf_size)
1225 int offset = copy_p - line_buf;
1226 line_buf_size *= 2;
1227 line_buf = (char *) xrealloc (line_buf, line_buf_size);
1228 copy_p = line_buf + offset;
1230 *copy_p++ = *s++;
1231 continue_outer: ;
1233 if (copy_p + 2 - line_buf >= line_buf_size)
1235 int offset = copy_p - line_buf;
1236 line_buf_size *= 2;
1237 line_buf = (char *) xrealloc (line_buf, line_buf_size);
1238 copy_p = line_buf + offset;
1240 *copy_p++ = '\n';
1241 *copy_p = '\0';
1243 return (got_unexpanded ? savestring (line_buf, copy_p - line_buf) : 0);
1246 /* Return the absolutized filename for the given relative
1247 filename. Note that if that filename is already absolute, it may
1248 still be returned in a modified form because this routine also
1249 eliminates redundant slashes and single dots and eliminates double
1250 dots to get a shortest possible filename from the given input
1251 filename. The absolutization of relative filenames is made by
1252 assuming that the given filename is to be taken as relative to
1253 the first argument (cwd) or to the current directory if cwd is
1254 NULL. */
1256 static char *
1257 abspath (cwd, rel_filename)
1258 const char *cwd;
1259 const char *rel_filename;
1261 /* Setup the current working directory as needed. */
1262 const char *cwd2 = (cwd) ? cwd : cwd_buffer;
1263 char *const abs_buffer
1264 = (char *) alloca (strlen (cwd2) + strlen (rel_filename) + 2);
1265 char *endp = abs_buffer;
1266 char *outp, *inp;
1268 /* Copy the filename (possibly preceded by the current working
1269 directory name) into the absolutization buffer. */
1272 const char *src_p;
1274 if (rel_filename[0] != '/')
1276 src_p = cwd2;
1277 while ((*endp++ = *src_p++))
1278 continue;
1279 *(endp-1) = '/'; /* overwrite null */
1281 src_p = rel_filename;
1282 while ((*endp++ = *src_p++))
1283 continue;
1286 /* Now make a copy of abs_buffer into abs_buffer, shortening the
1287 filename (by taking out slashes and dots) as we go. */
1289 outp = inp = abs_buffer;
1290 *outp++ = *inp++; /* copy first slash */
1291 #if defined (apollo) || defined (_WIN32) || defined (__INTERIX)
1292 if (inp[0] == '/')
1293 *outp++ = *inp++; /* copy second slash */
1294 #endif
1295 for (;;)
1297 if (!inp[0])
1298 break;
1299 else if (inp[0] == '/' && outp[-1] == '/')
1301 inp++;
1302 continue;
1304 else if (inp[0] == '.' && outp[-1] == '/')
1306 if (!inp[1])
1307 break;
1308 else if (inp[1] == '/')
1310 inp += 2;
1311 continue;
1313 else if ((inp[1] == '.') && (inp[2] == 0 || inp[2] == '/'))
1315 inp += (inp[2] == '/') ? 3 : 2;
1316 outp -= 2;
1317 while (outp >= abs_buffer && *outp != '/')
1318 outp--;
1319 if (outp < abs_buffer)
1321 /* Catch cases like /.. where we try to backup to a
1322 point above the absolute root of the logical file
1323 system. */
1325 notice ("%s: invalid file name: %s\n",
1326 pname, rel_filename);
1327 exit (FATAL_EXIT_CODE);
1329 *++outp = '\0';
1330 continue;
1333 *outp++ = *inp++;
1336 /* On exit, make sure that there is a trailing null, and make sure that
1337 the last character of the returned string is *not* a slash. */
1339 *outp = '\0';
1340 if (outp[-1] == '/')
1341 *--outp = '\0';
1343 /* Make a copy (in the heap) of the stuff left in the absolutization
1344 buffer and return a pointer to the copy. */
1346 return savestring (abs_buffer, outp - abs_buffer);
1349 /* Given a filename (and possibly a directory name from which the filename
1350 is relative) return a string which is the shortest possible
1351 equivalent for the corresponding full (absolutized) filename. The
1352 shortest possible equivalent may be constructed by converting the
1353 absolutized filename to be a relative filename (i.e. relative to
1354 the actual current working directory). However if a relative filename
1355 is longer, then the full absolute filename is returned.
1357 KNOWN BUG:
1359 Note that "simple-minded" conversion of any given type of filename (either
1360 relative or absolute) may not result in a valid equivalent filename if any
1361 subpart of the original filename is actually a symbolic link. */
1363 static const char *
1364 shortpath (cwd, filename)
1365 const char *cwd;
1366 const char *filename;
1368 char *rel_buffer;
1369 char *rel_buf_p;
1370 char *cwd_p = cwd_buffer;
1371 char *path_p;
1372 int unmatched_slash_count = 0;
1373 size_t filename_len = strlen (filename);
1375 path_p = abspath (cwd, filename);
1376 rel_buf_p = rel_buffer = (char *) xmalloc (filename_len);
1378 while (*cwd_p && (*cwd_p == *path_p))
1380 cwd_p++;
1381 path_p++;
1383 if (!*cwd_p && (!*path_p || *path_p == '/')) /* whole pwd matched */
1385 if (!*path_p) /* input *is* the current path! */
1386 return ".";
1387 else
1388 return ++path_p;
1390 else
1392 if (*path_p)
1394 --cwd_p;
1395 --path_p;
1396 while (*cwd_p != '/') /* backup to last slash */
1398 --cwd_p;
1399 --path_p;
1401 cwd_p++;
1402 path_p++;
1403 unmatched_slash_count++;
1406 /* Find out how many directory levels in cwd were *not* matched. */
1407 while (*cwd_p)
1408 if (*cwd_p++ == '/')
1409 unmatched_slash_count++;
1411 /* Now we know how long the "short name" will be.
1412 Reject it if longer than the input. */
1413 if (unmatched_slash_count * 3 + strlen (path_p) >= filename_len)
1414 return filename;
1416 /* For each of them, put a `../' at the beginning of the short name. */
1417 while (unmatched_slash_count--)
1419 /* Give up if the result gets to be longer
1420 than the absolute path name. */
1421 if (rel_buffer + filename_len <= rel_buf_p + 3)
1422 return filename;
1423 *rel_buf_p++ = '.';
1424 *rel_buf_p++ = '.';
1425 *rel_buf_p++ = '/';
1428 /* Then tack on the unmatched part of the desired file's name. */
1431 if (rel_buffer + filename_len <= rel_buf_p)
1432 return filename;
1434 while ((*rel_buf_p++ = *path_p++));
1436 --rel_buf_p;
1437 if (*(rel_buf_p-1) == '/')
1438 *--rel_buf_p = '\0';
1439 return rel_buffer;
1443 /* Lookup the given filename in the hash table for filenames. If it is a
1444 new one, then the hash table info pointer will be null. In this case,
1445 we create a new file_info record to go with the filename, and we initialize
1446 that record with some reasonable values. */
1448 /* FILENAME was const, but that causes a warning on AIX when calling stat.
1449 That is probably a bug in AIX, but might as well avoid the warning. */
1451 static file_info *
1452 find_file (filename, do_not_stat)
1453 char *filename;
1454 int do_not_stat;
1456 hash_table_entry *hash_entry_p;
1458 hash_entry_p = lookup (filename_primary, filename);
1459 if (hash_entry_p->fip)
1460 return hash_entry_p->fip;
1461 else
1463 struct stat stat_buf;
1464 file_info *file_p = (file_info *) xmalloc (sizeof (file_info));
1466 /* If we cannot get status on any given source file, give a warning
1467 and then just set its time of last modification to infinity. */
1469 if (do_not_stat)
1470 stat_buf.st_mtime = (time_t) 0;
1471 else
1473 if (my_stat (filename, &stat_buf) == -1)
1475 int errno_val = errno;
1476 notice ("%s: %s: can't get status: %s\n",
1477 pname, shortpath (NULL, filename),
1478 xstrerror (errno_val));
1479 stat_buf.st_mtime = (time_t) -1;
1483 hash_entry_p->fip = file_p;
1484 file_p->hash_entry = hash_entry_p;
1485 file_p->defs_decs = NULL;
1486 file_p->mtime = stat_buf.st_mtime;
1487 return file_p;
1491 /* Generate a fatal error because some part of the aux_info file is
1492 messed up. */
1494 static void
1495 aux_info_corrupted ()
1497 notice ("\n%s: fatal error: aux info file corrupted at line %d\n",
1498 pname, current_aux_info_lineno);
1499 exit (FATAL_EXIT_CODE);
1502 /* ??? This comment is vague. Say what the condition is for. */
1503 /* Check to see that a condition is true. This is kind of like an assert. */
1505 static void
1506 check_aux_info (cond)
1507 int cond;
1509 if (! cond)
1510 aux_info_corrupted ();
1513 /* Given a pointer to the closing right parenthesis for a particular formals
1514 list (in an aux_info file) find the corresponding left parenthesis and
1515 return a pointer to it. */
1517 static const char *
1518 find_corresponding_lparen (p)
1519 const char *p;
1521 const char *q;
1522 int paren_depth;
1524 for (paren_depth = 1, q = p-1; paren_depth; q--)
1526 switch (*q)
1528 case ')':
1529 paren_depth++;
1530 break;
1531 case '(':
1532 paren_depth--;
1533 break;
1536 return ++q;
1539 /* Given a line from an aux info file, and a time at which the aux info
1540 file it came from was created, check to see if the item described in
1541 the line comes from a file which has been modified since the aux info
1542 file was created. If so, return non-zero, else return zero. */
1544 static int
1545 referenced_file_is_newer (l, aux_info_mtime)
1546 const char *l;
1547 time_t aux_info_mtime;
1549 const char *p;
1550 file_info *fi_p;
1551 char *filename;
1553 check_aux_info (l[0] == '/');
1554 check_aux_info (l[1] == '*');
1555 check_aux_info (l[2] == ' ');
1558 const char *filename_start = p = l + 3;
1560 while (*p != ':')
1561 p++;
1562 filename = (char *) alloca ((size_t) (p - filename_start) + 1);
1563 strncpy (filename, filename_start, (size_t) (p - filename_start));
1564 filename[p-filename_start] = '\0';
1567 /* Call find_file to find the file_info record associated with the file
1568 which contained this particular def or dec item. Note that this call
1569 may cause a new file_info record to be created if this is the first time
1570 that we have ever known about this particular file. */
1572 fi_p = find_file (abspath (invocation_filename, filename), 0);
1574 return (fi_p->mtime > aux_info_mtime);
1577 /* Given a line of info from the aux_info file, create a new
1578 def_dec_info record to remember all of the important information about
1579 a function definition or declaration.
1581 Link this record onto the list of such records for the particular file in
1582 which it occurred in proper (descending) line number order (for now).
1584 If there is an identical record already on the list for the file, throw
1585 this one away. Doing so takes care of the (useless and troublesome)
1586 duplicates which are bound to crop up due to multiple inclusions of any
1587 given individual header file.
1589 Finally, link the new def_dec record onto the list of such records
1590 pertaining to this particular function name. */
1592 static void
1593 save_def_or_dec (l, is_syscalls)
1594 const char *l;
1595 int is_syscalls;
1597 const char *p;
1598 const char *semicolon_p;
1599 def_dec_info *def_dec_p = (def_dec_info *) xmalloc (sizeof (def_dec_info));
1601 #ifndef UNPROTOIZE
1602 def_dec_p->written = 0;
1603 #endif /* !defined (UNPROTOIZE) */
1605 /* Start processing the line by picking off 5 pieces of information from
1606 the left hand end of the line. These are filename, line number,
1607 new/old/implicit flag (new = ANSI prototype format), definition or
1608 declaration flag, and extern/static flag). */
1610 check_aux_info (l[0] == '/');
1611 check_aux_info (l[1] == '*');
1612 check_aux_info (l[2] == ' ');
1615 const char *filename_start = p = l + 3;
1616 char *filename;
1618 while (*p != ':')
1619 p++;
1620 filename = (char *) alloca ((size_t) (p - filename_start) + 1);
1621 strncpy (filename, filename_start, (size_t) (p - filename_start));
1622 filename[p-filename_start] = '\0';
1624 /* Call find_file to find the file_info record associated with the file
1625 which contained this particular def or dec item. Note that this call
1626 may cause a new file_info record to be created if this is the first time
1627 that we have ever known about this particular file.
1629 Note that we started out by forcing all of the base source file names
1630 (i.e. the names of the aux_info files with the .X stripped off) into the
1631 filenames hash table, and we simultaneously setup file_info records for
1632 all of these base file names (even if they may be useless later).
1633 The file_info records for all of these "base" file names (properly)
1634 act as file_info records for the "original" (i.e. un-included) files
1635 which were submitted to gcc for compilation (when the -aux-info
1636 option was used). */
1638 def_dec_p->file = find_file (abspath (invocation_filename, filename), is_syscalls);
1642 const char *line_number_start = ++p;
1643 char line_number[10];
1645 while (*p != ':')
1646 p++;
1647 strncpy (line_number, line_number_start, (size_t) (p - line_number_start));
1648 line_number[p-line_number_start] = '\0';
1649 def_dec_p->line = atoi (line_number);
1652 /* Check that this record describes a new-style, old-style, or implicit
1653 definition or declaration. */
1655 p++; /* Skip over the `:'. */
1656 check_aux_info ((*p == 'N') || (*p == 'O') || (*p == 'I'));
1658 /* Is this a new style (ANSI prototyped) definition or declaration? */
1660 def_dec_p->prototyped = (*p == 'N');
1662 #ifndef UNPROTOIZE
1664 /* Is this an implicit declaration? */
1666 def_dec_p->is_implicit = (*p == 'I');
1668 #endif /* !defined (UNPROTOIZE) */
1670 p++;
1672 check_aux_info ((*p == 'C') || (*p == 'F'));
1674 /* Is this item a function definition (F) or a declaration (C). Note that
1675 we treat item taken from the syscalls file as though they were function
1676 definitions regardless of what the stuff in the file says. */
1678 def_dec_p->is_func_def = ((*p++ == 'F') || is_syscalls);
1680 #ifndef UNPROTOIZE
1681 def_dec_p->definition = 0; /* Fill this in later if protoizing. */
1682 #endif /* !defined (UNPROTOIZE) */
1684 check_aux_info (*p++ == ' ');
1685 check_aux_info (*p++ == '*');
1686 check_aux_info (*p++ == '/');
1687 check_aux_info (*p++ == ' ');
1689 #ifdef UNPROTOIZE
1690 check_aux_info ((!strncmp (p, "static", 6)) || (!strncmp (p, "extern", 6)));
1691 #else /* !defined (UNPROTOIZE) */
1692 if (!strncmp (p, "static", 6))
1693 def_dec_p->is_static = -1;
1694 else if (!strncmp (p, "extern", 6))
1695 def_dec_p->is_static = 0;
1696 else
1697 check_aux_info (0); /* Didn't find either `extern' or `static'. */
1698 #endif /* !defined (UNPROTOIZE) */
1701 const char *ansi_start = p;
1703 p += 6; /* Pass over the "static" or "extern". */
1705 /* We are now past the initial stuff. Search forward from here to find
1706 the terminating semicolon that should immediately follow the entire
1707 ANSI format function declaration. */
1709 while (*++p != ';')
1710 continue;
1712 semicolon_p = p;
1714 /* Make a copy of the ansi declaration part of the line from the aux_info
1715 file. */
1717 def_dec_p->ansi_decl
1718 = dupnstr (ansi_start, (size_t) ((semicolon_p+1) - ansi_start));
1720 /* Backup and point at the final right paren of the final argument list. */
1722 p--;
1724 #ifndef UNPROTOIZE
1725 def_dec_p->f_list_chain = NULL;
1726 #endif /* !defined (UNPROTOIZE) */
1728 while (p != ansi_start && (p[-1] == ' ' || p[-1] == '\t')) p--;
1729 if (*p != ')')
1731 free_def_dec (def_dec_p);
1732 return;
1736 /* Now isolate a whole set of formal argument lists, one-by-one. Normally,
1737 there will only be one list to isolate, but there could be more. */
1739 def_dec_p->f_list_count = 0;
1741 for (;;)
1743 const char *left_paren_p = find_corresponding_lparen (p);
1744 #ifndef UNPROTOIZE
1746 f_list_chain_item *cip
1747 = (f_list_chain_item *) xmalloc (sizeof (f_list_chain_item));
1749 cip->formals_list
1750 = dupnstr (left_paren_p + 1, (size_t) (p - (left_paren_p+1)));
1752 /* Add the new chain item at the head of the current list. */
1754 cip->chain_next = def_dec_p->f_list_chain;
1755 def_dec_p->f_list_chain = cip;
1757 #endif /* !defined (UNPROTOIZE) */
1758 def_dec_p->f_list_count++;
1760 p = left_paren_p - 2;
1762 /* p must now point either to another right paren, or to the last
1763 character of the name of the function that was declared/defined.
1764 If p points to another right paren, then this indicates that we
1765 are dealing with multiple formals lists. In that case, there
1766 really should be another right paren preceding this right paren. */
1768 if (*p != ')')
1769 break;
1770 else
1771 check_aux_info (*--p == ')');
1776 const char *past_fn = p + 1;
1778 check_aux_info (*past_fn == ' ');
1780 /* Scan leftwards over the identifier that names the function. */
1782 while (is_id_char (*p))
1783 p--;
1784 p++;
1786 /* p now points to the leftmost character of the function name. */
1789 char *fn_string = (char *) alloca (past_fn - p + 1);
1791 strncpy (fn_string, p, (size_t) (past_fn - p));
1792 fn_string[past_fn-p] = '\0';
1793 def_dec_p->hash_entry = lookup (function_name_primary, fn_string);
1797 /* Look at all of the defs and decs for this function name that we have
1798 collected so far. If there is already one which is at the same
1799 line number in the same file, then we can discard this new def_dec_info
1800 record.
1802 As an extra assurance that any such pair of (nominally) identical
1803 function declarations are in fact identical, we also compare the
1804 ansi_decl parts of the lines from the aux_info files just to be on
1805 the safe side.
1807 This comparison will fail if (for instance) the user was playing
1808 messy games with the preprocessor which ultimately causes one
1809 function declaration in one header file to look differently when
1810 that file is included by two (or more) other files. */
1813 const def_dec_info *other;
1815 for (other = def_dec_p->hash_entry->ddip; other; other = other->next_for_func)
1817 if (def_dec_p->line == other->line && def_dec_p->file == other->file)
1819 if (strcmp (def_dec_p->ansi_decl, other->ansi_decl))
1821 notice ("%s:%d: declaration of function `%s' takes different forms\n",
1822 def_dec_p->file->hash_entry->symbol,
1823 def_dec_p->line,
1824 def_dec_p->hash_entry->symbol);
1825 exit (FATAL_EXIT_CODE);
1827 free_def_dec (def_dec_p);
1828 return;
1833 #ifdef UNPROTOIZE
1835 /* If we are doing unprotoizing, we must now setup the pointers that will
1836 point to the K&R name list and to the K&R argument declarations list.
1838 Note that if this is only a function declaration, then we should not
1839 expect to find any K&R style formals list following the ANSI-style
1840 formals list. This is because GCC knows that such information is
1841 useless in the case of function declarations (function definitions
1842 are a different story however).
1844 Since we are unprotoizing, we don't need any such lists anyway.
1845 All we plan to do is to delete all characters between ()'s in any
1846 case. */
1848 def_dec_p->formal_names = NULL;
1849 def_dec_p->formal_decls = NULL;
1851 if (def_dec_p->is_func_def)
1853 p = semicolon_p;
1854 check_aux_info (*++p == ' ');
1855 check_aux_info (*++p == '/');
1856 check_aux_info (*++p == '*');
1857 check_aux_info (*++p == ' ');
1858 check_aux_info (*++p == '(');
1861 const char *kr_names_start = ++p; /* Point just inside '('. */
1863 while (*p++ != ')')
1864 continue;
1865 p--; /* point to closing right paren */
1867 /* Make a copy of the K&R parameter names list. */
1869 def_dec_p->formal_names
1870 = dupnstr (kr_names_start, (size_t) (p - kr_names_start));
1873 check_aux_info (*++p == ' ');
1874 p++;
1876 /* p now points to the first character of the K&R style declarations
1877 list (if there is one) or to the star-slash combination that ends
1878 the comment in which such lists get embedded. */
1880 /* Make a copy of the K&R formal decls list and set the def_dec record
1881 to point to it. */
1883 if (*p == '*') /* Are there no K&R declarations? */
1885 check_aux_info (*++p == '/');
1886 def_dec_p->formal_decls = "";
1888 else
1890 const char *kr_decls_start = p;
1892 while (p[0] != '*' || p[1] != '/')
1893 p++;
1894 p--;
1896 check_aux_info (*p == ' ');
1898 def_dec_p->formal_decls
1899 = dupnstr (kr_decls_start, (size_t) (p - kr_decls_start));
1902 /* Handle a special case. If we have a function definition marked as
1903 being in "old" style, and if its formal names list is empty, then
1904 it may actually have the string "void" in its real formals list
1905 in the original source code. Just to make sure, we will get setup
1906 to convert such things anyway.
1908 This kludge only needs to be here because of an insurmountable
1909 problem with generating .X files. */
1911 if (!def_dec_p->prototyped && !*def_dec_p->formal_names)
1912 def_dec_p->prototyped = 1;
1915 /* Since we are unprotoizing, if this item is already in old (K&R) style,
1916 we can just ignore it. If that is true, throw away the itme now. */
1918 if (!def_dec_p->prototyped)
1920 free_def_dec (def_dec_p);
1921 return;
1924 #endif /* defined (UNPROTOIZE) */
1926 /* Add this record to the head of the list of records pertaining to this
1927 particular function name. */
1929 def_dec_p->next_for_func = def_dec_p->hash_entry->ddip;
1930 def_dec_p->hash_entry->ddip = def_dec_p;
1932 /* Add this new def_dec_info record to the sorted list of def_dec_info
1933 records for this file. Note that we don't have to worry about duplicates
1934 (caused by multiple inclusions of header files) here because we have
1935 already eliminated duplicates above. */
1937 if (!def_dec_p->file->defs_decs)
1939 def_dec_p->file->defs_decs = def_dec_p;
1940 def_dec_p->next_in_file = NULL;
1942 else
1944 int line = def_dec_p->line;
1945 const def_dec_info *prev = NULL;
1946 const def_dec_info *curr = def_dec_p->file->defs_decs;
1947 const def_dec_info *next = curr->next_in_file;
1949 while (next && (line < curr->line))
1951 prev = curr;
1952 curr = next;
1953 next = next->next_in_file;
1955 if (line >= curr->line)
1957 def_dec_p->next_in_file = curr;
1958 if (prev)
1959 ((NONCONST def_dec_info *) prev)->next_in_file = def_dec_p;
1960 else
1961 def_dec_p->file->defs_decs = def_dec_p;
1963 else /* assert (next == NULL); */
1965 ((NONCONST def_dec_info *) curr)->next_in_file = def_dec_p;
1966 /* assert (next == NULL); */
1967 def_dec_p->next_in_file = next;
1972 /* Set up the vector COMPILE_PARAMS which is the argument list for running GCC.
1973 Also set input_file_name_index and aux_info_file_name_index
1974 to the indices of the slots where the file names should go. */
1976 /* We initialize the vector by removing -g, -O, -S, -c, and -o options,
1977 and adding '-aux-info AUXFILE -S -o /dev/null INFILE' at the end. */
1979 static void
1980 munge_compile_params (params_list)
1981 const char *params_list;
1983 /* Build up the contents in a temporary vector
1984 that is so big that to has to be big enough. */
1985 const char **temp_params
1986 = (const char **) alloca ((strlen (params_list) + 8) * sizeof (char *));
1987 int param_count = 0;
1988 const char *param;
1990 temp_params[param_count++] = compiler_file_name;
1991 for (;;)
1993 while (ISSPACE ((const unsigned char)*params_list))
1994 params_list++;
1995 if (!*params_list)
1996 break;
1997 param = params_list;
1998 while (*params_list && !ISSPACE ((const unsigned char)*params_list))
1999 params_list++;
2000 if (param[0] != '-')
2001 temp_params[param_count++]
2002 = dupnstr (param, (size_t) (params_list - param));
2003 else
2005 switch (param[1])
2007 case 'g':
2008 case 'O':
2009 case 'S':
2010 case 'c':
2011 break; /* Don't copy these. */
2012 case 'o':
2013 while (ISSPACE ((const unsigned char)*params_list))
2014 params_list++;
2015 while (*params_list
2016 && !ISSPACE ((const unsigned char)*params_list))
2017 params_list++;
2018 break;
2019 default:
2020 temp_params[param_count++]
2021 = dupnstr (param, (size_t) (params_list - param));
2024 if (!*params_list)
2025 break;
2027 temp_params[param_count++] = "-aux-info";
2029 /* Leave room for the aux-info file name argument. */
2030 aux_info_file_name_index = param_count;
2031 temp_params[param_count++] = NULL;
2033 temp_params[param_count++] = "-S";
2034 temp_params[param_count++] = "-o";
2035 temp_params[param_count++] = "/dev/null";
2037 /* Leave room for the input file name argument. */
2038 input_file_name_index = param_count;
2039 temp_params[param_count++] = NULL;
2040 /* Terminate the list. */
2041 temp_params[param_count++] = NULL;
2043 /* Make a copy of the compile_params in heap space. */
2045 compile_params
2046 = (const char **) xmalloc (sizeof (char *) * (param_count+1));
2047 memcpy (compile_params, temp_params, sizeof (char *) * param_count);
2050 /* Do a recompilation for the express purpose of generating a new aux_info
2051 file to go with a specific base source file.
2053 The result is a boolean indicating success. */
2055 static int
2056 gen_aux_info_file (base_filename)
2057 const char *base_filename;
2059 if (!input_file_name_index)
2060 munge_compile_params ("");
2062 /* Store the full source file name in the argument vector. */
2063 compile_params[input_file_name_index] = shortpath (NULL, base_filename);
2064 /* Add .X to source file name to get aux-info file name. */
2065 compile_params[aux_info_file_name_index]
2066 = savestring2 (compile_params[input_file_name_index],
2067 strlen (compile_params[input_file_name_index]),
2068 ".X",
2071 if (!quiet_flag)
2072 notice ("%s: compiling `%s'\n",
2073 pname, compile_params[input_file_name_index]);
2076 char *errmsg_fmt, *errmsg_arg;
2077 int wait_status, pid;
2078 char *temp_base = choose_temp_base ();
2080 pid = pexecute (compile_params[0], (char * const *) compile_params,
2081 pname, temp_base, &errmsg_fmt, &errmsg_arg,
2082 PEXECUTE_FIRST | PEXECUTE_LAST | PEXECUTE_SEARCH);
2084 if (pid == -1)
2086 int errno_val = errno;
2087 fprintf (stderr, "%s: ", pname);
2088 fprintf (stderr, errmsg_fmt, errmsg_arg);
2089 fprintf (stderr, ": %s\n", xstrerror (errno_val));
2090 return 0;
2093 pid = pwait (pid, &wait_status, 0);
2094 if (pid == -1)
2096 notice ("%s: wait: %s\n", pname, xstrerror (errno));
2097 return 0;
2099 if (WIFSIGNALED (wait_status))
2101 notice ("%s: subprocess got fatal signal %d\n",
2102 pname, WTERMSIG (wait_status));
2103 return 0;
2105 if (WIFEXITED (wait_status))
2107 if (WEXITSTATUS (wait_status) != 0)
2109 notice ("%s: %s exited with status %d\n",
2110 pname, compile_params[0], WEXITSTATUS (wait_status));
2111 return 0;
2113 return 1;
2115 abort ();
2119 /* Read in all of the information contained in a single aux_info file.
2120 Save all of the important stuff for later. */
2122 static void
2123 process_aux_info_file (base_source_filename, keep_it, is_syscalls)
2124 const char *base_source_filename;
2125 int keep_it;
2126 int is_syscalls;
2128 size_t base_len = strlen (base_source_filename);
2129 char * aux_info_filename
2130 = (char *) alloca (base_len + strlen (aux_info_suffix) + 1);
2131 char *aux_info_base;
2132 char *aux_info_limit;
2133 char *aux_info_relocated_name;
2134 const char *aux_info_second_line;
2135 time_t aux_info_mtime;
2136 size_t aux_info_size;
2137 int must_create;
2139 /* Construct the aux_info filename from the base source filename. */
2141 strcpy (aux_info_filename, base_source_filename);
2142 strcat (aux_info_filename, aux_info_suffix);
2144 /* Check that the aux_info file exists and is readable. If it does not
2145 exist, try to create it (once only). */
2147 /* If file doesn't exist, set must_create.
2148 Likewise if it exists and we can read it but it is obsolete.
2149 Otherwise, report an error. */
2150 must_create = 0;
2152 /* Come here with must_create set to 1 if file is out of date. */
2153 start_over: ;
2155 if (my_access (aux_info_filename, R_OK) == -1)
2157 if (errno == ENOENT)
2159 if (is_syscalls)
2161 notice ("%s: warning: missing SYSCALLS file `%s'\n",
2162 pname, aux_info_filename);
2163 return;
2165 must_create = 1;
2167 else
2169 int errno_val = errno;
2170 notice ("%s: can't read aux info file `%s': %s\n",
2171 pname, shortpath (NULL, aux_info_filename),
2172 xstrerror (errno_val));
2173 errors++;
2174 return;
2177 #if 0 /* There is code farther down to take care of this. */
2178 else
2180 struct stat s1, s2;
2181 stat (aux_info_file_name, &s1);
2182 stat (base_source_file_name, &s2);
2183 if (s2.st_mtime > s1.st_mtime)
2184 must_create = 1;
2186 #endif /* 0 */
2188 /* If we need a .X file, create it, and verify we can read it. */
2189 if (must_create)
2191 if (!gen_aux_info_file (base_source_filename))
2193 errors++;
2194 return;
2196 if (my_access (aux_info_filename, R_OK) == -1)
2198 int errno_val = errno;
2199 notice ("%s: can't read aux info file `%s': %s\n",
2200 pname, shortpath (NULL, aux_info_filename),
2201 xstrerror (errno_val));
2202 errors++;
2203 return;
2208 struct stat stat_buf;
2210 /* Get some status information about this aux_info file. */
2212 if (my_stat (aux_info_filename, &stat_buf) == -1)
2214 int errno_val = errno;
2215 notice ("%s: can't get status of aux info file `%s': %s\n",
2216 pname, shortpath (NULL, aux_info_filename),
2217 xstrerror (errno_val));
2218 errors++;
2219 return;
2222 /* Check on whether or not this aux_info file is zero length. If it is,
2223 then just ignore it and return. */
2225 if ((aux_info_size = stat_buf.st_size) == 0)
2226 return;
2228 /* Get the date/time of last modification for this aux_info file and
2229 remember it. We will have to check that any source files that it
2230 contains information about are at least this old or older. */
2232 aux_info_mtime = stat_buf.st_mtime;
2234 if (!is_syscalls)
2236 /* Compare mod time with the .c file; update .X file if obsolete.
2237 The code later on can fail to check the .c file
2238 if it did not directly define any functions. */
2240 if (my_stat (base_source_filename, &stat_buf) == -1)
2242 int errno_val = errno;
2243 notice ("%s: can't get status of aux info file `%s': %s\n",
2244 pname, shortpath (NULL, base_source_filename),
2245 xstrerror (errno_val));
2246 errors++;
2247 return;
2249 if (stat_buf.st_mtime > aux_info_mtime)
2251 must_create = 1;
2252 goto start_over;
2258 int aux_info_file;
2260 /* Open the aux_info file. */
2262 if ((aux_info_file = my_open (aux_info_filename, O_RDONLY, 0444 )) == -1)
2264 int errno_val = errno;
2265 notice ("%s: can't open aux info file `%s' for reading: %s\n",
2266 pname, shortpath (NULL, aux_info_filename),
2267 xstrerror (errno_val));
2268 return;
2271 /* Allocate space to hold the aux_info file in memory. */
2273 aux_info_base = xmalloc (aux_info_size + 1);
2274 aux_info_limit = aux_info_base + aux_info_size;
2275 *aux_info_limit = '\0';
2277 /* Read the aux_info file into memory. */
2279 if (safe_read (aux_info_file, aux_info_base, aux_info_size) !=
2280 (int) aux_info_size)
2282 int errno_val = errno;
2283 notice ("%s: error reading aux info file `%s': %s\n",
2284 pname, shortpath (NULL, aux_info_filename),
2285 xstrerror (errno_val));
2286 free (aux_info_base);
2287 close (aux_info_file);
2288 return;
2291 /* Close the aux info file. */
2293 if (close (aux_info_file))
2295 int errno_val = errno;
2296 notice ("%s: error closing aux info file `%s': %s\n",
2297 pname, shortpath (NULL, aux_info_filename),
2298 xstrerror (errno_val));
2299 free (aux_info_base);
2300 close (aux_info_file);
2301 return;
2305 /* Delete the aux_info file (unless requested not to). If the deletion
2306 fails for some reason, don't even worry about it. */
2308 if (must_create && !keep_it)
2309 if (my_unlink (aux_info_filename) == -1)
2311 int errno_val = errno;
2312 notice ("%s: can't delete aux info file `%s': %s\n",
2313 pname, shortpath (NULL, aux_info_filename),
2314 xstrerror (errno_val));
2317 /* Save a pointer into the first line of the aux_info file which
2318 contains the filename of the directory from which the compiler
2319 was invoked when the associated source file was compiled.
2320 This information is used later to help create complete
2321 filenames out of the (potentially) relative filenames in
2322 the aux_info file. */
2325 char *p = aux_info_base;
2327 while (*p != ':')
2328 p++;
2329 p++;
2330 while (*p == ' ')
2331 p++;
2332 invocation_filename = p; /* Save a pointer to first byte of path. */
2333 while (*p != ' ')
2334 p++;
2335 *p++ = '/';
2336 *p++ = '\0';
2337 while (*p++ != '\n')
2338 continue;
2339 aux_info_second_line = p;
2340 aux_info_relocated_name = 0;
2341 if (invocation_filename[0] != '/')
2343 /* INVOCATION_FILENAME is relative;
2344 append it to BASE_SOURCE_FILENAME's dir. */
2345 char *dir_end;
2346 aux_info_relocated_name = xmalloc (base_len + (p-invocation_filename));
2347 strcpy (aux_info_relocated_name, base_source_filename);
2348 dir_end = strrchr (aux_info_relocated_name, '/');
2349 if (dir_end)
2350 dir_end++;
2351 else
2352 dir_end = aux_info_relocated_name;
2353 strcpy (dir_end, invocation_filename);
2354 invocation_filename = aux_info_relocated_name;
2360 const char *aux_info_p;
2362 /* Do a pre-pass on the lines in the aux_info file, making sure that all
2363 of the source files referenced in there are at least as old as this
2364 aux_info file itself. If not, go back and regenerate the aux_info
2365 file anew. Don't do any of this for the syscalls file. */
2367 if (!is_syscalls)
2369 current_aux_info_lineno = 2;
2371 for (aux_info_p = aux_info_second_line; *aux_info_p; )
2373 if (referenced_file_is_newer (aux_info_p, aux_info_mtime))
2375 free (aux_info_base);
2376 xfree (aux_info_relocated_name);
2377 if (keep_it && my_unlink (aux_info_filename) == -1)
2379 int errno_val = errno;
2380 notice ("%s: can't delete file `%s': %s\n",
2381 pname, shortpath (NULL, aux_info_filename),
2382 xstrerror (errno_val));
2383 return;
2385 must_create = 1;
2386 goto start_over;
2389 /* Skip over the rest of this line to start of next line. */
2391 while (*aux_info_p != '\n')
2392 aux_info_p++;
2393 aux_info_p++;
2394 current_aux_info_lineno++;
2398 /* Now do the real pass on the aux_info lines. Save their information in
2399 the in-core data base. */
2401 current_aux_info_lineno = 2;
2403 for (aux_info_p = aux_info_second_line; *aux_info_p;)
2405 char *unexpanded_line = unexpand_if_needed (aux_info_p);
2407 if (unexpanded_line)
2409 save_def_or_dec (unexpanded_line, is_syscalls);
2410 free (unexpanded_line);
2412 else
2413 save_def_or_dec (aux_info_p, is_syscalls);
2415 /* Skip over the rest of this line and get to start of next line. */
2417 while (*aux_info_p != '\n')
2418 aux_info_p++;
2419 aux_info_p++;
2420 current_aux_info_lineno++;
2424 free (aux_info_base);
2425 xfree (aux_info_relocated_name);
2428 #ifndef UNPROTOIZE
2430 /* Check an individual filename for a .c suffix. If the filename has this
2431 suffix, rename the file such that its suffix is changed to .C. This
2432 function implements the -C option. */
2434 static void
2435 rename_c_file (hp)
2436 const hash_table_entry *hp;
2438 const char *filename = hp->symbol;
2439 int last_char_index = strlen (filename) - 1;
2440 char *const new_filename = (char *) alloca (strlen (filename) + 1);
2442 /* Note that we don't care here if the given file was converted or not. It
2443 is possible that the given file was *not* converted, simply because there
2444 was nothing in it which actually required conversion. Even in this case,
2445 we want to do the renaming. Note that we only rename files with the .c
2446 suffix. */
2448 if (filename[last_char_index] != 'c' || filename[last_char_index-1] != '.')
2449 return;
2451 strcpy (new_filename, filename);
2452 new_filename[last_char_index] = 'C';
2454 if (my_link (filename, new_filename) == -1)
2456 int errno_val = errno;
2457 notice ("%s: warning: can't link file `%s' to `%s': %s\n",
2458 pname, shortpath (NULL, filename),
2459 shortpath (NULL, new_filename), xstrerror (errno_val));
2460 errors++;
2461 return;
2464 if (my_unlink (filename) == -1)
2466 int errno_val = errno;
2467 notice ("%s: warning: can't delete file `%s': %s\n",
2468 pname, shortpath (NULL, filename), xstrerror (errno_val));
2469 errors++;
2470 return;
2474 #endif /* !defined (UNPROTOIZE) */
2476 /* Take the list of definitions and declarations attached to a particular
2477 file_info node and reverse the order of the list. This should get the
2478 list into an order such that the item with the lowest associated line
2479 number is nearest the head of the list. When these lists are originally
2480 built, they are in the opposite order. We want to traverse them in
2481 normal line number order later (i.e. lowest to highest) so reverse the
2482 order here. */
2484 static void
2485 reverse_def_dec_list (hp)
2486 const hash_table_entry *hp;
2488 file_info *file_p = hp->fip;
2489 def_dec_info *prev = NULL;
2490 def_dec_info *current = (def_dec_info *)file_p->defs_decs;
2492 if (!current)
2493 return; /* no list to reverse */
2495 prev = current;
2496 if (! (current = (def_dec_info *)current->next_in_file))
2497 return; /* can't reverse a single list element */
2499 prev->next_in_file = NULL;
2501 while (current)
2503 def_dec_info *next = (def_dec_info *)current->next_in_file;
2505 current->next_in_file = prev;
2506 prev = current;
2507 current = next;
2510 file_p->defs_decs = prev;
2513 #ifndef UNPROTOIZE
2515 /* Find the (only?) extern definition for a particular function name, starting
2516 from the head of the linked list of entries for the given name. If we
2517 cannot find an extern definition for the given function name, issue a
2518 warning and scrounge around for the next best thing, i.e. an extern
2519 function declaration with a prototype attached to it. Note that we only
2520 allow such substitutions for extern declarations and never for static
2521 declarations. That's because the only reason we allow them at all is
2522 to let un-prototyped function declarations for system-supplied library
2523 functions get their prototypes from our own extra SYSCALLS.c.X file which
2524 contains all of the correct prototypes for system functions. */
2526 static const def_dec_info *
2527 find_extern_def (head, user)
2528 const def_dec_info *head;
2529 const def_dec_info *user;
2531 const def_dec_info *dd_p;
2532 const def_dec_info *extern_def_p = NULL;
2533 int conflict_noted = 0;
2535 /* Don't act too stupid here. Somebody may try to convert an entire system
2536 in one swell fwoop (rather than one program at a time, as should be done)
2537 and in that case, we may find that there are multiple extern definitions
2538 of a given function name in the entire set of source files that we are
2539 converting. If however one of these definitions resides in exactly the
2540 same source file as the reference we are trying to satisfy then in that
2541 case it would be stupid for us to fail to realize that this one definition
2542 *must* be the precise one we are looking for.
2544 To make sure that we don't miss an opportunity to make this "same file"
2545 leap of faith, we do a prescan of the list of records relating to the
2546 given function name, and we look (on this first scan) *only* for a
2547 definition of the function which is in the same file as the reference
2548 we are currently trying to satisfy. */
2550 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2551 if (dd_p->is_func_def && !dd_p->is_static && dd_p->file == user->file)
2552 return dd_p;
2554 /* Now, since we have not found a definition in the same file as the
2555 reference, we scan the list again and consider all possibilities from
2556 all files. Here we may get conflicts with the things listed in the
2557 SYSCALLS.c.X file, but if that happens it only means that the source
2558 code being converted contains its own definition of a function which
2559 could have been supplied by libc.a. In such cases, we should avoid
2560 issuing the normal warning, and defer to the definition given in the
2561 user's own code. */
2563 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2564 if (dd_p->is_func_def && !dd_p->is_static)
2566 if (!extern_def_p) /* Previous definition? */
2567 extern_def_p = dd_p; /* Remember the first definition found. */
2568 else
2570 /* Ignore definition just found if it came from SYSCALLS.c.X. */
2572 if (is_syscalls_file (dd_p->file))
2573 continue;
2575 /* Quietly replace the definition previously found with the one
2576 just found if the previous one was from SYSCALLS.c.X. */
2578 if (is_syscalls_file (extern_def_p->file))
2580 extern_def_p = dd_p;
2581 continue;
2584 /* If we get here, then there is a conflict between two function
2585 declarations for the same function, both of which came from the
2586 user's own code. */
2588 if (!conflict_noted) /* first time we noticed? */
2590 conflict_noted = 1;
2591 notice ("%s: conflicting extern definitions of '%s'\n",
2592 pname, head->hash_entry->symbol);
2593 if (!quiet_flag)
2595 notice ("%s: declarations of '%s' will not be converted\n",
2596 pname, head->hash_entry->symbol);
2597 notice ("%s: conflict list for '%s' follows:\n",
2598 pname, head->hash_entry->symbol);
2599 fprintf (stderr, "%s: %s(%d): %s\n",
2600 pname,
2601 shortpath (NULL, extern_def_p->file->hash_entry->symbol),
2602 extern_def_p->line, extern_def_p->ansi_decl);
2605 if (!quiet_flag)
2606 fprintf (stderr, "%s: %s(%d): %s\n",
2607 pname,
2608 shortpath (NULL, dd_p->file->hash_entry->symbol),
2609 dd_p->line, dd_p->ansi_decl);
2613 /* We want to err on the side of caution, so if we found multiple conflicting
2614 definitions for the same function, treat this as being that same as if we
2615 had found no definitions (i.e. return NULL). */
2617 if (conflict_noted)
2618 return NULL;
2620 if (!extern_def_p)
2622 /* We have no definitions for this function so do the next best thing.
2623 Search for an extern declaration already in prototype form. */
2625 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2626 if (!dd_p->is_func_def && !dd_p->is_static && dd_p->prototyped)
2628 extern_def_p = dd_p; /* save a pointer to the definition */
2629 if (!quiet_flag)
2630 notice ("%s: warning: using formals list from %s(%d) for function `%s'\n",
2631 pname,
2632 shortpath (NULL, dd_p->file->hash_entry->symbol),
2633 dd_p->line, dd_p->hash_entry->symbol);
2634 break;
2637 /* Gripe about unprototyped function declarations that we found no
2638 corresponding definition (or other source of prototype information)
2639 for.
2641 Gripe even if the unprototyped declaration we are worried about
2642 exists in a file in one of the "system" include directories. We
2643 can gripe about these because we should have at least found a
2644 corresponding (pseudo) definition in the SYSCALLS.c.X file. If we
2645 didn't, then that means that the SYSCALLS.c.X file is missing some
2646 needed prototypes for this particular system. That is worth telling
2647 the user about! */
2649 if (!extern_def_p)
2651 const char *file = user->file->hash_entry->symbol;
2653 if (!quiet_flag)
2654 if (in_system_include_dir (file))
2656 /* Why copy this string into `needed' at all?
2657 Why not just use user->ansi_decl without copying? */
2658 char *needed = (char *) alloca (strlen (user->ansi_decl) + 1);
2659 char *p;
2661 strcpy (needed, user->ansi_decl);
2662 p = (NONCONST char *) substr (needed, user->hash_entry->symbol)
2663 + strlen (user->hash_entry->symbol) + 2;
2664 /* Avoid having ??? in the string. */
2665 *p++ = '?';
2666 *p++ = '?';
2667 *p++ = '?';
2668 strcpy (p, ");");
2670 notice ("%s: %d: `%s' used but missing from SYSCALLS\n",
2671 shortpath (NULL, file), user->line,
2672 needed+7); /* Don't print "extern " */
2674 #if 0
2675 else
2676 notice ("%s: %d: warning: no extern definition for `%s'\n",
2677 shortpath (NULL, file), user->line,
2678 user->hash_entry->symbol);
2679 #endif
2682 return extern_def_p;
2685 /* Find the (only?) static definition for a particular function name in a
2686 given file. Here we get the function-name and the file info indirectly
2687 from the def_dec_info record pointer which is passed in. */
2689 static const def_dec_info *
2690 find_static_definition (user)
2691 const def_dec_info *user;
2693 const def_dec_info *head = user->hash_entry->ddip;
2694 const def_dec_info *dd_p;
2695 int num_static_defs = 0;
2696 const def_dec_info *static_def_p = NULL;
2698 for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
2699 if (dd_p->is_func_def && dd_p->is_static && (dd_p->file == user->file))
2701 static_def_p = dd_p; /* save a pointer to the definition */
2702 num_static_defs++;
2704 if (num_static_defs == 0)
2706 if (!quiet_flag)
2707 notice ("%s: warning: no static definition for `%s' in file `%s'\n",
2708 pname, head->hash_entry->symbol,
2709 shortpath (NULL, user->file->hash_entry->symbol));
2711 else if (num_static_defs > 1)
2713 notice ("%s: multiple static defs of `%s' in file `%s'\n",
2714 pname, head->hash_entry->symbol,
2715 shortpath (NULL, user->file->hash_entry->symbol));
2716 return NULL;
2718 return static_def_p;
2721 /* Find good prototype style formal argument lists for all of the function
2722 declarations which didn't have them before now.
2724 To do this we consider each function name one at a time. For each function
2725 name, we look at the items on the linked list of def_dec_info records for
2726 that particular name.
2728 Somewhere on this list we should find one (and only one) def_dec_info
2729 record which represents the actual function definition, and this record
2730 should have a nice formal argument list already associated with it.
2732 Thus, all we have to do is to connect up all of the other def_dec_info
2733 records for this particular function name to the special one which has
2734 the full-blown formals list.
2736 Of course it is a little more complicated than just that. See below for
2737 more details. */
2739 static void
2740 connect_defs_and_decs (hp)
2741 const hash_table_entry *hp;
2743 const def_dec_info *dd_p;
2744 const def_dec_info *extern_def_p = NULL;
2745 int first_extern_reference = 1;
2747 /* Traverse the list of definitions and declarations for this particular
2748 function name. For each item on the list, if it is a function
2749 definition (either old style or new style) then GCC has already been
2750 kind enough to produce a prototype for us, and it is associated with
2751 the item already, so declare the item as its own associated "definition".
2753 Also, for each item which is only a function declaration, but which
2754 nonetheless has its own prototype already (obviously supplied by the user)
2755 declare the item as its own definition.
2757 Note that when/if there are multiple user-supplied prototypes already
2758 present for multiple declarations of any given function, these multiple
2759 prototypes *should* all match exactly with one another and with the
2760 prototype for the actual function definition. We don't check for this
2761 here however, since we assume that the compiler must have already done
2762 this consistency checking when it was creating the .X files. */
2764 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2765 if (dd_p->prototyped)
2766 ((NONCONST def_dec_info *) dd_p)->definition = dd_p;
2768 /* Traverse the list of definitions and declarations for this particular
2769 function name. For each item on the list, if it is an extern function
2770 declaration and if it has no associated definition yet, go try to find
2771 the matching extern definition for the declaration.
2773 When looking for the matching function definition, warn the user if we
2774 fail to find one.
2776 If we find more that one function definition also issue a warning.
2778 Do the search for the matching definition only once per unique function
2779 name (and only when absolutely needed) so that we can avoid putting out
2780 redundant warning messages, and so that we will only put out warning
2781 messages when there is actually a reference (i.e. a declaration) for
2782 which we need to find a matching definition. */
2784 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2785 if (!dd_p->is_func_def && !dd_p->is_static && !dd_p->definition)
2787 if (first_extern_reference)
2789 extern_def_p = find_extern_def (hp->ddip, dd_p);
2790 first_extern_reference = 0;
2792 ((NONCONST def_dec_info *) dd_p)->definition = extern_def_p;
2795 /* Traverse the list of definitions and declarations for this particular
2796 function name. For each item on the list, if it is a static function
2797 declaration and if it has no associated definition yet, go try to find
2798 the matching static definition for the declaration within the same file.
2800 When looking for the matching function definition, warn the user if we
2801 fail to find one in the same file with the declaration, and refuse to
2802 convert this kind of cross-file static function declaration. After all,
2803 this is stupid practice and should be discouraged.
2805 We don't have to worry about the possibility that there is more than one
2806 matching function definition in the given file because that would have
2807 been flagged as an error by the compiler.
2809 Do the search for the matching definition only once per unique
2810 function-name/source-file pair (and only when absolutely needed) so that
2811 we can avoid putting out redundant warning messages, and so that we will
2812 only put out warning messages when there is actually a reference (i.e. a
2813 declaration) for which we actually need to find a matching definition. */
2815 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2816 if (!dd_p->is_func_def && dd_p->is_static && !dd_p->definition)
2818 const def_dec_info *dd_p2;
2819 const def_dec_info *static_def;
2821 /* We have now found a single static declaration for which we need to
2822 find a matching definition. We want to minimize the work (and the
2823 number of warnings), so we will find an appropriate (matching)
2824 static definition for this declaration, and then distribute it
2825 (as the definition for) any and all other static declarations
2826 for this function name which occur within the same file, and which
2827 do not already have definitions.
2829 Note that a trick is used here to prevent subsequent attempts to
2830 call find_static_definition for a given function-name & file
2831 if the first such call returns NULL. Essentially, we convert
2832 these NULL return values to -1, and put the -1 into the definition
2833 field for each other static declaration from the same file which
2834 does not already have an associated definition.
2835 This makes these other static declarations look like they are
2836 actually defined already when the outer loop here revisits them
2837 later on. Thus, the outer loop will skip over them. Later, we
2838 turn the -1's back to NULL's. */
2840 ((NONCONST def_dec_info *) dd_p)->definition =
2841 (static_def = find_static_definition (dd_p))
2842 ? static_def
2843 : (const def_dec_info *) -1;
2845 for (dd_p2 = dd_p->next_for_func; dd_p2; dd_p2 = dd_p2->next_for_func)
2846 if (!dd_p2->is_func_def && dd_p2->is_static
2847 && !dd_p2->definition && (dd_p2->file == dd_p->file))
2848 ((NONCONST def_dec_info *)dd_p2)->definition = dd_p->definition;
2851 /* Convert any dummy (-1) definitions we created in the step above back to
2852 NULL's (as they should be). */
2854 for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
2855 if (dd_p->definition == (def_dec_info *) -1)
2856 ((NONCONST def_dec_info *) dd_p)->definition = NULL;
2859 #endif /* !defined (UNPROTOIZE) */
2861 /* Give a pointer into the clean text buffer, return a number which is the
2862 original source line number that the given pointer points into. */
2864 static int
2865 identify_lineno (clean_p)
2866 const char *clean_p;
2868 int line_num = 1;
2869 const char *scan_p;
2871 for (scan_p = clean_text_base; scan_p <= clean_p; scan_p++)
2872 if (*scan_p == '\n')
2873 line_num++;
2874 return line_num;
2877 /* Issue an error message and give up on doing this particular edit. */
2879 static void
2880 declare_source_confusing (clean_p)
2881 const char *clean_p;
2883 if (!quiet_flag)
2885 if (clean_p == 0)
2886 notice ("%s: %d: warning: source too confusing\n",
2887 shortpath (NULL, convert_filename), last_known_line_number);
2888 else
2889 notice ("%s: %d: warning: source too confusing\n",
2890 shortpath (NULL, convert_filename),
2891 identify_lineno (clean_p));
2893 longjmp (source_confusion_recovery, 1);
2896 /* Check that a condition which is expected to be true in the original source
2897 code is in fact true. If not, issue an error message and give up on
2898 converting this particular source file. */
2900 static void
2901 check_source (cond, clean_p)
2902 int cond;
2903 const char *clean_p;
2905 if (!cond)
2906 declare_source_confusing (clean_p);
2909 /* If we think of the in-core cleaned text buffer as a memory mapped
2910 file (with the variable last_known_line_start acting as sort of a
2911 file pointer) then we can imagine doing "seeks" on the buffer. The
2912 following routine implements a kind of "seek" operation for the in-core
2913 (cleaned) copy of the source file. When finished, it returns a pointer to
2914 the start of a given (numbered) line in the cleaned text buffer.
2916 Note that protoize only has to "seek" in the forward direction on the
2917 in-core cleaned text file buffers, and it never needs to back up.
2919 This routine is made a little bit faster by remembering the line number
2920 (and pointer value) supplied (and returned) from the previous "seek".
2921 This prevents us from always having to start all over back at the top
2922 of the in-core cleaned buffer again. */
2924 static const char *
2925 seek_to_line (n)
2926 int n;
2928 if (n < last_known_line_number)
2929 abort ();
2931 while (n > last_known_line_number)
2933 while (*last_known_line_start != '\n')
2934 check_source (++last_known_line_start < clean_text_limit, 0);
2935 last_known_line_start++;
2936 last_known_line_number++;
2938 return last_known_line_start;
2941 /* Given a pointer to a character in the cleaned text buffer, return a pointer
2942 to the next non-whitespace character which follows it. */
2944 static const char *
2945 forward_to_next_token_char (ptr)
2946 const char *ptr;
2948 for (++ptr; ISSPACE ((const unsigned char)*ptr);
2949 check_source (++ptr < clean_text_limit, 0))
2950 continue;
2951 return ptr;
2954 /* Copy a chunk of text of length `len' and starting at `str' to the current
2955 output buffer. Note that all attempts to add stuff to the current output
2956 buffer ultimately go through here. */
2958 static void
2959 output_bytes (str, len)
2960 const char *str;
2961 size_t len;
2963 if ((repl_write_ptr + 1) + len >= repl_text_limit)
2965 size_t new_size = (repl_text_limit - repl_text_base) << 1;
2966 char *new_buf = (char *) xrealloc (repl_text_base, new_size);
2968 repl_write_ptr = new_buf + (repl_write_ptr - repl_text_base);
2969 repl_text_base = new_buf;
2970 repl_text_limit = new_buf + new_size;
2972 memcpy (repl_write_ptr + 1, str, len);
2973 repl_write_ptr += len;
2976 /* Copy all bytes (except the trailing null) of a null terminated string to
2977 the current output buffer. */
2979 static void
2980 output_string (str)
2981 const char *str;
2983 output_bytes (str, strlen (str));
2986 /* Copy some characters from the original text buffer to the current output
2987 buffer.
2989 This routine takes a pointer argument `p' which is assumed to be a pointer
2990 into the cleaned text buffer. The bytes which are copied are the `original'
2991 equivalents for the set of bytes between the last value of `clean_read_ptr'
2992 and the argument value `p'.
2994 The set of bytes copied however, comes *not* from the cleaned text buffer,
2995 but rather from the direct counterparts of these bytes within the original
2996 text buffer.
2998 Thus, when this function is called, some bytes from the original text
2999 buffer (which may include original comments and preprocessing directives)
3000 will be copied into the output buffer.
3002 Note that the request implied when this routine is called includes the
3003 byte pointed to by the argument pointer `p'. */
3005 static void
3006 output_up_to (p)
3007 const char *p;
3009 size_t copy_length = (size_t) (p - clean_read_ptr);
3010 const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
3012 if (copy_length == 0)
3013 return;
3015 output_bytes (copy_start, copy_length);
3016 clean_read_ptr = p;
3019 /* Given a pointer to a def_dec_info record which represents some form of
3020 definition of a function (perhaps a real definition, or in lieu of that
3021 perhaps just a declaration with a full prototype) return true if this
3022 function is one which we should avoid converting. Return false
3023 otherwise. */
3025 static int
3026 other_variable_style_function (ansi_header)
3027 const char *ansi_header;
3029 #ifdef UNPROTOIZE
3031 /* See if we have a stdarg function, or a function which has stdarg style
3032 parameters or a stdarg style return type. */
3034 return substr (ansi_header, "...") != 0;
3036 #else /* !defined (UNPROTOIZE) */
3038 /* See if we have a varargs function, or a function which has varargs style
3039 parameters or a varargs style return type. */
3041 const char *p;
3042 int len = strlen (varargs_style_indicator);
3044 for (p = ansi_header; p; )
3046 const char *candidate;
3048 if ((candidate = substr (p, varargs_style_indicator)) == 0)
3049 return 0;
3050 else
3051 if (!is_id_char (candidate[-1]) && !is_id_char (candidate[len]))
3052 return 1;
3053 else
3054 p = candidate + 1;
3056 return 0;
3057 #endif /* !defined (UNPROTOIZE) */
3060 /* Do the editing operation specifically for a function "declaration". Note
3061 that editing for function "definitions" are handled in a separate routine
3062 below. */
3064 static void
3065 edit_fn_declaration (def_dec_p, clean_text_p)
3066 const def_dec_info *def_dec_p;
3067 const char *volatile clean_text_p;
3069 const char *start_formals;
3070 const char *end_formals;
3071 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3072 size_t func_name_len = strlen (function_to_edit);
3073 const char *end_of_fn_name;
3075 #ifndef UNPROTOIZE
3077 const f_list_chain_item *this_f_list_chain_item;
3078 const def_dec_info *definition = def_dec_p->definition;
3080 /* If we are protoizing, and if we found no corresponding definition for
3081 this particular function declaration, then just leave this declaration
3082 exactly as it is. */
3084 if (!definition)
3085 return;
3087 /* If we are protoizing, and if the corresponding definition that we found
3088 for this particular function declaration defined an old style varargs
3089 function, then we want to issue a warning and just leave this function
3090 declaration unconverted. */
3092 if (other_variable_style_function (definition->ansi_decl))
3094 if (!quiet_flag)
3095 notice ("%s: %d: warning: varargs function declaration not converted\n",
3096 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3097 def_dec_p->line);
3098 return;
3101 #endif /* !defined (UNPROTOIZE) */
3103 /* Setup here to recover from confusing source code detected during this
3104 particular "edit". */
3106 save_pointers ();
3107 if (setjmp (source_confusion_recovery))
3109 restore_pointers ();
3110 notice ("%s: declaration of function `%s' not converted\n",
3111 pname, function_to_edit);
3112 return;
3115 /* We are editing a function declaration. The line number we did a seek to
3116 contains the comma or semicolon which follows the declaration. Our job
3117 now is to scan backwards looking for the function name. This name *must*
3118 be followed by open paren (ignoring whitespace, of course). We need to
3119 replace everything between that open paren and the corresponding closing
3120 paren. If we are protoizing, we need to insert the prototype-style
3121 formals lists. If we are unprotoizing, we need to just delete everything
3122 between the pairs of opening and closing parens. */
3124 /* First move up to the end of the line. */
3126 while (*clean_text_p != '\n')
3127 check_source (++clean_text_p < clean_text_limit, 0);
3128 clean_text_p--; /* Point to just before the newline character. */
3130 /* Now we can scan backwards for the function name. */
3134 for (;;)
3136 /* Scan leftwards until we find some character which can be
3137 part of an identifier. */
3139 while (!is_id_char (*clean_text_p))
3140 check_source (--clean_text_p > clean_read_ptr, 0);
3142 /* Scan backwards until we find a char that cannot be part of an
3143 identifier. */
3145 while (is_id_char (*clean_text_p))
3146 check_source (--clean_text_p > clean_read_ptr, 0);
3148 /* Having found an "id break", see if the following id is the one
3149 that we are looking for. If so, then exit from this loop. */
3151 if (!strncmp (clean_text_p+1, function_to_edit, func_name_len))
3153 char ch = *(clean_text_p + 1 + func_name_len);
3155 /* Must also check to see that the name in the source text
3156 ends where it should (in order to prevent bogus matches
3157 on similar but longer identifiers. */
3159 if (! is_id_char (ch))
3160 break; /* exit from loop */
3164 /* We have now found the first perfect match for the function name in
3165 our backward search. This may or may not be the actual function
3166 name at the start of the actual function declaration (i.e. we could
3167 have easily been mislead). We will try to avoid getting fooled too
3168 often by looking forward for the open paren which should follow the
3169 identifier we just found. We ignore whitespace while hunting. If
3170 the next non-whitespace byte we see is *not* an open left paren,
3171 then we must assume that we have been fooled and we start over
3172 again accordingly. Note that there is no guarantee, that even if
3173 we do see the open paren, that we are in the right place.
3174 Programmers do the strangest things sometimes! */
3176 end_of_fn_name = clean_text_p + strlen (def_dec_p->hash_entry->symbol);
3177 start_formals = forward_to_next_token_char (end_of_fn_name);
3179 while (*start_formals != '(');
3181 /* start_of_formals now points to the opening left paren which immediately
3182 follows the name of the function. */
3184 /* Note that there may be several formals lists which need to be modified
3185 due to the possibility that the return type of this function is a
3186 pointer-to-function type. If there are several formals lists, we
3187 convert them in left-to-right order here. */
3189 #ifndef UNPROTOIZE
3190 this_f_list_chain_item = definition->f_list_chain;
3191 #endif /* !defined (UNPROTOIZE) */
3193 for (;;)
3196 int depth;
3198 end_formals = start_formals + 1;
3199 depth = 1;
3200 for (; depth; check_source (++end_formals < clean_text_limit, 0))
3202 switch (*end_formals)
3204 case '(':
3205 depth++;
3206 break;
3207 case ')':
3208 depth--;
3209 break;
3212 end_formals--;
3215 /* end_formals now points to the closing right paren of the formals
3216 list whose left paren is pointed to by start_formals. */
3218 /* Now, if we are protoizing, we insert the new ANSI-style formals list
3219 attached to the associated definition of this function. If however
3220 we are unprotoizing, then we simply delete any formals list which
3221 may be present. */
3223 output_up_to (start_formals);
3224 #ifndef UNPROTOIZE
3225 if (this_f_list_chain_item)
3227 output_string (this_f_list_chain_item->formals_list);
3228 this_f_list_chain_item = this_f_list_chain_item->chain_next;
3230 else
3232 if (!quiet_flag)
3233 notice ("%s: warning: too many parameter lists in declaration of `%s'\n",
3234 pname, def_dec_p->hash_entry->symbol);
3235 check_source (0, end_formals); /* leave the declaration intact */
3237 #endif /* !defined (UNPROTOIZE) */
3238 clean_read_ptr = end_formals - 1;
3240 /* Now see if it looks like there may be another formals list associated
3241 with the function declaration that we are converting (following the
3242 formals list that we just converted. */
3245 const char *another_r_paren = forward_to_next_token_char (end_formals);
3247 if ((*another_r_paren != ')')
3248 || (*(start_formals = forward_to_next_token_char (another_r_paren)) != '('))
3250 #ifndef UNPROTOIZE
3251 if (this_f_list_chain_item)
3253 if (!quiet_flag)
3254 notice ("\n%s: warning: too few parameter lists in declaration of `%s'\n",
3255 pname, def_dec_p->hash_entry->symbol);
3256 check_source (0, start_formals); /* leave the decl intact */
3258 #endif /* !defined (UNPROTOIZE) */
3259 break;
3264 /* There does appear to be yet another formals list, so loop around
3265 again, and convert it also. */
3269 /* Edit a whole group of formals lists, starting with the rightmost one
3270 from some set of formals lists. This routine is called once (from the
3271 outside) for each function declaration which is converted. It is
3272 recursive however, and it calls itself once for each remaining formal
3273 list that lies to the left of the one it was originally called to work
3274 on. Thus, a whole set gets done in right-to-left order.
3276 This routine returns non-zero if it thinks that it should not be trying
3277 to convert this particular function definition (because the name of the
3278 function doesn't match the one expected). */
3280 static int
3281 edit_formals_lists (end_formals, f_list_count, def_dec_p)
3282 const char *end_formals;
3283 unsigned int f_list_count;
3284 const def_dec_info *def_dec_p;
3286 const char *start_formals;
3287 int depth;
3289 start_formals = end_formals - 1;
3290 depth = 1;
3291 for (; depth; check_source (--start_formals > clean_read_ptr, 0))
3293 switch (*start_formals)
3295 case '(':
3296 depth--;
3297 break;
3298 case ')':
3299 depth++;
3300 break;
3303 start_formals++;
3305 /* start_formals now points to the opening left paren of the formals list. */
3307 f_list_count--;
3309 if (f_list_count)
3311 const char *next_end;
3313 /* There should be more formal lists to the left of here. */
3315 next_end = start_formals - 1;
3316 check_source (next_end > clean_read_ptr, 0);
3317 while (ISSPACE ((const unsigned char)*next_end))
3318 check_source (--next_end > clean_read_ptr, 0);
3319 check_source (*next_end == ')', next_end);
3320 check_source (--next_end > clean_read_ptr, 0);
3321 check_source (*next_end == ')', next_end);
3322 if (edit_formals_lists (next_end, f_list_count, def_dec_p))
3323 return 1;
3326 /* Check that the function name in the header we are working on is the same
3327 as the one we would expect to find. If not, issue a warning and return
3328 non-zero. */
3330 if (f_list_count == 0)
3332 const char *expected = def_dec_p->hash_entry->symbol;
3333 const char *func_name_start;
3334 const char *func_name_limit;
3335 size_t func_name_len;
3337 for (func_name_limit = start_formals-1;
3338 ISSPACE ((const unsigned char)*func_name_limit); )
3339 check_source (--func_name_limit > clean_read_ptr, 0);
3341 for (func_name_start = func_name_limit++;
3342 is_id_char (*func_name_start);
3343 func_name_start--)
3344 check_source (func_name_start > clean_read_ptr, 0);
3345 func_name_start++;
3346 func_name_len = func_name_limit - func_name_start;
3347 if (func_name_len == 0)
3348 check_source (0, func_name_start);
3349 if (func_name_len != strlen (expected)
3350 || strncmp (func_name_start, expected, func_name_len))
3352 notice ("%s: %d: warning: found `%s' but expected `%s'\n",
3353 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3354 identify_lineno (func_name_start),
3355 dupnstr (func_name_start, func_name_len),
3356 expected);
3357 return 1;
3361 output_up_to (start_formals);
3363 #ifdef UNPROTOIZE
3364 if (f_list_count == 0)
3365 output_string (def_dec_p->formal_names);
3366 #else /* !defined (UNPROTOIZE) */
3368 unsigned f_list_depth;
3369 const f_list_chain_item *flci_p = def_dec_p->f_list_chain;
3371 /* At this point, the current value of f_list count says how many
3372 links we have to follow through the f_list_chain to get to the
3373 particular formals list that we need to output next. */
3375 for (f_list_depth = 0; f_list_depth < f_list_count; f_list_depth++)
3376 flci_p = flci_p->chain_next;
3377 output_string (flci_p->formals_list);
3379 #endif /* !defined (UNPROTOIZE) */
3381 clean_read_ptr = end_formals - 1;
3382 return 0;
3385 /* Given a pointer to a byte in the clean text buffer which points to
3386 the beginning of a line that contains a "follower" token for a
3387 function definition header, do whatever is necessary to find the
3388 right closing paren for the rightmost formals list of the function
3389 definition header. */
3391 static const char *
3392 find_rightmost_formals_list (clean_text_p)
3393 const char *clean_text_p;
3395 const char *end_formals;
3397 /* We are editing a function definition. The line number we did a seek
3398 to contains the first token which immediately follows the entire set of
3399 formals lists which are part of this particular function definition
3400 header.
3402 Our job now is to scan leftwards in the clean text looking for the
3403 right-paren which is at the end of the function header's rightmost
3404 formals list.
3406 If we ignore whitespace, this right paren should be the first one we
3407 see which is (ignoring whitespace) immediately followed either by the
3408 open curly-brace beginning the function body or by an alphabetic
3409 character (in the case where the function definition is in old (K&R)
3410 style and there are some declarations of formal parameters). */
3412 /* It is possible that the right paren we are looking for is on the
3413 current line (together with its following token). Just in case that
3414 might be true, we start out here by skipping down to the right end of
3415 the current line before starting our scan. */
3417 for (end_formals = clean_text_p; *end_formals != '\n'; end_formals++)
3418 continue;
3419 end_formals--;
3421 #ifdef UNPROTOIZE
3423 /* Now scan backwards while looking for the right end of the rightmost
3424 formals list associated with this function definition. */
3427 char ch;
3428 const char *l_brace_p;
3430 /* Look leftward and try to find a right-paren. */
3432 while (*end_formals != ')')
3434 if (ISSPACE ((unsigned char)*end_formals))
3435 while (ISSPACE ((unsigned char)*end_formals))
3436 check_source (--end_formals > clean_read_ptr, 0);
3437 else
3438 check_source (--end_formals > clean_read_ptr, 0);
3441 ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3442 /* Since we are unprotoizing an ANSI-style (prototyped) function
3443 definition, there had better not be anything (except whitespace)
3444 between the end of the ANSI formals list and the beginning of the
3445 function body (i.e. the '{'). */
3447 check_source (ch == '{', l_brace_p);
3450 #else /* !defined (UNPROTOIZE) */
3452 /* Now scan backwards while looking for the right end of the rightmost
3453 formals list associated with this function definition. */
3455 while (1)
3457 char ch;
3458 const char *l_brace_p;
3460 /* Look leftward and try to find a right-paren. */
3462 while (*end_formals != ')')
3464 if (ISSPACE ((const unsigned char)*end_formals))
3465 while (ISSPACE ((const unsigned char)*end_formals))
3466 check_source (--end_formals > clean_read_ptr, 0);
3467 else
3468 check_source (--end_formals > clean_read_ptr, 0);
3471 ch = *(l_brace_p = forward_to_next_token_char (end_formals));
3473 /* Since it is possible that we found a right paren before the starting
3474 '{' of the body which IS NOT the one at the end of the real K&R
3475 formals list (say for instance, we found one embedded inside one of
3476 the old K&R formal parameter declarations) we have to check to be
3477 sure that this is in fact the right paren that we were looking for.
3479 The one we were looking for *must* be followed by either a '{' or
3480 by an alphabetic character, while others *cannot* validly be followed
3481 by such characters. */
3483 if ((ch == '{') || ISALPHA ((unsigned char)ch))
3484 break;
3486 /* At this point, we have found a right paren, but we know that it is
3487 not the one we were looking for, so backup one character and keep
3488 looking. */
3490 check_source (--end_formals > clean_read_ptr, 0);
3493 #endif /* !defined (UNPROTOIZE) */
3495 return end_formals;
3498 #ifndef UNPROTOIZE
3500 /* Insert into the output file a totally new declaration for a function
3501 which (up until now) was being called from within the current block
3502 without having been declared at any point such that the declaration
3503 was visible (i.e. in scope) at the point of the call.
3505 We need to add in explicit declarations for all such function calls
3506 in order to get the full benefit of prototype-based function call
3507 parameter type checking. */
3509 static void
3510 add_local_decl (def_dec_p, clean_text_p)
3511 const def_dec_info *def_dec_p;
3512 const char *clean_text_p;
3514 const char *start_of_block;
3515 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3517 /* Don't insert new local explicit declarations unless explicitly requested
3518 to do so. */
3520 if (!local_flag)
3521 return;
3523 /* Setup here to recover from confusing source code detected during this
3524 particular "edit". */
3526 save_pointers ();
3527 if (setjmp (source_confusion_recovery))
3529 restore_pointers ();
3530 notice ("%s: local declaration for function `%s' not inserted\n",
3531 pname, function_to_edit);
3532 return;
3535 /* We have already done a seek to the start of the line which should
3536 contain *the* open curly brace which begins the block in which we need
3537 to insert an explicit function declaration (to replace the implicit one).
3539 Now we scan that line, starting from the left, until we find the
3540 open curly brace we are looking for. Note that there may actually be
3541 multiple open curly braces on the given line, but we will be happy
3542 with the leftmost one no matter what. */
3544 start_of_block = clean_text_p;
3545 while (*start_of_block != '{' && *start_of_block != '\n')
3546 check_source (++start_of_block < clean_text_limit, 0);
3548 /* Note that the line from the original source could possibly
3549 contain *no* open curly braces! This happens if the line contains
3550 a macro call which expands into a chunk of text which includes a
3551 block (and that block's associated open and close curly braces).
3552 In cases like this, we give up, issue a warning, and do nothing. */
3554 if (*start_of_block != '{')
3556 if (!quiet_flag)
3557 notice ("\n%s: %d: warning: can't add declaration of `%s' into macro call\n",
3558 def_dec_p->file->hash_entry->symbol, def_dec_p->line,
3559 def_dec_p->hash_entry->symbol);
3560 return;
3563 /* Figure out what a nice (pretty) indentation would be for the new
3564 declaration we are adding. In order to do this, we must scan forward
3565 from the '{' until we find the first line which starts with some
3566 non-whitespace characters (i.e. real "token" material). */
3569 const char *ep = forward_to_next_token_char (start_of_block) - 1;
3570 const char *sp;
3572 /* Now we have ep pointing at the rightmost byte of some existing indent
3573 stuff. At least that is the hope.
3575 We can now just scan backwards and find the left end of the existing
3576 indentation string, and then copy it to the output buffer. */
3578 for (sp = ep; ISSPACE ((const unsigned char)*sp) && *sp != '\n'; sp--)
3579 continue;
3581 /* Now write out the open { which began this block, and any following
3582 trash up to and including the last byte of the existing indent that
3583 we just found. */
3585 output_up_to (ep);
3587 /* Now we go ahead and insert the new declaration at this point.
3589 If the definition of the given function is in the same file that we
3590 are currently editing, and if its full ANSI declaration normally
3591 would start with the keyword `extern', suppress the `extern'. */
3594 const char *decl = def_dec_p->definition->ansi_decl;
3596 if ((*decl == 'e') && (def_dec_p->file == def_dec_p->definition->file))
3597 decl += 7;
3598 output_string (decl);
3601 /* Finally, write out a new indent string, just like the preceding one
3602 that we found. This will typically include a newline as the first
3603 character of the indent string. */
3605 output_bytes (sp, (size_t) (ep - sp) + 1);
3609 /* Given a pointer to a file_info record, and a pointer to the beginning
3610 of a line (in the clean text buffer) which is assumed to contain the
3611 first "follower" token for the first function definition header in the
3612 given file, find a good place to insert some new global function
3613 declarations (which will replace scattered and imprecise implicit ones)
3614 and then insert the new explicit declaration at that point in the file. */
3616 static void
3617 add_global_decls (file_p, clean_text_p)
3618 const file_info *file_p;
3619 const char *clean_text_p;
3621 const def_dec_info *dd_p;
3622 const char *scan_p;
3624 /* Setup here to recover from confusing source code detected during this
3625 particular "edit". */
3627 save_pointers ();
3628 if (setjmp (source_confusion_recovery))
3630 restore_pointers ();
3631 notice ("%s: global declarations for file `%s' not inserted\n",
3632 pname, shortpath (NULL, file_p->hash_entry->symbol));
3633 return;
3636 /* Start by finding a good location for adding the new explicit function
3637 declarations. To do this, we scan backwards, ignoring whitespace
3638 and comments and other junk until we find either a semicolon, or until
3639 we hit the beginning of the file. */
3641 scan_p = find_rightmost_formals_list (clean_text_p);
3642 for (;; --scan_p)
3644 if (scan_p < clean_text_base)
3645 break;
3646 check_source (scan_p > clean_read_ptr, 0);
3647 if (*scan_p == ';')
3648 break;
3651 /* scan_p now points either to a semicolon, or to just before the start
3652 of the whole file. */
3654 /* Now scan forward for the first non-whitespace character. In theory,
3655 this should be the first character of the following function definition
3656 header. We will put in the added declarations just prior to that. */
3658 scan_p++;
3659 while (ISSPACE ((const unsigned char)*scan_p))
3660 scan_p++;
3661 scan_p--;
3663 output_up_to (scan_p);
3665 /* Now write out full prototypes for all of the things that had been
3666 implicitly declared in this file (but only those for which we were
3667 actually able to find unique matching definitions). Avoid duplicates
3668 by marking things that we write out as we go. */
3671 int some_decls_added = 0;
3673 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3674 if (dd_p->is_implicit && dd_p->definition && !dd_p->definition->written)
3676 const char *decl = dd_p->definition->ansi_decl;
3678 /* If the function for which we are inserting a declaration is
3679 actually defined later in the same file, then suppress the
3680 leading `extern' keyword (if there is one). */
3682 if (*decl == 'e' && (dd_p->file == dd_p->definition->file))
3683 decl += 7;
3685 output_string ("\n");
3686 output_string (decl);
3687 some_decls_added = 1;
3688 ((NONCONST def_dec_info *) dd_p->definition)->written = 1;
3690 if (some_decls_added)
3691 output_string ("\n\n");
3694 /* Unmark all of the definitions that we just marked. */
3696 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
3697 if (dd_p->definition)
3698 ((NONCONST def_dec_info *) dd_p->definition)->written = 0;
3701 #endif /* !defined (UNPROTOIZE) */
3703 /* Do the editing operation specifically for a function "definition". Note
3704 that editing operations for function "declarations" are handled by a
3705 separate routine above. */
3707 static void
3708 edit_fn_definition (def_dec_p, clean_text_p)
3709 const def_dec_info *def_dec_p;
3710 const char *clean_text_p;
3712 const char *end_formals;
3713 const char *function_to_edit = def_dec_p->hash_entry->symbol;
3715 /* Setup here to recover from confusing source code detected during this
3716 particular "edit". */
3718 save_pointers ();
3719 if (setjmp (source_confusion_recovery))
3721 restore_pointers ();
3722 notice ("%s: definition of function `%s' not converted\n",
3723 pname, function_to_edit);
3724 return;
3727 end_formals = find_rightmost_formals_list (clean_text_p);
3729 /* end_of_formals now points to the closing right paren of the rightmost
3730 formals list which is actually part of the `header' of the function
3731 definition that we are converting. */
3733 /* If the header of this function definition looks like it declares a
3734 function with a variable number of arguments, and if the way it does
3735 that is different from that way we would like it (i.e. varargs vs.
3736 stdarg) then issue a warning and leave the header unconverted. */
3738 if (other_variable_style_function (def_dec_p->ansi_decl))
3740 if (!quiet_flag)
3741 notice ("%s: %d: warning: definition of %s not converted\n",
3742 shortpath (NULL, def_dec_p->file->hash_entry->symbol),
3743 identify_lineno (end_formals),
3744 other_var_style);
3745 output_up_to (end_formals);
3746 return;
3749 if (edit_formals_lists (end_formals, def_dec_p->f_list_count, def_dec_p))
3751 restore_pointers ();
3752 notice ("%s: definition of function `%s' not converted\n",
3753 pname, function_to_edit);
3754 return;
3757 /* Have to output the last right paren because this never gets flushed by
3758 edit_formals_list. */
3760 output_up_to (end_formals);
3762 #ifdef UNPROTOIZE
3764 const char *decl_p;
3765 const char *semicolon_p;
3766 const char *limit_p;
3767 const char *scan_p;
3768 int had_newlines = 0;
3770 /* Now write out the K&R style formal declarations, one per line. */
3772 decl_p = def_dec_p->formal_decls;
3773 limit_p = decl_p + strlen (decl_p);
3774 for (;decl_p < limit_p; decl_p = semicolon_p + 2)
3776 for (semicolon_p = decl_p; *semicolon_p != ';'; semicolon_p++)
3777 continue;
3778 output_string ("\n");
3779 output_string (indent_string);
3780 output_bytes (decl_p, (size_t) ((semicolon_p + 1) - decl_p));
3783 /* If there are no newlines between the end of the formals list and the
3784 start of the body, we should insert one now. */
3786 for (scan_p = end_formals+1; *scan_p != '{'; )
3788 if (*scan_p == '\n')
3790 had_newlines = 1;
3791 break;
3793 check_source (++scan_p < clean_text_limit, 0);
3795 if (!had_newlines)
3796 output_string ("\n");
3798 #else /* !defined (UNPROTOIZE) */
3799 /* If we are protoizing, there may be some flotsam & jetsam (like comments
3800 and preprocessing directives) after the old formals list but before
3801 the following { and we would like to preserve that stuff while effectively
3802 deleting the existing K&R formal parameter declarations. We do so here
3803 in a rather tricky way. Basically, we white out any stuff *except*
3804 the comments/pp-directives in the original text buffer, then, if there
3805 is anything in this area *other* than whitespace, we output it. */
3807 const char *end_formals_orig;
3808 const char *start_body;
3809 const char *start_body_orig;
3810 const char *scan;
3811 const char *scan_orig;
3812 int have_flotsam = 0;
3813 int have_newlines = 0;
3815 for (start_body = end_formals + 1; *start_body != '{';)
3816 check_source (++start_body < clean_text_limit, 0);
3818 end_formals_orig = orig_text_base + (end_formals - clean_text_base);
3819 start_body_orig = orig_text_base + (start_body - clean_text_base);
3820 scan = end_formals + 1;
3821 scan_orig = end_formals_orig + 1;
3822 for (; scan < start_body; scan++, scan_orig++)
3824 if (*scan == *scan_orig)
3826 have_newlines |= (*scan_orig == '\n');
3827 /* Leave identical whitespace alone. */
3828 if (!ISSPACE ((const unsigned char)*scan_orig))
3829 *((NONCONST char *)scan_orig) = ' '; /* identical - so whiteout */
3831 else
3832 have_flotsam = 1;
3834 if (have_flotsam)
3835 output_bytes (end_formals_orig + 1,
3836 (size_t) (start_body_orig - end_formals_orig) - 1);
3837 else
3838 if (have_newlines)
3839 output_string ("\n");
3840 else
3841 output_string (" ");
3842 clean_read_ptr = start_body - 1;
3844 #endif /* !defined (UNPROTOIZE) */
3847 /* Clean up the clean text buffer. Do this by converting comments and
3848 preprocessing directives into spaces. Also convert line continuations
3849 into whitespace. Also, whiteout string and character literals. */
3851 static void
3852 do_cleaning (new_clean_text_base, new_clean_text_limit)
3853 char *new_clean_text_base;
3854 char *new_clean_text_limit;
3856 char *scan_p;
3857 int non_whitespace_since_newline = 0;
3859 for (scan_p = new_clean_text_base; scan_p < new_clean_text_limit; scan_p++)
3861 switch (*scan_p)
3863 case '/': /* Handle comments. */
3864 if (scan_p[1] != '*')
3865 goto regular;
3866 non_whitespace_since_newline = 1;
3867 scan_p[0] = ' ';
3868 scan_p[1] = ' ';
3869 scan_p += 2;
3870 while (scan_p[1] != '/' || scan_p[0] != '*')
3872 if (!ISSPACE ((const unsigned char)*scan_p))
3873 *scan_p = ' ';
3874 if (++scan_p >= new_clean_text_limit)
3875 abort ();
3877 *scan_p++ = ' ';
3878 *scan_p = ' ';
3879 break;
3881 case '#': /* Handle pp directives. */
3882 if (non_whitespace_since_newline)
3883 goto regular;
3884 *scan_p = ' ';
3885 while (scan_p[1] != '\n' || scan_p[0] == '\\')
3887 if (!ISSPACE ((const unsigned char)*scan_p))
3888 *scan_p = ' ';
3889 if (++scan_p >= new_clean_text_limit)
3890 abort ();
3892 *scan_p++ = ' ';
3893 break;
3895 case '\'': /* Handle character literals. */
3896 non_whitespace_since_newline = 1;
3897 while (scan_p[1] != '\'' || scan_p[0] == '\\')
3899 if (scan_p[0] == '\\'
3900 && !ISSPACE ((const unsigned char)scan_p[1]))
3901 scan_p[1] = ' ';
3902 if (!ISSPACE ((const unsigned char)*scan_p))
3903 *scan_p = ' ';
3904 if (++scan_p >= new_clean_text_limit)
3905 abort ();
3907 *scan_p++ = ' ';
3908 break;
3910 case '"': /* Handle string literals. */
3911 non_whitespace_since_newline = 1;
3912 while (scan_p[1] != '"' || scan_p[0] == '\\')
3914 if (scan_p[0] == '\\'
3915 && !ISSPACE ((const unsigned char)scan_p[1]))
3916 scan_p[1] = ' ';
3917 if (!ISSPACE ((const unsigned char)*scan_p))
3918 *scan_p = ' ';
3919 if (++scan_p >= new_clean_text_limit)
3920 abort ();
3922 if (!ISSPACE ((const unsigned char)*scan_p))
3923 *scan_p = ' ';
3924 scan_p++;
3925 break;
3927 case '\\': /* Handle line continuations. */
3928 if (scan_p[1] != '\n')
3929 goto regular;
3930 *scan_p = ' ';
3931 break;
3933 case '\n':
3934 non_whitespace_since_newline = 0; /* Reset. */
3935 break;
3937 case ' ':
3938 case '\v':
3939 case '\t':
3940 case '\r':
3941 case '\f':
3942 case '\b':
3943 break; /* Whitespace characters. */
3945 default:
3946 regular:
3947 non_whitespace_since_newline = 1;
3948 break;
3953 /* Given a pointer to the closing right parenthesis for a particular formals
3954 list (in the clean text buffer) find the corresponding left parenthesis
3955 and return a pointer to it. */
3957 static const char *
3958 careful_find_l_paren (p)
3959 const char *p;
3961 const char *q;
3962 int paren_depth;
3964 for (paren_depth = 1, q = p-1; paren_depth; check_source (--q >= clean_text_base, 0))
3966 switch (*q)
3968 case ')':
3969 paren_depth++;
3970 break;
3971 case '(':
3972 paren_depth--;
3973 break;
3976 return ++q;
3979 /* Scan the clean text buffer for cases of function definitions that we
3980 don't really know about because they were preprocessed out when the
3981 aux info files were created.
3983 In this version of protoize/unprotoize we just give a warning for each
3984 one found. A later version may be able to at least unprotoize such
3985 missed items.
3987 Note that we may easily find all function definitions simply by
3988 looking for places where there is a left paren which is (ignoring
3989 whitespace) immediately followed by either a left-brace or by an
3990 upper or lower case letter. Whenever we find this combination, we
3991 have also found a function definition header.
3993 Finding function *declarations* using syntactic clues is much harder.
3994 I will probably try to do this in a later version though. */
3996 static void
3997 scan_for_missed_items (file_p)
3998 const file_info *file_p;
4000 static const char *scan_p;
4001 const char *limit = clean_text_limit - 3;
4002 static const char *backup_limit;
4004 backup_limit = clean_text_base - 1;
4006 for (scan_p = clean_text_base; scan_p < limit; scan_p++)
4008 if (*scan_p == ')')
4010 static const char *last_r_paren;
4011 const char *ahead_p;
4013 last_r_paren = scan_p;
4015 for (ahead_p = scan_p + 1; ISSPACE ((const unsigned char)*ahead_p); )
4016 check_source (++ahead_p < limit, limit);
4018 scan_p = ahead_p - 1;
4020 if (ISALPHA ((const unsigned char)*ahead_p) || *ahead_p == '{')
4022 const char *last_l_paren;
4023 const int lineno = identify_lineno (ahead_p);
4025 if (setjmp (source_confusion_recovery))
4026 continue;
4028 /* We know we have a function definition header. Now skip
4029 leftwards over all of its associated formals lists. */
4033 last_l_paren = careful_find_l_paren (last_r_paren);
4034 for (last_r_paren = last_l_paren-1;
4035 ISSPACE ((const unsigned char)*last_r_paren); )
4036 check_source (--last_r_paren >= backup_limit, backup_limit);
4038 while (*last_r_paren == ')');
4040 if (is_id_char (*last_r_paren))
4042 const char *id_limit = last_r_paren + 1;
4043 const char *id_start;
4044 size_t id_length;
4045 const def_dec_info *dd_p;
4047 for (id_start = id_limit-1; is_id_char (*id_start); )
4048 check_source (--id_start >= backup_limit, backup_limit);
4049 id_start++;
4050 backup_limit = id_start;
4051 if ((id_length = (size_t) (id_limit - id_start)) == 0)
4052 goto not_missed;
4055 char *func_name = (char *) alloca (id_length + 1);
4056 static const char * const stmt_keywords[]
4057 = { "if", "else", "do", "while", "for", "switch", "case", "return", 0 };
4058 const char * const *stmt_keyword;
4060 strncpy (func_name, id_start, id_length);
4061 func_name[id_length] = '\0';
4063 /* We must check here to see if we are actually looking at
4064 a statement rather than an actual function call. */
4066 for (stmt_keyword = stmt_keywords; *stmt_keyword; stmt_keyword++)
4067 if (!strcmp (func_name, *stmt_keyword))
4068 goto not_missed;
4070 #if 0
4071 notice ("%s: found definition of `%s' at %s(%d)\n",
4072 pname,
4073 func_name,
4074 shortpath (NULL, file_p->hash_entry->symbol),
4075 identify_lineno (id_start));
4076 #endif /* 0 */
4077 /* We really should check for a match of the function name
4078 here also, but why bother. */
4080 for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
4081 if (dd_p->is_func_def && dd_p->line == lineno)
4082 goto not_missed;
4084 /* If we make it here, then we did not know about this
4085 function definition. */
4087 notice ("%s: %d: warning: `%s' excluded by preprocessing\n",
4088 shortpath (NULL, file_p->hash_entry->symbol),
4089 identify_lineno (id_start), func_name);
4090 notice ("%s: function definition not converted\n",
4091 pname);
4093 not_missed: ;
4100 /* Do all editing operations for a single source file (either a "base" file
4101 or an "include" file). To do this we read the file into memory, keep a
4102 virgin copy there, make another cleaned in-core copy of the original file
4103 (i.e. one in which all of the comments and preprocessing directives have
4104 been replaced with whitespace), then use these two in-core copies of the
4105 file to make a new edited in-core copy of the file. Finally, rename the
4106 original file (as a way of saving it), and then write the edited version
4107 of the file from core to a disk file of the same name as the original.
4109 Note that the trick of making a copy of the original sans comments &
4110 preprocessing directives make the editing a whole lot easier. */
4112 static void
4113 edit_file (hp)
4114 const hash_table_entry *hp;
4116 struct stat stat_buf;
4117 const file_info *file_p = hp->fip;
4118 char *new_orig_text_base;
4119 char *new_orig_text_limit;
4120 char *new_clean_text_base;
4121 char *new_clean_text_limit;
4122 size_t orig_size;
4123 size_t repl_size;
4124 int first_definition_in_file;
4126 /* If we are not supposed to be converting this file, or if there is
4127 nothing in there which needs converting, just skip this file. */
4129 if (!needs_to_be_converted (file_p))
4130 return;
4132 convert_filename = file_p->hash_entry->symbol;
4134 /* Convert a file if it is in a directory where we want conversion
4135 and the file is not excluded. */
4137 if (!directory_specified_p (convert_filename)
4138 || file_excluded_p (convert_filename))
4140 if (!quiet_flag
4141 #ifdef UNPROTOIZE
4142 /* Don't even mention "system" include files unless we are
4143 protoizing. If we are protoizing, we mention these as a
4144 gentle way of prodding the user to convert his "system"
4145 include files to prototype format. */
4146 && !in_system_include_dir (convert_filename)
4147 #endif /* defined (UNPROTOIZE) */
4149 notice ("%s: `%s' not converted\n",
4150 pname, shortpath (NULL, convert_filename));
4151 return;
4154 /* Let the user know what we are up to. */
4156 if (nochange_flag)
4157 notice ("%s: would convert file `%s'\n",
4158 pname, shortpath (NULL, convert_filename));
4159 else
4160 notice ("%s: converting file `%s'\n",
4161 pname, shortpath (NULL, convert_filename));
4162 fflush (stderr);
4164 /* Find out the size (in bytes) of the original file. */
4166 /* The cast avoids an erroneous warning on AIX. */
4167 if (my_stat ((char *)convert_filename, &stat_buf) == -1)
4169 int errno_val = errno;
4170 notice ("%s: can't get status for file `%s': %s\n",
4171 pname, shortpath (NULL, convert_filename),
4172 xstrerror (errno_val));
4173 return;
4175 orig_size = stat_buf.st_size;
4177 /* Allocate a buffer to hold the original text. */
4179 orig_text_base = new_orig_text_base = (char *) xmalloc (orig_size + 2);
4180 orig_text_limit = new_orig_text_limit = new_orig_text_base + orig_size;
4182 /* Allocate a buffer to hold the cleaned-up version of the original text. */
4184 clean_text_base = new_clean_text_base = (char *) xmalloc (orig_size + 2);
4185 clean_text_limit = new_clean_text_limit = new_clean_text_base + orig_size;
4186 clean_read_ptr = clean_text_base - 1;
4188 /* Allocate a buffer that will hopefully be large enough to hold the entire
4189 converted output text. As an initial guess for the maximum size of the
4190 output buffer, use 125% of the size of the original + some extra. This
4191 buffer can be expanded later as needed. */
4193 repl_size = orig_size + (orig_size >> 2) + 4096;
4194 repl_text_base = (char *) xmalloc (repl_size + 2);
4195 repl_text_limit = repl_text_base + repl_size - 1;
4196 repl_write_ptr = repl_text_base - 1;
4199 int input_file;
4201 /* Open the file to be converted in READ ONLY mode. */
4203 if ((input_file = my_open (convert_filename, O_RDONLY, 0444)) == -1)
4205 int errno_val = errno;
4206 notice ("%s: can't open file `%s' for reading: %s\n",
4207 pname, shortpath (NULL, convert_filename),
4208 xstrerror (errno_val));
4209 return;
4212 /* Read the entire original source text file into the original text buffer
4213 in one swell fwoop. Then figure out where the end of the text is and
4214 make sure that it ends with a newline followed by a null. */
4216 if (safe_read (input_file, new_orig_text_base, orig_size) !=
4217 (int) orig_size)
4219 int errno_val = errno;
4220 close (input_file);
4221 notice ("\n%s: error reading input file `%s': %s\n",
4222 pname, shortpath (NULL, convert_filename),
4223 xstrerror (errno_val));
4224 return;
4227 close (input_file);
4230 if (orig_size == 0 || orig_text_limit[-1] != '\n')
4232 *new_orig_text_limit++ = '\n';
4233 orig_text_limit++;
4236 /* Create the cleaned up copy of the original text. */
4238 memcpy (new_clean_text_base, orig_text_base,
4239 (size_t) (orig_text_limit - orig_text_base));
4240 do_cleaning (new_clean_text_base, new_clean_text_limit);
4242 #if 0
4244 int clean_file;
4245 size_t clean_size = orig_text_limit - orig_text_base;
4246 char *const clean_filename = (char *) alloca (strlen (convert_filename) + 6 + 1);
4248 /* Open (and create) the clean file. */
4250 strcpy (clean_filename, convert_filename);
4251 strcat (clean_filename, ".clean");
4252 if ((clean_file = creat (clean_filename, 0666)) == -1)
4254 int errno_val = errno;
4255 notice ("%s: can't create/open clean file `%s': %s\n",
4256 pname, shortpath (NULL, clean_filename),
4257 xstrerror (errno_val));
4258 return;
4261 /* Write the clean file. */
4263 safe_write (clean_file, new_clean_text_base, clean_size, clean_filename);
4265 close (clean_file);
4267 #endif /* 0 */
4269 /* Do a simplified scan of the input looking for things that were not
4270 mentioned in the aux info files because of the fact that they were
4271 in a region of the source which was preprocessed-out (via #if or
4272 via #ifdef). */
4274 scan_for_missed_items (file_p);
4276 /* Setup to do line-oriented forward seeking in the clean text buffer. */
4278 last_known_line_number = 1;
4279 last_known_line_start = clean_text_base;
4281 /* Now get down to business and make all of the necessary edits. */
4284 const def_dec_info *def_dec_p;
4286 first_definition_in_file = 1;
4287 def_dec_p = file_p->defs_decs;
4288 for (; def_dec_p; def_dec_p = def_dec_p->next_in_file)
4290 const char *clean_text_p = seek_to_line (def_dec_p->line);
4292 /* clean_text_p now points to the first character of the line which
4293 contains the `terminator' for the declaration or definition that
4294 we are about to process. */
4296 #ifndef UNPROTOIZE
4298 if (global_flag && def_dec_p->is_func_def && first_definition_in_file)
4300 add_global_decls (def_dec_p->file, clean_text_p);
4301 first_definition_in_file = 0;
4304 /* Don't edit this item if it is already in prototype format or if it
4305 is a function declaration and we have found no corresponding
4306 definition. */
4308 if (def_dec_p->prototyped
4309 || (!def_dec_p->is_func_def && !def_dec_p->definition))
4310 continue;
4312 #endif /* !defined (UNPROTOIZE) */
4314 if (def_dec_p->is_func_def)
4315 edit_fn_definition (def_dec_p, clean_text_p);
4316 else
4317 #ifndef UNPROTOIZE
4318 if (def_dec_p->is_implicit)
4319 add_local_decl (def_dec_p, clean_text_p);
4320 else
4321 #endif /* !defined (UNPROTOIZE) */
4322 edit_fn_declaration (def_dec_p, clean_text_p);
4326 /* Finalize things. Output the last trailing part of the original text. */
4328 output_up_to (clean_text_limit - 1);
4330 /* If this is just a test run, stop now and just deallocate the buffers. */
4332 if (nochange_flag)
4334 free (new_orig_text_base);
4335 free (new_clean_text_base);
4336 free (repl_text_base);
4337 return;
4340 /* Change the name of the original input file. This is just a quick way of
4341 saving the original file. */
4343 if (!nosave_flag)
4345 char *new_filename
4346 = (char *) xmalloc (strlen (convert_filename) + strlen (save_suffix) + 2);
4348 strcpy (new_filename, convert_filename);
4349 strcat (new_filename, save_suffix);
4350 if (my_link (convert_filename, new_filename) == -1)
4352 int errno_val = errno;
4353 if (errno_val == EEXIST)
4355 if (!quiet_flag)
4356 notice ("%s: warning: file `%s' already saved in `%s'\n",
4357 pname,
4358 shortpath (NULL, convert_filename),
4359 shortpath (NULL, new_filename));
4361 else
4363 notice ("%s: can't link file `%s' to `%s': %s\n",
4364 pname,
4365 shortpath (NULL, convert_filename),
4366 shortpath (NULL, new_filename),
4367 xstrerror (errno_val));
4368 return;
4373 if (my_unlink (convert_filename) == -1)
4375 int errno_val = errno;
4376 notice ("%s: can't delete file `%s': %s\n",
4377 pname, shortpath (NULL, convert_filename),
4378 xstrerror (errno_val));
4379 return;
4383 int output_file;
4385 /* Open (and create) the output file. */
4387 if ((output_file = creat (convert_filename, 0666)) == -1)
4389 int errno_val = errno;
4390 notice ("%s: can't create/open output file `%s': %s\n",
4391 pname, shortpath (NULL, convert_filename),
4392 xstrerror (errno_val));
4393 return;
4396 /* Write the output file. */
4399 unsigned int out_size = (repl_write_ptr + 1) - repl_text_base;
4401 safe_write (output_file, repl_text_base, out_size, convert_filename);
4404 close (output_file);
4407 /* Deallocate the conversion buffers. */
4409 free (new_orig_text_base);
4410 free (new_clean_text_base);
4411 free (repl_text_base);
4413 /* Change the mode of the output file to match the original file. */
4415 /* The cast avoids an erroneous warning on AIX. */
4416 if (my_chmod ((char *)convert_filename, stat_buf.st_mode) == -1)
4418 int errno_val = errno;
4419 notice ("%s: can't change mode of file `%s': %s\n",
4420 pname, shortpath (NULL, convert_filename),
4421 xstrerror (errno_val));
4424 /* Note: We would try to change the owner and group of the output file
4425 to match those of the input file here, except that may not be a good
4426 thing to do because it might be misleading. Also, it might not even
4427 be possible to do that (on BSD systems with quotas for instance). */
4430 /* Do all of the individual steps needed to do the protoization (or
4431 unprotoization) of the files referenced in the aux_info files given
4432 in the command line. */
4434 static void
4435 do_processing ()
4437 const char * const *base_pp;
4438 const char * const * const end_pps
4439 = &base_source_filenames[n_base_source_files];
4441 #ifndef UNPROTOIZE
4442 int syscalls_len;
4443 #endif /* !defined (UNPROTOIZE) */
4445 /* One-by-one, check (and create if necessary), open, and read all of the
4446 stuff in each aux_info file. After reading each aux_info file, the
4447 aux_info_file just read will be automatically deleted unless the
4448 keep_flag is set. */
4450 for (base_pp = base_source_filenames; base_pp < end_pps; base_pp++)
4451 process_aux_info_file (*base_pp, keep_flag, 0);
4453 #ifndef UNPROTOIZE
4455 /* Also open and read the special SYSCALLS.c aux_info file which gives us
4456 the prototypes for all of the standard system-supplied functions. */
4458 if (nondefault_syscalls_dir)
4460 syscalls_absolute_filename
4461 = (char *) xmalloc (strlen (nondefault_syscalls_dir)
4462 + sizeof (syscalls_filename) + 1);
4463 strcpy (syscalls_absolute_filename, nondefault_syscalls_dir);
4465 else
4467 syscalls_absolute_filename
4468 = (char *) xmalloc (strlen (default_syscalls_dir)
4469 + sizeof (syscalls_filename) + 1);
4470 strcpy (syscalls_absolute_filename, default_syscalls_dir);
4473 syscalls_len = strlen (syscalls_absolute_filename);
4474 if (*(syscalls_absolute_filename + syscalls_len - 1) != '/')
4476 *(syscalls_absolute_filename + syscalls_len++) = '/';
4477 *(syscalls_absolute_filename + syscalls_len) = '\0';
4479 strcat (syscalls_absolute_filename, syscalls_filename);
4481 /* Call process_aux_info_file in such a way that it does not try to
4482 delete the SYSCALLS aux_info file. */
4484 process_aux_info_file (syscalls_absolute_filename, 1, 1);
4486 #endif /* !defined (UNPROTOIZE) */
4488 /* When we first read in all of the information from the aux_info files
4489 we saved in it descending line number order, because that was likely to
4490 be faster. Now however, we want the chains of def & dec records to
4491 appear in ascending line number order as we get further away from the
4492 file_info record that they hang from. The following line causes all of
4493 these lists to be rearranged into ascending line number order. */
4495 visit_each_hash_node (filename_primary, reverse_def_dec_list);
4497 #ifndef UNPROTOIZE
4499 /* Now do the "real" work. The following line causes each declaration record
4500 to be "visited". For each of these nodes, an attempt is made to match
4501 up the function declaration with a corresponding function definition,
4502 which should have a full prototype-format formals list with it. Once
4503 these match-ups are made, the conversion of the function declarations
4504 to prototype format can be made. */
4506 visit_each_hash_node (function_name_primary, connect_defs_and_decs);
4508 #endif /* !defined (UNPROTOIZE) */
4510 /* Now convert each file that can be converted (and needs to be). */
4512 visit_each_hash_node (filename_primary, edit_file);
4514 #ifndef UNPROTOIZE
4516 /* If we are working in cplusplus mode, try to rename all .c files to .C
4517 files. Don't panic if some of the renames don't work. */
4519 if (cplusplus_flag && !nochange_flag)
4520 visit_each_hash_node (filename_primary, rename_c_file);
4522 #endif /* !defined (UNPROTOIZE) */
4525 static struct option longopts[] =
4527 {"version", 0, 0, 'V'},
4528 {"file_name", 0, 0, 'p'},
4529 {"quiet", 0, 0, 'q'},
4530 {"silent", 0, 0, 'q'},
4531 {"force", 0, 0, 'f'},
4532 {"keep", 0, 0, 'k'},
4533 {"nosave", 0, 0, 'N'},
4534 {"nochange", 0, 0, 'n'},
4535 {"compiler-options", 1, 0, 'c'},
4536 {"exclude", 1, 0, 'x'},
4537 {"directory", 1, 0, 'd'},
4538 #ifdef UNPROTOIZE
4539 {"indent", 1, 0, 'i'},
4540 #else
4541 {"local", 0, 0, 'l'},
4542 {"global", 0, 0, 'g'},
4543 {"c++", 0, 0, 'C'},
4544 {"syscalls-dir", 1, 0, 'B'},
4545 #endif
4546 {0, 0, 0, 0}
4550 main (argc, argv)
4551 int argc;
4552 char **const argv;
4554 int longind;
4555 int c;
4556 const char *params = "";
4558 pname = strrchr (argv[0], '/');
4559 pname = pname ? pname+1 : argv[0];
4561 #ifdef HAVE_LC_MESSAGES
4562 setlocale (LC_MESSAGES, "");
4563 #endif
4564 (void) bindtextdomain (PACKAGE, localedir);
4565 (void) textdomain (PACKAGE);
4567 cwd_buffer = getpwd ();
4568 if (!cwd_buffer)
4570 notice ("%s: cannot get working directory: %s\n",
4571 pname, xstrerror(errno));
4572 exit (FATAL_EXIT_CODE);
4575 /* By default, convert the files in the current directory. */
4576 directory_list = string_list_cons (cwd_buffer, NULL);
4578 while ((c = getopt_long (argc, argv,
4579 #ifdef UNPROTOIZE
4580 "c:d:i:knNp:qvVx:",
4581 #else
4582 "B:c:Cd:gklnNp:qvVx:",
4583 #endif
4584 longopts, &longind)) != EOF)
4586 if (c == 0) /* Long option. */
4587 c = longopts[longind].val;
4588 switch (c)
4590 case 'p':
4591 compiler_file_name = optarg;
4592 break;
4593 case 'd':
4594 directory_list
4595 = string_list_cons (abspath (NULL, optarg), directory_list);
4596 break;
4597 case 'x':
4598 exclude_list = string_list_cons (optarg, exclude_list);
4599 break;
4601 case 'v':
4602 case 'V':
4603 version_flag = 1;
4604 break;
4605 case 'q':
4606 quiet_flag = 1;
4607 break;
4608 #if 0
4609 case 'f':
4610 force_flag = 1;
4611 break;
4612 #endif
4613 case 'n':
4614 nochange_flag = 1;
4615 keep_flag = 1;
4616 break;
4617 case 'N':
4618 nosave_flag = 1;
4619 break;
4620 case 'k':
4621 keep_flag = 1;
4622 break;
4623 case 'c':
4624 params = optarg;
4625 break;
4626 #ifdef UNPROTOIZE
4627 case 'i':
4628 indent_string = optarg;
4629 break;
4630 #else /* !defined (UNPROTOIZE) */
4631 case 'l':
4632 local_flag = 1;
4633 break;
4634 case 'g':
4635 global_flag = 1;
4636 break;
4637 case 'C':
4638 cplusplus_flag = 1;
4639 break;
4640 case 'B':
4641 nondefault_syscalls_dir = optarg;
4642 break;
4643 #endif /* !defined (UNPROTOIZE) */
4644 default:
4645 usage ();
4649 /* Set up compile_params based on -p and -c options. */
4650 munge_compile_params (params);
4652 n_base_source_files = argc - optind;
4654 /* Now actually make a list of the base source filenames. */
4656 base_source_filenames
4657 = (const char **) xmalloc ((n_base_source_files + 1) * sizeof (char *));
4658 n_base_source_files = 0;
4659 for (; optind < argc; optind++)
4661 const char *path = abspath (NULL, argv[optind]);
4662 int len = strlen (path);
4664 if (path[len-1] == 'c' && path[len-2] == '.')
4665 base_source_filenames[n_base_source_files++] = path;
4666 else
4668 notice ("%s: input file names must have .c suffixes: %s\n",
4669 pname, shortpath (NULL, path));
4670 errors++;
4674 #ifndef UNPROTOIZE
4675 /* We are only interested in the very first identifier token in the
4676 definition of `va_list', so if there is more junk after that first
4677 identifier token, delete it from the `varargs_style_indicator'. */
4679 const char *cp;
4681 for (cp = varargs_style_indicator;
4682 ISALNUM ((const unsigned char)*cp) || *cp == '_'; cp++)
4683 continue;
4684 if (*cp != 0)
4685 varargs_style_indicator = savestring (varargs_style_indicator,
4686 cp - varargs_style_indicator);
4688 #endif /* !defined (UNPROTOIZE) */
4690 if (errors)
4691 usage ();
4692 else
4694 if (version_flag)
4695 fprintf (stderr, "%s: %s\n", pname, version_string);
4696 do_processing ();
4699 exit (errors ? FATAL_EXIT_CODE : SUCCESS_EXIT_CODE);
4701 return 1;