home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Education Sampler 1992 [NeXTSTEP]
/
Education_1992_Sampler.iso
/
NeXT
/
GnuSource
/
cc-61.0.1
/
cc
/
protoize.c
< prev
next >
Wrap
C/C++ Source or Header
|
1991-06-03
|
138KB
|
4,252 lines
/* Protoize program - Written by Ron Guilmette at the Microelectronics
and Computer Technology Corporation (MCC). The author's current
E-mail address is rfg@ics.uci.edu.
Copyright (C) 1989 Free Software Foundation, Inc.
This file is part of GNU CC.
GNU CC is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.
GNU CC is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with GNU CC; see the file COPYING. If not, write to
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */
#if defined (__cplusplus)
extern "C" { /* Start of extern "C" section. */
#endif
#include "config.h"
#include <stdio.h>
#include <ctype.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/dir.h>
#include <sys/file.h>
#include <sys/param.h>
#include <sys/wait.h>
#include <setjmp.h>
/* Sometimes param.h defines these macros. */
#undef CHAR_BIT
#undef CHAR_MAX
#undef CHAR_MIN
#undef CLK_TCK
#undef INT_MAX
#undef INT_MIN
#undef LONG_MAX
#undef LONG_MIN
#undef SCHAR_MAX
#undef SCHAR_MIN
#undef SHRT_MAX
#undef SHRT_MIN
#undef UCHAR_MAX
#undef UINT_MAX
#undef ULONG_MAX
#undef USHRT_MAX
#include <limits.h>
#ifdef USG
#define vfork fork
#endif
/* Emulate getwd on non-BSD systems. */
#if defined(USG) || defined(VMS)
#define getwd(addr) getcwd (addr, MAXPATHLEN)
#endif
#if defined(USG) || defined(VMS)
#include <string.h>
#define rindex strrchr
#else
#include <strings.h>
#if !defined (BSD)
#define BSD 1
#endif
#endif
#if defined (__cplusplus)
} /* End of extern "C" section. */
#endif
/* Look for these where the `const' qualifier is intentionally cast aside. */
#define NONCONST
#if (!defined (__GNUC__) && !defined (__cplusplus)) || defined (DGUX)
#define inline
#endif
#if !defined(__STDC__) && !defined (__cplusplus)
#define const
#define volatile
#endif
/* The VAX C compiler doesn't believe in void pointers. */
#if defined(vax) && !defined(__STDC__) && !defined(__GNUC__) && !defined (__cplusplus)
typedef char * pointer_t;
typedef const char * const_pointer_t;
#else
typedef void * pointer_t;
typedef const void * const_pointer_t;
#endif
typedef void void_t;
/* String to identify this version. */
static const char * const version_string = "Version 1.07";
/* Suffix of aux_info files. */
static const char * const aux_info_suffix = ".X";
/* String to attach to pathnames for saved versions of original files. */
static const char * const save_suffix = ".save";
#ifndef UNPROTOIZE
/* File name of the file which contains descriptions of standard system
routines. Note that we never actually do anything with this file per se,
but we do read in its corresponding aux_info file. */
static const char * const syscalls_filename = "SYSCALLS.c";
/* Default place to find the above file. */
#ifdef STD_PROTO_DIR
static const char * const default_syscalls_dir = STD_PROTO_DIR;
#else
static const char * const default_syscalls_dir = "/usr/local/lib";
#endif
/* Variable to hold the complete absolutized pathname of the SYSCALLS.c.X
file. */
static char * syscalls_pathname;
#endif
/* Type of the structure that holds information about macro unexpansions. */
struct unexpansion_struct {
const char *expanded;
const char *contracted;
};
typedef struct unexpansion_struct unexpansion;
/* A table of conversions that may need to be made for some (stupid) older
operating systems where these types are preprocessor macros rather than
typedefs (as they really ought to be).
WARNING: The contracted forms must be as small (or smaller) as the
expanded forms, or else havoc will ensue. */
static const unexpansion unexpansions[] = {
{ "struct _iobuf", "FILE" },
{ 0, 0 }
};
/* Maximum length of lines in both the aux_info files and in the actual
source file. Lines had better not be bigger than this or we are in big
trouble. */
#define MAX_LINE_LEN 8192
#define INDENT_MAXCHARS MAX_LINE_LEN
/* Maximum number of options that can be passed to the C compiler. */
#define MAX_OPTIONS 2048
/* The number of "primary" slots in the hash tables for filenames and for
function names. This can be as big or as small as you like, except that
it must be a power of two. */
#define HASH_TABLE_SIZE (1 << 9)
/* Bit mask to use when computing hash values. */
static const int hash_mask = (HASH_TABLE_SIZE - 1);
/* Prefix of files that we should avoid trying to convert (unless the -f option
is in effect). */
static const char * const usr_include = "/usr/include/";
/* A list of other directory names that we should also avoid messing with
(unless the -f option is used). */
static const char * const gnu_include[] = {
"/gcc-include/",
"/g++-include/",
0
};
/* The name of the other style of variable-number-of-parameters functions
(i.e. the style that we want to leave unconverted because we don't yet
know how to convert them to this style. This string is used in warning
messages. */
/* Also define here the string that we can search for in the parameter lists
taken from the .X files which will unambiguously indicate that we have
found a varargs style function. */
#ifdef UNPROTOIZE
static const char * const other_var_style = "stdarg";
#else
static const char * const other_var_style = "varargs";
#if defined (m88k)
static const char * const varargs_style_indicator = "__va_1st_arg";
#elif defined (i860)
static const char * const varargs_style_indicator = ???;
#elif defined (mips)
static const char * const varargs_style_indicator = "_va_alist";
#elif defined (pyr)
static const char * const varargs_style_indicator = "__builtin_va_alist";
#elif defined (spur)
static const char * const varargs_style_indicator = "__va_regs";
#else
static const char * const varargs_style_indicator = "__builtin_va_alist";
#endif
#endif
/* System supplied externals. */
#if defined (__cplusplus)
extern "C" { /* Start of extern "C" section. */
#endif
extern int errno;
extern char * sys_errlist[];
#ifndef bcopy
extern int bcopy (const void *, void *, int);
#else
extern void *memcpy (void *, const void *, size_t);
#endif
#ifdef getwd
extern char *getcwd (char *, int);
#else
extern char *getwd (char *);
#endif
#ifndef vfork
extern int vfork (void);
#else
extern int fork (void);
#endif
#if defined (BSD) || defined (__MACH__)
#define WAIT_ARG_TYPE union wait
extern int wait (union wait *);
#else
#define WAIT_ARG_TYPE int
#endif
#if !defined (WIFEXITED) || defined (DGUX)
#undef WIFEXITED
#define WIFEXITED(status_word) ((*((int *)&status_word) & 0xff) == 0x00)
#endif
#if !defined (WEXITSTATUS) || defined (DGUX)
#undef WEXITSTATUS
#define WEXITSTATUS(status_word) ((*((int *)&status_word) & 0xff00) >> 8)
#endif
/* A prototype for open and stat would conflict with some
other declarations. */
/* extern int open (const char *, int, ...); */
/* extern int stat (const char *, struct stat *); */
extern int fprintf (FILE *, const char *, ...);
extern int printf (const char *, ...);
extern void_t exit (int);
extern pointer_t malloc (size_t);
extern pointer_t realloc (void *, size_t);
extern void_t free (void *);
extern int read (int, void *, size_t);
extern int write (int, const void *, size_t);
extern int close (int);
extern int link (const char *, const char *);
extern int unlink (const char *);
extern int fflush (FILE *);
extern int _flsbuf (unsigned int, FILE *);
extern int atoi (const char *);
extern int access (const char *, int);
extern int puts (const char *);
extern int fputs (const char *, FILE *);
extern int fputc (int, FILE *);
extern int execvp (const char *, const char **);
extern int setjmp (jmp_buf);
extern void_t longjmp (jmp_buf, int);
#if defined (__cplusplus)
} /* End of extern "C" section. */
#endif
/* The following two types are used to create hash tables. In this program,
there are two hash tables which are used to store and quickly lookup two
different classes of strings. The first type of strings stored in the
first hash table are absolute pathnames of files which protoize needs to
know about. The second type of strings (stored in the second hash table)
are function names. It is this second class of strings which really
inspired the use of the hash tables, because there may be a lot of them. */
typedef struct hash_table_entry_struct hash_table_entry;
/* Do some typedefs so that we don't have to write "struct" so often. */
typedef struct def_dec_info_struct def_dec_info;
typedef struct file_info_struct file_info;
typedef struct f_list_chain_item_struct f_list_chain_item;
/* In the struct below, note that the "_info" field has two different uses
depending on the type of hash table we are in (i.e. either the pathnames
hash table or the function names hash table). In the pathnames hash table
the info fields of the entries point to the file_info struct which is
associated with each pathname (1 per pathname). In the function names
hash table, the info field points to the head of a singly linked list of
def_dec_info entries which are all defs or decs of the function whose
name is pointed to by the "symbol" field. Keeping all of the defs/decs
for a given function name on a special list specifically for that function
name makes it quick and easy to find out all of the important information
about a given (named) function. */
struct hash_table_entry_struct {
hash_table_entry * hash_next; /* -> to secondary entries */
const char * symbol; /* -> to the hashed string */
union {
const def_dec_info * _ddip;
file_info * _fip;
} _info;
};
#define ddip _info._ddip
#define fip _info._fip
/* Define a type specifically for our two hash tables. */
typedef hash_table_entry hash_table[HASH_TABLE_SIZE];
/* The following struct holds all of the important information about any
single pathname (e.g. file) which we need to know about. */
struct file_info_struct {
const hash_table_entry * hash_entry; /* -> to associated hash entry */
const def_dec_info * defs_decs; /* -> to chain of defs/decs */
time_t mtime; /* Time of last modification. */
};
/* Due to the possibility that functions may return pointers to functions,
(which may themselves have their own parameter lists) and due to the
fact that returned pointers-to-functions may be of type "pointer-to-
function-returning-pointer-to-function" (ad nauseum) we have to keep
an entire chain of ANSI style formal parameter lists for each function.
Normally, for any given function, there will only be one formals list
on the chain, but you never know.
Note that the head of each chain of formals lists is pointed to by the
`f_list_chain' field of the corresponding def_dec_info record.
For any given chain, the item at the head of the chain is the *leftmost*
parameter list seen in the actual C language function declaration. If
there are other members of the chain, then these are linked in left-to-right
order from the head of the chain. */
struct f_list_chain_item_struct {
const f_list_chain_item * chain_next; /* -> to next item on chain */
const char * formals_list; /* -> to formals list string */
};
/* The following struct holds all of the important information about any
single function definition or declaration which we need to know about.
Note that for unprotoize we don't need to know very much because we
never even create records for stuff that we don't intend to convert
(like for instance defs and decs which are already in old K&R format
and "implicit" function declarations). */
struct def_dec_info_struct {
const def_dec_info * next_in_file; /* -> to rest of chain for file */
file_info * file; /* -> file_info for containing file */
int line; /* source line number of def/dec */
const char * ansi_decl; /* -> left end of ansi decl */
hash_table_entry * hash_entry; /* -> hash entry for function name */
unsigned int is_func_def; /* = 0 means this is a declaration */
const def_dec_info * next_for_func; /* -> to rest of chain for func name */
unsigned int f_list_count; /* count of formals lists we expect */
char prototyped; /* = 0 means already prototyped */
#ifndef UNPROTOIZE
const f_list_chain_item * f_list_chain; /* -> chain of formals lists */
const def_dec_info * definition; /* -> def/dec containing related def */
char is_static; /* = 0 means visiblilty is "extern" */
char is_implicit; /* != 0 for implicit func decl's */
char written; /* != 0 means written for implicit */
#else
const char * formal_names; /* -> to list of names of formals */
const char * formal_decls; /* -> to string of formal declartions */
#endif
};
/* Pointer to the tail component of the pathname by which this program was
invoked. Used everywhere in error and warning messages. */
static const char *pname;
/* Error counter. Will be non-zero if we should give up at the next convenient
stopping point. */
static int errors = 0;
/* Option flags. */
static int version_flag = 0; /* set by -V option */
static int quiet_flag = 0; /* set by -q option */
static int force_flag = 0; /* set by -f option */
static int nochange_flag = 0; /* set by -n option */
static int nosave_flag = 0; /* set by -N option */
static int keep_flag = 0; /* set by -k option */
static const char ** compile_params = 0; /* set by -c option */
#ifdef UNPROTOIZE
static const char *indent_string = " "; /* set by -i option */
#else
static int local_flag = 0; /* set by -l option */
static int global_flag = 0; /* set by -g option */
static int cplusplus_flag = 0; /* set by -C option */
static const char* nondefault_syscalls_dir = 0; /* set by -B option */
#endif
/* An index into the compile_params array where we should insert the filename
parameter when we are ready to exec the C compiler. A zero value indicates
that we have not yet called munge_compile_params(). */
static int filename_index = 0;
/* Count of command line arguments which were "filename" arguments. */
static int base_source_files = 0;
/* Points to a malloc'ed list of pointers to all of the filenames of base
source files which were specified on the command line. */
static const char **base_source_paths;
/* Line number of the line within the current aux_info file that we
are currently processing. Used for error messages in case the prototypes
info file is corrupted somehow. */
static int current_aux_info_lineno;
/* Pointer to the name of the source file currently being converted. */
static const char *convert_path;
/* Pointer to relative root string (taken from aux_info file) which indicates
where directory the user was in when he did the compilation step that
produced the containing aux_info file. */
static const char *invocation_path;
/* Pointer to the base of the input buffer that holds the original text for the
source file currently being converted. */
static const char *orig_text_base;
/* Pointer to the byte just beyond the end of the input buffer that holds the
original text for the source file currently being converted. */
static const char *orig_text_limit;
/* Pointer to the base of the input buffer that holds the cleaned text for the
source file currently being converted. */
static const char *clean_text_base;
/* Pointer to the byte just beyond the end of the input buffer that holds the
cleaned text for the source file currently being converted. */
static const char *clean_text_limit;
/* Pointer to the last byte in the cleaned text buffer that we have already
(virtually) copied to the output buffer (or decided to ignore). */
static const char * clean_read_ptr;
/* Pointer to the base of the output buffer that holds the replacement text
for the source file currently being converted. */
static char *repl_text_base;
/* Pointer to the byte just beyond the end of the output buffer that holds the
replacement text for the source file currently being converted. */
static char *repl_text_limit;
/* Pointer to the last byte which has been stored into the output buffer.
The next byte to be stored should be stored just past where this points
to. */
static char * repl_write_ptr;
/* Pointer into the cleaned text buffer for the source file we are currently
converting. This points to the first character of the line that we last
did a "seek_to_line()" to (see below). */
static const char *last_known_line_start;
/* Number of the line (in the cleaned text buffer) that we last did a
"seek_to_line()" to. Will be one if we just read a new source file
into the cleaned text buffer. */
static int last_known_line_number;
/* The pathnames hash table. */
static hash_table pathname_primary;
/* The function names hash table. */
static hash_table function_name_primary;
/* The place to keep the recovery address which is used only in cases where
we get hopelessly confused by something in the cleaned original text. */
static jmp_buf source_confusion_recovery;
/* An area to hold the current pathname (used by abspath). */
static char cwd_buffer[MAXPATHLEN];
/* A place to save the read pointer until we are sure that an individual
attempt at editing will succeed. */
static const char * saved_clean_read_ptr;
/* A place to save the write pointer until we are sure that an individual
attempt at editing will succeed. */
static char * saved_repl_write_ptr;
/* Forward declaration. */
static const char *shortpath (const char *cwd, const char *pathname);
/* Get setup to recover in case the edit we are about to do goes awry. */
void save_pointers (void)
{
saved_clean_read_ptr = clean_read_ptr;
saved_repl_write_ptr = repl_write_ptr;
}
/* Call this routine to recover our previous state whenever something looks
too confusing in the source code we are trying to edit. */
void restore_pointers (void)
{
clean_read_ptr = saved_clean_read_ptr;
repl_write_ptr = saved_repl_write_ptr;
}
/* Return true if the given character is a legal identifier character. */
inline static int
is_id_char (char ch)
{
return (isalnum (ch) || (ch == '_') || (ch == '$'));
}
/* Allocate some space, but check that the allocation was successful. */
static pointer_t
xmalloc (int byte_count)
{
pointer_t rv;
if ((rv = malloc (byte_count)) == NULL)
{
fprintf (stderr, "\n%s: fatal error: can't allocate %d more bytes of memory\n",
pname, byte_count);
exit (1);
return 0; /* avoid warnings */
}
else
return rv;
}
/* Reallocate some space, but check that the reallocation was successful. */
static pointer_t
xrealloc (pointer_t old_space, int byte_count)
{
pointer_t rv;
if ((rv = realloc (old_space, byte_count)) == NULL)
{
fprintf (stderr, "\n%s: fatal error: can't allocate %d more bytes of memory\n",
pname, byte_count);
exit (1);
return 0; /* avoid warnings */
}
else
return rv;
}
/* Deallocate the area pointed to by an arbitrary pointer, but first, strip
the `const' qualifier from it and also make sure that the pointer value
is non-null. */
static void_t
xfree (const_pointer_t p)
{
if (p)
free ((NONCONST pointer_t) p);
}
/* More 'friendly' abort that prints the line and file.
config.h can #define abort fancy_abort if you like that sort of thing. */
void
fancy_abort ()
{
fprintf (stderr, "%s: internal abort.\n", pname);
exit (1);
}
/* Make a duplicate of a given string in a newly allocated area. */
static char *
dupstr (const char *s)
{
return strcpy ((char *) xmalloc (strlen (s) + 1), s);
}
/* Make a duplicate of the first N bytes of a given string in a newly
allocated area. */
static char *
dupnstr (const char *s, int n)
{
char *ret_val = strncpy ((char *) xmalloc (n + 1), s, n);
ret_val[n] = '\0';
return ret_val;
}
/* Return a pointer to the first occurance of s2 within s1 or NULL if s2
does not occur within s1. Assume neither s1 nor s2 are null pointers. */
static const char *
substr (const char *s1, const char *const s2)
{
for (; *s1 ; s1++)
{
const char *p1;
const char *p2;
char c;
for (p1 = s1, p2 = s2; c = *p2; p1++, p2++)
if (*p1 != c)
goto outer;
return s1;
outer:
;
}
return 0;
}
/* Give a message indicating the proper way to invoke this program and then
exit with non-zero status. */
static void_t
usage (void)
{
#ifdef UNPROTOIZE
fprintf (stderr, "%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ pathname ... ]'\n",
#else
fprintf (stderr, "%s: usage '%s [ -VqfnkNlgC ] [ -B <diname> ] [ pathname ... ]'\n",
#endif
pname, pname);
exit (1);
}
/* Return true if the given pathname (assumed to be an absolute pathname)
designates a file residing anywhere beneath any one of the "system"
include directories. */
static int
in_system_include_dir (const char *path)
{
const char * const *gnu_include_p;
if (path[0] != '/')
abort (); /* Must be an absolutized pathname. */
if (!strncmp (path, usr_include, strlen (usr_include)))
return 1;
for (gnu_include_p = gnu_include; *gnu_include_p; gnu_include_p++)
if (substr (path, *gnu_include_p))
return 1;
return 0;
}
/* Return true if the given pathname designates a file that the user has
read access to and for which the user has write access to the containing
directory. */
static int
file_could_be_converted (const char *path)
{
char dir_name[MAXPATHLEN];
if (access (path, R_OK))
return 0;
{
char *dir_last_slash;
strcpy (dir_name, path);
dir_last_slash = rindex (dir_name, '/');
if (dir_last_slash)
*dir_last_slash = '\0';
else
abort (); /* Should have been an absolutized pathname. */
}
if (access (path, W_OK))
return 0;
return 1;
}
/* Return true if the given pathname designates a file that has its "sticky"
bit set. */
static int
file_is_sticky (const char *path)
{
struct stat stat_buf;
if (stat (path, &stat_buf) == -1)
{
fprintf (stderr, "%s: warning: can't get mode of file `%s': %s\n",
pname,
shortpath (NULL, path),
sys_errlist[errno]);
return 0;
}
return (stat_buf.st_mode & S_ISVTX);
}
/* Return true if the given pathname designates a file that we are allowed
to modify. Files which we should not attempt to modify are (a) "system"
include files, and (b) files which the user doesn't have write access to
or for which the sticky bit is set, and (c) files which reside in
directories which the user doesn't have write access to or for which the
sticky bit is set. Unless requested to be quiet, give warnings about
files that we will not try to convert for one reason or another. An
exception is made for "system" include files, which we never try to
convert and for which we don't issue the usual warnings. */
static int
file_normally_convertable (const char *path)
{
char dir_name[MAXPATHLEN];
if (in_system_include_dir (path))
return 0;
if (file_is_sticky (path))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: file `%s' is sticky\n",
pname, shortpath (NULL, path));
return 0;
}
{
char *dir_last_slash;
strcpy (dir_name, path);
dir_last_slash = rindex (dir_name, '/');
if (dir_last_slash)
*dir_last_slash = '\0';
else
abort (); /* Should have been an absolutized pathname. */
}
if (file_is_sticky (dir_name))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: directory containing file `%s' is sticky\n",
pname, shortpath (NULL, path));
return 0;
}
if (access (path, R_OK))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: no read access for file `%s'\n",
pname, shortpath (NULL, path));
return 0;
}
if (access (path, W_OK))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: no write access for file `%s'\n",
pname, shortpath (NULL, path));
return 0;
}
if (access (dir_name, W_OK))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: no write access for dir containing `%s'\n",
pname, shortpath (NULL, path));
return 0;
}
return 1;
}
#ifndef UNPROTOIZE
/* Return true if the given file_info struct refers to the special SYSCALLS.c.X
file. Return false otherwise. */
static int
is_syscalls_file (const file_info *fi_p)
{
return (substr (fi_p->hash_entry->symbol, syscalls_filename) != NULL);
}
#endif
/* Check to see if this file will need to have anything done to it on this
run. If there is nothing in the given file which both needs conversion
and for which we have the necessary stuff to do the conversion, return
false. Otherwise, return true.
Note that (for protoize) it is only valid to call this function *after*
the connections between declarations and definitions have all been made
by connect_defs_and_decs(). */
static int
needs_to_be_converted (const file_info *file_p)
{
const def_dec_info *ddp;
#ifndef UNPROTOIZE
if (is_syscalls_file (file_p))
return 0;
#endif
for (ddp = file_p->defs_decs; ddp; ddp = ddp->next_in_file)
if (
#ifndef UNPROTOIZE
/* ... and if we a protoizing and this function is in old style ... */
!ddp->prototyped
/* ... and if this a definition or is a decl with an associated def ... */
&& (ddp->is_func_def || (!ddp->is_func_def && ddp->definition))
#else
/* ... and if we are unprotoizing and this function is in new style ... */
ddp->prototyped
#endif
)
/* ... then the containing file needs converting. */
return -1;
return 0;
}
/* Given a hash table, apply some function to each node in the table. The
table to traverse is given as the "hash_tab_p" argument, and the
function to be applied to each node in the table is given as "func"
argument. */
static void_t
visit_each_hash_node (const hash_table_entry *hash_tab_p, void_t (*func) (const hash_table_entry *))
{
const hash_table_entry *primary;
for (primary = hash_tab_p; primary < &hash_tab_p[HASH_TABLE_SIZE]; primary++)
if (primary->symbol)
{
hash_table_entry *second;
(*func)(primary);
for (second = primary->hash_next; second; second = second->hash_next)
(*func) (second);
}
}
/* Initialize all of the fields of a new hash table entry, pointed
to by the "p" parameter. Note that the space to hold the entry
is assumed to have already been allocated before this routine is
called. */
static hash_table_entry *
add_symbol (hash_table_entry *p, const char *s)
{
p->hash_next = NULL;
p->symbol = dupstr (s);
p->ddip = NULL;
p->fip = NULL;
return p;
}
/* Look for a particular function name or pathname in the particular
hash table indicated by "hash_tab_p". If the name is not in the
given hash table, add it. Either way, return a pointer to the
hash table entry for the given name. */
static hash_table_entry *
lookup (hash_table_entry *hash_tab_p, const char *search_symbol)
{
int hash_value = 0;
const char *search_symbol_char_p = search_symbol;
hash_table_entry *p;
while (*search_symbol_char_p)
hash_value += *search_symbol_char_p++;
hash_value &= hash_mask;
p = &hash_tab_p[hash_value];
if (! p->symbol)
return add_symbol (p, search_symbol);
if (!strcmp (p->symbol, search_symbol))
return p;
while (p->hash_next)
{
p = p->hash_next;
if (!strcmp (p->symbol, search_symbol))
return p;
}
p->hash_next = (hash_table_entry *) xmalloc (sizeof (hash_table_entry));
p = p->hash_next;
return add_symbol (p, search_symbol);
}
/* Throw a def/dec record on the junk heap.
Also, since we are not using this record anymore, free up all of the
stuff it pointed to. */
inline static void_t
free_def_dec (def_dec_info *p)
{
xfree (p->ansi_decl);
#ifndef UNPROTOIZE
{
const f_list_chain_item * curr;
const f_list_chain_item * next;
for (curr = p->f_list_chain; curr; curr = next)
{
next = curr->chain_next;
xfree (curr);
}
}
#endif
xfree (p);
}
/* Unexpand as many macro symbol as we can find.
If the given line must be unexpanded, make a copy of it in the heap and
return a pointer to the unexpanded copy. Otherwise return NULL. */
static char *
unexpand_if_needed (const char *aux_info_line)
{
char line_buf[MAX_LINE_LEN];
const unexpansion* unexp_p;
int got_unexpanded = 0;
const char *s;
char *copy_p = line_buf;
/* Make a copy of the input string in line_buf, expanding as necessary. */
for (s = aux_info_line; *s != '\n'; )
{
for (unexp_p = unexpansions; unexp_p->expanded; unexp_p++)
{
const char *in_p = unexp_p->expanded;
int len = strlen (in_p);
if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
{
got_unexpanded = 1;
strcpy (copy_p, unexp_p->contracted);
copy_p += strlen (unexp_p->contracted);
/* Assume the there will not be another replacement required
within the text just replaced. */
s += len;
goto continue_outer;
}
}
*copy_p++ = *s++;
continue_outer: ;
}
*copy_p++ = '\n';
*copy_p++ = '\0';
return (got_unexpanded) ? dupstr (line_buf) : 0;
}
/* Return the absolutized pathname for the given relative
pathname. Note that if that pathname is already absolute, it may
still be returned in a modified form because this routine also
eliminates redundant slashes and single dots and eliminates double
dots to get a shortest possible pathname from the given input
pathname. The absolutization of relative pathnames is made by
assuming that the given pathname is to be taken as relative to
the first argument (cwd) or to the current directory if cwd is
NULL. */
static char *
abspath (const char *cwd, const char *rel_pathname)
{
char abs_buffer[MAXPATHLEN];
char *endp;
char *inp = abs_buffer;
char *outp = abs_buffer;
/* Setup the current working directory as needed. */
if (!cwd)
cwd = cwd_buffer;
/* Copy the pathname (possibly preceeded by the current working
directory name) into the absolutization buffer. */
endp = abs_buffer;
{
const char *src_p;
if (rel_pathname[0] != '/')
{
src_p = cwd;
while (*endp++ = *src_p++)
continue;
*(endp-1) = '/'; /* overwrite null */
}
src_p = rel_pathname;
while (*endp++ = *src_p++)
continue;
if (endp[-1] == '/')
*endp = '\0';
}
/* Now make a copy of abs_buffer into abs_buffer, shortening the
pathname (by taking out slashes and dots) as we go. */
*outp++ = *inp++; /* copy first slash */
for (;;)
{
if (!inp[0])
break;
else if (inp[0] == '/' && outp[-1] == '/')
{
inp++;
continue;
}
else if (inp[0] == '.' && outp[-1] == '/')
{
if (!inp[1])
break;
else if (inp[1] == '/')
{
inp += 2;
continue;
}
else if ((inp[1] == '.') && (inp[2] == 0 || inp[2] == '/'))
{
inp += (inp[2] == '/') ? 3 : 2;
outp -= 2;
while (outp >= abs_buffer && *outp != '/')
outp--;
if (outp < abs_buffer)
{
/* Catch cases like /.. where we try to backup to a
point above the absolute root of the logical file
system. */
fprintf (stderr, "%s: fatal error: invalid pathname: %s\n",
pname, rel_pathname);
exit (1);
}
*++outp = '\0';
continue;
}
}
*outp++ = *inp++;
}
/* On exit, make sure that there is a trailing null, and make sure that
the last character of the returned string is *not* a slash. */
*outp = '\0';
if (outp[-1] == '/')
*--outp = '\0';
/* Make a copy (in the heap) of the stuff left in the absolutization
buffer and return a pointer to the copy. */
return dupstr (abs_buffer);
}
/* Given a pathname (and possibly a directory name from which the pathname
is relative) return a string which is the shortest possible
equivalent for the corresponding full (absolutized) pathname. The
shortest possible equivalent may be constructed by converting the
absolutized pathname to be a relative pathname (i.e. relative to
the actual current working directory). However if a relative pathname
is longer, then the full absolute pathname is returned.
KNOWN BUG:
Note that "simple-minded" conversion of any given type of pathname (either
relative or absolute) may not result in a valid equivalent pathname if any
subpart of the original pathname is actually a symbolic link. */
static const char *
shortpath (const char *cwd, const char *pathname)
{
static char cwd_buffer[MAXPATHLEN];
char rel_buffer[MAXPATHLEN];
char *rel_buf_p = rel_buffer;
char *cwd_p = cwd_buffer;
char *path_p;
int unmatched_slash_count = 0;
path_p = abspath (cwd, pathname);
if (!cwd_buffer[0])
getwd (cwd_buffer);
while (*cwd_p && (*cwd_p == *path_p))
{
cwd_p++;
path_p++;
}
if (!*cwd_p) /* whole pwd matched */
{
if (!*path_p) /* input *is* the current path! */
return ".";
else
return ++path_p;
}
else
{
if (*path_p)
{
--cwd_p;
--path_p;
while (*cwd_p != '/') /* backup to last slash */
{
--cwd_p;
--path_p;
}
cwd_p++;
path_p++;
unmatched_slash_count++;
}
while (*cwd_p)
if (*cwd_p++ == '/')
unmatched_slash_count++;
while (unmatched_slash_count--)
{
*rel_buf_p++ = '.';
*rel_buf_p++ = '.';
*rel_buf_p++ = '/';
}
while (*rel_buf_p++ = *path_p++)
continue;
--rel_buf_p;
if (*(rel_buf_p-1) == '/')
*--rel_buf_p = '\0';
if (strlen (rel_buffer) > strlen (pathname))
return pathname;
else
return dupstr (rel_buffer);
}
return 0; /* Prevent warnings for old versions of GCC. */
}
/* Lookup the given pathname in the hash table for pathnames. If it is a
new one, then the hash table info pointer will be null. In this case,
we create a new file_info record to go with the pathname, and we initialize
that record with some reasonable values. */
static file_info *
find_file (const char *pathname, int do_not_stat)
{
hash_table_entry *hash_entry_p;
hash_entry_p = lookup (pathname_primary, pathname);
if (hash_entry_p->fip)
return hash_entry_p->fip;
else
{
struct stat stat_buf;
file_info *file_p = (file_info *) xmalloc (sizeof (file_info));
/* If we cannot get status on any given source file, give a warning
and then just set its time of last modification to infinity. */
if (do_not_stat)
stat_buf.st_mtime = (time_t) 0;
else
{
if (stat (pathname, &stat_buf) == -1)
{
fprintf (stderr, "%s: error: can't get status of `%s': %s\n",
pname, shortpath (NULL, pathname), sys_errlist[errno]);
stat_buf.st_mtime = (time_t) -1;
}
}
hash_entry_p->fip = file_p;
file_p->hash_entry = hash_entry_p;
file_p->defs_decs = NULL;
file_p->mtime = stat_buf.st_mtime;
return file_p;
}
}
/* Generate a fatal error because some part of the aux_info file is
messed up. */
static void_t
aux_info_corrupted (void)
{
fprintf (stderr, "\n%s: fatal error: aux info file corrupted at line %d\n",
pname, current_aux_info_lineno);
exit (1);
}
/* Check to see that a condition is true. This is kind of like an assert(). */
inline static void_t
check_aux_info (int cond)
{
if (! cond)
aux_info_corrupted ();
}
/* Given a pointer to the closing right parenthesis for a particular formals
list (in a aux_info file) find the corresponding left parenthesis and
return a pointer to it. */
static const char *
find_corresponding_lparen (const char *p)
{
const char *q;
int paren_depth;
for (paren_depth = 1, q = p-1; paren_depth; q--)
{
switch (*q)
{
case ')':
paren_depth++;
break;
case '(':
paren_depth--;
break;
}
}
return ++q;
}
/* Given a line from an aux info file, and a time at which the aux info
file it came from was created, check to see if the item described in
the line comes from a file which has been modified since the aux info
file was created. If so, return non-zero, else return zero. */
static int
referenced_file_is_newer (const char *l, time_t aux_info_mtime)
{
const char *p;
file_info *fi_p;
char filename[MAXPATHLEN];
check_aux_info (l[0] == '/');
check_aux_info (l[1] == '*');
check_aux_info (l[2] == ' ');
{
const char *filename_start = p = l + 3;
while (*p != ':')
p++;
strncpy (filename, filename_start, p - filename_start);
filename[p-filename_start] = '\0';
}
/* Call find_file to find the file_info record associated with the file
which contained this particular def or dec item. Note that this call
may cause a new file_info record to be created if this is the first time
that we have ever known about this particular file. */
fi_p = find_file (abspath (invocation_path, filename), 0);
return (fi_p->mtime > aux_info_mtime);
}
/* Given a line of info from the aux_info file, create a new
def_dec_info record to remember all of the important information about
a function definition or declaration.
Link this record onto the list of such records for the particular file in
which it occured in proper (descending) line number order (for now).
If there is an identical record already on the list for the file, throw
this one away. Doing so takes care of the (useless and troublesome)
duplicates which are bound to crop up due to multiple inclusions of any
given individual header file.
Finally, link the new def_dec record onto the list of such records
pertaining to this particular function name. */
static void_t
save_def_or_dec (const char *l, int is_syscalls)
{
const char *p;
const char *semicolon_p;
def_dec_info *def_dec_p = (def_dec_info *) xmalloc (sizeof (def_dec_info));
#ifndef UNPROTOIZE
def_dec_p->written = 0;
#endif
/* Start processing the line by picking off 5 pieces of information from
the left hand end of the line. These are filename, line number,
new/old/implicit flag (new = ANSI prototype format), definition or
declaration flag, and extern/static flag). */
check_aux_info (l[0] == '/');
check_aux_info (l[1] == '*');
check_aux_info (l[2] == ' ');
{
const char *filename_start = p = l + 3;
char filename[MAXPATHLEN];
while (*p != ':')
p++;
strncpy (filename, filename_start, p - filename_start);
filename[p-filename_start] = '\0';
/* Call find_file to find the file_info record associated with the file
which contained this particular def or dec item. Note that this call
may cause a new file_info record to be created if this is the first time
that we have ever known about this particular file.
Note that we started out by forcing all of the base source file pathnames
(i.e. the names of the aux_info files with the .X stripped off) into the
pathnames hash table, and we simultaneously setup file_info records for
all of these base file pathnames (even if they may be useless later).
The file_info records for all of these "base" file pathnames (properly)
act as file_info records for the "original" (i.e. un-included) files
which were submitted to gcc for compilation (when the -fgen-aux-info
option was used). */
def_dec_p->file = find_file (abspath (invocation_path, filename), is_syscalls);
}
{
const char *line_number_start = ++p;
char line_number[10];
while (*p != ':')
p++;
strncpy (line_number, line_number_start, p - line_number_start);
line_number[p-line_number_start] = '\0';
def_dec_p->line = atoi (line_number);
}
/* Check that this record describes a new-style, old-style, or implicit
definition or declaration. */
p++; /* Skip over the `:'. */
check_aux_info ((*p == 'N') || (*p == 'O') || (*p == 'I'));
/* Is this a new style (ANSI prototyped) definition or declaration? */
def_dec_p->prototyped = (*p == 'N');
#ifndef UNPROTOIZE
/* Is this an implicit declaration? */
def_dec_p->is_implicit = (*p == 'I');
#endif
p++;
check_aux_info ((*p == 'C') || (*p == 'F'));
/* Is this item a function definition (F) or a declaration (C). Note that
we treat item taken from the syscalls file as though they were function
definitions regardless of what the stuff in the file says. */
def_dec_p->is_func_def = ((*p++ == 'F') || is_syscalls);
#ifndef UNPROTOIZE
def_dec_p->definition = 0; /* Fill this in later if protoizing. */
#endif
check_aux_info (*p++ == ' ');
check_aux_info (*p++ == '*');
check_aux_info (*p++ == '/');
check_aux_info (*p++ == ' ');
#ifdef UNPROTOIZE
check_aux_info ((!strncmp (p, "static", 6)) || (!strncmp (p, "extern", 6)));
#else
if (!strncmp (p, "static", 6))
def_dec_p->is_static = -1;
else if (!strncmp (p, "extern", 6))
def_dec_p->is_static = 0;
else
check_aux_info (0); /* Didn't find either `extern' or `static'. */
#endif
{
const char *ansi_start = p;
p += 6; /* Pass over the "static" or "extern". */
/* We are now past the initial stuff. Search forward from here to find
the terminating semicolon that should immediately follow the entire
ANSI format function declaration. */
while (*++p != ';')
continue;
semicolon_p = p;
/* Make a copy of the ansi declaration part of the line from the aux_info
file. */
def_dec_p->ansi_decl = dupnstr (ansi_start, (semicolon_p+1) - ansi_start);
}
/* Backup and point at the final right paren of the final argument list. */
p--;
/* Now isolate a whole set of formal argument lists, one-by-one. Normally,
there will only be one list to isolate, but there could be more. */
def_dec_p->f_list_count = 0;
#ifndef UNPROTOIZE
def_dec_p->f_list_chain = NULL;
#endif
for (;;)
{
const char *left_paren_p = find_corresponding_lparen (p);
#ifndef UNPROTOIZE
{
f_list_chain_item *cip =
(f_list_chain_item *) xmalloc (sizeof (f_list_chain_item));
cip->formals_list = dupnstr (left_paren_p+1, p - (left_paren_p+1));
/* Add the new chain item at the head of the current list. */
cip->chain_next = def_dec_p->f_list_chain;
def_dec_p->f_list_chain = cip;
}
#endif
def_dec_p->f_list_count++;
p = left_paren_p - 2;
/* p must now point either to another right paren, or to the last
character of the name of the function that was declared/defined.
If p points to another right paren, then this indicates that we
are dealing with multiple formals lists. In that case, there
really should be another right paren preceeding this right paren. */
if (*p != ')')
break;
else
check_aux_info (*--p == ')');
}
{
const char *past_fn = p + 1;
char fn_string[1024];
check_aux_info (*past_fn == ' ');
/* Scan leftwards over the identifier that names the function. */
while (is_id_char (*p))
p--;
p++;
/* p now points to the leftmost character of the function name. */
strncpy (fn_string, p, past_fn - p);
fn_string[past_fn-p] = '\0';
def_dec_p->hash_entry = lookup (function_name_primary, fn_string);
}
/* Look at all of the defs and decs for this function name that we have
collected so far. If there is already one which is at the same
line number in the same file, then we can discard this new def_dec_info
record.
As an extra assurance that any such pair of (nominally) identical
function declarations are in fact identical, we also compare the
ansi_decl parts of the lines from the aux_info files just to be on
the safe side.
This comparison will fail if (for instance) the user was playing
messy games with the preprocessor which ultimately causes one
function declaration in one header file to look differently when
that file is included by two (or more) other files. */
{
const def_dec_info *other;
for (other = def_dec_p->hash_entry->ddip; other; other = other->next_for_func)
{
if (def_dec_p->line == other->line && def_dec_p->file == other->file)
{
if (strcmp (def_dec_p->ansi_decl, other->ansi_decl))
{
fprintf (stderr, "%s: error: declaration of function `%s' at %s(%d) takes different forms\n",
pname,
def_dec_p->hash_entry->symbol,
def_dec_p->file->hash_entry->symbol,
def_dec_p->line);
exit (1);
}
free_def_dec (def_dec_p);
return;
}
}
}
#ifdef UNPROTOIZE
/* If we are doing unprotoizing, we must now setup the pointers that will
point to the K&R name list and to the K&R argument declarations list.
Note that if this is only a function declaration, then we should not
expect to find any K&R style formals list following the ANSI-style
formals list. This is because GCC knows that such information is
useless in the case of function declarations (function definitions
are a different story however).
Since we are unprotoizing, we don't need any such lists anyway.
All we plan to do is to delete all characters between ()'s in any
case. */
def_dec_p->formal_names = NULL;
def_dec_p->formal_decls = NULL;
if (def_dec_p->is_func_def)
{
p = semicolon_p;
check_aux_info (*++p == ' ');
check_aux_info (*++p == '/');
check_aux_info (*++p == '*');
check_aux_info (*++p == ' ');
check_aux_info (*++p == '(');
{
const char *kr_names_start = ++p; /* Point just inside '('. */
while (*p++ != ')')
continue;
p--; /* point to closing right paren */
/* Make a copy of the K&R parameter names list. */
def_dec_p->formal_names = dupnstr (kr_names_start, p - kr_names_start);
}
check_aux_info (*++p == ' ');
p++;
/* p now points to the first character of the K&R style declarations
list (if there is one) or to the star-slash combination that ends
the comment in which such lists get embedded. */
/* Make a copy of the K&R formal decls list and set the def_dec record
to point to it. */
if (*p == '*') /* Are there no K&R declarations? */
{
check_aux_info (*++p == '/');
def_dec_p->formal_decls = "";
}
else
{
const char *kr_decls_start = p;
while (p[0] != '*' || p[1] != '/')
p++;
p--;
check_aux_info (*p == ' ');
def_dec_p->formal_decls = dupnstr (kr_decls_start, p-kr_decls_start);
}
/* Handle a special case. If we have a function definition marked as
being in "old" style, and if it's formal names list is empty, then
it may actually have the string "void" in its real formals list
in the original source code. Just to make sure, we will get setup
to convert such things anyway.
This kludge only needs to be here because of an insurmountable
problem with generating .X files. */
if (!def_dec_p->prototyped && !*def_dec_p->formal_names)
def_dec_p->prototyped = 1;
}
/* Since we are unprotoizing, if this item is already in old (K&R) style,
we can just ignore it. If that is true, throw away the itme now. */
if (!def_dec_p->prototyped)
{
free_def_dec (def_dec_p);
return;
}
#endif
/* Add this record to the head of the list of records pertaining to this
particular function name. */
def_dec_p->next_for_func = def_dec_p->hash_entry->ddip;
def_dec_p->hash_entry->ddip = def_dec_p;
/* Add this new def_dec_info record to the sorted list of def_dec_info
records for this file. Note that we don't have to worry about duplicates
(caused by multiple inclusions of header files) here because we have
already eliminated duplicates above. */
if (!def_dec_p->file->defs_decs)
{
def_dec_p->file->defs_decs = def_dec_p;
def_dec_p->next_in_file = NULL;
}
else
{
int line = def_dec_p->line;
const def_dec_info *prev = NULL;
const def_dec_info *curr = def_dec_p->file->defs_decs;
const def_dec_info *next = curr->next_in_file;
while (next && (line < curr->line))
{
prev = curr;
curr = next;
next = next->next_in_file;
}
if (line >= curr->line)
{
def_dec_p->next_in_file = curr;
if (prev)
((NONCONST def_dec_info *) prev)->next_in_file = def_dec_p;
else
def_dec_p->file->defs_decs = def_dec_p;
}
else /* assert (next == NULL); */
{
((NONCONST def_dec_info *) curr)->next_in_file = def_dec_p;
/* assert (next == NULL); */
def_dec_p->next_in_file = next;
}
}
}
/* Rewrite the options list used to recompile base source files. All we are
really doing here is removing -g, -O, -S, -c, and -o options, and then
adding a final group of options like '-fgen-aux-info -S -o /dev/null'. */
static void
munge_compile_params (const char *params_list)
{
const char *temp_params[MAX_OPTIONS];
int param_count = 0;
const char *param;
temp_params[param_count++] = "gcc";
for (;;)
{
while (isspace (*params_list))
params_list++;
if (!*params_list)
break;
param = params_list;
while (*params_list && !isspace (*params_list))
params_list++;
if (param[0] != '-')
temp_params[param_count++] = dupnstr (param, params_list - param);
else
{
switch (param[1])
{
case 'g':
case 'O':
case 'S':
case 'c':
break; /* Don't copy these. */
case 'o':
while (isspace (*params_list))
params_list++;
while (*params_list && !isspace (*params_list))
params_list++;
break;
default:
temp_params[param_count++] = dupnstr (param, params_list - param);
}
}
if (!*params_list)
break;
}
temp_params[param_count++] = "-fgen-aux-info";
temp_params[param_count++] = "-S";
temp_params[param_count++] = "-o";
temp_params[param_count++] = "/dev/null";
/* Leave room for the filename argument and a terminating null pointer. */
temp_params[filename_index = param_count++] = NULL;
temp_params[param_count++] = NULL;
/* Make a copy of the compile_params in heap space. */
compile_params = xmalloc (sizeof (char *) * (param_count+1));
bcopy (temp_params, compile_params, sizeof (char *) * param_count);
}
/* Do a recompilation for the express purpose of generating a new aux_info
file to go with a specific base source file. */
static int
gen_aux_info_file (const char *base_pathname)
{
int child_pid;
if (!filename_index)
munge_compile_params ("");
compile_params[filename_index] = shortpath (NULL, base_pathname);
if (!quiet_flag)
fprintf (stderr, "%s: compiling `%s'\n",
pname, compile_params[filename_index]);
if (child_pid = vfork ())
{
if (child_pid == -1)
{
fprintf (stderr, "%s: error: could not fork process: %s\n",
pname, sys_errlist[errno]);
return 0;
}
#if 0
/* Print out the command line that the other process is now executing. */
if (!quiet_flag)
{
const char **arg;
fputs ("\t", stderr);
for (arg = compile_params; *arg; arg++)
{
fputs (*arg, stderr);
fputc (' ', stderr);
}
fputc ('\n', stderr);
fflush (stderr);
}
#endif
{
WAIT_ARG_TYPE wait_status;
if (wait (&wait_status) == -1)
{
fprintf (stderr, "%s: error: wait for process failed: %s\n",
pname, sys_errlist[errno]);
return 0;
}
if (!WIFEXITED (wait_status))
{
kill (child_pid, 9);
return 0;
}
return (WEXITSTATUS (wait_status) == 0) ? 1 : 0;
}
}
else
{
if (execvp (compile_params[0], compile_params))
{
fprintf (stderr, "%s: error: execvp returned: %s\n",
pname, sys_errlist[errno]);
exit (errno);
}
return 1; /* Never executed. */
}
}
/* Read in all of the information contained in a single aux_info file.
Save all of the important stuff for later. */
static void_t
process_aux_info_file (const char *base_source_pathname, int keep_it, int is_syscalls)
{
char aux_info_pathname[MAXPATHLEN];
char *aux_info_base;
char *aux_info_limit;
const char *aux_info_second_line;
time_t aux_info_mtime;
size_t aux_info_size;
/* Construct the aux_info pathname from the base source pathname. */
strcpy (aux_info_pathname, base_source_pathname);
strcat (aux_info_pathname, aux_info_suffix);
/* Check that the aux_info file exists and is readable. If it does not
exist, try to create it (once only). */
start_over: ;
{
int retries = 0;
retry:
if (access (aux_info_pathname, R_OK) == -1)
{
if (errno == ENOENT && retries == 0)
{
if (is_syscalls)
{
fprintf (stderr, "%s: warning: missing SYSCALLS file `%s'\n",
pname, aux_info_pathname);
return;
}
if (!gen_aux_info_file (base_source_pathname))
return;
retries++;
goto retry;
}
else
{
fprintf (stderr, "%s: error: can't read aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_pathname), sys_errlist[errno]);
errors++;
return;
}
}
}
{
struct stat stat_buf;
/* Get some status information about this aux_info file. */
if (stat (aux_info_pathname, &stat_buf) == -1)
{
fprintf (stderr, "%s: error: can't get status of aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_pathname), sys_errlist[errno]);
errors++;
return;
}
/* Check on whether or not this aux_info file is zero length. If it is,
then just ignore it and return. */
if ((aux_info_size = stat_buf.st_size) == 0)
return;
/* Get the date/time of last modification for this aux_info file and
remember it. We will have to check that any source files that it
contains information about are at least this old or older. */
aux_info_mtime = stat_buf.st_mtime;
}
{
int aux_info_file;
/* Open the aux_info file. */
if ((aux_info_file = open (aux_info_pathname, O_RDONLY )) == -1)
{
fprintf (stderr, "%s: error: can't open aux info file `%s' for reading: %s\n",
pname, shortpath (NULL, aux_info_pathname), sys_errlist[errno]);
return;
}
/* Allocate space to hold the aux_info file in memory. */
aux_info_base = xmalloc (aux_info_size + 1);
aux_info_limit = aux_info_base + aux_info_size;
*aux_info_limit = '\0';
/* Read the aux_info file into memory. */
if (read (aux_info_file, aux_info_base, aux_info_size) != aux_info_size)
{
fprintf (stderr, "%s: error: while reading aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_pathname), sys_errlist[errno]);
free (aux_info_base);
close (aux_info_file);
return;
}
/* Close the aux info file. */
if (close (aux_info_file))
{
fprintf (stderr, "%s: error: while closing aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_pathname), sys_errlist[errno]);
free (aux_info_base);
close (aux_info_file);
return;
}
}
/* Delete the aux_info file (unless requested not to). If the deletion
fails for some reason, don't even worry about it. */
if (!keep_it)
if (unlink (aux_info_pathname) == -1)
fprintf (stderr, "%s: error: can't delete aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_pathname), sys_errlist[errno]);
/* Save a pointer into the first line of the aux_info file which
contains the pathname of the directory from which the compiler
was invoked when the associated source file was compiled.
This information is used later to help create complete
pathnames out of the (potentially) relative pathnames in
the aux_info file. */
{
char *p = aux_info_base;
while (*p != ':')
p++;
p++;
while (*p == ' ')
p++;
invocation_path = p; /* Save a pointer to first byte of path. */
while (*p != ' ')
p++;
*p++ = '/';
*p++ = '\0';
while (*p++ != '\n')
continue;
aux_info_second_line = p;
}
{
const char *aux_info_p;
/* Do a pre-pass on the lines in the aux_info file, making sure that all
of the source files referenced in there are at least as old as this
aux_info file itself. If not, go back and regenerate the aux_info
file anew. Don't do any of this for the syscalls file. */
if (!is_syscalls)
{
current_aux_info_lineno = 2;
for (aux_info_p = aux_info_second_line; *aux_info_p; )
{
if (referenced_file_is_newer (aux_info_p, aux_info_mtime))
{
free (aux_info_base);
if (unlink (aux_info_pathname) == -1)
{
fprintf (stderr, "%s: error: can't delete file `%s': %s\n",
pname,
shortpath (NULL, aux_info_pathname),
sys_errlist[errno]);
return;
}
goto start_over;
}
/* Skip over the rest of this line to start of next line. */
while (*aux_info_p != '\n')
aux_info_p++;
aux_info_p++;
current_aux_info_lineno++;
}
}
/* Now do the real pass on the aux_info lines. Save their information in
the in-core data base. */
current_aux_info_lineno = 2;
for (aux_info_p = aux_info_second_line; *aux_info_p;)
{
char *unexpanded_line = unexpand_if_needed (aux_info_p);
if (unexpanded_line)
{
save_def_or_dec (unexpanded_line, is_syscalls);
free (unexpanded_line);
}
else
save_def_or_dec (aux_info_p, is_syscalls);
/* Skip over the rest of this line and get to start of next line. */
while (*aux_info_p != '\n')
aux_info_p++;
aux_info_p++;
current_aux_info_lineno++;
}
}
free (aux_info_base);
}
#ifndef UNPROTOIZE
/* Check an individual filename for a .c suffix. If the filename has this
suffix, rename the file such that its suffix is changed to .C. This
function implements the -C option. */
static void_t
rename_c_file (const hash_table_entry *hp)
{
const char *pathname = hp->symbol;
int last_char_index = strlen (pathname) - 1;
char new_pathname[MAXPATHLEN];
/* Note that we don't care here if the given file was converted or not. It
is possible that the given file was *not* converted, simply because there
was nothing in it which actually required conversion. Even in this case,
we want to do the renaming. Note that we only rename files with the .c
suffix. */
if (pathname[last_char_index] != 'c' || pathname[last_char_index-1] != '.')
return;
strcpy (new_pathname, pathname);
new_pathname[last_char_index] = 'C';
if (link (pathname, new_pathname) == -1)
{
fprintf (stderr, "%s: warning: can't link file `%s' to `%s': %s\n",
pname, shortpath (NULL, pathname),
shortpath (NULL, new_pathname), sys_errlist[errno]);
errors++;
return;
}
if (unlink (pathname) == -1)
{
fprintf (stderr, "%s: warning: can't delete file `%s': %s\n",
pname, shortpath (NULL, pathname), sys_errlist[errno]);
errors++;
return;
}
}
#endif
/* Take the list of definitions and declarations attached to a particular
file_info node and reverse the order of the list. This should get the
list into an order such that the item with the lowest associated line
number is nearest the head of the list. When these lists are originally
built, they are in the opposite order. We want to traverse them in
normal line number order later (i.e. lowest to highest) so reverse the
order here. */
static void_t
reverse_def_dec_list (const hash_table_entry *hp)
{
file_info *file_p = hp->fip;
const def_dec_info *prev = NULL;
const def_dec_info *current = file_p->defs_decs;
if (!( current = file_p->defs_decs))
return; /* no list to reverse */
prev = current;
if (! (current = current->next_in_file))
return; /* can't reverse a single list element */
((NONCONST def_dec_info *) prev)->next_in_file = NULL;
while (current)
{
const def_dec_info *next = current->next_in_file;
((NONCONST def_dec_info *) current)->next_in_file = prev;
prev = current;
current = next;
}
file_p->defs_decs = prev;
}
#ifndef UNPROTOIZE
/* Find the (only?) extern definition for a particular function name, starting
from the head of the linked list of entries for the given name. If we
cannot find an extern definition for the given function name, issue a
warning and scrounge around for the next best thing, i.e. an extern
function declaration with a prototype attached to it. Note that we only
allow such substitutions for extern declarations and never for static
declarations. That's because the only reason we allow them at all is
to let un-prototyped function declarations for system-supplied library
functions get their prototypes from our own extra SYSCALLS.c.X file which
contains all of the correct prototypes for system functions. */
static const def_dec_info *
find_extern_def (const def_dec_info *head, const def_dec_info *user)
{
const def_dec_info *dd_p;
const def_dec_info *extern_def_p = NULL;
int conflict_noted = 0;
/* Don't act too stupid here. Somebody may try to convert an entire system
in one swell fwoop (rather than one program at a time, as should be done)
and in that case, we may find that there are multiple extern definitions
of a given function name in the entire set of source files that we are
converting. If however one of these definitions resides in exactly the
same source file as the reference we are trying to satisfy then in that
case it would be stupid for us to fail to realize that this one definition
*must* be the precise one we are looking for.
To make sure that we don't miss an opportunity to make this "same file"
leap of faith, we do a prescan of the list of records relating to the
given function name, and we look (on this first scan) *only* for a
definition of the function which is in the same file as the reference
we are currently trying to satisfy. */
for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
if (dd_p->is_func_def && !dd_p->is_static && dd_p->file == user->file)
return dd_p;
/* Now, since we have not found a definition in the same file as the
reference, we scan the list again and consider all possibilities from
all files. Here we may get conflicts with the things listed in the
SYSCALLS.c.X file, but if that happens it only means that the source
code being converted contains its own definition of a function which
could have been supplied by libc.a. In such cases, we should avoid
issuing the normal warning, and defer to the definition given in the
user's own code. */
for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
if (dd_p->is_func_def && !dd_p->is_static)
{
if (!extern_def_p) /* Previous definition? */
extern_def_p = dd_p; /* Remember the first definition found. */
else
{
/* Ignore definition just found if it came from SYSCALLS.c.X. */
if (is_syscalls_file (dd_p->file))
continue;
/* Quietly replace the definition previously found with the one
just found if the previous one was from SYSCALLS.c.X. */
if (is_syscalls_file (extern_def_p->file))
{
extern_def_p = dd_p;
continue;
}
/* If we get here, then there is a conflict between two function
declarations for the same function, both of which came from the
user's own code. */
if (!conflict_noted) /* first time we noticed? */
{
conflict_noted = 1;
fprintf (stderr, "%s: error: conflicting extern definitions of '%s'\n",
pname, head->hash_entry->symbol);
if (!quiet_flag)
{
fprintf (stderr, "%s: declarations of '%s' will not be converted\n",
pname, head->hash_entry->symbol);
fprintf (stderr, "%s: conflict list for '%s' follows:\n",
pname, head->hash_entry->symbol);
fprintf (stderr, "%s: %s(%d): %s\n",
pname,
shortpath (NULL, extern_def_p->file->hash_entry->symbol),
extern_def_p->line,
extern_def_p->ansi_decl);
}
}
if (!quiet_flag)
fprintf (stderr, "%s: %s(%d): %s\n",
pname,
shortpath (NULL, dd_p->file->hash_entry->symbol),
dd_p->line,
dd_p->ansi_decl);
}
}
/* We want to err on the side of caution, so if we found multiple conflicting
definitions for the same function, treat this as being that same as if we
had found no definitions (i.e. return NULL). */
if (conflict_noted)
return NULL;
if (!extern_def_p)
{
/* We have no definitions for this function so do the next best thing.
Search for an extern declaration already in prototype form. */
for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
if (!dd_p->is_func_def && !dd_p->is_static && dd_p->prototyped)
{
extern_def_p = dd_p; /* save a pointer to the definition */
if (!quiet_flag)
fprintf (stderr, "%s: warning: using formals list from %s(%d) for function `%s'\n",
pname,
shortpath (NULL, dd_p->file->hash_entry->symbol),
dd_p->line, dd_p->hash_entry->symbol);
break;
}
/* Gripe about unprototyped function declarations that we found no
corresponding definition (or other source of prototype information)
for.
Gripe even if the unprototyped declaration we are worried about
exists in a file in one of the "system" include directories. We
can gripe about these because we should have at least found a
corresponding (pseudo) definition in the SYSCALLS.c.X file. If we
didn't, then that means that the SYSCALLS.c.X file is missing some
needed prototypes for this particular system. That is worth telling
the user about! */
if (!extern_def_p)
{
const char *file = user->file->hash_entry->symbol;
if (!quiet_flag)
if (in_system_include_dir (file))
{
char needed[MAX_LINE_LEN];
char *p;
strcpy (needed, user->ansi_decl);
p = (NONCONST char *) substr (needed, user->hash_entry->symbol)
+ strlen (user->hash_entry->symbol) + 2;
strcpy (p, "???);");
fprintf (stderr, "%s: please add `%s' to SYSCALLS (see %s(%d))\n",
pname,
needed+7, /* Don't print "extern " */
shortpath (NULL, file),
user->line);
}
else
fprintf (stderr, "%s: warning: no extern definition for `%s' (see %s(%d))\n",
pname,
user->hash_entry->symbol,
shortpath (NULL, file),
user->line);
}
}
return extern_def_p;
}
/* Find the (only?) static definition for a particular function name in a
given file. Here we get the function-name and the file info indirectly
from the def_dec_info record pointer which is passed in. */
static const def_dec_info *
find_static_definition (const def_dec_info *user)
{
const def_dec_info *head = user->hash_entry->ddip;
const def_dec_info *dd_p;
int num_static_defs = 0;
const def_dec_info *static_def_p = NULL;
for (dd_p = head; dd_p; dd_p = dd_p->next_for_func)
if (dd_p->is_func_def && dd_p->is_static && (dd_p->file == user->file))
{
static_def_p = dd_p; /* save a pointer to the definition */
num_static_defs++;
}
if (num_static_defs == 0)
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: no static definition for `%s' in file `%s'\n",
pname,
head->hash_entry->symbol,
shortpath (NULL, user->file->hash_entry->symbol));
}
else if (num_static_defs > 1)
{
fprintf (stderr, "%s: error: multiple static defs of `%s' in file `%s'\n",
pname,
head->hash_entry->symbol,
shortpath (NULL, user->file->hash_entry->symbol));
return NULL;
}
return static_def_p;
}
/* Find good prototype style formal argument lists for all of the function
declarations which didn't have them before now.
To do this we consider each function name one at a time. For each function
name, we look at the items on the linked list of def_dec_info records for
that particular name.
Somewhere on this list we should find one (and only one) def_dec_info
record which represents the actual function definition, and this record
should have a nice formal argument list already associated with it.
Thus, all we have to do is to connect up all of the other def_dec_info
records for this particular function name to the special one which has
the full-blown formals list.
Of course it is a little more complicated than just that. See below for
more details. */
static void_t
connect_defs_and_decs (const hash_table_entry *hp)
{
const def_dec_info *dd_p;
const def_dec_info *extern_def_p = NULL;
int first_extern_reference = 1;
/* Traverse the list of definitions and declarations for this particular
function name. For each item on the list, if it is a function
definition (either old style or new style) then GCC has already been
kind enough to produce a prototype for us, and it is associated with
the item already, so declare the item as its own associated "definition".
Also, for each item which is only a function declaration, but which
nonetheless has its own prototype already (obviously supplied by the user)
declare the item as it's own definition.
Note that when/if there are multiple user-supplied prototypes already
present for multiple declarations of any given function, these multiple
prototypes *should* all match exactly with one another and with the
prototype for the actual function definition. We don't check for this
here however, since we assume that the compiler must have already done
this consistancy checking when it was creating the .X files. */
for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
if (dd_p->prototyped)
((NONCONST def_dec_info *) dd_p)->definition = dd_p;
/* Traverse the list of definitions and declarations for this particular
function name. For each item on the list, if it is an extern function
declaration and if it has no associated definition yet, go try to find
the matching extern definition for the declaration.
When looking for the matching function definition, warn the user if we
fail to find one.
If we find more that one function definition also issue a warning.
Do the search for the matching definition only once per unique function
name (and only when absolutely needed) so that we can avoid putting out
redundant warning messages, and so that we will only put out warning
messages when there is actually a reference (i.e. a declaration) for
which we need to find a matching definition. */
for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
if (!dd_p->is_func_def && !dd_p->is_static && !dd_p->definition)
{
if (first_extern_reference)
{
extern_def_p = find_extern_def (hp->ddip, dd_p);
first_extern_reference = 0;
}
((NONCONST def_dec_info *) dd_p)->definition = extern_def_p;
}
/* Traverse the list of definitions and declarations for this particular
function name. For each item on the list, if it is a static function
declaration and if it has no associated definition yet, go try to find
the matching static definition for the declaration within the same file.
When looking for the matching function definition, warn the user if we
fail to find one in the same file with the declaration, and refuse to
convert this kind of cross-file static function declaration. After all,
this is stupid practice and should be discouraged.
We don't have to worry about the possibility that there is more than one
matching function definition in the given file because that would have
been flagged as an error by the compiler.
Do the search for the matching definition only once per unique
function-name/source-file pair (and only when absolutely needed) so that
we can avoid putting out redundant warning messages, and so that we will
only put out warning messages when there is actually a reference (i.e. a
declaration) for which we actually need to find a matching definition. */
for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
if (!dd_p->is_func_def && dd_p->is_static && !dd_p->definition)
{
const def_dec_info *dd_p2;
const def_dec_info *static_def;
/* We have now found a single static declaration for which we need to
find a matching definition. We want to minimize the work (and the
number of warnings), so we will find an appropriate (matching)
static definition for this declaration, and then distribute it
(as the definition for) any and all other static declarations
for this function name which occur within the same file, and which
do not already have definitions.
Note that a trick is used here to prevent subsequent attempts to
call find_static_definition() for a given function-name & file
if the first such call returns NULL. Essentially, we convert
these NULL return values to -1, and put the -1 into the definition
field for each other static declaration from the same file which
does not already have an associated definition.
This makes these other static declarations look like they are
actually defined already when the outer loop here revisits them
later on. Thus, the outer loop will skip over them. Later, we
turn the -1's back to NULL's. */
((NONCONST def_dec_info *) dd_p)->definition =
(static_def = find_static_definition (dd_p))
? static_def
: (const def_dec_info *) -1;
for (dd_p2 = dd_p->next_for_func; dd_p2; dd_p2 = dd_p2->next_for_func)
if (!dd_p2->is_func_def && dd_p2->is_static
&& !dd_p2->definition && (dd_p2->file == dd_p->file))
((NONCONST def_dec_info *)dd_p2)->definition = dd_p->definition;
}
/* Convert any dummy (-1) definitions we created in the step above back to
NULL's (as they should be). */
for (dd_p = hp->ddip; dd_p; dd_p = dd_p->next_for_func)
if (dd_p->definition == (def_dec_info *) -1)
((NONCONST def_dec_info *) dd_p)->definition = NULL;
}
#endif
/* Give a pointer into the clean text buffer, return a number which is the
original source line number that the given pointer points into. */
static int
identify_lineno (const char *clean_p)
{
int line_num = 1;
const char *scan_p;
for (scan_p = clean_text_base; scan_p <= clean_p; scan_p++)
if (*scan_p == '\n')
line_num++;
return line_num;
}
/* Issue an error message and give up on doing this particular edit. */
static void_t
declare_source_confusing (const char *clean_p)
{
if (!quiet_flag)
{
if (clean_p == 0)
fprintf (stderr, "%s: warning: source too confusing near %s(%d)\n",
pname, shortpath (NULL, convert_path), last_known_line_number);
else
fprintf (stderr, "%s: warning: source too confusing at %s(%d)\n",
pname, shortpath (NULL, convert_path), identify_lineno (clean_p));
}
longjmp (source_confusion_recovery, 1);
}
/* Check that a condition which is expected to be true in the original source
code is in fact true. If not, issue an error message and give up on
converting this particular source file. */
inline static void_t
check_source (int cond, const char *clean_p)
{
if (!cond)
declare_source_confusing (clean_p);
}
/* If we think of the in-core cleaned text buffer as a memory mapped
file (with the variable last_known_line_start acting as sort of a
file pointer) then we can imagine doing "seeks" on the buffer. The
following routine implements a kind of "seek" operation for the in-core
(cleaned) copy of the source file. When finished, it returns a pointer to
the start of a given (numbered) line in the cleaned text buffer.
Note that protoize only has to "seek" in the forward direction on the
in-core cleaned text file buffers, and it never needs to back up.
This routine is made a little bit faster by remembering the line number
(and pointer value) supplied (and returned) from the previous "seek".
This prevents us from always having to start all over back at the top
of the in-core cleaned buffer again. */
static const char *
seek_to_line (int n)
{
if (n < last_known_line_number)
abort ();
while (n > last_known_line_number)
{
while (*last_known_line_start != '\n')
check_source (++last_known_line_start < clean_text_limit, 0);
last_known_line_start++;
last_known_line_number++;
}
return last_known_line_start;
}
/* Given a pointer to a character in the cleaned text buffer, return a pointer
to the next non-whitepace character which follows it. */
static const char *
forward_to_next_token_char (const char *ptr)
{
for (++ptr; isspace (*ptr); check_source (++ptr < clean_text_limit, 0))
continue;
return ptr;
}
/* Copy a chunk of text of length `len' and starting at `str' to the current
output buffer. Note that all attempts to add stuff to the current output
buffer ultimately go through here. */
static void_t
output_bytes (const char *str, int len)
{
if ((repl_write_ptr + 1) + len >= repl_text_limit)
{
size_t new_size = (repl_text_limit - repl_text_base) << 1;
char *new_buf = (char *) xrealloc (repl_text_base, new_size);
repl_write_ptr = new_buf + (repl_write_ptr - repl_text_base);
repl_text_base = new_buf;
repl_text_limit = new_buf + new_size;
}
bcopy (str, repl_write_ptr + 1, len);
repl_write_ptr += len;
}
/* Copy all bytes (except the trailing null) of a null terminated string to
the current output buffer. */
static void_t
output_string (const char *str)
{
output_bytes (str, strlen (str));
}
/* Copy some characters from the original text buffer to the current output
buffer.
This routine takes a pointer argument `p' which is assumed to be a pointer
into the cleaned text buffer. The bytes which are copied are the `original'
equivalents for the set of bytes between the last value of `clean_read_ptr'
and the argument value `p'.
The set of bytes copied however, comes *not* from the cleaned text buffer,
but rather from the direct counterparts of these bytes within the original
text buffer.
Thus, when this function is called, some bytes from the original text
buffer (which may include original comments and preprocessing directives)
will be copied into the output buffer.
Note that the request implide when this routine is called includes the
byte pointed to by the argument pointer `p'. */
static void_t
output_up_to (const char *p)
{
int copy_length = p - clean_read_ptr;
const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
if (copy_length == 0)
return;
if (copy_length < 0)
abort ();
output_bytes (copy_start, copy_length);
clean_read_ptr = p;
}
/* Given a pointer to a def_dec_info record which represents some form of
definition of a function (perhaps a real definition, or in lieu of that
perhaps just a declaration with a full prototype) return true if this
function is one which we should avoid converting. Return false
otherwise. */
static int
other_variable_style_function (const char *ansi_header)
{
#ifdef UNPROTOIZE
/* See if we have a stdarg function, or a function which has stdarg style
parameters or a stdarg style return type. */
return (int) substr (ansi_header, "...");
#else
/* See if we have a varargs function, or a function which has varargs style
parameters or a varargs style return type. */
const char *p;
int len = strlen (varargs_style_indicator);
for (p = ansi_header; p; )
{
const char *candidate;
if ((candidate = substr (p, varargs_style_indicator)) == 0)
return 0;
else
if (!is_id_char (candidate[-1]) && !is_id_char (candidate[len]))
return 1;
else
p = candidate + 1;
}
return 0;
#endif
}
/* Do the editing operation specifically for a function "declaration". Note
that editing for function "definitions" are handled in a separate routine
below. */
static void_t
edit_fn_declaration (const def_dec_info *def_dec_p,
const char *volatile clean_text_p)
{
const char *start_formals;
const char *end_formals;
const char *function_to_edit = def_dec_p->hash_entry->symbol;
int func_name_len = strlen (function_to_edit);
const char *end_of_fn_name;
#ifndef UNPROTOIZE
const f_list_chain_item *this_f_list_chain_item;
const def_dec_info *definition = def_dec_p->definition;
/* If we are protoizing, and if we found no corresponding definition for
this particular function declaration, then just leave this declaration
exactly as it is. */
if (!definition)
return;
/* If we are protoizing, and if the corresponding definition that we found
for this particular function declaration defined an old style varargs
function, then we want to issue a warning and just leave this function
declaration unconverted. */
if (other_variable_style_function (definition->ansi_decl))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: varargs function declaration at %s(%d) not converted\n",
pname,
shortpath (NULL, def_dec_p->file->hash_entry->symbol),
def_dec_p->line);
return;
}
#endif
/* Setup here to recover from confusing source code detected during this
particular "edit". */
save_pointers ();
if (setjmp (source_confusion_recovery))
{
restore_pointers ();
fprintf (stderr, "%s: declaration of function `%s' not converted\n",
pname, function_to_edit);
return;
}
/* We are editing a function declaration. The line number we did a seek to
contains the comma or semicolon which follows the declaration. Our job
now is to scan backwards looking for the function name. This name *must*
be followed by open paren (ignoring whitespace, of course). We need to
replace everything between that open paren and the corresponding closing
paren. If we are protoizing, we need to insert the prototype-style
formals lists. If we are unprotoizing, we need to just delete everything
between the pairs of opening and closing parens. */
/* First move up to the end of the line. */
while (*clean_text_p != '\n')
check_source (++clean_text_p < clean_text_limit, 0);
clean_text_p--; /* Point to just before the newline character. */
/* Now we can scan backwards for the function name. */
do
{
for (;;)
{
/* Scan leftwards until we find some character which can be
part of an identifier. */
while (!is_id_char (*clean_text_p))
check_source (--clean_text_p > clean_read_ptr, 0);
/* Scan backwards until we find a char that cannot be part of an
identifier. */
while (is_id_char (*clean_text_p))
check_source (--clean_text_p > clean_read_ptr, 0);
/* Having found an "id break", see if the following id is the one
that we are looking for. If so, then exit from this loop. */
if (!strncmp (clean_text_p+1, function_to_edit, func_name_len))
{
char ch = *(clean_text_p + 1 + func_name_len);
/* Must also check to see that the name in the source text
ends where it should (in order to prevent bogus matches
on similar but longer identifiers. */
if (! is_id_char (ch))
break; /* exit from loop */
}
}
/* We have now found the first perfect match for the function name in
our backward search. This may or may not be the actual function
name at the start of the actual function declaration (i.e. we could
have easily been mislead). We will try to avoid getting fooled too
often by looking forward for the open paren which should follow the
identifier we just found. We ignore whitespace while hunting. If
the next non-whitespace byte we see is *not* an open left paren,
then we must assume that we have been fooled and we start over
again accordingly. Note that there is no guarrantee, that even if
we do see the open paren, that we are in the right place.
Programmers do the strangest things sometimes! */
end_of_fn_name = clean_text_p + strlen (def_dec_p->hash_entry->symbol);
start_formals = forward_to_next_token_char (end_of_fn_name);
}
while (*start_formals != '(');
/* start_of_formals now points to the opening left paren which immediately
follows the name of the function. */
/* Note that there may be several formals lists which need to be modified
due to the possibility that the return type of this function is a
pointer-to-function type. If there are several formals lists, we
convert them in left-to-right order here. */
#ifndef UNPROTOIZE
this_f_list_chain_item = definition->f_list_chain;
#endif
for (;;)
{
{
int depth;
end_formals = start_formals + 1;
depth = 1;
for (; depth; check_source (++end_formals < clean_text_limit, 0))
{
switch (*end_formals)
{
case '(':
depth++;
break;
case ')':
depth--;
break;
}
}
end_formals--;
}
/* end_formals now points to the closing right paren of the formals
list whose left paren is pointed to by start_formals. */
/* Now, if we are protoizing, we insert the new ANSI-style formals list
attached to the associated definition of this function. If however
we are unprotoizing, then we simply delete any formals list which
may be present. */
output_up_to (start_formals);
#ifndef UNPROTOIZE
if (this_f_list_chain_item)
{
output_string (this_f_list_chain_item->formals_list);
this_f_list_chain_item = this_f_list_chain_item->chain_next;
}
else
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: too many parameter lists in declaration of `%s'\n",
pname, def_dec_p->hash_entry->symbol);
check_source (0, end_formals); /* leave the declaration intact */
}
#endif
clean_read_ptr = end_formals - 1;
/* Now see if it looks like there may be another formals list associated
with the function declaration that we are converting (following the
formals list that we just converted. */
{
const char *another_r_paren = forward_to_next_token_char (end_formals);
if ((*another_r_paren != ')')
|| (*(start_formals = forward_to_next_token_char (another_r_paren)) != '('))
{
#ifndef UNPROTOIZE
if (this_f_list_chain_item)
{
if (!quiet_flag)
fprintf (stderr, "\n%s: warning: too few parameter lists in declaration of `%s'\n",
pname, def_dec_p->hash_entry->symbol);
check_source (0, start_formals); /* leave the decl intact */
}
#endif
break;
}
}
/* There does appear to be yet another formals list, so loop around
again, and convert it also. */
}
}
/* Edit a whole group of formals lists, starting with the rightmost one
from some set of formals lists. This routine is called once (from the
outside) for each function declaration which is converted. It is
recursive however, and it calls itself once for each remaining formal
list that lies to the left of the one it was originally called to work
on. Thus, a whole set gets done in right-to-left order.
This routine returns non-zero if it thinks that it should not be trying
to convert this particular function definition (because the name of the
function doesn't match the one expected). */
static int
edit_formals_lists (const char *end_formals, int f_list_count, const def_dec_info *def_dec_p)
{
const char *start_formals;
int depth;
start_formals = end_formals - 1;
depth = 1;
for (; depth; check_source (--start_formals > clean_read_ptr, 0))
{
switch (*start_formals)
{
case '(':
depth--;
break;
case ')':
depth++;
break;
}
}
start_formals++;
/* start_formals now points to the opening left paren of the formals list. */
f_list_count--;
if (f_list_count)
{
const char *next_end;
/* There should be more formal lists to the left of here. */
next_end = start_formals - 1;
check_source (next_end > clean_read_ptr, 0);
while (isspace (*next_end))
check_source (--next_end > clean_read_ptr, 0);
check_source (*next_end == ')', next_end);
check_source (--next_end > clean_read_ptr, 0);
check_source (*next_end == ')', next_end);
if (edit_formals_lists (next_end, f_list_count, def_dec_p))
return 1;
}
/* Check that the function name in the header we are working on is the same
as the one we would expect to find. If not, issue a warning and return
non-zero. */
if (f_list_count == 0)
{
const char *expected = def_dec_p->hash_entry->symbol;
const char *func_name_start;
const char *func_name_limit;
size_t func_name_len;
for (func_name_limit = start_formals-1; isspace (*func_name_limit); )
check_source (--func_name_limit > clean_read_ptr, 0);
for (func_name_start = func_name_limit++;
is_id_char (*func_name_start);
func_name_start--)
check_source (func_name_start > clean_read_ptr, 0);
func_name_start++;
func_name_len = func_name_limit - func_name_start;
if (func_name_len == 0)
check_source (0, func_name_start);
if (func_name_len != strlen (expected)
|| strncmp (func_name_start, expected, func_name_len))
{
fprintf (stderr, "%s: warning: found `%s' at %s(%d) but expected `%s'\n",
pname,
dupnstr (func_name_start, func_name_len),
shortpath (NULL, def_dec_p->file->hash_entry->symbol),
identify_lineno (func_name_start),
expected);
return 1;
}
}
output_up_to (start_formals);
#ifdef UNPROTOIZE
if (f_list_count == 0)
output_string (def_dec_p->formal_names);
#else
{
int f_list_depth;
const f_list_chain_item *flci_p = def_dec_p->f_list_chain;
/* At this point, the current value of f_list count says how many
links we have to follow through the f_list_chain to get to the
particular formals list that we need to output next. */
for (f_list_depth = 0; f_list_depth < f_list_count; f_list_depth++)
flci_p = flci_p->chain_next;
output_string (flci_p->formals_list);
}
#endif
clean_read_ptr = end_formals - 1;
return 0;
}
/* Given a pointer to a byte in the clean text buffer which points to the
beginning of a line that contains a "follower" token for a function
definition header, do whatever is necessary to find the right closing
paren for the rightmost formals list of the function definition header.
*/
static const char *
find_rightmost_formals_list (const char *clean_text_p)
{
const char *end_formals;
/* We are editing a function definition. The line number we did a seek
to contains the first token which immediately follows the entire set of
formals lists which are part of this particular function definition
header.
Our job now is to scan leftwards in the clean text looking for the
right-paren which is at the end of the function header's rightmost
formals list.
If we ignore whitespace, this right paren should be the first one we
see which is (ignoring whitespace) immediately followed either by the
open curly-brace beginning the function body or by an alphabetic
character (in the case where the function definition is in old (K&R)
style and there are some declarations of formal parameters). */
/* It is possible that the right paren we are looking for is on the
current line (together with its following token). Just in case that
might be true, we start out here by skipping down to the right end of
the current line before starting our scan. */
for (end_formals = clean_text_p; *end_formals != '\n'; end_formals++)
continue;
end_formals--;
/* Now scan backwards while looking for the right end of the rightmost
formals list associated with this function definition. */
for (;;)
{
char ch;
const char *l_brace_p;
/* Look leftward and try to find a right-paren. */
while (*end_formals != ')')
{
if (isspace (*end_formals))
while (isspace (*end_formals))
check_source (--end_formals > clean_read_ptr, 0);
else
check_source (--end_formals > clean_read_ptr, 0);
}
ch = *(l_brace_p = forward_to_next_token_char (end_formals));
#ifdef UNPROTOIZE
/* Since we are unprotoizing an ANSI-style (prototyped) function
definition, there had better not be anything (except whitespace)
between the end of the ANSI formals list and the beginning of the
function body (i.e. the '{'). */
check_source (ch == '{', l_brace_p);
break;
#else
/* Since it is possible that we found a right paren before the starting
'{' of the body which IS NOT the one at the end of the real K&R
formals list (say for instance, we found one embedded inside one of
the old K&R formal parameter declarations) we have to check to be
sure that this is in fact the right paren that we were looking for.
The one we were looking for *must* be followed by either a '{' or
by an alphabetic character, while others *cannot* legally be followed
by such characters. */
if ((ch == '{') || isalpha (ch))
break;
/* At this point, we have found a right paren, but we know that it is
not the one we were looking for, so backup one character and keep
looking. */
check_source (--end_formals > clean_read_ptr, 0);
#endif
}
return end_formals;
}
#ifndef UNPROTOIZE
/* Insert into the output file a totally new declaration for a function
which (up until now) was being called from within the current block
without having been declared at any point such that the declaration
was visible (i.e. in scope) at the point of the call.
We need to add in explicit declarations for all such function calls
in order to get the full benefit of prototype-based function call
parameter type checking. */
static void_t
add_local_decl (const def_dec_info *def_dec_p, const char *clean_text_p)
{
const char *start_of_block;
const char *function_to_edit = def_dec_p->hash_entry->symbol;
/* Don't insert new local explicit declarations unless explicitly requested
to do so. */
if (!local_flag)
return;
/* Setup here to recover from confusing source code detected during this
particular "edit". */
save_pointers ();
if (setjmp (source_confusion_recovery))
{
restore_pointers ();
fprintf (stderr, "%s: local declaration for function `%s' not inserted\n",
pname, function_to_edit);
return;
}
/* We have already done a seek to the start of the line which should
contain *the* open curly brace which begins the block in which we need
to insert an explicit function declaration (to replace the implicit one).
Now we scan that line, starting from the left, until we find the
open curly brace we are looking for. Note that there may actually be
multiple open curly braces on the given line, but we will be happy
with the leftmost one no matter what. */
start_of_block = clean_text_p;
while (*start_of_block != '{' && *start_of_block != '\n')
check_source (++start_of_block < clean_text_limit, 0);
/* Note that the line from the original source could possibly
contain *no* open curly braces! This happens if the line contains
a macro call which expands into a chunk of text which includes a
block (and that block's associated open and close curly braces).
In cases like this, we give up, issue a warning, and do nothing. */
if (*start_of_block != '{')
{
if (!quiet_flag)
fprintf (stderr,
"\n%s: warning: can't add declaration of `%s' into macro call at %s(%d)\n",
pname,
def_dec_p->hash_entry->symbol,
def_dec_p->file->hash_entry->symbol,
def_dec_p->line);
return;
}
/* Figure out what a nice (pretty) indentation would be for the new
declaration we are adding. In order to do this, we must scan forward
from the '{' until we find the first line which starts with some
non-whitespace characters (i.e. real "token" material). */
{
const char *ep = forward_to_next_token_char (start_of_block) - 1;
const char *sp;
/* Now we have ep pointing at the rightmost byte of some existing indent
stuff. At least that is the hope.
We can now just scan backwards and find the left end of the existing
indentation string, and then copy it to the output buffer. */
for (sp = ep; isspace (*sp) && *sp != '\n'; sp--)
continue;
/* Now write out the open { which began this block, and any following
trash up to and including the last byte of the existing indent that
we just found. */
output_up_to (ep);
/* Now we go ahead and insert the new declaration at this point.
If the definition of the given function is in the same file that we
are currently editing, and if its full ANSI declaration normally
would start with the keyword `extern', suppress the `extern'. */
{
const char *decl = def_dec_p->definition->ansi_decl;
if ((*decl == 'e') && (def_dec_p->file == def_dec_p->definition->file))
decl += 7;
output_string (decl);
}
/* Finally, write out a new indent string, just like the preceeding one
that we found. This will typically include a newline as the first
character of the indent string. */
output_bytes (sp, (ep - sp) + 1);
}
}
/* Given a pointer to a file_info record, and a pointer to the beginning
of a line (in the clean text buffer) which is assumed to contain the
first "follower" token for the first function definition header in the
given file, find a good place to insert some new global function
declarations (which will replace scattered and imprecise implicit ones)
and then insert the new explicit declaration at that point in the file. */
static void_t
add_global_decls (const file_info *file_p, const char *clean_text_p)
{
const def_dec_info *dd_p;
const char *scan_p;
/* Setup here to recover from confusing source code detected during this
particular "edit". */
save_pointers ();
if (setjmp (source_confusion_recovery))
{
restore_pointers ();
fprintf (stderr, "%s: global declarations for file `%s' not inserted\n",
pname, shortpath (NULL, file_p->hash_entry->symbol));
return;
}
/* Start by finding a good location for adding the new explicit function
declarations. To do this, we scan backwards, ignoring whitespace
and comments and other junk until we find either a semicolon, or until
we hit the beginning of the file. */
scan_p = find_rightmost_formals_list (clean_text_p);
for (;; --scan_p)
{
if (scan_p < clean_text_base)
break;
check_source (scan_p > clean_read_ptr, 0);
if (*scan_p == ';')
break;
}
/* scan_p now points either to a semicolon, or to just before the start
of the whole file. */
/* Now scan forward for the first non-whitespace character. In theory,
this should be the first character of the following function definition
header. We will put in the added declarations just prior to that. */
scan_p++;
while (isspace (*scan_p))
scan_p++;
scan_p--;
output_up_to (scan_p);
/* Now write out full prototypes for all of the things that had been
implicitly declared in this file (but only those for which we were
actually able to find unique matching definitions). Avoid duplicates
by marking things that we write out as we go. */
{
int some_decls_added = 0;
for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
if (dd_p->is_implicit && dd_p->definition && !dd_p->definition->written)
{
const char *decl = dd_p->definition->ansi_decl;
/* If the function for which we are inserting a declaration is
actually defined later in the same file, then suppress the
leading `extern' keyword (if there is one). */
if (*decl == 'e' && (dd_p->file == dd_p->definition->file))
decl += 7;
output_string ("\n");
output_string (decl);
some_decls_added = 1;
((NONCONST def_dec_info *) dd_p->definition)->written = 1;
}
if (some_decls_added)
output_string ("\n\n");
}
/* Unmark all of the definitions that we just marked. */
for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
if (dd_p->definition)
((NONCONST def_dec_info *) dd_p->definition)->written = 0;
}
#endif
/* Do the editing operation specifically for a function "definition". Note
that editing operations for function "declarations" are handled by a
separate routine above. */
static void_t
edit_fn_definition (const def_dec_info *def_dec_p, const char *clean_text_p)
{
const char *end_formals;
const char *function_to_edit = def_dec_p->hash_entry->symbol;
/* Setup here to recover from confusing source code detected during this
particular "edit". */
save_pointers ();
if (setjmp (source_confusion_recovery))
{
restore_pointers ();
fprintf (stderr, "%s: definition of function `%s' not converted\n",
pname, function_to_edit);
return;
}
end_formals = find_rightmost_formals_list (clean_text_p);
/* end_of_formals now points to the closing right paren of the rightmost
formals list which is actually part of the `header' of the function
definition that we are converting. */
/* If the header of this function definition looks like it declares a
function with a variable number of arguments, and if the way it does
that is different from that way we would like it (i.e. varargs vs.
stdarg) then issue a warning and leave the header unconverted. */
if (other_variable_style_function (def_dec_p->ansi_decl))
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: %s function definition at %s(%d) not converted\n",
pname,
other_var_style,
shortpath (NULL, def_dec_p->file->hash_entry->symbol),
identify_lineno (end_formals));
output_up_to (end_formals);
return;
}
if (edit_formals_lists (end_formals, def_dec_p->f_list_count, def_dec_p))
{
restore_pointers ();
fprintf (stderr, "%s: definition of function `%s' not converted\n",
pname, function_to_edit);
return;
}
/* Have to output the last right paren because this never gets flushed by
edit_formals_list. */
output_up_to (end_formals);
#ifdef UNPROTOIZE
{
const char *decl_p;
const char *semicolon_p;
const char *limit_p;
const char *scan_p;
int had_newlines = 0;
/* Now write out the K&R style formal declarations, one per line. */
decl_p = def_dec_p->formal_decls;
limit_p = decl_p + strlen (decl_p);
for (;decl_p < limit_p; decl_p = semicolon_p + 2)
{
for (semicolon_p = decl_p; *semicolon_p != ';'; semicolon_p++)
continue;
output_string ("\n");
output_string (indent_string);
output_bytes (decl_p, (semicolon_p + 1) - decl_p);
}
/* If there are no newlines between the end of the formals list and the
start of the body, we should insert one now. */
for (scan_p = end_formals+1; *scan_p != '{'; )
{
if (*scan_p == '\n')
{
had_newlines = 1;
break;
}
check_source (++scan_p < clean_text_limit, 0);
}
if (!had_newlines)
output_string ("\n");
}
#else
/* If we are protoizing, there may be some flotsum & jetsum (like comments
and preprocessing directives) after the old formals list but before
the following { and we would like to preserve that stuff while effectively
deleting the existing K&R formal parameter declarations. We do so here
in a rather tricky way. Basically, we white out any stuff *except*
the comments/pp-directives in the original text buffer, then, if there
is anything in this area *other* than whitespace, we output it. */
{
const char *end_formals_orig;
const char *start_body;
const char *start_body_orig;
const char *scan;
const char *scan_orig;
int have_flotsum = 0;
int have_newlines = 0;
for (start_body = end_formals + 1; *start_body != '{';)
check_source (++start_body < clean_text_limit, 0);
end_formals_orig = orig_text_base + (end_formals - clean_text_base);
start_body_orig = orig_text_base + (start_body - clean_text_base);
scan = end_formals + 1;
scan_orig = end_formals_orig + 1;
for (; scan < start_body; scan++, scan_orig++)
{
if (*scan == *scan_orig)
{
have_newlines |= (*scan_orig == '\n');
/* Leave identical whitespace alone. */
if (!isspace (*scan_orig))
*((NONCONST char *)scan_orig) = ' '; /* identical - so whiteout */
}
else
have_flotsum = 1;
}
if (have_flotsum)
output_bytes (end_formals_orig + 1, start_body_orig-end_formals_orig-1);
else
if (have_newlines)
output_string ("\n");
else
output_string (" ");
clean_read_ptr = start_body - 1;
}
#endif
}
/* Clean up the clean text buffer. Do this by converting comments and
preprocessor directives into spaces. Also convert line continuations
into whitespace. Also, whiteout string and character literals. */
static void
do_cleaning (char *new_clean_text_base, char *new_clean_text_limit)
{
char *scan_p;
int non_whitespace_since_newline = 0;
for (scan_p = new_clean_text_base; scan_p < new_clean_text_limit; scan_p++)
{
switch (*scan_p)
{
case '/': /* Handle comments. */
if (scan_p[1] != '*')
goto regular;
non_whitespace_since_newline = 1;
scan_p[0] = ' ';
scan_p[1] = ' ';
scan_p += 2;
while (scan_p[1] != '/' || scan_p[0] != '*')
{
if (!isspace (*scan_p))
*scan_p = ' ';
if (++scan_p >= new_clean_text_limit)
abort ();
}
*scan_p++ = ' ';
*scan_p = ' ';
break;
case '#': /* Handle pp directives. */
if (non_whitespace_since_newline)
goto regular;
*scan_p = ' ';
while (scan_p[1] != '\n' || scan_p[0] == '\\')
{
if (!isspace (*scan_p))
*scan_p = ' ';
if (++scan_p >= new_clean_text_limit)
abort ();
}
*scan_p++ = ' ';
break;
case '\'': /* Handle character literals. */
non_whitespace_since_newline = 1;
while (scan_p[1] != '\'' || scan_p[0] == '\\')
{
if (scan_p[0] == '\\' && !isspace (scan_p[1]))
scan_p[1] = ' ';
if (!isspace (*scan_p))
*scan_p = ' ';
if (++scan_p >= new_clean_text_limit)
abort ();
}
*scan_p++ = ' ';
break;
case '"': /* Handle string literals. */
non_whitespace_since_newline = 1;
while (scan_p[1] != '"' || scan_p[0] == '\\')
{
if (scan_p[0] == '\\' && !isspace (scan_p[1]))
scan_p[1] = ' ';
if (!isspace (*scan_p))
*scan_p = ' ';
if (++scan_p >= new_clean_text_limit)
abort ();
}
*scan_p++ = ' ';
break;
case '\\': /* Handle line continuations. */
if (scan_p[1] != '\n')
goto regular;
*scan_p = ' ';
break;
case '\n':
non_whitespace_since_newline = 0; /* Reset. */
break;
case ' ':
case '\v':
case '\t':
case '\r':
case '\f':
case '\b':
break; /* Whitespace characters. */
default:
regular:
non_whitespace_since_newline = 1;
break;
}
}
}
/* Given a pointer to the closing right parenthesis for a particular formals
list (in the clean text buffer) find the corresponding left parenthesis
and return a pointer to it. */
static const char *
careful_find_l_paren (const char *p)
{
const char *q;
int paren_depth;
for (paren_depth = 1, q = p-1; paren_depth; check_source (--q >= clean_text_base, 0))
{
switch (*q)
{
case ')':
paren_depth++;
break;
case '(':
paren_depth--;
break;
}
}
return ++q;
}
/* Scan the clean text buffer for cases of function definitions that we
don't really know about because they were preprocessed out when the
aux info files were created.
In this version of protoize/unprotoize we just give a warning for each
one found. A later version may be able to at least unprotoize such
missed items.
Note that we may easily find all function definitions simply by
looking for places where there is a left paren which is (ignoring
whitespace) immediately followed by either a left-brace or by an
upper or lower case letter. Whenever we find this combination, we
have also found a function definition header.
Finding function *declarations* using syntactic clues is much harder.
I will probably try to do this in a later version though. */
static void
scan_for_missed_items (const file_info *file_p)
{
static const char *scan_p;
const char *limit = clean_text_limit - 3;
static const char *backup_limit;
backup_limit = clean_text_base - 1;
for (scan_p = clean_text_base; scan_p < limit; scan_p++)
{
if (*scan_p == ')')
{
static const char *last_r_paren;
const char *ahead_p;
last_r_paren = scan_p;
for (ahead_p = scan_p + 1; isspace (*ahead_p); )
check_source (++ahead_p < limit, limit);
scan_p = ahead_p - 1;
if (isalpha (*ahead_p) || *ahead_p == '{')
{
const char *last_l_paren;
const int lineno = identify_lineno (ahead_p);
if (setjmp (source_confusion_recovery))
continue;
/* We know we have a function definition header. Now skip
leftwards over all of its associated formals lists. */
do
{
last_l_paren = careful_find_l_paren (last_r_paren);
for (last_r_paren = last_l_paren-1; isspace (*last_r_paren); )
check_source (--last_r_paren >= backup_limit, backup_limit);
}
while (*last_r_paren == ')');
if (is_id_char (*last_r_paren))
{
char func_name[MAX_LINE_LEN];
const char *id_limit = last_r_paren + 1;
const char *id_start;
int id_length;
const def_dec_info *dd_p;
for (id_start = id_limit-1; is_id_char (*id_start); )
check_source (--id_start >= backup_limit, backup_limit);
id_start++;
backup_limit = id_start;
if ((id_length = id_limit - id_start) == 0)
goto not_missed;
strncpy (func_name, id_start, id_length);
func_name[id_length] = '\0';
{
static const char * const stmt_keywords[] =
{ "if", "while", "for", "switch", "return", 0 };
const char * const *stmt_keyword;
/* We must check here to see if we are actually looking at
a statement rather than an actual function call. */
for (stmt_keyword = stmt_keywords; *stmt_keyword; stmt_keyword++)
if (!strcmp (func_name, *stmt_keyword))
goto not_missed;
}
#if 0
fprintf (stderr, "%s: found definition of `%s' at %s(%d)\n",
pname,
func_name,
shortpath (NULL, file_p->hash_entry->symbol),
identify_lineno (id_start));
#endif
/* We really should check for a match of the function name
here also, but why bother. */
for (dd_p = file_p->defs_decs; dd_p; dd_p = dd_p->next_in_file)
if (dd_p->is_func_def && dd_p->line == lineno)
goto not_missed;
/* If we make it here, then we did not know about this
function definition. */
fprintf (stderr, "%s: warning: `%s' at %s(%d) was #if 0\n",
pname,
func_name,
shortpath (NULL, file_p->hash_entry->symbol),
identify_lineno (id_start));
fprintf (stderr, "%s: function definition not converted\n",
pname);
not_missed: ;
}
}
}
}
}
/* Do all editing operations for a single source file (either a "base" file
or an "include" file). To do this we read the file into memory, keep a
virgin copy there, make another cleaned in-core copy of the original file
(i.e. one in which all of the comments and preprocessor directives have
been replaced with whitespace), then use these two in-core copies of the
file to make a new edited in-core copy of the file. Finally, rename the
original file (as a way of saving it), and then write the edited version
of the file from core to a disk file of the same name as the original.
Note that the trick of making a copy of the original sans comments &
preprocessor directives make the editing a whole lot easier. */
static void_t
edit_file (const hash_table_entry *hp)
{
struct stat stat_buf;
const file_info *file_p = hp->fip;
char *new_orig_text_base;
char *new_orig_text_limit;
char *new_clean_text_base;
char *new_clean_text_limit;
size_t orig_size;
size_t repl_size;
int first_definition_in_file;
/* If we are not supposed to be converting this file, or if there is
nothing in there which needs converting, just skip this file. */
if (!needs_to_be_converted (file_p))
return;
convert_path = file_p->hash_entry->symbol;
/* If this file should not be converted for any reason, don't even try
unless the -f option was used. Even with the -f flag, don't bother
with those files that we could not read & replace even if we tried. */
if ((!file_normally_convertable (convert_path) && !force_flag)
|| !file_could_be_converted (convert_path))
{
if (!quiet_flag
#ifdef UNPROTOIZE
/* Don't even mention "system" include files unless we are
protoizing. If we are protoizing, we mention these as a
gentile way of prodding the user to convert his "system"
include files to prototype format. */
&& !in_system_include_dir (convert_path)
#endif
)
fprintf (stderr, "%s: file `%s' not converted\n",
pname, shortpath (NULL, convert_path));
return;
}
/* Let the user know what we are up to. */
if (nochange_flag)
puts (shortpath (NULL, convert_path));
else
{
fprintf (stderr, "%s: converting file `%s'\n",
pname, shortpath (NULL, convert_path));
fflush (stderr);
}
/* Find out the size (in bytes) of the original file. */
if (stat (convert_path, &stat_buf) == -1)
{
fprintf (stderr, "%s: error: can't get status for file `%s': %s\n",
pname, shortpath (NULL, convert_path), sys_errlist[errno]);
return;
}
orig_size = stat_buf.st_size;
/* Allocate a buffer to hold the original text. */
orig_text_base = new_orig_text_base = (char *) xmalloc (orig_size + 2);
orig_text_limit = new_orig_text_limit = new_orig_text_base + orig_size;
/* Allocate a buffer to hold the cleaned-up version of the original text. */
clean_text_base = new_clean_text_base = (char *) xmalloc (orig_size + 2);
clean_text_limit = new_clean_text_limit = new_clean_text_base + orig_size;
clean_read_ptr = clean_text_base - 1;
/* Allocate a buffer that will hopefully be large enough to hold the entire
converted output text. As an initial guess for the maximum size of the
output buffer, use 125% of the size of the original + some extra. This
buffer can be expanded later as needed. */
repl_size = orig_size + (orig_size >> 2) + 4096;
repl_text_base = (char *) xmalloc (repl_size + 2);
repl_text_limit = repl_text_base + repl_size - 1;
repl_write_ptr = repl_text_base - 1;
{
int input_file;
/* Open the file to be converted in READ ONLY mode. */
if ((input_file = open (convert_path, O_RDONLY)) == -1)
{
fprintf (stderr, "%s: error: can't open file `%s' for reading: %s\n",
pname, shortpath (NULL, convert_path), sys_errlist[errno]);
return;
}
/* Read the entire original source text file into the original text buffer
in one swell fwoop. Then figure out where the end of the text is and
make sure that it ends with a newline followed by a null. */
if (read (input_file, new_orig_text_base, orig_size) != orig_size)
{
close (input_file);
fprintf (stderr, "\n%s: error: while reading input file `%s': %s\n",
pname, shortpath (NULL, convert_path), sys_errlist[errno]);
return;
}
close (input_file);
}
if (orig_size == 0 || orig_text_limit[-1] != '\n')
{
*new_orig_text_limit++ = '\n';
orig_text_limit++;
}
/* Create the cleaned up copy of the original text. */
bcopy (orig_text_base, new_clean_text_base, orig_text_limit - orig_text_base);
do_cleaning (new_clean_text_base, new_clean_text_limit);
#if 0
{
int clean_file;
size_t clean_size = orig_text_limit - orig_text_base;
char clean_path[MAXPATHLEN];
/* Open (and create) the clean file. */
strcpy (clean_path, convert_path);
strcat (clean_path, ".clean");
if ((clean_file = open (clean_path, O_CREAT | O_WRONLY, 0666)) == -1)
{
fprintf (stderr, "%s: error: can't create/open clean file `%s': %s\n",
pname,
shortpath (NULL, clean_path),
sys_errlist[errno]);
return;
}
/* Write the clean file. */
if (write (clean_file, new_clean_text_base, clean_size) != clean_size)
fprintf (stderr, "%s: error: while writing file `%s': %s\n",
pname, shortpath (NULL, clean_path), sys_errlist[errno]);
close (clean_file);
}
#endif
/* Do a simplified scan of the input looking for things that were not
mentioned in the aux info files because of the fact that they were
in a region of the source which was preprocessed-out (via #if or
via #ifdef). */
scan_for_missed_items (file_p);
/* Setup to do line-oriented forward seeking in the clean text buffer. */
last_known_line_number = 1;
last_known_line_start = clean_text_base;
/* Now get down to business and make all of the necessary edits. */
{
const def_dec_info *def_dec_p;
first_definition_in_file = 1;
def_dec_p = file_p->defs_decs;
for (; def_dec_p; def_dec_p = def_dec_p->next_in_file)
{
const char *clean_text_p = seek_to_line (def_dec_p->line);
/* clean_text_p now points to the first character of the line which
contains the `terminator' for the declaration or definition that
we are about to process. */
#ifndef UNPROTOIZE
if (global_flag && def_dec_p->is_func_def && first_definition_in_file)
{
add_global_decls (def_dec_p->file, clean_text_p);
first_definition_in_file = 0;
}
/* Don't edit this item if it is already in prototype format or if it
is a function declaration and we have found no corresponding
definition. */
if (def_dec_p->prototyped
|| (!def_dec_p->is_func_def && !def_dec_p->definition))
continue;
#endif
if (def_dec_p->is_func_def)
edit_fn_definition (def_dec_p, clean_text_p);
else
#ifndef UNPROTOIZE
if (def_dec_p->is_implicit)
add_local_decl (def_dec_p, clean_text_p);
else
#endif
edit_fn_declaration (def_dec_p, clean_text_p);
}
}
/* Finalize things. Output the last trailing part of the original text. */
output_up_to (clean_text_limit - 1);
/* If this is just a test run, stop now and just deallocate the buffers. */
if (nochange_flag)
{
free (new_orig_text_base);
free (new_clean_text_base);
free (repl_text_base);
return;
}
/* Change the name of the original input file. This is just a quick way of
saving the original file. */
if (!nosave_flag)
{
char *new_path =
(char *) xmalloc (strlen (convert_path) + strlen (save_suffix) + 2);
strcpy (new_path, convert_path);
strcat (new_path, save_suffix);
if (link (convert_path, new_path) == -1)
{
if (errno == EEXIST)
{
if (!quiet_flag)
fprintf (stderr, "%s: warning: file `%s' already saved in `%s'\n",
pname,
shortpath (NULL, convert_path),
shortpath (NULL, new_path));
}
else
{
fprintf (stderr, "%s: error: can't link file `%s' to `%s': %s\n",
pname,
shortpath (NULL, convert_path),
shortpath (NULL, new_path),
sys_errlist[errno]);
return;
}
}
}
if (unlink (convert_path) == -1)
{
fprintf (stderr, "%s: error: can't delete file `%s': %s\n",
pname,
shortpath (NULL, convert_path),
sys_errlist[errno]);
return;
}
{
int output_file;
/* Open (and create) the output file. */
if ((output_file = open (convert_path, O_CREAT | O_WRONLY, 0777)) == -1)
{
fprintf (stderr, "%s: error: can't create/open output file `%s': %s\n",
pname,
shortpath (NULL, convert_path),
sys_errlist[errno]);
return;
}
/* Write the output file. */
{
unsigned int out_size = (repl_write_ptr + 1) - repl_text_base;
if (write (output_file, repl_text_base, out_size) != out_size)
fprintf (stderr, "%s: error: while writing file `%s': %s\n",
pname, shortpath (NULL, convert_path), sys_errlist[errno]);
}
close (output_file);
}
/* Deallocate the conversion buffers. */
free (new_orig_text_base);
free (new_clean_text_base);
free (repl_text_base);
/* Change the mode of the output file to match the original file. */
if (chmod (convert_path, stat_buf.st_mode) == -1)
fprintf (stderr, "%s: error: can't change mode of file `%s': %s\n",
pname, shortpath (NULL, convert_path), sys_errlist[errno]);
/* Note: We would try to change the owner and group of the output file
to match those of the input file here, except that may not be a good
thing to do because it might be misleading. Also, it might not even
be possible to do that (on BSD systems with quotas for instance). */
}
/* Do all of the individual steps needed to do the protoization (or
unprotoization) of the files referenced in the aux_info files given
in the command line. */
static void_t
do_processing (void)
{
const char * const *base_pp;
const char * const * const end_pps = &base_source_paths[base_source_files];
#ifndef UNPROTOIZE
int syscalls_len;
#endif
/* One-by-one, check (and create if necessary), open, and read all of the
stuff in each aux_info file. After reading each aux_info file, the
aux_info_file just read will be automatically deleted unless the
keep_flag is set. */
for (base_pp = base_source_paths; base_pp < end_pps; base_pp++)
process_aux_info_file (*base_pp, keep_flag, 0);
#ifndef UNPROTOIZE
/* Also open and read the special SYSCALLS.c aux_info file which gives us
the prototypes for all of the standard system-supplied functions. */
if (nondefault_syscalls_dir)
{
syscalls_pathname
= (char *) xmalloc (strlen (nondefault_syscalls_dir)
+ strlen (syscalls_filename) + 1);
strcpy (syscalls_pathname, nondefault_syscalls_dir);
}
else
{
syscalls_pathname
= (char *) xmalloc (strlen (default_syscalls_dir)
+ strlen (syscalls_filename) + 1);
strcpy (syscalls_pathname, default_syscalls_dir);
}
syscalls_len = strlen (syscalls_pathname);
if (*(syscalls_pathname + syscalls_len - 1) != '/')
{
*(syscalls_pathname + syscalls_len++) = '/';
*(syscalls_pathname + syscalls_len) = '\0';
}
strcat (syscalls_pathname, syscalls_filename);
/* Call process_aux_info_file in such a way that it does not try to
delete the SYSCALLS aux_info file. */
process_aux_info_file (syscalls_pathname, 1, 1);
#endif
/* When we first read in all of the information from the aux_info files
we saved in it decending line number order, because that was likely to
be faster. Now however, we want the chains of def & dec records to
appear in ascending line number order as we get further away from the
file_info record that they hang from. The following line causes all of
these lists to be rearranged into ascending line number order. */
visit_each_hash_node (pathname_primary, reverse_def_dec_list);
#ifndef UNPROTOIZE
/* Now do the "real" work. The following line causes each declaration record
to be "visited". For each of these nodes, an attempt is made to match
up the function declaration with a corresponding function definition,
which should have a full prototype-format formals list with it. Once
these match-ups are made, the conversion of the function declarations
to prototype format can be made. */
visit_each_hash_node (function_name_primary, connect_defs_and_decs);
#endif
/* Now convert each file that can be converted (and needs to be). */
visit_each_hash_node (pathname_primary, edit_file);
#ifndef UNPROTOIZE
/* If we are working in cplusplus mode, try to rename all .c files to .C
files. Don't panic if some of the renames don't work. */
if (cplusplus_flag && !nochange_flag)
visit_each_hash_node (pathname_primary, rename_c_file);
#endif
}
int
main (int argc, const char **const argv)
{
const char **argn;
const char *option_letter_p;
pname = (pname = rindex (argv[0], '/')) ? pname+1 : argv[0];
/* On DG/UX 4.10 (AViiON) both getcwd and getwd are supplied, but I don't
trust either of them. Internally, they call fclose() without having
first called fopen(). */
getwd (cwd_buffer);
for (argn = argv+1; *argn; argn++)
{
if (**argn != '-')
base_source_files++; /* Just count the possible filename args for now. */
else
{
option_letter_p = (*argn) + 1;
if (!*option_letter_p)
{
fprintf (stderr, "%s: fatal error: invalid option: -\n", pname);
errors = -1;
}
for (;*option_letter_p; option_letter_p++)
{
switch (*option_letter_p)
{
case 'V':
version_flag = 1;
break;
case 'q':
quiet_flag = 1;
break;
case 'f':
force_flag = 1;
break;
case 'n':
nochange_flag = 1;
keep_flag = 1;
break;
case 'N':
nosave_flag = 1;
break;
case 'k':
keep_flag = 1;
break;
case 'c':
if (*++argn)
{
munge_compile_params (*argn);
*argn = NULL;
}
else
{
fprintf (stderr, "%s: fatal error: missing argument for -c option\n",
pname);
errors++;
}
if (*(option_letter_p + 1))
{
fprintf (stderr, "%s: fatal error: -c must be last in a group\n",
pname);
errors++;
}
break;
#ifdef UNPROTOIZE
case 'i':
if (*++argn)
{
indent_string = *argn;
*argn = NULL;
}
else
{
fprintf (stderr, "%s: fatal error: missing argument for -i option\n",
pname);
errors++;
}
if (*(option_letter_p + 1))
{
fprintf (stderr, "%s: fatal error: -i must be last in a group\n",
pname);
errors++;
}
break;
#else
case 'l':
local_flag = 1;
break;
case 'g':
global_flag = 1;
break;
case 'C':
cplusplus_flag = 1;
break;
case 'B':
if (*++argn)
{
nondefault_syscalls_dir = *argn;
*argn = NULL;
}
else
{
fprintf (stderr, "%s: fatal error: missing argument for -B option\n",
pname);
errors++;
}
if (*(option_letter_p + 1))
{
fprintf (stderr, "%s: fatal error: -B must be last in a group\n",
pname);
errors++;
}
break;
#endif
default:
fprintf (stderr, "%s: fatal error: invalid option: -%c\n",
pname, *option_letter_p);
errors = -1;
}
}
}
}
/* Now actually make a list of the base source pathnames. */
base_source_paths =
(const char **) xmalloc ((base_source_files + 1) * sizeof (char *));
base_source_files = 0;
argc--;
for (argn = argv+1; argc; argn++, argc--)
{
if (*argn && **argn != '-')
{
const char *path = abspath (NULL, *argn);
int len = strlen (path);
if (path[len-1] == 'c' && path[len-2] == '.')
base_source_paths[base_source_files++] = path;
else
{
fprintf (stderr, "%s: fatal error: input pathnames must have .c suffixes: %s\n",
pname, shortpath (NULL, path));
errors++;
}
}
}
if (errors)
usage ();
else
{
if (version_flag)
fprintf (stderr, "%s: %s\n", pname, version_string);
do_processing ();
}
if (errors)
exit (1);
else
exit (0);
return 1;
}