blob: 3b50836774abc2f9748da0ae2e1d1a2bb3a74391 [file] [log] [blame]
/* Protoize program - Original version by Ron Guilmette (rfg@segfault.us.com).
Copyright (C) 1989, 92-96, 1997 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, 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA. */
/* Any reasonable C++ compiler should have all of the same features
as __STDC__ plus more, so make sure that __STDC__ is defined if
__cplusplus is defined. */
#if defined(__cplusplus) && !defined(__STDC__)
#define __STDC__ 1
#endif /* defined(__cplusplus) && !defined(__STDC__) */
#if defined(__GNUC__) || defined (__GNUG__)
#define VOLATILE volatile
#else
#define VOLATILE
#endif
#ifndef __STDC__
#define const
#define volatile
#endif
#include "config.h"
#if 0
/* Users are not supposed to use _POSIX_SOURCE to say the
system is a POSIX system. That is not what _POSIX_SOURCE means! -- rms */
/* If the user asked for POSIX via _POSIX_SOURCE, turn on POSIX code. */
#if defined(_POSIX_SOURCE) && !defined(POSIX)
#define POSIX
#endif
#endif /* 0 */
#ifdef POSIX /* We should be able to define _POSIX_SOURCE unconditionally,
but some systems respond in buggy ways to it,
including SunOS 4.1.1. Which we don't classify as POSIX. */
/* In case this is a POSIX system with an ANSI C compiler,
ask for definition of all POSIX facilities. */
#undef _POSIX_SOURCE
#define _POSIX_SOURCE
#endif
#ifdef HAVE_VARARGS_H
#include <varargs.h>
#else
#ifdef HAVE_SYS_VARARGS_H
#include <sys/varargs.h>
#endif
#endif
/* On some systems stdio.h includes stdarg.h;
we must bring in varargs.h first. */
#include <stdio.h>
#include <ctype.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/stat.h>
#if ! defined (_WIN32) || defined (__CYGWIN32__)
#if defined(POSIX) || defined(CONCURRENT)
#include <dirent.h>
#else
#include <sys/dir.h>
#endif
#endif
#include <setjmp.h>
#include "gansidecl.h"
/* Include getopt.h for the sake of getopt_long.
We don't need the declaration of getopt, and it could conflict
with something from a system header file, so effectively nullify that. */
#define getopt getopt_loser
#include "getopt.h"
#undef getopt
#ifndef errno
extern int errno;
#endif
#ifndef HAVE_STRERROR
extern int sys_nerr;
extern char *sys_errlist[];
#else
extern char *strerror();
#endif
extern char *version_string;
/* Systems which are compatible only with POSIX 1003.1-1988 (but *not*
with POSIX 1003.1-1990), e.g. Ultrix 4.2, might not have
const qualifiers in the prototypes in the system include files.
Unfortunately, this can lead to GCC issuing lots of warnings for
calls to the following functions. To eliminate these warnings we
provide the following #defines. */
#define my_access(file,flag) access((char *)file, flag)
#define my_stat(file,pkt) stat((char *)file, pkt)
#define my_link(file1, file2) link((char *)file1, (char *)file2)
#define my_unlink(file) unlink((char *)file)
#define my_open(file, mode, flag) open((char *)file, mode, flag)
#define my_chmod(file, mode) chmod((char *)file, mode)
extern char *getpwd ();
extern char *choose_temp_base PROTO ((void));
extern int pexecute PROTO ((const char *, char * const *, const char *,
const char *, char **, char **, int));
extern int pwait PROTO ((int, int *, int));
/* Flag arguments to pexecute. */
#define PEXECUTE_FIRST 1
#define PEXECUTE_LAST 2
#define PEXECUTE_SEARCH 4
/* Aliases for pointers to void.
These were made to facilitate compilation with old brain-dead DEC C
compilers which didn't properly grok `void*' types. */
#ifdef __STDC__
typedef void * pointer_type;
typedef const void * const_pointer_type;
#else
typedef char * pointer_type;
typedef char * const_pointer_type;
#endif
#if defined(POSIX)
#include <stdlib.h>
#include <unistd.h>
#include <signal.h>
#include <fcntl.h>
#include <sys/wait.h>
#else /* !defined(POSIX) */
#define R_OK 4 /* Test for Read permission */
#define W_OK 2 /* Test for Write permission */
#define X_OK 1 /* Test for eXecute permission */
#define F_OK 0 /* Test for existence of File */
#ifndef O_RDONLY
#define O_RDONLY 0
#endif
#ifndef O_WRONLY
#define O_WRONLY 1
#endif
#ifndef WIFSIGNALED
#define WIFSIGNALED(S) (((S) & 0xff) != 0 && ((S) & 0xff) != 0x7f)
#endif
#ifndef WTERMSIG
#define WTERMSIG(S) ((S) & 0x7f)
#endif
#ifndef WIFEXITED
#define WIFEXITED(S) (((S) & 0xff) == 0)
#endif
#ifndef WEXITSTATUS
#define WEXITSTATUS(S) (((S) & 0xff00) >> 8)
#endif
/* Declaring stat or __flsbuf with a prototype
causes conflicts with system headers on some systems. */
#ifndef abort
typedef void voidfn ();
extern VOLATILE voidfn abort;
#endif
extern int creat ();
#if 0 /* These conflict with stdio.h on some systems. */
extern int fprintf (FILE *, const char *, ...);
extern int printf (const char *, ...);
extern int open (const char *, int, ...);
#endif /* 0 */
extern void exit ();
extern void free ();
extern int read ();
extern int write ();
extern int close ();
extern int fflush ();
extern int atoi ();
extern int puts ();
extern int fputs ();
extern int fputc ();
#if !defined(_WIN32)
extern int link ();
#endif
extern int unlink ();
extern int access ();
#if 0 /* size_t from sys/types.h may fail to match GCC.
If so, we would get a warning from this. */
extern size_t strlen ()
#endif
/* Fork is not declared because the declaration caused a conflict
on the HPPA. */
#if !(defined (USG) || defined (VMS))
#define fork vfork
#endif /* (defined (USG) || defined (VMS)) */
#endif /* !defined (POSIX) */
extern char *rindex ();
/* Look for these where the `const' qualifier is intentionally cast aside. */
#define NONCONST
/* Define a STRINGIFY macro that's right for ANSI or traditional C. */
#ifdef __STDC__
#define STRINGIFY(STRING) #STRING
#else
#define STRINGIFY(STRING) "STRING"
#endif
/* Define a default place to find the SYSCALLS.X file. */
#ifndef STD_PROTO_DIR
#define STD_PROTO_DIR "/usr/local/lib"
#endif /* !defined (STD_PROTO_DIR) */
/* Suffix of aux_info files. */
static const char * const aux_info_suffix = ".X";
/* String to attach to filenames 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 syscalls_filename[] = "SYSCALLS.c";
/* Default place to find the above file. */
static const char * const default_syscalls_dir = STD_PROTO_DIR;
/* Variable to hold the complete absolutized filename of the SYSCALLS.c.X
file. */
static char * syscalls_absolute_filename;
#endif /* !defined (UNPROTOIZE) */
/* 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 }
};
/* 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);
/* Make a table of default system include directories
just as it is done in cccp.c. */
#ifndef STANDARD_INCLUDE_DIR
#define STANDARD_INCLUDE_DIR "/usr/include"
#endif
#ifndef LOCAL_INCLUDE_DIR
#define LOCAL_INCLUDE_DIR "/usr/local/include"
#endif
struct default_include { const char *fname; int x1, x2; } include_defaults[]
#ifdef INCLUDE_DEFAULTS
= INCLUDE_DEFAULTS;
#else
= {
/* Pick up GNU C++ specific include files. */
{ GPLUSPLUS_INCLUDE_DIR, 1, 1 },
#ifdef CROSS_COMPILE
/* This is the dir for fixincludes. Put it just before
the files that we fix. */
{ GCC_INCLUDE_DIR, 0, 0 },
/* For cross-compilation, this dir name is generated
automatically in Makefile.in. */
{ CROSS_INCLUDE_DIR, 0, 0 },
/* This is another place that the target system's headers might be. */
{ TOOL_INCLUDE_DIR, 0, 0 },
#else /* not CROSS_COMPILE */
/* This should be /use/local/include and should come before
the fixincludes-fixed header files. */
{ LOCAL_INCLUDE_DIR, 0, 1 },
/* This is here ahead of GCC_INCLUDE_DIR because assert.h goes here.
Likewise, behind LOCAL_INCLUDE_DIR, where glibc puts its assert.h. */
{ TOOL_INCLUDE_DIR, 0, 0 },
/* This is the dir for fixincludes. Put it just before
the files that we fix. */
{ GCC_INCLUDE_DIR, 0, 0 },
/* Some systems have an extra dir of include files. */
#ifdef SYSTEM_INCLUDE_DIR
{ SYSTEM_INCLUDE_DIR, 0, 0 },
#endif
{ STANDARD_INCLUDE_DIR, 0, 0},
#endif /* not CROSS_COMPILE */
{ 0, 0, 0}
};
#endif /* no INCLUDE_DEFAULTS */
/* Datatype for lists of directories or filenames. */
struct string_list
{
char *name;
struct string_list *next;
};
/* List of directories in which files should be converted. */
struct string_list *directory_list;
/* List of file names which should not be converted.
A file is excluded if the end of its name, following a /,
matches one of the names in this list. */
struct string_list *exclude_list;
/* 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 /* !defined (UNPROTOIZE) */
static const char * const other_var_style = "varargs";
/* Note that this is a string containing the expansion of va_alist.
But in `main' we discard all but the first token. */
static const char *varargs_style_indicator = STRINGIFY (va_alist);
#endif /* !defined (UNPROTOIZE) */
/* 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 filenames 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 filenames
hash table or the function names hash table). In the filenames hash table
the info fields of the entries point to the file_info struct which is
associated with each filename (1 per filename). 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 filename (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 visibility is "extern" */
char is_implicit; /* != 0 for implicit func decl's */
char written; /* != 0 means written for implicit */
#else /* !defined (UNPROTOIZE) */
const char * formal_names; /* -> to list of names of formals */
const char * formal_decls; /* -> to string of formal declarations */
#endif /* !defined (UNPROTOIZE) */
};
/* Pointer to the tail component of the filename 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. */
/* ??? These comments should say what the flag mean as well as the options
that set them. */
/* File name to use for running gcc. Allows GCC 2 to be named
something other than gcc. */
static const char *compiler_file_name = "gcc";
static int version_flag = 0; /* Print our version number. */
static int quiet_flag = 0; /* Don't print messages normally. */
static int nochange_flag = 0; /* Don't convert, just say what files
we would have converted. */
static int nosave_flag = 0; /* Don't save the old version. */
static int keep_flag = 0; /* Don't delete the .X files. */
static const char ** compile_params = 0; /* Option string for gcc. */
#ifdef UNPROTOIZE
static const char *indent_string = " "; /* Indentation for newly
inserted parm decls. */
#else /* !defined (UNPROTOIZE) */
static int local_flag = 0; /* Insert new local decls (when?). */
static int global_flag = 0; /* set by -g option */
static int cplusplus_flag = 0; /* Rename converted files to *.C. */
static const char *nondefault_syscalls_dir = 0; /* Dir to look for
SYSCALLS.c.X in. */
#endif /* !defined (UNPROTOIZE) */
/* An index into the compile_params array where we should insert the source
file name when we are ready to exec the C compiler. A zero value indicates
that we have not yet called munge_compile_params. */
static int input_file_name_index = 0;
/* An index into the compile_params array where we should insert the filename
for the aux info file, when we run the C compiler. */
static int aux_info_file_name_index = 0;
/* Count of command line arguments which were "filename" arguments. */
static int n_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_filenames;
/* 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_filename;
/* 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_filename;
/* 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 filenames hash table. */
static hash_table filename_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;
/* A pointer to the current directory filename (used by abspath). */
static char *cwd_buffer;
/* 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 ();
char *
my_strerror(e)
int e;
{
#ifdef HAVE_STRERROR
return strerror(e);
#else
static char buffer[30];
if (!e)
return "";
if (e > 0 && e < sys_nerr)
return sys_errlist[e];
sprintf (buffer, "Unknown error %d", e);
return buffer;
#endif
}
/* Allocate some space, but check that the allocation was successful. */
/* alloca.c uses this, so don't make it static. */
pointer_type
xmalloc (byte_count)
size_t byte_count;
{
pointer_type rv;
rv = (pointer_type) malloc (byte_count);
if (rv == NULL)
{
fprintf (stderr, "\n%s: virtual memory exceeded\n", pname);
exit (FATAL_EXIT_CODE);
return 0; /* avoid warnings */
}
else
return rv;
}
/* Reallocate some space, but check that the reallocation was successful. */
pointer_type
xrealloc (old_space, byte_count)
pointer_type old_space;
size_t byte_count;
{
pointer_type rv;
rv = (pointer_type) realloc (old_space, byte_count);
if (rv == NULL)
{
fprintf (stderr, "\n%s: virtual memory exceeded\n", pname);
exit (FATAL_EXIT_CODE);
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. */
void
xfree (p)
const_pointer_type p;
{
if (p)
free ((NONCONST pointer_type) p);
}
/* Make a copy of a string INPUT with size SIZE. */
static char *
savestring (input, size)
const char *input;
unsigned int size;
{
char *output = (char *) xmalloc (size + 1);
strcpy (output, input);
return output;
}
/* Make a copy of the concatenation of INPUT1 and INPUT2. */
static char *
savestring2 (input1, size1, input2, size2)
const char *input1;
unsigned int size1;
const char *input2;
unsigned int size2;
{
char *output = (char *) xmalloc (size1 + size2 + 1);
strcpy (output, input1);
strcpy (&output[size1], input2);
return output;
}
/* 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 (FATAL_EXIT_CODE);
}
/* Make a duplicate of the first N bytes of a given string in a newly
allocated area. */
static char *
dupnstr (s, n)
const char *s;
size_t n;
{
char *ret_val = (char *) xmalloc (n + 1);
strncpy (ret_val, s, n);
ret_val[n] = '\0';
return ret_val;
}
/* Return a pointer to the first occurrence 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 (s1, s2)
const char *s1;
const char *const s2;
{
for (; *s1 ; s1++)
{
const char *p1;
const char *p2;
int c;
for (p1 = s1, p2 = s2; c = *p2; p1++, p2++)
if (*p1 != c)
goto outer;
return s1;
outer:
;
}
return 0;
}
/* Read LEN bytes at PTR from descriptor DESC, for file FILENAME,
retrying if necessary. Return the actual number of bytes read. */
static int
safe_read (desc, ptr, len)
int desc;
char *ptr;
int len;
{
int left = len;
while (left > 0) {
int nchars = read (desc, ptr, left);
if (nchars < 0)
{
#ifdef EINTR
if (errno == EINTR)
continue;
#endif
return nchars;
}
if (nchars == 0)
break;
ptr += nchars;
left -= nchars;
}
return len - left;
}
/* Write LEN bytes at PTR to descriptor DESC,
retrying if necessary, and treating any real error as fatal. */
static void
safe_write (desc, ptr, len, out_fname)
int desc;
char *ptr;
int len;
char *out_fname;
{
while (len > 0) {
int written = write (desc, ptr, len);
if (written < 0)
{
#ifdef EINTR
if (errno == EINTR)
continue;
#endif
fprintf (stderr, "%s: error writing file `%s': %s\n",
pname, shortpath (NULL, out_fname), my_strerror(errno));
return;
}
ptr += written;
len -= written;
}
}
/* Get setup to recover in case the edit we are about to do goes awry. */
void
save_pointers ()
{
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 ()
{
clean_read_ptr = saved_clean_read_ptr;
repl_write_ptr = saved_repl_write_ptr;
}
/* Return true if the given character is a valid identifier character. */
static int
is_id_char (ch)
char ch;
{
return (isalnum (ch) || (ch == '_') || (ch == '$'));
}
/* Give a message indicating the proper way to invoke this program and then
exit with non-zero status. */
static void
usage ()
{
#ifdef UNPROTOIZE
fprintf (stderr, "%s: usage '%s [ -VqfnkN ] [ -i <istring> ] [ filename ... ]'\n",
pname, pname);
#else /* !defined (UNPROTOIZE) */
fprintf (stderr, "%s: usage '%s [ -VqfnkNlgC ] [ -B <dirname> ] [ filename ... ]'\n",
pname, pname);
#endif /* !defined (UNPROTOIZE) */
exit (FATAL_EXIT_CODE);
}
/* Return true if the given filename (assumed to be an absolute filename)
designates a file residing anywhere beneath any one of the "system"
include directories. */
static int
in_system_include_dir (path)
const char *path;
{
struct default_include *p;
if (path[0] != '/')
abort (); /* Must be an absolutized filename. */
for (p = include_defaults; p->fname; p++)
if (!strncmp (path, p->fname, strlen (p->fname))
&& path[strlen (p->fname)] == '/')
return 1;
return 0;
}
#if 0
/* Return true if the given filename 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 *const dir_name = (char *) alloca (strlen (path) + 1);
if (my_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 filename. */
}
if (my_access (path, W_OK))
return 0;
return 1;
}
/* Return true if the given filename 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,
and (c) files which reside in directories which the user doesn't have
write access to. 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_convertible (const char *path)
{
char *const dir_name = alloca (strlen (path) + 1);
if (in_system_include_dir (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 filename. */
}
if (my_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 (my_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 (my_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;
}
#endif /* 0 */
#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 (fi_p)
const file_info *fi_p;
{
char const *f = fi_p->hash_entry->symbol;
size_t fl = strlen (f), sysl = sizeof (syscalls_filename) - 1;
return sysl <= fl && strcmp (f + fl - sysl, syscalls_filename) == 0;
}
#endif /* !defined (UNPROTOIZE) */
/* 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 (file_p)
const file_info *file_p;
{
const def_dec_info *ddp;
#ifndef UNPROTOIZE
if (is_syscalls_file (file_p))
return 0;
#endif /* !defined (UNPROTOIZE) */
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 /* defined (UNPROTOIZE) */
/* ... and if we are unprotoizing and this function is in new style ... */
ddp->prototyped
#endif /* defined (UNPROTOIZE) */
)
/* ... then the containing file needs converting. */
return -1;
return 0;
}
/* Return 1 if the file name NAME is in a directory
that should be converted. */
static int
directory_specified_p (name)
const char *name;
{
struct string_list *p;
for (p = directory_list; p; p = p->next)
if (!strncmp (name, p->name, strlen (p->name))
&& name[strlen (p->name)] == '/')
{
const char *q = name + strlen (p->name) + 1;
/* If there are more slashes, it's in a subdir, so
this match doesn't count. */
while (*q)
if (*q++ == '/')
goto lose;
return 1;
lose: ;
}
return 0;
}
/* Return 1 if the file named NAME should be excluded from conversion. */
static int
file_excluded_p (name)
const char *name;
{
struct string_list *p;
int len = strlen (name);
for (p = exclude_list; p; p = p->next)
if (!strcmp (name + len - strlen (p->name), p->name)
&& name[len - strlen (p->name) - 1] == '/')
return 1;
return 0;
}
/* Construct a new element of a string_list.
STRING is the new element value, and REST holds the remaining elements. */
static struct string_list *
string_list_cons (string, rest)
char *string;
struct string_list *rest;
{
struct string_list *temp
= (struct string_list *) xmalloc (sizeof (struct string_list));
temp->next = rest;
temp->name = string;
return temp;
}
/* ??? The GNU convention for mentioning function args in its comments
is to capitalize them. So change "hash_tab_p" to HASH_TAB_P below.
Likewise for all the other functions. */
/* 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
visit_each_hash_node (hash_tab_p, func)
const hash_table_entry *hash_tab_p;
void (*func)();
{
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 (p, s)
hash_table_entry *p;
const char *s;
{
p->hash_next = NULL;
p->symbol = savestring (s, strlen (s));
p->ddip = NULL;
p->fip = NULL;
return p;
}
/* Look for a particular function name or filename 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_tab_p, search_symbol)
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. */
static void
free_def_dec (p)
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 /* !defined (UNPROTOIZE) */
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 (aux_info_line)
const char *aux_info_line;
{
static char *line_buf = 0;
static int line_buf_size = 0;
const unexpansion *unexp_p;
int got_unexpanded = 0;
const char *s;
char *copy_p = line_buf;
if (line_buf == 0)
{
line_buf_size = 1024;
line_buf = (char *) xmalloc (line_buf_size);
}
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;
size_t len = strlen (in_p);
if (*s == *in_p && !strncmp (s, in_p, len) && !is_id_char (s[len]))
{
int size = strlen (unexp_p->contracted);
got_unexpanded = 1;
if (copy_p + size - line_buf >= line_buf_size)
{
int offset = copy_p - line_buf;
line_buf_size *= 2;
line_buf_size += size;
line_buf = (char *) xrealloc (line_buf, line_buf_size);
copy_p = line_buf + offset;
}
strcpy (copy_p, unexp_p->contracted);
copy_p += size;
/* Assume the there will not be another replacement required
within the text just replaced. */
s += len;
goto continue_outer;
}
}
if (copy_p - line_buf == line_buf_size)
{
int offset = copy_p - line_buf;
line_buf_size *= 2;
line_buf = (char *) xrealloc (line_buf, line_buf_size);
copy_p = line_buf + offset;
}
*copy_p++ = *s++;
continue_outer: ;
}
if (copy_p + 2 - line_buf >= line_buf_size)
{
int offset = copy_p - line_buf;
line_buf_size *= 2;
line_buf = (char *) xrealloc (line_buf, line_buf_size);
copy_p = line_buf + offset;
}
*copy_p++ = '\n';
*copy_p = '\0';
return (got_unexpanded ? savestring (line_buf, copy_p - line_buf) : 0);
}
/* Return the absolutized filename for the given relative
filename. Note that if that filename 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 filename from the given input
filename. The absolutization of relative filenames is made by
assuming that the given filename is to be taken as relative to
the first argument (cwd) or to the current directory if cwd is
NULL. */
static char *
abspath (cwd, rel_filename)
const char *cwd;
const char *rel_filename;
{
/* Setup the current working directory as needed. */
const char *cwd2 = (cwd) ? cwd : cwd_buffer;
char *const abs_buffer
= (char *) alloca (strlen (cwd2) + strlen (rel_filename) + 2);
char *endp = abs_buffer;
char *outp, *inp;
/* Copy the filename (possibly preceded by the current working
directory name) into the absolutization buffer. */
{
const char *src_p;
if (rel_filename[0] != '/')
{
src_p = cwd2;
while (*endp++ = *src_p++)
continue;
*(endp-1) = '/'; /* overwrite null */
}
src_p = rel_filename;
while (*endp++ = *src_p++)
continue;
}
/* Now make a copy of abs_buffer into abs_buffer, shortening the
filename (by taking out slashes and dots) as we go. */
outp = inp = abs_buffer;
*outp++ = *inp++; /* copy first slash */
#ifdef apollo
if (inp[0] == '/')
*outp++ = *inp++; /* copy second slash */
#endif
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: invalid file name: %s\n",
pname, rel_filename);
exit (FATAL_EXIT_CODE);
}
*++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 savestring (abs_buffer, outp - abs_buffer);
}
/* Given a filename (and possibly a directory name from which the filename
is relative) return a string which is the shortest possible
equivalent for the corresponding full (absolutized) filename. The
shortest possible equivalent may be constructed by converting the
absolutized filename to be a relative filename (i.e. relative to
the actual current working directory). However if a relative filename
is longer, then the full absolute filename is returned.
KNOWN BUG:
Note that "simple-minded" conversion of any given type of filename (either
relative or absolute) may not result in a valid equivalent filename if any
subpart of the original filename is actually a symbolic link. */
static const char *
shortpath (cwd, filename)
const char *cwd;
const char *filename;
{
char *rel_buffer;
char *rel_buf_p;
char *cwd_p = cwd_buffer;
char *path_p;
int unmatched_slash_count = 0;
size_t filename_len = strlen (filename);
path_p = abspath (cwd, filename);
rel_buf_p = rel_buffer = (char *) xmalloc (filename_len);
while (*cwd_p && (*cwd_p == *path_p))
{
cwd_p++;
path_p++;
}
if (!*cwd_p && (!*path_p || *path_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++;
}
/* Find out how many directory levels in cwd were *not* matched. */
while (*cwd_p)
if (*cwd_p++ == '/')
unmatched_slash_count++;
/* Now we know how long the "short name" will be.
Reject it if longer than the input. */
if (unmatched_slash_count * 3 + strlen (path_p) >= filename_len)
return filename;
/* For each of them, put a `../' at the beginning of the short name. */
while (unmatched_slash_count--)
{
/* Give up if the result gets to be longer
than the absolute path name. */
if (rel_buffer + filename_len <= rel_buf_p + 3)
return filename;
*rel_buf_p++ = '.';
*rel_buf_p++ = '.';
*rel_buf_p++ = '/';
}
/* Then tack on the unmatched part of the desired file's name. */
do
{
if (rel_buffer + filename_len <= rel_buf_p)
return filename;
}
while (*rel_buf_p++ = *path_p++);
--rel_buf_p;
if (*(rel_buf_p-1) == '/')
*--rel_buf_p = '\0';
return rel_buffer;
}
}
/* Lookup the given filename in the hash table for filenames. 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 filename, and we initialize
that record with some reasonable values. */
/* FILENAME was const, but that causes a warning on AIX when calling stat.
That is probably a bug in AIX, but might as well avoid the warning. */
static file_info *
find_file (filename, do_not_stat)
char *filename;
int do_not_stat;
{
hash_table_entry *hash_entry_p;
hash_entry_p = lookup (filename_primary, filename);
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 (my_stat (filename, &stat_buf) == -1)
{
fprintf (stderr, "%s: %s: can't get status: %s\n",
pname, shortpath (NULL, filename), my_strerror(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
aux_info_corrupted ()
{
fprintf (stderr, "\n%s: fatal error: aux info file corrupted at line %d\n",
pname, current_aux_info_lineno);
exit (FATAL_EXIT_CODE);
}
/* ??? This comment is vague. Say what the condition is for. */
/* Check to see that a condition is true. This is kind of like an assert. */
static void
check_aux_info (cond)
int cond;
{
if (! cond)
aux_info_corrupted ();
}
/* Given a pointer to the closing right parenthesis for a particular formals
list (in an aux_info file) find the corresponding left parenthesis and
return a pointer to it. */
static const char *
find_corresponding_lparen (p)
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 (l, aux_info_mtime)
const char *l;
time_t aux_info_mtime;
{
const char *p;
file_info *fi_p;
char *filename;
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++;
filename = (char *) alloca ((size_t) (p - filename_start) + 1);
strncpy (filename, filename_start, (size_t) (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_filename, 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 occurred 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
save_def_or_dec (l, is_syscalls)
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 /* !defined (UNPROTOIZE) */
/* 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;
while (*p != ':')
p++;
filename = (char *) alloca ((size_t) (p - filename_start) + 1);
strncpy (filename, filename_start, (size_t) (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 names
(i.e. the names of the aux_info files with the .X stripped off) into the
filenames hash table, and we simultaneously setup file_info records for
all of these base file names (even if they may be useless later).
The file_info records for all of these "base" file names (properly)
act as file_info records for the "original" (i.e. un-included) files
which were submitted to gcc for compilation (when the -aux-info
option was used). */
def_dec_p->file = find_file (abspath (invocation_filename, filename), is_syscalls);
}
{
const char *line_number_start = ++p;
char line_number[10];
while (*p != ':')
p++;
strncpy (line_number, line_number_start, (size_t) (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 /* !defined (UNPROTOIZE) */
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 /* !defined (UNPROTOIZE) */
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 /* !defined (UNPROTOIZE) */
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 /* !defined (UNPROTOIZE) */
{
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, (size_t) ((semicolon_p+1) - ansi_start));
/* Backup and point at the final right paren of the final argument list. */
p--;
#ifndef UNPROTOIZE
def_dec_p->f_list_chain = NULL;
#endif /* !defined (UNPROTOIZE) */
while (p != ansi_start && (p[-1] == ' ' || p[-1] == '\t')) p--;
if (*p != ')')
{
free_def_dec (def_dec_p);
return;
}
}
/* 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;
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, (size_t) (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 /* !defined (UNPROTOIZE) */
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 preceding this right paren. */
if (*p != ')')
break;
else
check_aux_info (*--p == ')');
}
{
const char *past_fn = p + 1;
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. */
{
char *fn_string = (char *) alloca (past_fn - p + 1);
strncpy (fn_string, p, (size_t) (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:%d: declaration of function `%s' takes different forms\n",
def_dec_p->file->hash_entry->symbol,
def_dec_p->line,
def_dec_p->hash_entry->symbol);
exit (FATAL_EXIT_CODE);
}
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, (size_t) (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, (size_t) (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 /* defined (UNPROTOIZE) */
/* 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;
}
}
}
/* Set up the vector COMPILE_PARAMS which is the argument list for running GCC.
Also set input_file_name_index and aux_info_file_name_index
to the indices of the slots where the file names should go. */
/* We initialize the vector by removing -g, -O, -S, -c, and -o options,
and adding '-aux-info AUXFILE -S -o /dev/null INFILE' at the end. */
static void
munge_compile_params (params_list)
const char *params_list;
{
/* Build up the contents in a temporary vector
that is so big that to has to be big enough. */
const char **temp_params
= (const char **) alloca ((strlen (params_list) + 8) * sizeof (char *));
int param_count = 0;
const char *param;
temp_params[param_count++] = compiler_file_name;
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, (size_t) (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, (size_t) (params_list - param));
}
}
if (!*params_list)
break;
}
temp_params[param_count++] = "-aux-info";
/* Leave room for the aux-info file name argument. */
aux_info_file_name_index = param_count;
temp_params[param_count++] = NULL;
temp_params[param_count++] = "-S";
temp_params[param_count++] = "-o";
temp_params[param_count++] = "/dev/null";
/* Leave room for the input file name argument. */
input_file_name_index = param_count;
temp_params[param_count++] = NULL;
/* Terminate the list. */
temp_params[param_count++] = NULL;
/* Make a copy of the compile_params in heap space. */
compile_params
= (const char **) xmalloc (sizeof (char *) * (param_count+1));
memcpy (compile_params, temp_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.
The result is a boolean indicating success. */
static int
gen_aux_info_file (base_filename)
const char *base_filename;
{
if (!input_file_name_index)
munge_compile_params ("");
/* Store the full source file name in the argument vector. */
compile_params[input_file_name_index] = shortpath (NULL, base_filename);
/* Add .X to source file name to get aux-info file name. */
compile_params[aux_info_file_name_index]
= savestring2 (compile_params[input_file_name_index],
strlen (compile_params[input_file_name_index]),
".X",
2);
if (!quiet_flag)
fprintf (stderr, "%s: compiling `%s'\n",
pname, compile_params[input_file_name_index]);
{
char *errmsg_fmt, *errmsg_arg;
int wait_status, pid;
char *temp_base = choose_temp_base ();
pid = pexecute (compile_params[0], (char * const *) compile_params,
pname, temp_base, &errmsg_fmt, &errmsg_arg,
PEXECUTE_FIRST | PEXECUTE_LAST | PEXECUTE_SEARCH);
if (pid == -1)
{
int errno_val = errno;
fprintf (stderr, "%s: ", pname);
fprintf (stderr, errmsg_fmt, errmsg_arg);
fprintf (stderr, ": %s\n", my_strerror (errno_val));
return 0;
}
pid = pwait (pid, &wait_status, 0);
if (pid == -1)
{
fprintf (stderr, "%s: wait: %s\n", pname, my_strerror (errno));
return 0;
}
if (WIFSIGNALED (wait_status))
{
fprintf (stderr, "%s: subprocess got fatal signal %d\n",
pname, WTERMSIG (wait_status));
return 0;
}
if (WIFEXITED (wait_status))
{
if (WEXITSTATUS (wait_status) != 0)
{
fprintf (stderr, "%s: %s exited with status %d\n",
pname, compile_params[0], WEXITSTATUS (wait_status));
return 0;
}
return 1;
}
abort ();
}
}
/* Read in all of the information contained in a single aux_info file.
Save all of the important stuff for later. */
static void
process_aux_info_file (base_source_filename, keep_it, is_syscalls)
const char *base_source_filename;
int keep_it;
int is_syscalls;
{
size_t base_len = strlen (base_source_filename);
char * aux_info_filename
= (char *) alloca (base_len + strlen (aux_info_suffix) + 1);
char *aux_info_base;
char *aux_info_limit;
char *aux_info_relocated_name;
const char *aux_info_second_line;
time_t aux_info_mtime;
size_t aux_info_size;
int must_create;
/* Construct the aux_info filename from the base source filename. */
strcpy (aux_info_filename, base_source_filename);
strcat (aux_info_filename, aux_info_suffix);
/* Check that the aux_info file exists and is readable. If it does not
exist, try to create it (once only). */
/* If file doesn't exist, set must_create.
Likewise if it exists and we can read it but it is obsolete.
Otherwise, report an error. */
must_create = 0;
/* Come here with must_create set to 1 if file is out of date. */
start_over: ;
if (my_access (aux_info_filename, R_OK) == -1)
{
if (errno == ENOENT)
{
if (is_syscalls)
{
fprintf (stderr, "%s: warning: missing SYSCALLS file `%s'\n",
pname, aux_info_filename);
return;
}
must_create = 1;
}
else
{
fprintf (stderr, "%s: can't read aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(errno));
errors++;
return;
}
}
#if 0 /* There is code farther down to take care of this. */
else
{
struct stat s1, s2;
stat (aux_info_file_name, &s1);
stat (base_source_file_name, &s2);
if (s2.st_mtime > s1.st_mtime)
must_create = 1;
}
#endif /* 0 */
/* If we need a .X file, create it, and verify we can read it. */
if (must_create)
{
if (!gen_aux_info_file (base_source_filename))
{
errors++;
return;
}
if (my_access (aux_info_filename, R_OK) == -1)
{
fprintf (stderr, "%s: can't read aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(errno));
errors++;
return;
}
}
{
struct stat stat_buf;
/* Get some status information about this aux_info file. */
if (my_stat (aux_info_filename, &stat_buf) == -1)
{
fprintf (stderr, "%s: can't get status of aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(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;
if (!is_syscalls)
{
/* Compare mod time with the .c file; update .X file if obsolete.
The code later on can fail to check the .c file
if it did not directly define any functions. */
if (my_stat (base_source_filename, &stat_buf) == -1)
{
fprintf (stderr, "%s: can't get status of aux info file `%s': %s\n",
pname, shortpath (NULL, base_source_filename),
my_strerror(errno));
errors++;
return;
}
if (stat_buf.st_mtime > aux_info_mtime)
{
must_create = 1;
goto start_over;
}
}
}
{
int aux_info_file;
/* Open the aux_info file. */
if ((aux_info_file = my_open (aux_info_filename, O_RDONLY, 0444 )) == -1)
{
fprintf (stderr, "%s: can't open aux info file `%s' for reading: %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(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 (safe_read (aux_info_file, aux_info_base, aux_info_size) != aux_info_size)
{
fprintf (stderr, "%s: error reading aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(errno));
free (aux_info_base);
close (aux_info_file);
return;
}
/* Close the aux info file. */
if (close (aux_info_file))
{
fprintf (stderr, "%s: error closing aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(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 (must_create && !keep_it)
if (my_unlink (aux_info_filename) == -1)
fprintf (stderr, "%s: can't delete aux info file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(errno));
/* Save a pointer into the first line of the aux_info file which
contains the filename 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
filenames out of the (potentially) relative filenames in
the aux_info file. */
{
char *p = aux_info_base;
while (*p != ':')
p++;
p++;
while (*p == ' ')
p++;
invocation_filename = p; /* Save a pointer to first byte of path. */
while (*p != ' ')
p++;
*p++ = '/';
*p++ = '\0';
while (*p++ != '\n')
continue;
aux_info_second_line = p;
aux_info_relocated_name = 0;
if (invocation_filename[0] != '/')
{
/* INVOCATION_FILENAME is relative;
append it to BASE_SOURCE_FILENAME's dir. */
char *dir_end;
aux_info_relocated_name = xmalloc (base_len + (p-invocation_filename));
strcpy (aux_info_relocated_name, base_source_filename);
dir_end = rindex (aux_info_relocated_name, '/');
if (dir_end)
dir_end++;
else
dir_end = aux_info_relocated_name;
strcpy (dir_end, invocation_filename);
invocation_filename = aux_info_relocated_name;
}
}
{
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);
xfree (aux_info_relocated_name);
if (keep_it && my_unlink (aux_info_filename) == -1)
{
fprintf (stderr, "%s: can't delete file `%s': %s\n",
pname, shortpath (NULL, aux_info_filename),
my_strerror(errno));
return;
}
must_create = 1;
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);
xfree (aux_info_relocated_name);
}
#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
rename_c_file (hp)
const hash_table_entry *hp;
{
const char *filename = hp->symbol;
int last_char_index = strlen (filename) - 1;
char *const new_filename = (char *) alloca (strlen (filename) + 1);
/* 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 (filename[last_char_index] != 'c' || filename[last_char_index-1] != '.')
return;
strcpy (new_filename, filename);
new_filename[last_char_index] = 'C';
if (my_link (filename, new_filename) == -1)
{
fprintf (stderr, "%s: warning: can't link file `%s' to `%s': %s\n",
pname, shortpath (NULL, filename),
shortpath (NULL, new_filename), my_strerror(errno));
errors++;
return;
}
if (my_unlink (filename) == -1)
{
fprintf (stderr, "%s: warning: can't delete file `%s': %s\n",
pname, shortpath (NULL, filename), my_strerror(errno));
errors++;
return;
}
}
#endif /* !defined (UNPROTOIZE) */
/* 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
reverse_def_dec_list (hp)
const hash_table_entry *hp;
{
file_info *file_p = hp->fip;
def_dec_info *prev = NULL;
def_dec_info *current = (def_dec_info *)file_p->defs_decs;
if (!current)
return; /* no list to reverse */
prev = current;
if (! (current = (def_dec_info *)current->next_in_file))
return; /* can't reverse a single list element */
prev->next_in_file = NULL;
while (current)
{
def_dec_info *next = (def_dec_info *)current->next_in_file;
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 (head, user)
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: 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))
{
/* Why copy this string into `needed' at all?
Why not just use user->ansi_decl without copying? */
char *needed = (char *) alloca (strlen (user->ansi_decl) + 1);
char *p;
strcpy (needed, user->ansi_decl);
p = (NONCONST char *) substr (needed, user->hash_entry->symbol)
+ strlen (user->hash_entry->symbol) + 2;
/* Avoid having ??? in the string. */
*p++ = '?';
*p++ = '?';
*p++ = '?';
strcpy (p, ");");
fprintf (stderr, "%s: %d: `%s' used but missing from SYSCALLS\n",
shortpath (NULL, file), user->line,
needed+7); /* Don't print "extern " */
}
#if 0
else
fprintf (stderr, "%s: %d: warning: no extern definition for `%s'\n",
shortpath (NULL, file), user->line,
user->hash_entry->symbol);
#endif
}
}
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 (user)
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: 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
connect_defs_and_decs (hp)
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 consistency 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 /* !defined (UNPROTOIZE) */
/* 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 (clean_p)
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
declare_source_confusing (clean_p)
const char *clean_p;
{
if (!quiet_flag)
{
if (clean_p == 0)
fprintf (stderr, "%s: %d: warning: source too confusing\n",
shortpath (NULL, convert_filename), last_known_line_number);
else
fprintf (stderr, "%s: %d: warning: source too confusing\n",
shortpath (NULL, convert_filename),
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. */
static void
check_source (cond, clean_p)
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 (n)
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-whitespace character which follows it. */
static const char *
forward_to_next_token_char (ptr)
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
output_bytes (str, len)
const char *str;
size_t 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;
}
memcpy (repl_write_ptr + 1, str, len);
repl_write_ptr += len;
}
/* Copy all bytes (except the trailing null) of a null terminated string to
the current output buffer. */
static void
output_string (str)
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 implied when this routine is called includes the
byte pointed to by the argument pointer `p'. */
static void
output_up_to (p)
const char *p;
{
size_t copy_length = (size_t) (p - clean_read_ptr);
const char *copy_start = orig_text_base+(clean_read_ptr-clean_text_base)+1;
if (copy_length == 0)
return;
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 (ansi_header)
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 substr (ansi_header, "...") != 0;
#else /* !defined (UNPROTOIZE) */
/* 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 /* !defined (UNPROTOIZE) */
}
/* Do the editing operation specifically for a function "declaration". Note
that editing for function "definitions" are handled in a separate routine
below. */
static void
edit_fn_declaration (def_dec_p, clean_text_p)
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;
size_t 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: %d: warning: varargs function declaration not converted\n",
shortpath (NULL, def_dec_p->file->hash_entry->symbol),
def_dec_p->line);
return;
}
#endif /* !defined (UNPROTOIZE) */
/* 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 guarantee, 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 /* !defined (UNPROTOIZE) */
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 /* !defined (UNPROTOIZE) */
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 /* !defined (UNPROTOIZE) */
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 (end_formals, f_list_count, def_dec_p)
const char *end_formals;
unsigned 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++;