libvips/libsrc/iofuncs/util.c

1073 lines
21 KiB
C
Raw Normal View History

2007-08-29 18:23:50 +02:00
/* Some basic util functions.
*/
/*
Copyright (C) 1991-2003 The National Gallery
This program 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 of the License, or
(at your option) any later version.
This program 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 this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
/*
These files are distributed with VIPS - http://www.vips.ecs.soton.ac.uk
*/
/*
#define DEBUG
*/
#ifdef HAVE_CONFIG_H
#include <config.h>
#endif /*HAVE_CONFIG_H*/
#include <vips/intl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <ctype.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/stat.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /*HAVE_UNISTD_H*/
2008-08-15 23:45:18 +02:00
#include <fcntl.h>
2007-08-29 18:23:50 +02:00
2008-08-15 10:40:05 +02:00
#ifdef OS_WIN32
#include <windows.h>
#endif /*OS_WIN32*/
2007-08-29 18:23:50 +02:00
#include <assert.h>
#include <vips/vips.h>
#ifdef WITH_DMALLOC
#include <dmalloc.h>
#endif /*WITH_DMALLOC*/
/* Temp buffer for snprintf() layer on old systems.
*/
#define MAX_BUF (32768)
/* Test two lists for eqality.
*/
gboolean
im_slist_equal( GSList *l1, GSList *l2 )
{
while( l1 && l2 ) {
if( l1->data != l2->data )
return( FALSE );
l1 = l1->next;
l2 = l2->next;
}
if( l1 || l2 )
return( FALSE );
return( TRUE );
}
/* Map over an slist. _copy() the list in case the callback changes it.
*/
void *
im_slist_map2( GSList *list, VSListMap2Fn fn, void *a, void *b )
{
GSList *copy;
GSList *i;
void *result;
copy = g_slist_copy( list );
result = NULL;
for( i = copy; i && !(result = fn( i->data, a, b )); i = i->next )
;
g_slist_free( copy );
return( result );
}
void *
im_slist_map4( GSList *list,
VSListMap4Fn fn, void *a, void *b, void *c, void *d )
{
GSList *copy;
GSList *i;
void *result;
copy = g_slist_copy( list );
result = NULL;
for( i = copy;
i && !(result = fn( i->data, a, b, c, d )); i = i->next )
;
g_slist_free( copy );
return( result );
}
/* Map backwards. We _reverse() rather than recurse and unwind to save stack.
*/
void *
im_slist_map2_rev( GSList *list, VSListMap2Fn fn, void *a, void *b )
{
GSList *copy;
GSList *i;
void *result;
copy = g_slist_copy( list );
copy = g_slist_reverse( copy );
result = NULL;
for( i = copy; i && !(result = fn( i->data, a, b )); i = i->next )
;
g_slist_free( copy );
return( result );
}
void *
im_map_equal( void *a, void *b )
{
if( a == b )
return( a );
return( NULL );
}
void *
im_slist_fold2( GSList *list, void *start, VSListFold2Fn fn, void *a, void *b )
{
void *c;
GSList *this, *next;
for( c = start, this = list; this; this = next ) {
next = this->next;
if( !(c = fn( this->data, c, a, b )) )
return( NULL );
}
return( c );
}
static void
im_slist_free_all_cb( void * thing, void * dummy )
{
im_free( thing );
}
/* Free a g_slist of things which need im_free()ing.
*/
void
im_slist_free_all( GSList *list )
{
g_slist_foreach( list, im_slist_free_all_cb, NULL );
g_slist_free( list );
}
/* Remove all occurences of an item from a list.
*/
GSList *
im_slist_filter( GSList *list, VSListMap2Fn fn, void *a, void *b )
{
GSList *tmp;
GSList *prev;
prev = NULL;
tmp = list;
while( tmp ) {
if( fn( tmp->data, a, b ) ) {
GSList *next = tmp->next;
if( prev )
prev->next = next;
if( list == tmp )
list = next;
tmp->next = NULL;
g_slist_free( tmp );
tmp = next;
}
else {
prev = tmp;
tmp = tmp->next;
}
}
return( list );
2008-08-21 10:32:24 +02:00
}
typedef struct {
void *a;
void *b;
VSListMap2Fn fn;
void *result;
} Pair;
static gboolean
2008-08-24 12:19:04 +02:00
im_hash_table_predicate( const char *key, void *value, Pair *pair )
2008-08-21 10:32:24 +02:00
{
2008-08-24 12:19:04 +02:00
return( (pair->result == pair->fn( value, pair->a, pair->b )) );
2008-08-21 10:32:24 +02:00
}
/* Like slist map, but for a hash table.
*/
void *
im_hash_table_map( GHashTable *hash, VSListMap2Fn fn, void *a, void *b )
{
Pair pair;
pair.a = a;
pair.b = b;
pair.fn = fn;
pair.result = NULL;
g_hash_table_find( hash, (GHRFunc) im_hash_table_predicate, &pair );
return( pair.result );
2007-08-29 18:23:50 +02:00
}
/* Like strncpy(), but always NULL-terminate, and don't pad with NULLs.
*/
char *
im_strncpy( char *dest, const char *src, int n )
{
int i;
assert( n > 0 );
for( i = 0; i < n - 1; i++ )
if( !(dest[i] = src[i]) )
break;
dest[i] = '\0';
return( dest );
}
/* Find the rightmost occurrence of needle in haystack.
*/
char *
im_strrstr( const char *haystack, const char *needle )
{
int haystack_len = strlen( haystack );
int needle_len = strlen( needle );
int i;
for( i = haystack_len - needle_len; i >= 0; i-- )
if( strncmp( needle, haystack + i, needle_len ) == 0 )
return( (char *) haystack + i );
return( NULL );
}
/* strdup local to a descriptor.
*/
char *
im_strdup( IMAGE *im, const char *str )
{
int l = strlen( str );
char *buf;
if( !(buf = (char *) im_malloc( im, l + 1 )) )
return( NULL );
strcpy( buf, str );
return( buf );
}
/* Test for string b ends string a.
*/
gboolean
im_ispostfix( const char *a, const char *b )
{
int m = strlen( a );
int n = strlen( b );
if( n > m )
return( FALSE );
return( strcmp( a + m - n, b ) == 0 );
}
/* Test for string a starts string b.
*/
gboolean
im_isprefix( const char *a, const char *b )
{
int n = strlen( a );
int m = strlen( b );
int i;
if( m < n )
return( FALSE );
for( i = 0; i < n; i++ )
if( a[i] != b[i] )
return( FALSE );
return( TRUE );
}
/* Like strtok(). Give a string and a list of break characters. Then:
* - skip initial break characters
* - EOS? return NULL
* - skip a series of non-break characters
* - write a '\0' over the next break character and return a pointer to the
* char after that
*
* The idea is that this can be used in loops as the iterator. Example:
*
* char *p = " 1 2 3 "; // mutable
* char *q;
* int i;
* int v[...];
*
* for( i = 0; (q = im_break_token( p, " " )); i++, p = q )
* v[i] = atoi( p );
*
* will set
* v[0] = 1;
* v[1] = 2;
* v[2] = 3;
*
* or with just one pointer, provided your atoi() is OK with trailing chars
* and you know there is at least one item there
*
* char *p = " 1 2 3 "; // mutable
* int i;
* int v[...];
*
* for( i = 0; p; p = im_break_token( p, " " ) )
* v[i] = atoi( p );
*/
char *
im_break_token( char *str, const char *brk )
{
char *p;
/* Is the string empty? If yes, return NULL immediately.
*/
if( !str || !*str )
return( NULL );
/* Skip initial break characters.
*/
p = str + strspn( str, brk );
/* No item?
*/
if( !*p )
return( NULL );
/* We have a token ... search for the first break character after the
* token.
*/
p += strcspn( p, brk );
/* Is there string left?
*/
if( *p ) {
/* Write in an end-of-string mark and return the start of the
* next token.
*/
*p++ = '\0';
p += strspn( p, brk );
}
return( p );
}
/* Wrapper over (v)snprintf() ... missing on old systems.
*/
int
im_vsnprintf( char *str, size_t size, const char *format, va_list ap )
{
#ifdef HAVE_VSNPRINTF
return( vsnprintf( str, size, format, ap ) );
#else /*HAVE_VSNPRINTF*/
/* Bleurg!
*/
int n;
static char buf[MAX_BUF];
if( size > MAX_BUF )
error_exit( "panic: buffer overflow "
"(request to write %d bytes to buffer of %d bytes)",
size, MAX_BUF );
n = vsprintf( buf, format, ap );
if( n > MAX_BUF )
error_exit( "panic: buffer overflow "
"(%d bytes written to buffer of %d bytes)",
n, MAX_BUF );
im_strncpy( str, buf, size );
return( n );
#endif /*HAVE_VSNPRINTF*/
}
int
im_snprintf( char *str, size_t size, const char *format, ... )
{
va_list ap;
int n;
va_start( ap, format );
n = im_vsnprintf( str, size, format, ap );
va_end( ap );
return( n );
}
/* Split filename into name / mode components. name and mode should both be
* FILENAME_MAX chars.
*
* We look for the ':' splitting the name and mode by searching for the
* rightmost occurence of the regexp ".[A-Za-z0-9]+:". Example: consider the
* horror that is
*
* c:\silly:dir:name\fr:ed.tif:jpeg:95,,,,c:\icc\srgb.icc
*
*/
void
im_filename_split( const char *path, char *name, char *mode )
{
char *p;
im_strncpy( name, path, FILENAME_MAX );
/* Search back towards start stopping at each ':' char.
*/
for( p = name + strlen( name ) - 1; p > name; p -= 1 )
if( *p == ':' ) {
char *q;
for( q = p - 1; isalnum( *q ) && q > name; q -= 1 )
;
if( *q == '.' )
break;
}
if( *p == ':' ) {
im_strncpy( mode, p + 1, FILENAME_MAX );
*p = '\0';
}
else
strcpy( mode, "" );
}
/* Skip any leading path stuff. Horrible: if this is a filename which came
* from win32 and we're a *nix machine, it'll have '\\' not '/' as the
* separator :-(
*
* Try to fudge this ... if the file doesn't contain any of our native
* separators, look for the opposite one as well. If there are none of those
* either, just return the filename.
*/
const char *
im_skip_dir( const char *path )
{
char name[FILENAME_MAX];
char mode[FILENAME_MAX];
const char *p;
const char *q;
const char native_dir_sep = G_DIR_SEPARATOR;
const char non_native_dir_sep = native_dir_sep == '/' ? '\\' : '/';
/* Remove any trailing save modifiers: we don't want '/' or '\' in the
* modifier confusing us.
*/
im_filename_split( path, name, mode );
/* The '\0' char at the end of the string.
*/
p = name + strlen( name );
/* Search back for the first native dir sep, or failing that, the first
* non-native dir sep.
*/
for( q = p; q > name && q[-1] != native_dir_sep; q-- )
;
if( q == name )
for( q = p; q > name && q[-1] != non_native_dir_sep; q-- )
;
return( path + (q - name) );
}
/* Extract suffix from filename, ignoring any mode string. Suffix should be
2008-08-15 10:40:05 +02:00
* FILENAME_MAX chars. Include the "." character, if any.
2007-08-29 18:23:50 +02:00
*/
void
im_filename_suffix( const char *path, char *suffix )
{
char name[FILENAME_MAX];
char mode[FILENAME_MAX];
char *p;
im_filename_split( path, name, mode );
if( (p = strrchr( name, '.' )) )
2008-08-15 10:40:05 +02:00
strcpy( suffix, p );
2007-08-29 18:23:50 +02:00
else
strcpy( suffix, "" );
}
/* Does a filename have one of a set of suffixes. Ignore case.
*/
int
im_filename_suffix_match( const char *path, const char *suffixes[] )
{
char suffix[FILENAME_MAX];
const char **p;
im_filename_suffix( path, suffix );
for( p = suffixes; *p; p++ )
if( g_ascii_strcasecmp( suffix, *p ) == 0 )
return( 1 );
return( 0 );
}
/* p points to the start of a buffer ... move it on through the buffer (ready
* for the next call), and return the current option (or NULL for option
* missing). ',' characters inside options can be escaped with a '\'.
*/
char *
im_getnextoption( char **in )
{
char *p = *in;
char *q = p;
if( !p || !*p )
return( NULL );
/* Find the next ',' not prefixed with a '\'.
*/
while( (p = strchr( p, ',' )) && p[-1] == '\\' )
p += 1;
if( p ) {
/* Another option follows this one .. set up to pick that out
* next time.
*/
*p = '\0';
*in = p + 1;
}
else {
/* This is the last one.
*/
*in = NULL;
}
if( strlen( q ) > 0 )
return( q );
else
return( NULL );
}
/* Get a suboption string, or NULL.
*/
char *
im_getsuboption( const char *buf )
{
char *p, *q, *r;
if( !(p = strchr( buf, ':' )) )
/* No suboption.
*/
return( NULL );
/* Step over the ':'.
*/
p += 1;
/* Need to unescape any \, pairs. Shift stuff down one if we find one.
*/
for( q = p; *q; q++ )
if( q[0] == '\\' && q[1] == ',' )
for( r = q; *r; r++ )
r[0] = r[1];
return( p );
}
/* Make something local to an image descriptor ... pass in a constructor
* and a destructor, plus three args.
*/
void *
im_local( IMAGE *im,
im_construct_fn cons, im_callback_fn dest, void *a, void *b, void *c )
{
void *obj;
if( !im ) {
im_errormsg( "im_local: NULL image descriptor" );
return( NULL );
}
if( !(obj = cons( a, b, c )) )
return( NULL );
if( im_add_close_callback( im, (im_callback_fn) dest, obj, a ) ) {
dest( obj, a );
return( NULL );
}
return( obj );
}
/* Make an array of things local to a descriptor ... eg. make 6 local temp
* images.
*/
int
im_local_array( IMAGE *im, void **out, int n,
im_construct_fn cons, im_callback_fn dest, void *a, void *b, void *c )
{
int i;
for( i = 0; i < n; i++ )
if( !(out[i] = im_local( im, cons, dest, a, b, c )) )
return( -1 );
return( 0 );
}
/* Get file length ... 64-bitally. -1 for error.
*/
gint64
im_file_length( int fd )
{
#ifdef OS_WIN32
struct _stati64 st;
if( _fstati64( fd, &st ) == -1 ) {
#else /*!OS_WIN32*/
struct stat st;
if( fstat( fd, &st ) == -1 ) {
#endif /*OS_WIN32*/
im_error_system( errno, "im_file_length",
2008-11-02 23:11:01 +01:00
"%s", _( "unable to get file stats" ) );
2007-08-29 18:23:50 +02:00
return( -1 );
}
return( st.st_size );
}
/* Wrap write() up
*/
int
im__write( int fd, const void *buf, size_t count )
{
do {
size_t nwritten = write( fd, buf, count );
if( nwritten == (size_t) -1 ) {
im_error_system( errno, "im__write",
2008-11-02 23:11:01 +01:00
"%s", _( "write failed" ) );
2007-08-29 18:23:50 +02:00
return( -1 );
}
buf = (void *) ((char *) buf + nwritten);
count -= nwritten;
} while( count > 0 );
return( 0 );
}
/* Load up a file as a string.
*/
char *
im__file_read( FILE *fp, const char *name, unsigned int *length_out )
{
long len;
size_t read;
char *str;
/* Find length.
*/
fseek( fp, 0L, 2 );
len = ftell( fp );
if( len > 20 * 1024 * 1024 ) {
/* Seems crazy!
*/
im_error( "im__file_read", _( "\"%s\" too long" ), name );
return( NULL );
}
if( len == -1 ) {
int size;
/* Can't get length: read in chunks and realloc() to end of
* file.
*/
str = NULL;
len = 0;
size = 0;
do {
size += 1024;
if( !(str = realloc( str, size )) ) {
im_error( "im__file_read",
2008-11-02 23:11:01 +01:00
"%s", _( "out of memory" ) );
2007-08-29 18:23:50 +02:00
return( NULL );
}
/* -1 to allow space for an extra NULL we add later.
*/
read = fread( str + len, sizeof( char ),
(size - len - 1) / sizeof( char ),
fp );
len += read;
} while( !feof( fp ) );
#ifdef DEBUG
printf( "read %d bytes from unseekable stream\n", len );
#endif /*DEBUG*/
}
else {
/* Allocate memory and fill.
*/
if( !(str = im_malloc( NULL, len + 1 )) )
return( NULL );
rewind( fp );
read = fread( str, sizeof( char ), (size_t) len, fp );
if( read != (size_t) len ) {
im_free( str );
im_error( "im__file_read",
_( "error reading from file \"%s\"" ), name );
return( NULL );
}
}
str[len] = '\0';
if( length_out )
*length_out = len;
return( str );
}
/* Load from a filename as a string.
*/
char *
im__file_read_name( const char *name, unsigned int *length_out )
{
FILE *fp;
char *buffer;
#ifdef BINARY_OPEN
if( !(fp = fopen( name, "rb" )) ) {
#else /*BINARY_OPEN*/
if( !(fp = fopen( name, "r" )) ) {
#endif /*BINARY_OPEN*/
im_error( "im__file_read_name",
_( "unable to open file \"%s\"" ), name );
return( NULL );
}
if( !(buffer = im__file_read( fp, name, length_out )) ) {
fclose( fp );
return( NULL );
}
fclose( fp );
return( buffer );
}
/* Alloc/free a GValue.
*/
static GValue *
im__gvalue_new( GType type )
{
GValue *value;
value = g_new0( GValue, 1 );
g_value_init( value, type );
return( value );
}
static GValue *
im__gvalue_copy( GValue *value )
{
GValue *value_copy;
value_copy = im__gvalue_new( G_VALUE_TYPE( value ) );
g_value_copy( value, value_copy );
return( value_copy );
}
static void
im__gvalue_free( GValue *value )
{
g_value_unset( value );
g_free( value );
}
GValue *
im__gvalue_ref_string_new( const char *text )
{
GValue *value;
value = im__gvalue_new( IM_TYPE_REF_STRING );
im_ref_string_set( value, text );
return( value );
}
/* Free a GSList of GValue.
*/
void
im__gslist_gvalue_free( GSList *list )
{
g_slist_foreach( list, (GFunc) im__gvalue_free, NULL );
g_slist_free( list );
}
/* Copy a GSList of GValue.
*/
GSList *
im__gslist_gvalue_copy( const GSList *list )
{
GSList *copy;
const GSList *p;
copy = NULL;
for( p = list; p; p = p->next )
copy = g_slist_prepend( copy,
im__gvalue_copy( (GValue *) p->data ) );
copy = g_slist_reverse( copy );
return( copy );
}
/* Merge two GSList of GValue ... append to a all elements in b which are not
* in a. Return the new value of a. Works for any vips refcounted type
* (string, blob, etc.).
*/
GSList *
im__gslist_gvalue_merge( GSList *a, const GSList *b )
{
const GSList *i, *j;
GSList *tail;
tail = NULL;
for( i = b; i; i = i->next ) {
GValue *value = (GValue *) i->data;
assert( G_VALUE_TYPE( value ) == IM_TYPE_REF_STRING );
for( j = a; j; j = j->next ) {
GValue *value2 = (GValue *) j->data;
assert( G_VALUE_TYPE( value2 ) == IM_TYPE_REF_STRING );
/* Just do a pointer compare ... good enough 99.9% of
* the time.
*/
if( im_ref_string_get( value ) ==
im_ref_string_get( value2 ) )
break;
}
if( !j )
tail = g_slist_prepend( tail,
im__gvalue_copy( value ) );
}
a = g_slist_concat( a, g_slist_reverse( tail ) );
return( a );
}
/* Make a char* from GSList of GValue. Each GValue should be a ref_string.
* free the result. Empty list -> "", not NULL. Join strings with '\n'.
*/
char *
im__gslist_gvalue_get( const GSList *list )
{
const GSList *p;
size_t length;
char *all;
char *q;
/* Need to estimate length first.
*/
length = 0;
for( p = list; p; p = p->next ) {
GValue *value = (GValue *) p->data;
assert( G_VALUE_TYPE( value ) == IM_TYPE_REF_STRING );
/* +1 for the newline we will add for each item.
*/
length += im_ref_string_get_length( value ) + 1;
}
if( length == 0 )
return( NULL );
/* More than 10MB of history? Madness!
*/
assert( length < 10 * 1024 * 1024 );
/* +1 for '\0'.
*/
if( !(all = im_malloc( NULL, length + 1 )) )
return( NULL );
q = all;
for( p = list; p; p = p->next ) {
GValue *value = (GValue *) p->data;
strcpy( q, im_ref_string_get( value ) );
q += im_ref_string_get_length( value );
strcpy( q, "\n" );
q += 1;
}
assert( q - all == length );
return( all );
}
2008-08-15 10:40:05 +02:00
/* Need our own seek(), since lseek() on win32 can't do long files.
*/
int
im__seek( int fd, gint64 pos )
{
#ifdef OS_WIN32
{
HANDLE hFile = (HANDLE) _get_osfhandle( fd );
LARGE_INTEGER p;
p.QuadPart = pos;
if( !SetFilePointerEx( hFile, p, NULL, FILE_BEGIN ) ) {
im_error_system( GetLastError(), "im__seek",
2008-11-02 23:11:01 +01:00
"%s", _( "unable to seek" ) );
2008-08-15 10:40:05 +02:00
return( -1 );
}
}
#else /*!OS_WIN32*/
if( lseek( fd, pos, SEEK_SET ) == (off_t) -1 ) {
2008-11-02 23:11:01 +01:00
im_error( "im__seek", "%s", _( "unable to seek" ) );
2008-08-15 10:40:05 +02:00
return( -1 );
}
#endif /*OS_WIN32*/
return( 0 );
}
/* Need our own ftruncate(), since ftruncate() on win32 can't do long files.
DANGER ... this moves the file pointer to the end of file on win32,
but not on *nix; don't make any assumptions about the file pointer
position after calling this
*/
int
im__ftruncate( int fd, gint64 pos )
{
#ifdef OS_WIN32
{
HANDLE hFile = (HANDLE) _get_osfhandle( fd );
LARGE_INTEGER p;
p.QuadPart = pos;
if( im__seek( fd, pos ) )
return( -1 );
if( !SetEndOfFile( hFile ) ) {
im_error_system( GetLastError(), "im__ftruncate",
2008-11-02 23:11:01 +01:00
"%s", _( "unable to truncate" ) );
2008-08-15 10:40:05 +02:00
return( -1 );
}
}
#else /*!OS_WIN32*/
if( ftruncate( fd, pos ) ) {
im_error_system( errno, "im__ftruncate",
2008-11-02 23:11:01 +01:00
"%s", _( "unable to truncate" ) );
2008-08-15 10:40:05 +02:00
return( -1 );
}
#endif /*OS_WIN32*/
return( 0 );
}
/* Like fwrite(), but returns non-zero on error and sets error message.
*/
int
im__file_write( void *data, size_t size, size_t nmemb, FILE *stream )
{
2008-08-23 09:48:47 +02:00
size_t n;
2008-08-15 10:40:05 +02:00
2008-08-23 09:48:47 +02:00
if( !data )
return( 0 );
if( (n = fwrite( data, size, nmemb, stream )) != nmemb ) {
im_error( "im__file_write",
_( "writing error (%zd out of %zd blocks written) "
"... disc full?" ), n, nmemb );
return( -1 );
}
return( 0 );
}
2008-08-15 10:40:05 +02:00
/* Check whether arch corresponds to native byte order.
*/
gboolean
im_isnative( im_arch_type arch )
{
switch ( arch ) {
case IM_ARCH_NATIVE: return( TRUE );
case IM_ARCH_BYTE_SWAPPED: return( FALSE );
case IM_ARCH_LSB_FIRST: return( !im_amiMSBfirst() );
case IM_ARCH_MSB_FIRST: return( im_amiMSBfirst() );
}
abort();
}
2008-08-15 23:45:18 +02:00
/* Read a few bytes from the start of a file. For sniffing file types.
*/
int
im__get_bytes( const char *filename, unsigned char buf[], int len )
{
int fd;
/* File may not even exist (for tmp images for example!)
* so no hasty messages. And the file might be truncated, so no error
* on read either.
*/
#ifdef BINARY_OPEN
if( (fd = open( filename, O_RDONLY | O_BINARY )) == -1 )
#else /*BINARY_OPEN*/
if( (fd = open( filename, O_RDONLY )) == -1 )
#endif /*BINARY_OPEN*/
return( 0 );
if( read( fd, buf, len ) != len ) {
close( fd );
return( 0 );
}
close( fd );
2008-08-15 10:40:05 +02:00
2008-08-15 23:45:18 +02:00
return( 1 );
}