2023-07-30 19:18:50 +00:00
// -----------------------------------------------------------------------------
// file
#if 0 // ifdef _WIN32
# include <winsock2.h>
# if is(tcc)
# define CP_UTF8 65001
int WINAPI MultiByteToWideChar ( ) ;
int WINAPI WideCharToMultiByte ( ) ;
# endif
// widen() ? string1252() ? string16() ? stringw() ?
wchar_t * widen ( const char * utf8 ) { // wide strings (win only)
int chars = MultiByteToWideChar ( CP_UTF8 , 0 , utf8 , - 1 , NULL , 0 ) ;
char * buf = va ( " %.*s " , ( int ) ( chars * sizeof ( wchar_t ) ) , " " ) ;
return MultiByteToWideChar ( CP_UTF8 , 0 , utf8 , - 1 , ( void * ) buf , chars ) , ( wchar_t * ) buf ;
}
# define open8(path,mode) ifdef(win, _wopen(widen(path)) , open(path, mode) )
# define fopen8(path,mode) ifdef(win, _wfopen(widen(path),widen(mode)) , fopen(path,mode) )
# define remove8(path) ifdef(win, _wremove(widen(path)) , remove(path) )
# define rename8(path) ifdef(win, _wrename(widen(path)) , rename(path) )
# define stat8(path,st) ifdef(win, _wstat(widen(path),st) , stat(path,st) ) // _stati64()
# define stat8_t ifdef(win, _stat , stat_t ) // struct _stati64
# endif
char * file_name ( const char * pathfile ) {
char * s = strrchr ( pathfile , ' / ' ) , * t = strrchr ( pathfile , ' \\ ' ) ;
return va ( " %s " , s > t ? s + 1 : t ? t + 1 : pathfile ) ;
}
char * file_base ( const char * pathfile ) {
char * s = file_name ( pathfile ) ;
char * e = file_ext ( pathfile ) ;
return s [ strlen ( s ) - strlen ( e ) ] = ' \0 ' , s ;
}
char * file_pathabs ( const char * pathfile ) {
char * out = va ( " %*.s " , DIR_MAX + 1 , " " ) ;
# if is(win32)
_fullpath ( out , pathfile , DIR_MAX ) ;
# else
realpath ( pathfile , out ) ;
# endif
return out ;
}
char * file_path ( const char * pathfile ) {
return va ( " %.*s " , ( int ) ( strlen ( pathfile ) - strlen ( file_name ( pathfile ) ) ) , pathfile ) ;
}
char * file_load ( const char * filename , int * len ) { // @todo: 32 counters/thread enough?
static __thread array ( char ) buffer [ 32 ] = { 0 } , * invalid [ 1 ] ;
static __thread unsigned i = 0 ; i = ( i + 1 ) % 32 ;
FILE * fp = filename [ 0 ] ? fopen ( filename , " rb " ) : NULL ;
if ( fp ) {
fseek ( fp , 0L , SEEK_END ) ;
size_t sz = ftell ( fp ) ;
fseek ( fp , 0L , SEEK_SET ) ;
array_resize ( buffer [ i ] , sz + 1 ) ;
sz * = fread ( buffer [ i ] , sz , 1 , fp ) = = 1 ;
buffer [ i ] [ sz ] = 0 ;
if ( len ) * len = ( int ) sz ;
fclose ( fp ) ;
return buffer [ i ] ; // @fixme: return 0 on error instead?
}
if ( len ) * len = 0 ;
return 0 ;
}
char * file_read ( const char * filename ) { // @todo: fix leaks
return file_load ( filename , NULL ) ;
}
bool file_write ( const char * name , const void * ptr , int len ) {
bool ok = 0 ;
for ( FILE * fp = name & & ptr & & len > = 0 ? fopen ( name , " wb " ) : NULL ; fp ; fclose ( fp ) , fp = 0 ) {
ok = fwrite ( ptr , len , 1 , fp ) = = 1 ;
}
return ok ;
}
bool file_append ( const char * name , const void * ptr , int len ) {
bool ok = 0 ;
for ( FILE * fp = name & & ptr & & len > = 0 ? fopen ( name , " a+b " ) : NULL ; fp ; fclose ( fp ) , fp = 0 ) {
ok = fwrite ( ptr , len , 1 , fp ) = = 1 ;
}
return ok ;
}
2023-10-14 19:47:24 +00:00
static // not exposed
bool file_stat ( const char * fname , struct stat * st ) {
2023-07-30 19:18:50 +00:00
// remove ending slashes. win32+tcc does not like them.
int l = strlen ( fname ) , m = l ;
while ( l & & ( fname [ l - 1 ] = = ' / ' | | fname [ l - 1 ] = = ' \\ ' ) ) - - l ;
fname = l = = m ? fname : va ( " %.*s " , l , fname ) ;
return stat ( fname , st ) > = 0 ;
}
2023-08-26 13:33:29 +00:00
uint64_t file_stamp ( const char * fname ) {
2023-07-30 19:18:50 +00:00
struct stat st ;
return ! file_stat ( fname , & st ) ? 0ULL : st . st_mtime ;
}
2023-08-26 13:33:29 +00:00
uint64_t file_stamp10 ( const char * fname ) {
time_t mtime = ( time_t ) file_stamp ( fname ) ;
struct tm * ti = localtime ( & mtime ) ;
return atoi64 ( va ( " %04d%02d%02d%02d%02d%02d " , ti - > tm_year + 1900 , ti - > tm_mon + 1 , ti - > tm_mday , ti - > tm_hour , ti - > tm_min , ti - > tm_sec ) ) ;
}
2023-07-30 19:18:50 +00:00
uint64_t file_size ( const char * fname ) {
struct stat st ;
return ! file_stat ( fname , & st ) ? 0ULL : st . st_size ;
}
bool file_directory ( const char * pathfile ) {
struct stat st ;
return ! file_stat ( pathfile , & st ) ? 0 : S_IFDIR = = ( st . st_mode & S_IFMT ) ;
}
bool file_exist ( const char * fname ) {
struct stat st ;
return ! file_stat ( fname , & st ) ? false : true ;
}
char * file_normalize ( const char * name ) {
char * copy = va ( " %s " , name ) , * s = copy , c ;
# if is(win32)
for ( int i = 0 ; copy [ i ] ; + + i ) { if ( copy [ i ] = = ' / ' ) copy [ i ] = ' \\ ' ; if ( copy [ i ] = = ' \' ' ) copy [ i ] = ' \" ' ; }
# else
for ( int i = 0 ; copy [ i ] ; + + i ) { if ( copy [ i ] = = ' \\ ' ) copy [ i ] = ' / ' ; if ( copy [ i ] = = ' \" ' ) copy [ i ] = ' \' ' ; }
# endif
return copy ;
}
#if 0
char * file_normalize ( const char * name ) {
char * copy = va ( " %s " , name ) , * s = copy , c ;
// lowercases+digits+underscores+slashes only. anything else is truncated.
for ( ; * name ; + + name ) {
/**/ if ( * name > = ' a ' & & * name < = ' z ' ) * s + + = * name ;
else if ( * name > = ' A ' & & * name < = ' Z ' ) * s + + = * name - ' A ' + ' a ' ;
else if ( * name > = ' 0 ' & & * name < = ' 9 ' ) * s + + = * name ;
else if ( * name = = ' / ' | | * name = = ' \\ ' ) * s + + = ' / ' ;
else if ( * name < = ' ' | | * name = = ' . ' ) * s + + = ' _ ' ;
} * s + + = 0 ;
// remove dupe slashes
for ( name = s = copy , c = ' / ' ; * name ; ) {
while ( * name & & * name ! = c ) * s + + = * name + + ;
if ( * name ) * s + + = c ;
while ( * name & & * name = = c ) name + + ;
} * s + + = 0 ;
// remove dupe underlines
for ( name = s = copy , c = ' _ ' ; * name ; ) {
while ( * name & & * name ! = c ) * s + + = * name + + ;
if ( * name ) * s + + = c ;
while ( * name & & * name = = c ) name + + ;
} * s + + = 0 ;
return copy ;
}
char * file_normalize_with_folder ( const char * name ) {
char * s = file_normalize ( name ) ;
char * slash = strrchr ( s , ' / ' ) ; if ( slash ) * slash = 0 ;
char * penultimate = strrchr ( s , ' / ' ) ; if ( slash ) * slash = ' / ' ;
return penultimate ? penultimate + 1 : /*slash ? slash+1 :*/ s ;
}
# endif
char * file_ext ( const char * name ) {
char * b = file_name ( name ) , * s = strchr ( b , ' . ' ) ; //strrchr(name, '.');
return va ( " %s " , s ? s : " " ) ; // , name );
}
char * file_id ( const char * pathfile ) {
char * dir = file_path ( pathfile ) ; for ( int i = 0 ; dir [ i ] ; + + i ) dir [ i ] = tolower ( dir [ i ] ) ;
char * base = file_name ( pathfile ) ; for ( int i = 0 ; base [ i ] ; + + i ) base [ i ] = tolower ( base [ i ] ) ;
#if 0 // extensionless, larry.mid and larry.txt will collide, diffuse.tga and diffuse.png will match
char * ext = strchr ( base , ' . ' ) ; if ( ext ) ext [ 0 ] = ' \0 ' ; // remove all extensions
# else // extensionless for audio/images only (materials: diffuse.tga and diffuse.png will match)
char * ext = strrchr ( base , ' . ' ) ; //if (ext) ext[0] = '\0'; // remove all extensions
if ( ext ) if ( strstr ( " .jpg.png.bmp.tga.hdr " " . " , ext ) | | strstr ( " .ogg.mp3.wav.mod.xm.flac " " . " , ext ) | | strstr ( " .mp4.ogv.avi.mkv.wmv.mpg.mpeg " " . " , ext ) ) {
ext = strchr ( base , ' . ' ) ;
ext [ 0 ] = ' \0 ' ; //strcpy(ext, "_xxx");
}
# endif
// if (!dir[0]) return base;
char * stem = va ( " %s/%s " , dir , base ) ; // file_name(dir);
// /path2/path1/file2_file1 -> file1_file2/path1/path2
int len = 0 ;
int ids_count = 0 ;
char ids [ 64 ] [ 64 ] = { 0 } ;
// split path stems
2023-10-01 06:49:08 +00:00
for each_substring ( stem , " / \\ @ " , key ) {
2023-07-30 19:18:50 +00:00
int tokens_count = 0 ;
char * tokens [ 64 ] = { 0 } ;
// split tokens
for each_substring ( key , " []() _ " , it) {
tokens [ tokens_count + + ] = va ( " %s " , it ) ;
}
// sort alphabetically
if ( tokens_count > 1 ) qsort ( tokens , tokens_count , sizeof ( char * ) , strcmp_qsort ) ;
// concat sorted token1_token2_...
char built [ 256 ] ; * built = 0 ;
for ( int i = 0 ; i < tokens_count ; + + i ) {
strlcat ( built , " _ " , 256 ) ;
strlcat ( built , tokens [ i ] , 256 ) ;
}
strncpy ( ids [ ids_count ] , & built [ 1 ] , 64 ) ;
len + = strlen ( ids [ ids_count + + ] ) ;
}
// concat in inverse order: file/path1/path2/...
char buffer [ DIR_MAX ] ; buffer [ 0 ] = 0 ;
for ( int it = ids_count ; - - it > = 0 ; ) {
strcat ( buffer , ids [ it ] ) ;
strcat ( buffer , " / " ) ;
}
return va ( " %s " , buffer ) ;
}
2023-10-14 19:47:24 +00:00
array ( char * ) file_list ( const char * pathmasks ) {
2023-10-15 11:16:35 +00:00
static __thread array ( char * ) list = 0 ; // @fixme: add 16 slots
2023-07-30 19:18:50 +00:00
for ( int i = 0 ; i < array_count ( list ) ; + + i ) {
FREE ( list [ i ] ) ;
}
2023-10-15 11:16:35 +00:00
array_resize ( list , 0 ) ;
2023-07-30 19:18:50 +00:00
2023-10-14 19:47:24 +00:00
for each_substring ( pathmasks , " ; " , pathmask ) {
char * cwd = 0 , * masks = 0 ;
char * slash = strrchr ( pathmask , ' / ' ) ;
if ( ! slash ) cwd = " ./ " , masks = pathmask ;
else {
masks = va ( " %s " , slash + 1 ) ;
cwd = pathmask , slash [ 1 ] = ' \0 ' ;
}
if ( ! masks [ 0 ] ) masks = " * " ;
2023-10-11 19:01:06 +00:00
ASSERT ( strend ( cwd , " / " ) , " Error: dirs like '%s' must end with slash " , cwd ) ;
2023-10-10 10:15:14 +00:00
dir * d = dir_open ( cwd , strstr ( masks , " ** " ) ? " r " : " " ) ;
2023-10-10 08:52:29 +00:00
if ( d ) {
for ( int i = 0 ; i < dir_count ( d ) ; + + i ) {
if ( dir_file ( d , i ) ) {
// dir_name() should return full normalized paths "C:/prj/v4k/demos/art/fx/fxBloom.fs". should exclude system dirs as well
char * entry = dir_name ( d , i ) ;
char * fname = file_name ( entry ) ;
int allowed = 0 ;
for each_substring ( masks , " ; " , mask ) {
allowed | = strmatch ( fname , mask ) ;
}
if ( ! allowed ) continue ;
// if( strstr(fname, "/.") ) continue; // @fixme: still needed? useful?
2023-07-30 19:18:50 +00:00
// insert copy
2023-10-10 08:52:29 +00:00
char * copy = STRDUP ( entry ) ;
2023-07-30 19:18:50 +00:00
array_push ( list , copy ) ;
}
}
2023-10-10 08:52:29 +00:00
dir_close ( d ) ;
2023-07-30 19:18:50 +00:00
}
2023-10-11 19:01:06 +00:00
}
2023-10-10 08:52:29 +00:00
2023-10-15 11:16:35 +00:00
array_sort ( list , strcmp ) ;
2023-10-14 19:47:24 +00:00
return list ;
2023-07-30 19:18:50 +00:00
}
bool file_move ( const char * src , const char * dst ) {
bool ok = file_exist ( src ) & & ! file_exist ( dst ) & & 0 = = rename ( src , dst ) ;
return ok ;
}
bool file_delete ( const char * pathfile ) {
if ( file_exist ( pathfile ) ) {
for ( int i = 0 ; i < 10 ; + + i ) {
bool ok = 0 = = unlink ( pathfile ) ;
if ( ok ) return true ;
sleep_ms ( 10 ) ;
}
return false ;
}
return true ;
}
bool file_copy ( const char * src , const char * dst ) {
int ok = 0 , BUFSIZE = 1 < < 20 ; // 1 MiB
2023-10-13 10:59:44 +00:00
static __thread char * buffer = 0 ; do_once buffer = REALLOC ( 0 , BUFSIZE ) ; // @leak
2023-07-30 19:18:50 +00:00
for ( FILE * in = fopen ( src , " rb " ) ; in ; fclose ( in ) , in = 0 ) {
for ( FILE * out = fopen ( dst , " wb " ) ; out ; fclose ( out ) , out = 0 , ok = 1 ) {
for ( int n ; ! ! ( n = fread ( buffer , 1 , BUFSIZE , in ) ) ; ) {
if ( fwrite ( buffer , 1 , n , out ) ! = n )
return fclose ( in ) , fclose ( out ) , false ;
}
}
}
return ok ;
}
char * file_tempname ( ) {
static __thread int id ;
2023-08-10 14:30:56 +00:00
return va ( " %s/v4k-temp.%s.%p.%d " , app_temp ( ) , getenv ( ifdef ( win32 , " username " , " USER " ) ) , & id , rand ( ) ) ;
2023-07-30 19:18:50 +00:00
}
FILE * file_temp ( void ) {
const char * fname = file_tempname ( ) ;
FILE * fp = fopen ( fname , " w+b " ) ;
if ( fp ) unlink ( fname ) ;
return fp ;
}
char * file_counter ( const char * name ) {
static __thread char outfile [ DIR_MAX ] , init = 0 ;
static __thread map ( char * , int ) ext_counters ;
if ( ! init ) map_init ( ext_counters , less_str , hash_str ) , init = ' \1 ' ;
char * base = va ( " %s " , name ) , * ext = file_ext ( name ) ;
if ( ext & & ext [ 0 ] ) * strstr ( base , ext ) = ' \0 ' ;
int * counter = map_find_or_add ( ext_counters , ext , 0 ) ;
while ( * counter > = 0 ) {
* counter = * counter + 1 ;
sprintf ( outfile , " %s(%03d)%s " , base , * counter , ext ) ;
if ( ! file_exist ( outfile ) ) {
return va ( " %s " , outfile ) ;
}
}
return 0 ;
}
enum { MD5_HASHLEN = 16 } ;
enum { SHA1_HASHLEN = 20 } ;
enum { CRC32_HASHLEN = 4 } ;
void * file_sha1 ( const char * file ) { // 20bytes
hash_state hs = { 0 } ;
sha1_init ( & hs ) ;
for ( FILE * fp = fopen ( file , " rb " ) ; fp ; fclose ( fp ) , fp = 0 ) {
char buf [ 8192 ] ;
for ( int inlen ; ( inlen = fread ( buf , 1 , sizeof ( buf ) , fp ) ) > 0 ; ) {
sha1_process ( & hs , ( const unsigned char * ) buf , inlen ) ;
}
}
unsigned char * hash = va ( " %.*s " , SHA1_HASHLEN , " " ) ;
sha1_done ( & hs , hash ) ;
return hash ;
}
void * file_md5 ( const char * file ) { // 16bytes
hash_state hs = { 0 } ;
md5_init ( & hs ) ;
for ( FILE * fp = fopen ( file , " rb " ) ; fp ; fclose ( fp ) , fp = 0 ) {
char buf [ 8192 ] ;
for ( int inlen ; ( inlen = fread ( buf , 1 , sizeof ( buf ) , fp ) ) > 0 ; ) {
md5_process ( & hs , ( const unsigned char * ) buf , inlen ) ;
}
}
unsigned char * hash = va ( " %.*s " , MD5_HASHLEN , " " ) ;
md5_done ( & hs , hash ) ;
return hash ;
}
void * file_crc32 ( const char * file ) { // 4bytes
unsigned crc = 0 ;
for ( FILE * fp = fopen ( file , " rb " ) ; fp ; fclose ( fp ) , fp = 0 ) {
char buf [ 8192 ] ;
for ( int inlen ; ( inlen = fread ( buf , 1 , sizeof ( buf ) , fp ) ) > 0 ; ) {
crc = zip__crc32 ( crc , buf , inlen ) ; // unsigned int stbiw__crc32(unsigned char *buffer, int len)
}
}
unsigned char * hash = va ( " %.*s " , ( int ) sizeof ( crc ) , " " ) ;
memcpy ( hash , & crc , sizeof ( crc ) ) ;
return hash ;
}
#if 0
void * crc32_mem ( const void * ptr , int inlen ) { // 4bytes
unsigned hash = 0 ;
hash = zip__crc32 ( hash , ptr , inlen ) ; // unsigned int stbiw__crc32(unsigned char *buffer, int len)
return hash ;
}
void * md5_mem ( const void * ptr , int inlen ) { // 16bytes
hash_state hs = { 0 } ;
md5_init ( & hs ) ;
md5_process ( & hs , ( const unsigned char * ) ptr , inlen ) ;
unsigned char * hash = va ( " %.*s " , MD5_HASHLEN , " " ) ;
md5_done ( & hs , hash ) ;
return hash ;
}
void * sha1_mem ( const void * ptr , int inlen ) { // 20bytes
hash_state hs = { 0 } ;
sha1_init ( & hs ) ;
sha1_process ( & hs , ( const unsigned char * ) ptr , inlen ) ;
unsigned char * hash = va ( " %.*s " , SHA1_HASHLEN , " " ) ;
sha1_done ( & hs , hash ) ;
return hash ;
}
2023-10-09 06:07:17 +00:00
unsigned crc32_mem ( unsigned h , const void * ptr_ , unsigned len ) {
// based on public domain code by Karl Malbrain
const uint8_t * ptr = ( const uint8_t * ) ptr_ ;
if ( ! ptr ) return 0 ;
const unsigned tbl [ 16 ] = {
0x00000000 , 0x1db71064 , 0x3b6e20c8 , 0x26d930ac , 0x76dc4190 , 0x6b6b51f4 , 0x4db26158 , 0x5005713c ,
0xedb88320 , 0xf00f9344 , 0xd6d6a3e8 , 0xcb61b38c , 0x9b64c2b0 , 0x86d3d2d4 , 0xa00ae278 , 0xbdbdf21c } ;
for ( h = ~ h ; len - - ; ) { uint8_t b = * ptr + + ; h = ( h > > 4 ) ^ tbl [ ( h & 15 ) ^ ( b & 15 ) ] ; h = ( h > > 4 ) ^ tbl [ ( h & 15 ) ^ ( b > > 4 ) ] ; }
return ~ h ;
}
uint64_t crc64_mem ( uint64_t h , const void * ptr , uint64_t len ) {
// based on public domain code by Lasse Collin
// also, use poly64 0xC96C5795D7870F42 for crc64-ecma
static uint64_t crc64_table [ 256 ] ;
static uint64_t poly64 = UINT64_C ( 0x95AC9329AC4BC9B5 ) ;
if ( poly64 ) {
for ( int b = 0 ; b < 256 ; + + b ) {
uint64_t r = b ;
for ( int i = 0 ; i < 8 ; + + i ) {
r = r & 1 ? ( r > > 1 ) ^ poly64 : r > > 1 ;
}
crc64_table [ b ] = r ;
//printf("%016llx\n", crc64_table[b]);
}
poly64 = 0 ;
}
const uint8_t * buf = ( const uint8_t * ) ptr ;
uint64_t crc = ~ h ; // ~crc;
while ( len ! = 0 ) {
crc = crc64_table [ ( uint8_t ) crc ^ * buf + + ] ^ ( crc > > 8 ) ;
- - len ;
}
return ~ crc ;
}
// https://en.wikipedia.org/wiki/MurmurHash
static inline uint32_t murmur3_scramble ( uint32_t k ) {
return k * = 0xcc9e2d51 , k = ( k < < 15 ) | ( k > > 17 ) , k * = 0x1b873593 ;
}
uint32_t murmur3_mem ( const uint8_t * key , size_t len , uint32_t seed ) {
uint32_t h = seed ;
uint32_t k ;
/* Read in groups of 4. */
for ( size_t i = len > > 2 ; i ; i - - ) {
// Here is a source of differing results across endiannesses.
// A swap here has no effects on hash properties though.
k = * ( ( uint32_t * ) key ) ;
key + = sizeof ( uint32_t ) ;
h ^ = murmur3_scramble ( k ) ;
h = ( h < < 13 ) | ( h > > 19 ) ;
h = h * 5 + 0xe6546b64 ;
}
/* Read the rest. */
k = 0 ;
for ( size_t i = len & 3 ; i ; i - - ) {
k < < = 8 ;
k | = key [ i - 1 ] ;
}
// A swap is *not* necessary here because the preceeding loop already
// places the low bytes in the low places according to whatever endianness
// we use. Swaps only apply when the memory is copied in a chunk.
h ^ = murmur3_scramble ( k ) ;
/* Finalize. */
h ^ = len ;
h ^ = h > > 16 ;
h * = 0x85ebca6b ;
h ^ = h > > 13 ;
h * = 0xc2b2ae35 ;
h ^ = h > > 16 ;
return h ;
}
2023-07-30 19:18:50 +00:00
# endif
// -----------------------------------------------------------------------------
// storage (emscripten only)
void storage_mount ( const char * folder ) {
# if is(ems)
emscripten_run_script ( va ( " FS.mkdir('%s'); FS.mount(IDBFS, {}, '%s'); " , folder , folder ) ) ;
# else
( void ) folder ;
# endif
}
void storage_read ( ) {
# if is(ems)
EM_ASM (
/* sync from persisted state into memory */
FS . syncfs ( true , function ( err ) {
assert ( ! err ) ;
} ) ;
) ;
# endif
}
void storage_flush ( ) {
# if is(ems)
EM_ASM (
FS . syncfs ( false , function ( err ) {
assert ( ! err ) ;
} ) ;
) ;
# endif
}
// -----------------------------------------------------------------------------
// compressed archives
// return list of files inside zipfile
array ( char * ) file_zip_list ( const char * zipfile ) {
static __thread array ( char * ) list [ 16 ] = { 0 } ;
static __thread int count = 0 ;
count = ( count + 1 ) % 16 ;
array_resize ( list [ count ] , 0 ) ;
for ( zip * z = zip_open ( zipfile , " rb " ) ; z ; zip_close ( z ) , z = 0 ) {
for ( unsigned i = 0 ; i < zip_count ( z ) ; + + i ) {
array_push ( list [ count ] , zip_name ( z , i ) ) ;
}
}
return list [ count ] ;
}
// extract single file content from zipfile
array ( char ) file_zip_extract ( const char * zipfile , const char * filename ) {
static __thread array ( char ) list [ 16 ] = { 0 } ;
static __thread int count = 0 ;
array ( char ) out = list [ count = ( count + 1 ) % 16 ] ;
array_resize ( out , 0 ) ;
for ( zip * z = zip_open ( zipfile , " rb " ) ; z ; zip_close ( z ) , z = 0 ) {
int index = zip_find ( z , filename ) ; // convert entry to index. returns <0 if not found.
if ( index < 0 ) return zip_close ( z ) , out ;
unsigned outlen = zip_size ( z , index ) ;
unsigned excess = zip_excess ( z , index ) ;
array_resize ( out , outlen + 1 + excess ) ;
unsigned ret = zip_extract_inplace ( z , index , out , array_count ( out ) ) ;
if ( ret ) { out [ outlen ] = ' \0 ' ; array_resize ( out , outlen ) ; } else { array_resize ( out , 0 ) ; }
}
return out ;
}
// append single file into zipfile. compress with DEFLATE|6. Other compressors are also supported (try LZMA|5, ULZ|9, LZ4X|3, etc.)
bool file_zip_append ( const char * zipfile , const char * filename , int clevel ) {
bool ok = false ;
for ( zip * z = zip_open ( zipfile , " a+b " ) ; z ; zip_close ( z ) , z = 0 ) {
for ( FILE * fp = fopen ( filename , " rb " ) ; fp ; fclose ( fp ) , fp = 0 ) {
ok = zip_append_file ( z , filename , " " , fp , clevel ) ;
}
}
return ok ;
}
// append mem blob into zipfile. compress with DEFLATE|6. Other compressors are also supported (try LZMA|5, ULZ|9, LZ4X|3, etc.)
// @fixme: implement zip_append_mem() and use that instead
bool file_zip_appendmem ( const char * zipfile , const char * entryname , const void * ptr , unsigned len , int clevel ) {
bool ok = false ;
if ( ptr )
for ( zip * z = zip_open ( zipfile , " a+b " ) ; z ; zip_close ( z ) , z = 0 ) {
ok = zip_append_mem ( z , entryname , " " , ptr , len , clevel ) ;
}
return ok ;
}
// -----------------------------------------------------------------------------
// archives
enum { is_zip , is_tar , is_pak , is_dir } ;
typedef struct archive_dir {
char * path ;
union {
int type ;
int size ; // for cache only
} ;
union {
void * archive ;
void * data ; // for cache only
zip * zip_archive ;
tar * tar_archive ;
pak * pak_archive ;
} ;
struct archive_dir * next ;
} archive_dir ;
static archive_dir * dir_mount ;
2023-09-08 20:19:06 +00:00
static archive_dir * dir_cache ;
# ifndef MAX_CACHED_FILES // @todo: should this be MAX_CACHED_SIZE (in MiB) instead?
# define MAX_CACHED_FILES 32 // @todo: should we cache the cooked contents instead? ie, stbi() result instead of file.png?
# endif
2023-07-30 19:18:50 +00:00
struct vfs_entry {
const char * name ;
const char * id ;
unsigned size ;
} ;
2023-10-15 11:16:35 +00:00
static array ( struct vfs_entry ) vfs_hints ; // mounted raw assets
static array ( struct vfs_entry ) vfs_entries ; // mounted cooked assets
2023-07-30 19:18:50 +00:00
static bool vfs_mount_hints ( const char * path ) ;
static
void vfs_reload ( ) {
const char * app = app_name ( ) ;
array_resize ( vfs_hints , 0 ) ; // @leak
array_resize ( vfs_entries , 0 ) ; // @leak
2023-09-08 20:19:06 +00:00
// mount virtual filesystems later (mounting order matters: low -> to -> high priority)
2023-09-11 09:07:59 +00:00
# if defined(EMSCRIPTEN)
2023-09-11 10:23:33 +00:00
vfs_mount ( " index.zip " ) ;
# else
2023-10-13 10:59:44 +00:00
// mount fused executables
vfs_mount ( va ( " %s%s%s " , app_path ( ) , app_name ( ) , ifdef ( win32 , " .exe " , " " ) ) ) ;
2023-10-14 19:47:24 +00:00
// mount all zipfiles
for each_array ( file_list ( " *.zip " ) , char * , file ) vfs_mount ( file ) ;
2023-09-08 20:19:06 +00:00
# endif
2023-07-30 19:18:50 +00:00
// vfs_resolve() will use these art_folder locations as hints when cook-on-demand is in progress.
// cook-on-demand will not be able to resolve a virtual pathfile if there are no cooked assets on disk,
// unless there is a record of what's actually on disk somewhere, and that's where the hints belong to.
2023-10-14 20:00:30 +00:00
if ( COOK_ON_DEMAND )
2023-07-30 19:18:50 +00:00
for each_substring ( ART , " , " , art_folder ) {
vfs_mount_hints ( art_folder ) ;
}
}
2023-10-13 10:59:44 +00:00
2023-10-14 19:47:24 +00:00
# define ARK1 0x41724B31 // 'ArK1' in le, 0x314B7241 41 72 4B 31 otherwise
2023-10-13 10:59:44 +00:00
# define ARK1_PADDING (512 - 40) // 472
# define ARK_PRINTF(f,...) 0 // printf(f,__VA_ARGS__)
# define ARK_SWAP32(x) (x)
# define ARK_SWAP64(x) (x)
# define ARK_REALLOC REALLOC
static uint64_t ark_fget64 ( FILE * in ) { uint64_t v ; fread ( & v , 1 , 8 , in ) ; return ARK_SWAP64 ( v ) ; }
void ark_list ( const char * infile , zip * * z ) {
for ( FILE * in = fopen ( infile , " rb " ) ; in ; fclose ( in ) , in = 0 )
while ( ! feof ( in ) ) {
if ( 0 ! = ( ftell ( in ) % ARK1_PADDING ) ) fseek ( in , ARK1_PADDING - ( ftell ( in ) % ARK1_PADDING ) , SEEK_CUR ) ;
ARK_PRINTF ( " Reading at #%d \n " , ( int ) ftell ( in ) ) ;
uint64_t mark = ark_fget64 ( in ) ;
if ( mark ! = ARK1 ) continue ;
uint64_t stamp = ark_fget64 ( in ) ;
uint64_t datalen = ark_fget64 ( in ) ;
uint64_t datahash = ark_fget64 ( in ) ;
uint64_t namelen = ark_fget64 ( in ) ;
* z = zip_open_handle ( in , " rb " ) ;
return ;
}
}
2023-07-30 19:18:50 +00:00
static
bool vfs_mount_ ( const char * path , array ( struct vfs_entry ) * entries ) {
2023-10-14 20:00:30 +00:00
const char * path_bak = path ;
2023-07-30 19:18:50 +00:00
zip * z = NULL ; tar * t = NULL ; pak * p = NULL ; dir * d = NULL ;
int is_folder = ( ' / ' = = path [ strlen ( path ) - 1 ] ) ;
if ( is_folder ) d = dir_open ( path , " rb " ) ;
if ( is_folder & & ! d ) return 0 ;
if ( ! is_folder ) z = zip_open ( path , " rb " ) ;
if ( ! is_folder & & ! z ) t = tar_open ( path , " rb " ) ;
if ( ! is_folder & & ! z & & ! t ) p = pak_open ( path , " rb " ) ;
2023-10-13 10:59:44 +00:00
if ( ! is_folder & & ! z & & ! t & & ! p ) ark_list ( path , & z ) ; // last resort. try as .ark
2023-07-30 19:18:50 +00:00
if ( ! is_folder & & ! z & & ! t & & ! p ) return 0 ;
// normalize input -> "././" to ""
while ( path [ 0 ] = = ' . ' & & path [ 1 ] = = ' / ' ) path + = 2 ;
path = STRDUP ( path ) ;
if ( z | | t | | p ) {
// save local path for archives, so we can subtract those from upcoming requests
if ( strrchr ( path , ' / ' ) ) strrchr ( path , ' / ' ) [ 1 ] = ' \0 ' ;
} else if ( d ) 0 [ ( char * ) path ] = 0 ;
// append to mounted archives
archive_dir * prev = dir_mount , zero = { 0 } ;
* ( dir_mount = REALLOC ( 0 , sizeof ( archive_dir ) ) ) = zero ;
dir_mount - > next = prev ;
dir_mount - > path = ( char * ) path ;
dir_mount - > archive = z ? ( void * ) z : t ? ( void * ) t : p ? ( void * ) p : ( void * ) d ;
dir_mount - > type = is_folder ? is_dir : z ? is_zip : t ? is_tar : p ? is_pak : - 1 ;
ASSERT ( dir_mount - > type > = 0 & & dir_mount - > type < 4 ) ;
// append list of files to internal listing
for ( archive_dir * dir = dir_mount ; dir ; dir = 0 ) { // for(archive_dir *dir = dir_mount; dir; dir = dir->next) {
assert ( dir - > type > = 0 & & dir - > type < 4 ) ;
unsigned ( * fn_count [ 4 ] ) ( void * ) = { ( void * ) zip_count , ( void * ) tar_count , ( void * ) pak_count , ( void * ) dir_count } ;
char * ( * fn_name [ 4 ] ) ( void * , unsigned index ) = { ( void * ) zip_name , ( void * ) tar_name , ( void * ) pak_name , ( void * ) dir_name } ;
unsigned ( * fn_size [ 4 ] ) ( void * , unsigned index ) = { ( void * ) zip_size , ( void * ) tar_size , ( void * ) pak_size , ( void * ) dir_size } ;
for ( unsigned idx = 0 , end = fn_count [ dir - > type ] ( dir - > archive ) ; idx < end ; + + idx ) {
assert ( idx < end ) ;
const char * filename = STRDUP ( fn_name [ dir - > type ] ( dir - > archive , idx ) ) ;
const char * fileid = STRDUP ( file_id ( filename ) ) ;
unsigned filesize = fn_size [ dir - > type ] ( dir - > archive , idx ) ;
// printf("%u) %s %u [%s]\n", idx, filename, filesize, fileid);
// append to list
array_push ( * entries , ( struct vfs_entry ) { filename , fileid , filesize } ) ;
}
2023-10-14 20:00:30 +00:00
PRINTF ( " Mounted VFS volume '%s' (%u entries) \n " , path_bak , fn_count [ dir - > type ] ( dir - > archive ) ) ;
2023-07-30 19:18:50 +00:00
}
return 1 ;
}
static
bool vfs_mount_hints ( const char * path ) {
return vfs_mount_ ( path , & vfs_hints ) ;
}
bool vfs_mount ( const char * path ) {
return vfs_mount_ ( path , & vfs_entries ) ;
}
2023-10-15 11:16:35 +00:00
array ( char * ) vfs_list ( const char * masks ) {
static __thread array ( char * ) list = 0 ; // @fixme: add 16 slots
2023-07-30 19:18:50 +00:00
for ( int i = 0 ; i < array_count ( list ) ; + + i ) {
FREE ( list [ i ] ) ;
}
2023-10-15 11:16:35 +00:00
array_resize ( list , 0 ) ;
2023-07-30 19:18:50 +00:00
for each_substring ( masks , " ; " , it ) {
if ( COOK_ON_DEMAND ) // edge case: any game using only vfs api + cook-on-demand flag will never find any file
2023-10-14 19:47:24 +00:00
for each_array ( file_list ( it ) , char * , item ) {
2023-07-30 19:18:50 +00:00
// insert copy
2023-10-14 19:47:24 +00:00
char * copy = STRDUP ( item ) ;
2023-07-30 19:18:50 +00:00
array_push ( list , copy ) ;
}
it = va ( " */%s " , it ) ;
// int recurse = !!strstr(it, "**"); // @fixme: support non-recursive
for ( unsigned i = 0 ; i < array_count ( vfs_entries ) ; + + i ) {
const char * name = vfs_entries [ i ] . name ;
if ( strmatch ( name , it ) ) {
// insert copy
char * copy = STRDUP ( name ) ;
array_push ( list , copy ) ;
}
}
}
// sort alphabetically then remove dupes
array_sort ( list , strcmp ) ;
array_unique ( list , strcmp_qsort ) ;
2023-10-15 11:16:35 +00:00
return list ;
2023-07-30 19:18:50 +00:00
}
static
2023-10-07 17:34:09 +00:00
char * vfs_unpack ( const char * pathfile , int * size ) { // must FREE() after use
2023-07-30 19:18:50 +00:00
// @todo: add cache here
char * data = NULL ;
for ( archive_dir * dir = dir_mount ; dir & & ! data ; dir = dir - > next ) {
if ( dir - > type = = is_dir ) {
#if 0 // sandboxed
char buf [ DIR_MAX ] ;
snprintf ( buf , sizeof ( buf ) , " %s%s " , dir - > path , pathfile ) ;
data = file_load ( buf , size ) ;
# endif
} else {
int ( * fn_find [ 3 ] ) ( void * , const char * ) = { ( void * ) zip_find , ( void * ) tar_find , ( void * ) pak_find } ;
void * ( * fn_unpack [ 3 ] ) ( void * , unsigned ) = { ( void * ) zip_extract , ( void * ) tar_extract , ( void * ) pak_extract } ;
unsigned ( * fn_size [ 3 ] ) ( void * , unsigned ) = { ( void * ) zip_size , ( void * ) tar_size , ( void * ) pak_size } ;
#if 0
const char * cleanup = pathfile + strbegi ( pathfile , dir - > path ) * strlen ( dir - > path ) ;
while ( cleanup [ 0 ] = = ' / ' ) + + cleanup ;
# else
const char * cleanup = pathfile ;
# endif
int index = fn_find [ dir - > type ] ( dir - > archive , cleanup ) ;
data = fn_unpack [ dir - > type ] ( dir - > archive , index ) ;
if ( size ) * size = fn_size [ dir - > type ] ( dir - > archive , index ) ;
}
// printf("%c trying %s in %s ...\n", data ? 'Y':'N', pathfile, dir->path);
}
//wait_ms(1000); // <-- simulate slow hdd
return data ;
}
const char * vfs_resolve ( const char * pathfile ) {
// we dont resolve absolute paths. they dont belong to the vfs
// if( pathfile[0] == '/' || pathfile[0] == '\\' || pathfile[1] == ':' ) return pathfile;
char * id = file_id ( pathfile ) ;
// find best match (vfs_entries first)
for ( int i = array_count ( vfs_entries ) ; - - i > = 0 ; ) {
if ( strbegi ( vfs_entries [ i ] . id , id ) ) {
return vfs_entries [ i ] . name ;
}
}
// find best match (vfs_hints later)
for ( int i = array_count ( vfs_hints ) ; - - i > = 0 ; ) {
if ( strbegi ( vfs_hints [ i ] . id , id ) ) {
return vfs_hints [ i ] . name ;
}
}
return pathfile ;
}
char * vfs_load ( const char * pathfile , int * size_out ) { // @todo: fix leaks, vfs_unpack()
// @fixme: handle \\?\ absolute path (win)
if ( ! pathfile [ 0 ] ) return file_load ( pathfile , size_out ) ;
while ( pathfile [ 0 ] = = ' . ' & & ( pathfile [ 1 ] = = ' / ' | | pathfile [ 1 ] = = ' \\ ' ) ) pathfile + = 2 ;
2023-08-10 14:30:56 +00:00
// if (pathfile[0] == '/' || pathfile[1] == ':') return file_load(pathfile, size_out); // @fixme: handle current cooked /home/V4K or C:/V4K path cases within zipfiles
2023-07-30 19:18:50 +00:00
if ( size_out ) * size_out = 0 ;
if ( strend ( pathfile , " / " ) ) return 0 ; // it's a dir
static __thread map ( char * , int ) misses = 0 , * init = 0 ; if ( ! init ) init = misses , map_init ( misses , less_str , hash_str ) ;
int * found = map_find_or_add_allocated_key ( misses , STRDUP ( pathfile ) , - 1 ) ; // [-1]non-init,[false]could not cook,[true]cooked
if ( found & & * found = = 0 ) {
return 0 ;
}
//{
// exclude garbage from material names
// @todo: exclude double slashs in paths
char * base = file_name ( pathfile ) ; if ( strchr ( base , ' + ' ) ) base = strchr ( base , ' + ' ) + 1 ;
if ( base [ 0 ] = = ' \0 ' ) return 0 ; // it's a dir
char * folder = file_path ( pathfile ) ;
pathfile = va ( " %s%s " , folder , base ) ;
// solve virtual path
pathfile = va ( " %s " , vfs_resolve ( pathfile ) ) ;
base = file_name ( pathfile ) ;
if ( base [ 0 ] = = ' \0 ' ) return 0 ; // it's a dir
folder = file_path ( pathfile ) ;
2023-10-10 08:52:29 +00:00
// ease folders reading by shortening them: /home/rlyeh/prj/v4k/art/demos/audio/coin.wav -> demos/audio/coin.wav
// or C:/prj/v4k/engine/art/fonts/B612-BoldItalic.ttf -> fonts/B612-BoldItalic.ttf
2023-10-14 19:47:24 +00:00
static __thread array ( char * ) art_paths = 0 ;
if ( ! art_paths ) for each_substring ( ART , " , " , stem ) array_push ( art_paths , STRDUP ( stem ) ) ;
2023-10-10 08:52:29 +00:00
char * pretty_folder = " " ;
if ( folder ) for ( int i = 0 ; i < array_count ( art_paths ) ; + + i ) {
if ( strbeg ( folder , art_paths [ i ] ) ) { pretty_folder = folder + strlen ( art_paths [ i ] ) ; break ; }
}
2023-07-30 19:18:50 +00:00
//}
int size = 0 ;
void * ptr = 0 ;
#if 0
// clean pathfile
while ( pathfile [ 0 ] = = ' . ' & & pathfile [ 1 ] = = ' / ' ) pathfile + = 2 ;
while ( pathfile [ 0 ] = = ' / ' ) + + pathfile ;
# endif
const char * lookup_id = /*file_normalize_with_folder*/ ( pathfile ) ;
// search (last item)
2023-10-13 10:59:44 +00:00
static __thread char last_item [ 256 ] = { 0 } ;
static __thread void * last_ptr = 0 ;
static __thread int last_size = 0 ;
2023-07-30 19:18:50 +00:00
if ( ! strcmpi ( lookup_id , last_item ) ) {
ptr = last_ptr ;
size = last_size ;
}
// search (cache)
if ( ! ptr ) {
ptr = cache_lookup ( lookup_id , & size ) ;
}
if ( ptr ) {
PRINTF ( " Loading VFS (%s)%s (cached) \n " , pretty_folder , base ) ;
} else {
PRINTF ( " Loading VFS (%s)%s \n " , pretty_folder , base ) ;
}
// read cooked asset from mounted disks
if ( ! ptr ) {
ptr = vfs_unpack ( pathfile , & size ) ;
// asset not found? maybe it has not been cooked yet at this point (see --cook-on-demand)
2023-09-25 04:29:29 +00:00
if ( ! ptr & & COOK_ON_DEMAND ) {
2023-07-30 19:18:50 +00:00
static thread_mutex_t mutex , * init = 0 ; if ( ! init ) thread_mutex_init ( init = & mutex ) ;
thread_mutex_lock ( & mutex ) ;
// this block saves some boot time (editor --cook-on-demand: boot 1.50s -> 0.90s)
# if 1 // EXPERIMENTAL_DONT_COOK_NON_EXISTING_ASSETS
static set ( char * ) disk = 0 ;
2023-10-14 19:47:24 +00:00
if ( ! disk ) { set_init_str ( disk ) ; for each_substring ( ART , " , " , art_folder ) for each_array ( file_list ( va ( " %s** " , art_folder ) ) , char * , item ) set_insert ( disk , STRDUP ( item ) ) ; } // art_folder ends with '/'
2023-07-30 19:18:50 +00:00
int found = ! ! set_find ( disk , ( char * ) pathfile ) ;
if ( found )
# endif
{
// technically, we should only cook `base` asset at this point. however, cooks on demand can be very
// expensive. not only because we're invoking an external tool/cook app in here, which is scanning all the
// cook folders at every call, but also because there is this expensive vfs_reload() call at end of current scope.
// so, in order to minimize the number of upcoming cook requests, we cook more stuff than needed at this point;
// just in anticipation of what's likely going to happen in the next frames.
// so, when asked to cook "models/model.fbx" we actually:
// - do cook "models/model.* (so, "model.tga","model.png","model.jpg","model.mtl",etc. are also cooked)
// - do cook "models/*.fbx" as well
char * dir = file_path ( pathfile + ART_SKIP_ROOT ) ;
char * group1 = dir [ 0 ] ? va ( " \" */%s%s.* \" " , dir , file_base ( pathfile ) ) : base ; // -> "*/models/model.*"
char * group2 = dir [ 0 ] ? va ( " \" */%s*%s \" " , dir , file_ext ( pathfile ) ) : " " ; // -> "*/models/*.fbx"
char * cmd = va ( " %scook " ifdef ( osx , " .osx " , ifdef ( linux , " .linux " , " .exe " ) ) " %s %s --cook-ini=%s --cook-additive --cook-jobs=1 --quiet " , TOOLS , group1 , group2 , COOK_INI ) ;
// cook groups
int rc = system ( cmd ) ;
if ( rc < 0 ) PANIC ( " cannot invoke `%scook` (return code %d) " , TOOLS , rc ) ;
vfs_reload ( ) ; // @todo: optimize me. it is waaay inefficent to reload the whole VFS layout after cooking a single asset
}
thread_mutex_unlock ( & mutex ) ;
// finally, try again
pathfile = va ( " %s " , vfs_resolve ( pathfile ) ) ;
ptr = vfs_unpack ( pathfile , & size ) ;
}
if ( ptr ) {
cache_insert ( lookup_id , ptr , size ) ;
}
}
if ( ptr & & size )
if ( ptr ! = last_ptr ) {
snprintf ( last_item , 256 , " %s " , lookup_id ) ;
last_ptr = ptr ;
last_size = size ;
}
2023-09-11 10:23:33 +00:00
// yet another last resort: redirect vfs_load() calls to file_load()
// (for environments without tools or cooked assets)
if ( ! ptr ) {
2023-10-15 11:16:35 +00:00
if ( ! have_tools ( ) ) {
2023-09-11 10:23:33 +00:00
ptr = file_load ( pathfile , size_out ) ;
}
}
2023-07-30 19:18:50 +00:00
if ( ! ptr ) {
PRINTF ( " Loading %s (not found) \n " , pathfile ) ;
}
* found = ptr ? true : false ;
if ( size_out ) * size_out = ptr ? size : 0 ;
return ptr ;
}
char * vfs_read ( const char * pathfile ) {
return vfs_load ( pathfile , NULL ) ;
}
int vfs_size ( const char * pathfile ) {
int sz ;
return vfs_load ( pathfile , & sz ) , sz ;
}
FILE * vfs_handle ( const char * pathfile ) {
// @fixme: non-compressible assets (adpcm wavs,mp3,ogg,mp4,avi,...) are taking a suboptimal code path here.
// no need to unzip them. just seek within the zipfile and return the *fp at that position
int sz ;
char * buf = vfs_load ( pathfile , & sz ) ;
FILE * fp = fmemopen ( buf ? buf : " " , buf ? sz : 0 , " r+b " ) ;
ASSERT ( fp , " cannot create tempfile " ) ;
return fp ;
}
#if 0
const char * vfs_extract ( const char * pathfile ) { // extract a vfs file into the local (temp) filesystem
#if 0
FILE * fp = vfs_handle ( pathfile ) ;
return fp ? pathfile_from_handle ( fp ) : " " ;
# else
int sz ;
char * buf = vfs_load ( pathfile , & sz ) ;
if ( ! buf ) return " " ;
// pool of temp files. recycles after every loop
enum { MAX_TEMP_FILES = 16 } ;
static __thread char temps [ MAX_TEMP_FILES ] [ DIR_MAX ] = { 0 } ;
static __thread int i = 0 ;
if ( temps [ i ] [ 0 ] ) unlink ( temps [ i ] ) ;
i = ( i + 1 ) % MAX_TEMP_FILES ;
if ( ! temps [ i ] [ 0 ] ) snprintf ( temps [ i ] , DIR_MAX , " %s " , file_tempname ( ) ) ;
char * name = temps [ i ] ;
FILE * tmp = fopen ( name , " wb " ) ; //unlink(name);
ASSERT ( tmp , " cannot create tempfile %s " , name ) ;
fwrite ( buf ? buf : " " , 1 , buf ? sz : 0 , tmp ) ;
fclose ( tmp ) ;
return name ;
# endif
}
# endif
// -----------------------------------------------------------------------------
// cache
2023-10-14 19:47:24 +00:00
static thread_mutex_t cache_mutex ; AUTORUN { thread_mutex_init ( & cache_mutex ) ; }
2023-07-30 19:18:50 +00:00
void * cache_lookup ( const char * pathfile , int * size ) { // find key->value
if ( ! MAX_CACHED_FILES ) return 0 ;
2023-10-14 19:47:24 +00:00
void * data = 0 ;
thread_mutex_lock ( & cache_mutex ) ;
2023-07-30 19:18:50 +00:00
for ( archive_dir * dir = dir_cache ; dir ; dir = dir - > next ) {
if ( ! strcmp ( dir - > path , pathfile ) ) {
if ( size ) * size = dir - > size ;
2023-10-14 19:47:24 +00:00
data = dir - > data ;
break ;
2023-07-30 19:18:50 +00:00
}
}
2023-10-14 19:47:24 +00:00
thread_mutex_unlock ( & cache_mutex ) ;
return data ;
2023-07-30 19:18:50 +00:00
}
void * cache_insert ( const char * pathfile , void * ptr , int size ) { // append key/value; return LRU or NULL
if ( ! MAX_CACHED_FILES ) return 0 ;
if ( ! ptr | | ! size ) return 0 ;
2023-10-13 10:59:44 +00:00
// keep cached files within limits
2023-10-14 19:47:24 +00:00
thread_mutex_lock ( & cache_mutex ) ;
2023-10-13 10:59:44 +00:00
2023-07-30 19:18:50 +00:00
// append to cache
archive_dir zero = { 0 } , * old = dir_cache ;
* ( dir_cache = REALLOC ( 0 , sizeof ( archive_dir ) ) ) = zero ;
dir_cache - > next = old ;
dir_cache - > path = STRDUP ( pathfile ) ;
dir_cache - > size = size ;
dir_cache - > data = REALLOC ( 0 , size + 1 ) ;
memcpy ( dir_cache - > data , ptr , size ) ; size [ ( char * ) dir_cache - > data ] = 0 ; // copy+terminator
2023-10-13 10:59:44 +00:00
void * found = 0 ;
2023-07-30 19:18:50 +00:00
static int added = 0 ;
if ( added < MAX_CACHED_FILES ) {
+ + added ;
} else {
// remove oldest cache entry
for ( archive_dir * prev = dir_cache , * dir = prev ; dir ; prev = dir , dir = dir - > next ) {
if ( ! dir - > next ) {
prev - > next = 0 ; // break link
2023-10-13 10:59:44 +00:00
found = dir - > data ;
2023-07-30 19:18:50 +00:00
dir - > path = REALLOC ( dir - > path , 0 ) ;
dir - > data = REALLOC ( dir - > data , 0 ) ;
dir = REALLOC ( dir , 0 ) ;
2023-10-13 10:59:44 +00:00
break ;
2023-07-30 19:18:50 +00:00
}
}
}
2023-10-13 10:59:44 +00:00
2023-10-14 19:47:24 +00:00
thread_mutex_unlock ( & cache_mutex ) ;
2023-10-13 10:59:44 +00:00
return found ;
2023-07-30 19:18:50 +00:00
}
// ----------------------------------------------------------------------------
// ini
/* ini+, extended ini format
// - rlyeh, public domain
//
// # spec
//
// ; line comment
// [user] ; map section name (optional)
// name=john ; key and value (mapped here as user.name=john)
// +surname=doe jr. ; sub-key and value (mapped here as user.name.surname=doe jr.)
// age=30 ; numbers
/ / color = 240 ; key and value \
// color=253 ; key and value |> array: color[0], color[1] and color[2]
// color=255 ; key and value /
// color= ; remove key/value(s)
// color=white ; recreate key; color[1] and color[2] no longer exist
// [] ; unmap section
// -note=keys may start with symbols (except plus and semicolon)
// -note=linefeeds are either \r, \n or \r\n.
// -note=utf8 everywhere.
*/
static
char * ini_parse ( const char * s ) {
char * map = 0 ;
int mapcap = 0 , maplen = 0 ;
enum { DEL , REM , TAG , KEY , SUB , VAL } fsm = DEL ;
const char * cut [ 6 ] = { 0 } , * end [ 6 ] = { 0 } ;
while ( * s ) {
while ( * s & & ( * s = = ' ' | | * s = = ' \t ' | | * s = = ' \r ' | | * s = = ' \n ' ) ) + + s ;
/**/ if ( * s = = ' ; ' ) cut [ fsm = REM ] = + + s ;
else if ( * s = = ' [ ' ) cut [ fsm = TAG ] = + + s ;
else if ( * s = = ' + ' ) cut [ fsm = SUB ] = + + s ;
else if ( * s = = ' = ' ) cut [ fsm = VAL ] = + + s ;
else if ( * s > ' ' & & * s < = ' z ' & & * s ! = ' ] ' ) cut [ fsm = KEY ] = cut [ SUB ] = end [ SUB ] = s ;
else { if ( * s ) + + s ; continue ; }
/**/ if ( fsm = = REM ) { while ( * s & & * s ! = ' \r ' & & * s ! = ' \n ' ) + + s ; }
else if ( fsm = = TAG ) { while ( * s & & * s ! = ' \r ' & & * s ! = ' \n ' & & * s ! = ' ] ' ) + + s ; end [ TAG ] = s ; }
else if ( fsm = = KEY ) { while ( * s & & * s > ' ' & & * s < = ' z ' & & * s ! = ' = ' ) + + s ; end [ KEY ] = s ; }
else if ( fsm = = SUB ) { while ( * s & & * s > ' ' & & * s < = ' z ' & & * s ! = ' = ' ) + + s ; end [ SUB ] = s ; }
else if ( fsm = = VAL ) { while ( * s & & * s > = ' ' & & * s < = 127 & & * s ! = ' ; ' ) + + s ; end [ VAL ] = s ;
while ( end [ VAL ] [ - 1 ] < = ' ' ) { - - end [ VAL ] ; }
char buf [ 256 ] = { 0 } , * key = buf ;
if ( end [ TAG ] - cut [ TAG ] ) key + = sprintf ( key , " %.*s. " , ( int ) ( end [ TAG ] - cut [ TAG ] ) , cut [ TAG ] ) ;
if ( end [ KEY ] - cut [ KEY ] ) key + = sprintf ( key , " %.*s " , ( int ) ( end [ KEY ] - cut [ KEY ] ) , cut [ KEY ] ) ;
if ( end [ SUB ] - cut [ SUB ] ) key + = sprintf ( key , " .%.*s " , ( int ) ( end [ SUB ] - cut [ SUB ] ) , cut [ SUB ] ) ;
int reqlen = ( key - buf ) + 1 + ( end [ VAL ] - cut [ VAL ] ) + 1 + 1 ;
if ( ( reqlen + maplen ) > = mapcap ) map = REALLOC ( map , mapcap + = reqlen + 512 ) ;
sprintf ( map + maplen , " %.*s%c%.*s%c%c " , ( int ) ( key - buf ) , buf , 0 , ( int ) ( end [ VAL ] - cut [ VAL ] ) , cut [ VAL ] , 0 , 0 ) ;
maplen + = reqlen - 1 ;
}
}
return map ;
}
// @todo: evaluate alt api
// int count = ini_count(filename);
// char *key = ini_key(filename, id);
// char *val = ini_val(filename, id);
void ini_destroy ( ini_t x ) {
for each_map ( x , char * , k , char * , v ) {
FREE ( k ) ;
FREE ( v ) ;
}
map_free ( x ) ;
}
ini_t ini_from_mem ( const char * data ) {
if ( ! data | | ! data [ 0 ] ) return 0 ;
char * kv = ini_parse ( data ) ;
if ( ! kv ) return 0 ;
ini_t map = 0 ;
map_init ( map , less_str , hash_str ) ;
for ( char * iter = kv ; iter [ 0 ] ; ) {
char * key = iter ; while ( * iter + + ) ;
char * val = iter ; while ( * iter + + ) ;
char * * found = map_find ( map , key ) ;
if ( ! found ) map_insert ( map , STRDUP ( key ) , STRDUP ( val ) ) ;
assert ( map_find ( map , key ) ) ;
}
FREE ( kv ) ;
return map ;
}
ini_t ini ( const char * filename ) {
return ini_from_mem ( file_read ( filename ) ) ;
}
bool ini_write ( const char * filename , const char * section , const char * key , const char * value ) {
// this is a little hacky {
char * data = file_read ( filename ) ;
if ( data & & data [ 0 ] ) {
char * begin = strrchr ( data , ' [ ' ) ;
char * end = strrchr ( data , ' ] ' ) ;
if ( begin & & end & & begin < end ) {
char * last_section = va ( " %.*s " , ( int ) ( end - begin - 1 ) , begin + 1 ) ;
if ( ! strcmpi ( last_section , section ) ) section = 0 ;
}
}
// }
char * s = va ( " %s%s=%s \r \n " , section ? va ( " [%s] \r \n " , section ) : " " , key , value ) ;
return file_append ( filename , s , strlen ( s ) ) ;
}