Sweden-Number/tools/winedump/nls.c

366 lines
12 KiB
C

/*
* Dump a NLS file
*
* Copyright 2020 Alexandre Julliard
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
*/
#include "config.h"
#include "wine/port.h"
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include "windef.h"
#include "winedump.h"
static const void *read_data( unsigned int *pos, unsigned int size )
{
const void *ret = PRD( *pos, size );
*pos += size;
return ret;
}
static unsigned short casemap( const unsigned short *table, unsigned int len, unsigned short ch )
{
unsigned int off = table[ch >> 8] + ((ch >> 4) & 0x0f);
if (off >= len) return 0;
off = table[off] + (ch & 0x0f);
if (off >= len) return 0;
return ch + table[off];
}
static void dump_casemap(void)
{
int i, ch;
unsigned int pos = 0, upper_len, lower_len;
const unsigned short *header, *upper, *lower;
if (!(header = read_data( &pos, 2 * sizeof(*header) ))) return;
upper_len = header[1];
if (!(upper = read_data( &pos, upper_len * sizeof(*upper) )))
{
printf( "Invalid len %04x\n", header[1] );
return;
}
lower_len = dump_total_len / sizeof(*lower) - 2 - upper_len;
if (!(lower = read_data( &pos, lower_len * sizeof(*lower) ))) return;
printf( "Magic: %04x\n", header[0] );
printf( "Upper-case table:\n" );
for (i = 0; i < 0x10000; i++)
{
if (!(i % 16)) printf( "\n%04x:", i );
ch = casemap( upper, upper_len, i );
if (ch == i) printf( " ...." );
else printf( " %04x", ch );
}
printf( "\n\nLower-case table:\n" );
for (i = 0; i < 0x10000; i++)
{
if (!(i % 16)) printf( "\n%04x:", i );
ch = casemap( lower, lower_len, i );
if (ch == i) printf( " ...." );
else printf( " %04x", ch );
}
printf( "\n\n" );
}
static void dump_codepage(void)
{
unsigned int i, j, uni2cp_offset, pos = 0;
const unsigned short *header, *ptr;
if (!(header = read_data( &pos, 13 * sizeof(*header) ))) return;
printf( "Codepage: %03u\n", header[1] );
printf( "Char size: %u\n", header[2] );
printf( "Default char A: %04x / %04x\n", header[3], header[5] );
printf( "Default char W: %04x / %04x\n", header[4], header[6] );
if (header[2] == 2)
{
printf( "Lead bytes: " );
for (i = 0; i < 12; i++)
{
unsigned char val = ((unsigned char *)(header + 7))[i];
if (!val) break;
printf( "%c%02x", (i % 2) ? '-' : ' ', val );
}
printf( "\n" );
}
printf( "\nCharacter map:\n" );
pos = header[0] * sizeof(*ptr);
if (!(ptr = read_data( &pos, sizeof(*ptr) ))) return;
uni2cp_offset = pos / sizeof(*ptr) + *ptr;
if (!(ptr = read_data( &pos, 256 * sizeof(*ptr) ))) return;
for (i = 0; i < 256; i++)
{
if (!(i % 16)) printf( "\n%02x:", i );
printf( " %04x", ptr[i] );
}
printf( "\n" );
if (!(ptr = read_data( &pos, sizeof(*ptr) ))) return;
if (*ptr == 256)
{
if (!(ptr = read_data( &pos, 256 * sizeof(*ptr) ))) return;
printf( "\nGlyph table:\n" );
for (i = 0; i < 256; i++)
{
if (!(i % 16)) printf( "\n%02x:", i );
printf( " %04x", ptr[i] );
}
printf( "\n" );
}
if (!(ptr = read_data( &pos, sizeof(*ptr) ))) return;
if (*ptr)
{
if (!(ptr = read_data( &pos, (uni2cp_offset - pos) * sizeof(*ptr) ))) return;
for (i = 0; i < 256; i++)
{
if (!ptr[i] || ptr[i] > pos - 256) continue;
for (j = 0; j < 256; j++)
{
if (!(j % 16)) printf( "\n%02x%02x:", i, j );
printf( " %04x", ptr[ptr[i] + j] );
}
}
printf( "\n" );
}
printf( "\nUnicode table:\n" );
pos = uni2cp_offset * sizeof(*ptr);
if (header[2] == 2)
{
if (!(ptr = read_data( &pos, 65536 * sizeof(*ptr) ))) return;
for (i = 0; i < 65536; i++)
{
if (!(i % 16)) printf( "\n%04x:", i );
printf( " %04x", ptr[i] );
}
printf( "\n" );
}
else
{
const unsigned char *uni2cp;
if (!(uni2cp = read_data( &pos, 65536 ))) return;
for (i = 0; i < 65536; i++)
{
if (!(i % 16)) printf( "\n%04x:", i );
printf( " %02x", uni2cp[i] );
}
printf( "\n" );
}
printf( "\n" );
}
struct norm_table
{
WCHAR name[13]; /* 00 file name */
USHORT checksum[3]; /* 1a checksum? */
USHORT version[4]; /* 20 Unicode version */
USHORT form; /* 28 normalization form */
USHORT len_factor; /* 2a factor for length estimates */
USHORT unknown1; /* 2c */
USHORT decomp_size; /* 2e decomposition hash size */
USHORT comp_size; /* 30 composition hash size */
USHORT unknown2; /* 32 */
USHORT classes; /* 34 combining classes table offset */
USHORT props_level1; /* 36 char properties table level 1 offset */
USHORT props_level2; /* 38 char properties table level 2 offset */
USHORT decomp_hash; /* 3a decomposition hash table offset */
USHORT decomp_map; /* 3c decomposition character map table offset */
USHORT decomp_seq; /* 3e decomposition character sequences offset */
USHORT comp_hash; /* 40 composition hash table offset */
USHORT comp_seq; /* 42 composition character sequences offset */
};
static int offset_scale = 1; /* older versions use byte offsets */
#define GET_TABLE(info,table) ((const void *)((const BYTE *)info + (info->table * offset_scale)))
static unsigned int get_utf16( const WCHAR *str )
{
if (str[0] >= 0xd800 && str[0] <= 0xdbff &&
str[1] >= 0xdc00 && str[1] <= 0xdfff)
return 0x10000 + ((str[0] & 0x3ff) << 10) + (str[1] & 0x3ff);
return str[0];
}
static BYTE rol( BYTE val, BYTE count )
{
return (val << count) | (val >> (8 - count));
}
static unsigned char get_char_props( const struct norm_table *info, unsigned int ch )
{
const BYTE *level1 = GET_TABLE( info, props_level1 );
const BYTE *level2 = GET_TABLE( info, props_level2 );
BYTE off = level1[ch / 128];
if (!off || off >= 0xfb) return rol( off, 5 );
return level2[(off - 1) * 128 + ch % 128];
}
static const WCHAR *get_decomposition( const struct norm_table *info,
unsigned int ch, unsigned int *ret_len )
{
const USHORT *hash_table = GET_TABLE( info, decomp_hash );
const WCHAR *seq = GET_TABLE(info, decomp_seq );
const WCHAR *ret;
unsigned int i, pos, end, len, hash;
*ret_len = 1 + (ch >= 0x10000);
if (!info->decomp_size) return NULL;
hash = ch % info->decomp_size;
pos = hash_table[hash];
if (pos >> 13)
{
if (get_char_props( info, ch ) != 0xbf) return NULL;
ret = seq + (pos & 0x1fff);
len = pos >> 13;
}
else
{
const struct { WCHAR src; USHORT dst; } *pairs = GET_TABLE( info, decomp_map );
/* find the end of the hash bucket */
for (i = hash + 1; i < info->decomp_size; i++) if (!(hash_table[i] >> 13)) break;
if (i < info->decomp_size) end = hash_table[i];
else for (end = pos; pairs[end].src; end++) ;
for ( ; pos < end; pos++)
{
if (pairs[pos].src != (WCHAR)ch) continue;
ret = seq + (pairs[pos].dst & 0x1fff);
len = pairs[pos].dst >> 13;
break;
}
if (pos >= end) return NULL;
}
if (len == 7) while (ret[len]) len++;
*ret_len = len;
return ret;
}
static int cmp_compos( const void *a, const void *b )
{
int ret = ((unsigned int *)a)[0] - ((unsigned int *)b)[0];
if (!ret) ret = ((unsigned int *)a)[1] - ((unsigned int *)b)[1];
return ret;
}
static void dump_norm(void)
{
const struct norm_table *info;
const BYTE *classes;
unsigned int i;
char name[13];
if (!(info = PRD( 0, sizeof(*info) ))) return;
for (i = 0; i < sizeof(name); i++) name[i] = info->name[i];
printf( "Name: %s\n", name );
switch (info->form)
{
case 1: printf( "Form: NFC\n" ); break;
case 2: printf( "Form: NFD\n" ); break;
case 5: printf( "Form: NFKC\n" ); break;
case 6: printf( "Form: NFKD\n" ); break;
case 13: printf( "Form: IDNA\n" ); break;
default: printf( "Form: %u\n", info->form ); break;
}
printf( "Version: %u.%u.%u\n", info->version[0], info->version[1], info->version[2] );
printf( "Factor: %u\n", info->len_factor );
if (info->classes == sizeof(*info) / 2) offset_scale = 2;
classes = GET_TABLE( info, classes );
printf( "\nCharacter classes:\n" );
for (i = 0; i < 0x110000; i++)
{
BYTE flags = get_char_props( info, i );
if (!(i % 16)) printf( "\n%06x:", i );
if (!flags || (flags & 0x3f) == 0x3f)
{
static const char *flagstr[4] = { ".....", "Undef", "QC=No", "Inval" };
printf( " %s", flagstr[flags >> 6] );
}
else
{
static const char flagschar[4] = ".+*M";
BYTE class = classes[flags & 0x3f];
printf( " %c.%03u", flagschar[flags >> 6], class );
}
}
printf( "\n\nDecompositions:\n\n" );
for (i = 0; i < 0x110000; i++)
{
unsigned int j, len;
const WCHAR *decomp = get_decomposition( info, i, &len );
if (!decomp) continue;
printf( "%04x ->", i );
for (j = 0; j < len; j++)
{
unsigned int ch = get_utf16( decomp + j );
printf( " %04x", ch );
if (ch >= 0x10000) j++;
}
printf( "\n" );
}
if (info->comp_size)
{
unsigned int pos, len = (dump_total_len - info->comp_seq * offset_scale) / sizeof(WCHAR);
const WCHAR *seq = GET_TABLE( info, comp_seq );
unsigned int *map = malloc( len * sizeof(*map) );
printf( "\nCompositions:\n\n" );
/* ignore hash table, simply dump all the sequences */
for (i = pos = 0; i < len; pos += 3)
{
map[pos] = get_utf16( seq + i );
i += 1 + (map[pos] >= 0x10000);
map[pos+1] = get_utf16( seq + i );
i += 1 + (map[pos+1] >= 0x10000);
map[pos+2] = get_utf16( seq + i );
i += 1 + (map[pos+2] >= 0x10000);
}
qsort( map, pos / 3, 3 * sizeof(*map), cmp_compos );
for (i = 0; i < pos; i += 3) printf( "%04x %04x -> %04x\n", map[i], map[i + 1], map[i + 2] );
free( map );
}
printf( "\n" );
}
void nls_dump(void)
{
const char *name = strrchr( globals.input_name, '/' );
if (name) name++;
else name = globals.input_name;
if (!strcasecmp( name, "l_intl.nls" )) return dump_casemap();
if (!strncasecmp( name, "c_", 2 )) return dump_codepage();
if (!strncasecmp( name, "norm", 4 )) return dump_norm();
fprintf( stderr, "Unrecognized file name '%s'\n", globals.input_name );
}
enum FileSig get_kind_nls(void)
{
if (strlen( globals.input_name ) < 5) return SIG_UNKNOWN;
if (strcasecmp( globals.input_name + strlen(globals.input_name) - 4, ".nls" )) return SIG_UNKNOWN;
return SIG_NLS;
}