2008-12-31 12:27:24 +01:00
|
|
|
/*
|
|
|
|
* X Files parsing
|
|
|
|
*
|
|
|
|
* Copyright 2008 Christian Costa
|
|
|
|
*
|
|
|
|
* 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"
|
2014-03-31 11:44:56 +02:00
|
|
|
#include "wine/port.h"
|
2008-12-31 12:27:24 +01:00
|
|
|
#include "wine/debug.h"
|
|
|
|
|
|
|
|
#define COBJMACROS
|
|
|
|
|
|
|
|
#include "winbase.h"
|
|
|
|
#include "wingdi.h"
|
|
|
|
|
|
|
|
#include "d3dxof_private.h"
|
|
|
|
#include "dxfile.h"
|
|
|
|
|
|
|
|
#include <stdio.h>
|
|
|
|
|
2010-02-01 08:18:24 +01:00
|
|
|
WINE_DEFAULT_DEBUG_CHANNEL(d3dxof_parsing);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
2011-06-08 21:38:58 +02:00
|
|
|
#define MAKEFOUR(a,b,c,d) ((DWORD)a + ((DWORD)b << 8) + ((DWORD)c << 16) + ((DWORD)d << 24))
|
|
|
|
#define XOFFILE_FORMAT_MAGIC MAKEFOUR('x','o','f',' ')
|
|
|
|
#define XOFFILE_FORMAT_VERSION_302 MAKEFOUR('0','3','0','2')
|
|
|
|
#define XOFFILE_FORMAT_VERSION_303 MAKEFOUR('0','3','0','3')
|
|
|
|
#define XOFFILE_FORMAT_BINARY MAKEFOUR('b','i','n',' ')
|
|
|
|
#define XOFFILE_FORMAT_TEXT MAKEFOUR('t','x','t',' ')
|
|
|
|
#define XOFFILE_FORMAT_BINARY_MSZIP MAKEFOUR('b','z','i','p')
|
|
|
|
#define XOFFILE_FORMAT_TEXT_MSZIP MAKEFOUR('t','z','i','p')
|
|
|
|
#define XOFFILE_FORMAT_COMPRESSED MAKEFOUR('c','m','p',' ')
|
|
|
|
#define XOFFILE_FORMAT_FLOAT_BITS_32 MAKEFOUR('0','0','3','2')
|
|
|
|
#define XOFFILE_FORMAT_FLOAT_BITS_64 MAKEFOUR('0','0','6','4')
|
|
|
|
|
2013-04-28 22:46:09 +02:00
|
|
|
#define TOKEN_ERROR 0xffff
|
|
|
|
#define TOKEN_NONE 0
|
2008-12-31 12:27:24 +01:00
|
|
|
#define TOKEN_NAME 1
|
|
|
|
#define TOKEN_STRING 2
|
|
|
|
#define TOKEN_INTEGER 3
|
|
|
|
#define TOKEN_GUID 5
|
|
|
|
#define TOKEN_INTEGER_LIST 6
|
|
|
|
#define TOKEN_FLOAT_LIST 7
|
|
|
|
#define TOKEN_OBRACE 10
|
|
|
|
#define TOKEN_CBRACE 11
|
|
|
|
#define TOKEN_OPAREN 12
|
|
|
|
#define TOKEN_CPAREN 13
|
|
|
|
#define TOKEN_OBRACKET 14
|
|
|
|
#define TOKEN_CBRACKET 15
|
|
|
|
#define TOKEN_OANGLE 16
|
|
|
|
#define TOKEN_CANGLE 17
|
|
|
|
#define TOKEN_DOT 18
|
|
|
|
#define TOKEN_COMMA 19
|
|
|
|
#define TOKEN_SEMICOLON 20
|
|
|
|
#define TOKEN_TEMPLATE 31
|
|
|
|
#define TOKEN_WORD 40
|
|
|
|
#define TOKEN_DWORD 41
|
|
|
|
#define TOKEN_FLOAT 42
|
|
|
|
#define TOKEN_DOUBLE 43
|
|
|
|
#define TOKEN_CHAR 44
|
|
|
|
#define TOKEN_UCHAR 45
|
|
|
|
#define TOKEN_SWORD 46
|
|
|
|
#define TOKEN_SDWORD 47
|
|
|
|
#define TOKEN_VOID 48
|
|
|
|
#define TOKEN_LPSTR 49
|
|
|
|
#define TOKEN_UNICODE 50
|
|
|
|
#define TOKEN_CSTRING 51
|
|
|
|
#define TOKEN_ARRAY 52
|
|
|
|
|
|
|
|
#define CLSIDFMT "<%08X-%04X-%04X-%02X%02X-%02X%02X%02X%02X%02X%02X>"
|
|
|
|
|
2011-06-08 21:38:58 +02:00
|
|
|
/* FOURCC to string conversion for debug messages */
|
|
|
|
static const char *debugstr_fourcc(DWORD fourcc)
|
|
|
|
{
|
|
|
|
if (!fourcc) return "'null'";
|
|
|
|
return wine_dbg_sprintf ("\'%c%c%c%c\'",
|
|
|
|
(char)(fourcc), (char)(fourcc >> 8),
|
|
|
|
(char)(fourcc >> 16), (char)(fourcc >> 24));
|
|
|
|
}
|
|
|
|
|
2014-03-31 11:44:55 +02:00
|
|
|
static const char* get_primitive_string(DWORD token)
|
2009-01-07 12:35:08 +01:00
|
|
|
{
|
|
|
|
switch(token)
|
|
|
|
{
|
|
|
|
case TOKEN_WORD:
|
|
|
|
return "WORD";
|
|
|
|
case TOKEN_DWORD:
|
|
|
|
return "DWORD";
|
|
|
|
case TOKEN_FLOAT:
|
|
|
|
return "FLOAT";
|
|
|
|
case TOKEN_DOUBLE:
|
|
|
|
return "DOUBLE";
|
|
|
|
case TOKEN_CHAR:
|
|
|
|
return "CHAR";
|
|
|
|
case TOKEN_UCHAR:
|
|
|
|
return "UCHAR";
|
|
|
|
case TOKEN_SWORD:
|
|
|
|
return "SWORD";
|
|
|
|
case TOKEN_SDWORD:
|
|
|
|
return "SDWORD";
|
|
|
|
case TOKEN_VOID:
|
|
|
|
return "VOID";
|
|
|
|
case TOKEN_LPSTR:
|
|
|
|
return "STRING";
|
|
|
|
case TOKEN_UNICODE:
|
|
|
|
return "UNICODE";
|
|
|
|
case TOKEN_CSTRING:
|
|
|
|
return "CSTRING ";
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2011-06-08 21:39:05 +02:00
|
|
|
static void dump_template(xtemplate* templates_array, xtemplate* ptemplate)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2013-01-03 23:46:51 +01:00
|
|
|
ULONG j, k;
|
2008-12-31 12:27:24 +01:00
|
|
|
GUID* clsid;
|
|
|
|
|
|
|
|
clsid = &ptemplate->class_id;
|
|
|
|
|
|
|
|
DPRINTF("template %s\n", ptemplate->name);
|
|
|
|
DPRINTF("{\n");
|
|
|
|
DPRINTF(CLSIDFMT "\n", clsid->Data1, clsid->Data2, clsid->Data3, clsid->Data4[0],
|
|
|
|
clsid->Data4[1], clsid->Data4[2], clsid->Data4[3], clsid->Data4[4], clsid->Data4[5], clsid->Data4[6], clsid->Data4[7]);
|
|
|
|
for (j = 0; j < ptemplate->nb_members; j++)
|
|
|
|
{
|
|
|
|
if (ptemplate->members[j].nb_dims)
|
|
|
|
DPRINTF("array ");
|
|
|
|
if (ptemplate->members[j].type == TOKEN_NAME)
|
|
|
|
DPRINTF("%s ", templates_array[ptemplate->members[j].idx_template].name);
|
|
|
|
else
|
|
|
|
DPRINTF("%s ", get_primitive_string(ptemplate->members[j].type));
|
|
|
|
DPRINTF("%s", ptemplate->members[j].name);
|
|
|
|
for (k = 0; k < ptemplate->members[j].nb_dims; k++)
|
|
|
|
{
|
|
|
|
if (ptemplate->members[j].dim_fixed[k])
|
|
|
|
DPRINTF("[%d]", ptemplate->members[j].dim_value[k]);
|
|
|
|
else
|
|
|
|
DPRINTF("[%s]", ptemplate->members[ptemplate->members[j].dim_value[k]].name);
|
|
|
|
}
|
|
|
|
DPRINTF(";\n");
|
|
|
|
}
|
|
|
|
if (ptemplate->open)
|
|
|
|
DPRINTF("[...]\n");
|
2011-08-03 11:39:17 +02:00
|
|
|
else if (ptemplate->nb_children)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2011-08-03 11:39:17 +02:00
|
|
|
DPRINTF("[%s", ptemplate->children[0]);
|
|
|
|
for (j = 1; j < ptemplate->nb_children; j++)
|
|
|
|
DPRINTF(",%s", ptemplate->children[j]);
|
2008-12-31 12:27:24 +01:00
|
|
|
DPRINTF("]\n");
|
|
|
|
}
|
|
|
|
DPRINTF("}\n");
|
|
|
|
}
|
|
|
|
|
2011-06-11 20:20:34 +02:00
|
|
|
static BOOL read_bytes(parse_buffer * buf, LPVOID data, DWORD size)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
if (buf->rem_bytes < size)
|
|
|
|
return FALSE;
|
|
|
|
memcpy(data, buf->buffer, size);
|
|
|
|
buf->buffer += size;
|
|
|
|
buf->rem_bytes -= size;
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
2009-01-11 16:41:56 +01:00
|
|
|
static void rewind_bytes(parse_buffer * buf, DWORD size)
|
|
|
|
{
|
|
|
|
buf->buffer -= size;
|
|
|
|
buf->rem_bytes += size;
|
|
|
|
}
|
|
|
|
|
2011-06-08 21:38:58 +02:00
|
|
|
HRESULT parse_header(parse_buffer * buf, BYTE ** decomp_buffer_ptr)
|
|
|
|
{
|
|
|
|
/* X File common header:
|
|
|
|
* 0-3 -> Magic Number (format identifier)
|
|
|
|
* 4-7 -> Format Version
|
|
|
|
* 8-11 -> Format Type (text or binary, decompressed or compressed)
|
|
|
|
* 12-15 -> Float Size (32 or 64 bits) */
|
|
|
|
DWORD header[4];
|
|
|
|
|
|
|
|
if (!read_bytes(buf, header, sizeof(header)))
|
|
|
|
return DXFILEERR_BADFILETYPE;
|
|
|
|
|
|
|
|
if (TRACE_ON(d3dxof_parsing))
|
|
|
|
{
|
|
|
|
char string[17];
|
|
|
|
memcpy(string, header, 16);
|
|
|
|
string[16] = 0;
|
|
|
|
TRACE("header = '%s'\n", string);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (header[0] != XOFFILE_FORMAT_MAGIC)
|
|
|
|
return DXFILEERR_BADFILETYPE;
|
|
|
|
|
|
|
|
if (header[1] != XOFFILE_FORMAT_VERSION_302 && header[1] != XOFFILE_FORMAT_VERSION_303)
|
|
|
|
return DXFILEERR_BADFILEVERSION;
|
|
|
|
|
|
|
|
if (header[2] != XOFFILE_FORMAT_BINARY && header[2] != XOFFILE_FORMAT_TEXT &&
|
|
|
|
header[2] != XOFFILE_FORMAT_BINARY_MSZIP && header[2] != XOFFILE_FORMAT_TEXT_MSZIP)
|
|
|
|
{
|
|
|
|
WARN("File type %s unknown\n", debugstr_fourcc(header[2]));
|
|
|
|
return DXFILEERR_BADFILETYPE;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (header[3] != XOFFILE_FORMAT_FLOAT_BITS_32 && header[3] != XOFFILE_FORMAT_FLOAT_BITS_64)
|
|
|
|
return DXFILEERR_BADFILEFLOATSIZE;
|
|
|
|
|
|
|
|
buf->txt = header[2] == XOFFILE_FORMAT_TEXT || header[2] == XOFFILE_FORMAT_TEXT_MSZIP;
|
|
|
|
|
|
|
|
if (header[2] == XOFFILE_FORMAT_BINARY_MSZIP || header[2] == XOFFILE_FORMAT_TEXT_MSZIP)
|
|
|
|
{
|
|
|
|
/* Extended header for compressed data:
|
2012-02-17 00:19:12 +01:00
|
|
|
* 16-19 -> size of decompressed file including xof header,
|
|
|
|
* 20-21 -> size of first decompressed MSZIP chunk, 22-23 -> size of first compressed MSZIP chunk
|
|
|
|
* 24-xx -> compressed MSZIP data chunk
|
|
|
|
* xx-xx -> size of next decompressed MSZIP chunk, xx-xx -> size of next compressed MSZIP chunk
|
|
|
|
* xx-xx -> compressed MSZIP data chunk
|
|
|
|
* .............................................................................................. */
|
2011-06-08 21:38:58 +02:00
|
|
|
int err;
|
2012-02-17 00:19:12 +01:00
|
|
|
DWORD decomp_file_size;
|
|
|
|
WORD decomp_chunk_size;
|
|
|
|
WORD comp_chunk_size;
|
2011-06-08 21:38:58 +02:00
|
|
|
LPBYTE decomp_buffer;
|
|
|
|
|
2012-02-17 00:19:12 +01:00
|
|
|
if (!read_bytes(buf, &decomp_file_size, sizeof(decomp_file_size)))
|
2011-06-16 17:47:52 +02:00
|
|
|
return DXFILEERR_BADFILETYPE;
|
2011-06-08 21:38:58 +02:00
|
|
|
|
2012-02-17 00:19:12 +01:00
|
|
|
TRACE("Compressed format %s detected: decompressed file size with xof header = %d\n",
|
|
|
|
debugstr_fourcc(header[2]), decomp_file_size);
|
|
|
|
|
|
|
|
/* Does not take xof header into account */
|
|
|
|
decomp_file_size -= 16;
|
2011-06-08 21:38:58 +02:00
|
|
|
|
2012-02-17 00:19:12 +01:00
|
|
|
decomp_buffer = HeapAlloc(GetProcessHeap(), 0, decomp_file_size);
|
2011-06-08 21:38:58 +02:00
|
|
|
if (!decomp_buffer)
|
|
|
|
{
|
|
|
|
ERR("Out of memory\n");
|
|
|
|
return DXFILEERR_BADALLOC;
|
|
|
|
}
|
2012-02-17 00:19:12 +01:00
|
|
|
*decomp_buffer_ptr = decomp_buffer;
|
|
|
|
|
|
|
|
while (buf->rem_bytes)
|
2011-06-08 21:38:58 +02:00
|
|
|
{
|
2012-02-17 00:19:12 +01:00
|
|
|
if (!read_bytes(buf, &decomp_chunk_size, sizeof(decomp_chunk_size)))
|
|
|
|
return DXFILEERR_BADFILETYPE;
|
|
|
|
if (!read_bytes(buf, &comp_chunk_size, sizeof(comp_chunk_size)))
|
|
|
|
return DXFILEERR_BADFILETYPE;
|
|
|
|
|
|
|
|
TRACE("Process chunk: compressed_size = %d, decompressed_size = %d\n",
|
|
|
|
comp_chunk_size, decomp_chunk_size);
|
|
|
|
|
|
|
|
err = mszip_decompress(comp_chunk_size, decomp_chunk_size, (char*)buf->buffer, (char*)decomp_buffer);
|
|
|
|
if (err)
|
|
|
|
{
|
|
|
|
WARN("Error while decompressing MSZIP chunk %d\n", err);
|
2011-06-08 21:38:58 +02:00
|
|
|
HeapFree(GetProcessHeap(), 0, decomp_buffer);
|
|
|
|
return DXFILEERR_BADALLOC;
|
2012-02-17 00:19:12 +01:00
|
|
|
}
|
|
|
|
buf->rem_bytes -= comp_chunk_size;
|
|
|
|
buf->buffer += comp_chunk_size;
|
|
|
|
decomp_buffer += decomp_chunk_size;
|
2011-06-08 21:38:58 +02:00
|
|
|
}
|
2012-02-17 00:19:12 +01:00
|
|
|
|
|
|
|
if ((decomp_buffer - *decomp_buffer_ptr) != decomp_file_size)
|
|
|
|
ERR("Size of all decompressed chunks (%u) does not match decompressed file size (%u)\n",
|
|
|
|
(DWORD)(decomp_buffer - *decomp_buffer_ptr), decomp_file_size);
|
|
|
|
|
2011-06-08 21:38:58 +02:00
|
|
|
/* Use decompressed data */
|
2012-02-17 00:19:12 +01:00
|
|
|
buf->buffer = *decomp_buffer_ptr;
|
|
|
|
buf->rem_bytes = decomp_file_size;
|
2011-06-08 21:38:58 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
TRACE("Header is correct\n");
|
|
|
|
|
|
|
|
return S_OK;
|
|
|
|
}
|
|
|
|
|
2008-12-31 12:27:24 +01:00
|
|
|
static void dump_TOKEN(WORD token)
|
|
|
|
{
|
|
|
|
#define DUMP_TOKEN(t) case t: TRACE(#t "\n"); break
|
|
|
|
switch(token)
|
|
|
|
{
|
|
|
|
DUMP_TOKEN(TOKEN_NAME);
|
|
|
|
DUMP_TOKEN(TOKEN_STRING);
|
|
|
|
DUMP_TOKEN(TOKEN_INTEGER);
|
|
|
|
DUMP_TOKEN(TOKEN_GUID);
|
|
|
|
DUMP_TOKEN(TOKEN_INTEGER_LIST);
|
|
|
|
DUMP_TOKEN(TOKEN_FLOAT_LIST);
|
|
|
|
DUMP_TOKEN(TOKEN_OBRACE);
|
|
|
|
DUMP_TOKEN(TOKEN_CBRACE);
|
|
|
|
DUMP_TOKEN(TOKEN_OPAREN);
|
|
|
|
DUMP_TOKEN(TOKEN_CPAREN);
|
|
|
|
DUMP_TOKEN(TOKEN_OBRACKET);
|
|
|
|
DUMP_TOKEN(TOKEN_CBRACKET);
|
|
|
|
DUMP_TOKEN(TOKEN_OANGLE);
|
|
|
|
DUMP_TOKEN(TOKEN_CANGLE);
|
|
|
|
DUMP_TOKEN(TOKEN_DOT);
|
|
|
|
DUMP_TOKEN(TOKEN_COMMA);
|
|
|
|
DUMP_TOKEN(TOKEN_SEMICOLON);
|
|
|
|
DUMP_TOKEN(TOKEN_TEMPLATE);
|
|
|
|
DUMP_TOKEN(TOKEN_WORD);
|
|
|
|
DUMP_TOKEN(TOKEN_DWORD);
|
|
|
|
DUMP_TOKEN(TOKEN_FLOAT);
|
|
|
|
DUMP_TOKEN(TOKEN_DOUBLE);
|
|
|
|
DUMP_TOKEN(TOKEN_CHAR);
|
|
|
|
DUMP_TOKEN(TOKEN_UCHAR);
|
|
|
|
DUMP_TOKEN(TOKEN_SWORD);
|
|
|
|
DUMP_TOKEN(TOKEN_SDWORD);
|
|
|
|
DUMP_TOKEN(TOKEN_VOID);
|
|
|
|
DUMP_TOKEN(TOKEN_LPSTR);
|
|
|
|
DUMP_TOKEN(TOKEN_UNICODE);
|
|
|
|
DUMP_TOKEN(TOKEN_CSTRING);
|
|
|
|
DUMP_TOKEN(TOKEN_ARRAY);
|
|
|
|
default:
|
|
|
|
if (0)
|
|
|
|
TRACE("Unknown token %d\n", token);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
#undef DUMP_TOKEN
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_space(char c)
|
|
|
|
{
|
|
|
|
switch (c)
|
|
|
|
{
|
|
|
|
case 0x00:
|
|
|
|
case 0x0D:
|
|
|
|
case 0x0A:
|
|
|
|
case ' ':
|
|
|
|
case '\t':
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_operator(char c)
|
|
|
|
{
|
|
|
|
switch(c)
|
|
|
|
{
|
|
|
|
case '{':
|
|
|
|
case '}':
|
|
|
|
case '[':
|
|
|
|
case ']':
|
|
|
|
case '(':
|
|
|
|
case ')':
|
|
|
|
case '<':
|
|
|
|
case '>':
|
|
|
|
case ',':
|
|
|
|
case ';':
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline BOOL is_separator(char c)
|
|
|
|
{
|
|
|
|
return is_space(c) || is_operator(c);
|
|
|
|
}
|
|
|
|
|
|
|
|
static WORD get_operator_token(char c)
|
|
|
|
{
|
|
|
|
switch(c)
|
|
|
|
{
|
|
|
|
case '{':
|
|
|
|
return TOKEN_OBRACE;
|
|
|
|
case '}':
|
|
|
|
return TOKEN_CBRACE;
|
|
|
|
case '[':
|
|
|
|
return TOKEN_OBRACKET;
|
|
|
|
case ']':
|
|
|
|
return TOKEN_CBRACKET;
|
|
|
|
case '(':
|
|
|
|
return TOKEN_OPAREN;
|
|
|
|
case ')':
|
|
|
|
return TOKEN_CPAREN;
|
|
|
|
case '<':
|
|
|
|
return TOKEN_OANGLE;
|
|
|
|
case '>':
|
|
|
|
return TOKEN_CANGLE;
|
|
|
|
case ',':
|
|
|
|
return TOKEN_COMMA;
|
|
|
|
case ';':
|
|
|
|
return TOKEN_SEMICOLON;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_keyword(parse_buffer* buf, const char* keyword)
|
|
|
|
{
|
2009-02-10 14:07:19 +01:00
|
|
|
char tmp[8]; /* longest keyword size (template) */
|
2008-12-31 12:27:24 +01:00
|
|
|
DWORD len = strlen(keyword);
|
2009-02-10 14:07:19 +01:00
|
|
|
|
|
|
|
if (!read_bytes(buf, tmp, len))
|
|
|
|
return FALSE;
|
|
|
|
if (strncasecmp(tmp, keyword, len))
|
|
|
|
{
|
|
|
|
rewind_bytes(buf, len);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!read_bytes(buf, tmp, 1))
|
|
|
|
return TRUE;
|
|
|
|
if (is_separator(tmp[0]))
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2009-01-11 16:41:56 +01:00
|
|
|
rewind_bytes(buf, 1);
|
2008-12-31 12:27:24 +01:00
|
|
|
return TRUE;
|
|
|
|
}
|
2009-01-11 16:41:56 +01:00
|
|
|
rewind_bytes(buf, len+1);
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static WORD get_keyword_token(parse_buffer* buf)
|
|
|
|
{
|
|
|
|
if (is_keyword(buf, "template"))
|
|
|
|
return TOKEN_TEMPLATE;
|
|
|
|
if (is_keyword(buf, "WORD"))
|
|
|
|
return TOKEN_WORD;
|
|
|
|
if (is_keyword(buf, "DWORD"))
|
|
|
|
return TOKEN_DWORD;
|
|
|
|
if (is_keyword(buf, "FLOAT"))
|
|
|
|
return TOKEN_FLOAT;
|
|
|
|
if (is_keyword(buf, "DOUBLE"))
|
|
|
|
return TOKEN_DOUBLE;
|
|
|
|
if (is_keyword(buf, "CHAR"))
|
|
|
|
return TOKEN_CHAR;
|
|
|
|
if (is_keyword(buf, "UCHAR"))
|
|
|
|
return TOKEN_UCHAR;
|
|
|
|
if (is_keyword(buf, "SWORD"))
|
|
|
|
return TOKEN_SWORD;
|
|
|
|
if (is_keyword(buf, "SDWORD"))
|
|
|
|
return TOKEN_SDWORD;
|
|
|
|
if (is_keyword(buf, "VOID"))
|
|
|
|
return TOKEN_VOID;
|
|
|
|
if (is_keyword(buf, "STRING"))
|
|
|
|
return TOKEN_LPSTR;
|
|
|
|
if (is_keyword(buf, "UNICODE"))
|
|
|
|
return TOKEN_UNICODE;
|
|
|
|
if (is_keyword(buf, "CSTRING"))
|
|
|
|
return TOKEN_CSTRING;
|
|
|
|
if (is_keyword(buf, "array"))
|
|
|
|
return TOKEN_ARRAY;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_guid(parse_buffer* buf)
|
|
|
|
{
|
|
|
|
char tmp[50];
|
|
|
|
DWORD pos = 1;
|
|
|
|
GUID class_id;
|
|
|
|
DWORD tab[10];
|
|
|
|
int ret;
|
|
|
|
|
2011-04-20 22:20:01 +02:00
|
|
|
if (buf->rem_bytes < 38 || *buf->buffer != '<')
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
tmp[0] = '<';
|
2011-04-20 22:20:11 +02:00
|
|
|
while (pos < sizeof(tmp) - 2 && *(buf->buffer+pos) != '>')
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
tmp[pos] = *(buf->buffer+pos);
|
|
|
|
pos++;
|
|
|
|
}
|
|
|
|
tmp[pos++] = '>';
|
|
|
|
tmp[pos] = 0;
|
|
|
|
if (pos != 38 /* <+36+> */)
|
|
|
|
{
|
|
|
|
TRACE("Wrong guid %s (%d)\n", tmp, pos);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
buf->buffer += pos;
|
|
|
|
buf->rem_bytes -= pos;
|
|
|
|
|
|
|
|
ret = sscanf(tmp, CLSIDFMT, &class_id.Data1, tab, tab+1, tab+2, tab+3, tab+4, tab+5, tab+6, tab+7, tab+8, tab+9);
|
|
|
|
if (ret != 11)
|
|
|
|
{
|
|
|
|
TRACE("Wrong guid %s (%d)\n", tmp, pos);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
TRACE("Found guid %s (%d)\n", tmp, pos);
|
|
|
|
|
|
|
|
class_id.Data2 = tab[0];
|
|
|
|
class_id.Data3 = tab[1];
|
|
|
|
class_id.Data4[0] = tab[2];
|
|
|
|
class_id.Data4[1] = tab[3];
|
|
|
|
class_id.Data4[2] = tab[4];
|
|
|
|
class_id.Data4[3] = tab[5];
|
|
|
|
class_id.Data4[4] = tab[6];
|
|
|
|
class_id.Data4[5] = tab[7];
|
|
|
|
class_id.Data4[6] = tab[8];
|
|
|
|
class_id.Data4[7] = tab[9];
|
|
|
|
|
|
|
|
*(GUID*)buf->value = class_id;
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_name(parse_buffer* buf)
|
|
|
|
{
|
2011-04-20 22:20:11 +02:00
|
|
|
char tmp[512];
|
2008-12-31 12:27:24 +01:00
|
|
|
DWORD pos = 0;
|
|
|
|
char c;
|
2013-10-04 23:32:16 +02:00
|
|
|
BOOL error = FALSE;
|
2011-04-20 22:20:01 +02:00
|
|
|
while (pos < buf->rem_bytes && !is_separator(c = *(buf->buffer+pos)))
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
if (!(((c >= 'a') && (c <= 'z')) || ((c >= 'A') && (c <= 'Z')) || ((c >= '0') && (c <= '9')) || (c == '_') || (c == '-')))
|
2013-10-04 23:32:16 +02:00
|
|
|
error = TRUE;
|
2011-04-20 22:20:11 +02:00
|
|
|
if (pos < sizeof(tmp))
|
|
|
|
tmp[pos] = c;
|
|
|
|
pos++;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
2011-04-20 22:20:11 +02:00
|
|
|
tmp[min(pos, sizeof(tmp) - 1)] = 0;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
if (error)
|
|
|
|
{
|
|
|
|
TRACE("Wrong name %s\n", tmp);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
buf->buffer += pos;
|
|
|
|
buf->rem_bytes -= pos;
|
|
|
|
|
|
|
|
TRACE("Found name %s\n", tmp);
|
|
|
|
strcpy((char*)buf->value, tmp);
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_float(parse_buffer* buf)
|
|
|
|
{
|
2011-04-20 22:20:11 +02:00
|
|
|
char tmp[512];
|
2008-12-31 12:27:24 +01:00
|
|
|
DWORD pos = 0;
|
|
|
|
char c;
|
|
|
|
float decimal;
|
2013-10-18 12:35:08 +02:00
|
|
|
BOOL dot = FALSE;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
2011-04-20 22:20:01 +02:00
|
|
|
while (pos < buf->rem_bytes && !is_separator(c = *(buf->buffer+pos)))
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
if (!((!pos && (c == '-')) || ((c >= '0') && (c <= '9')) || (!dot && (c == '.'))))
|
|
|
|
return FALSE;
|
|
|
|
if (c == '.')
|
|
|
|
dot = TRUE;
|
2011-04-20 22:20:11 +02:00
|
|
|
if (pos < sizeof(tmp))
|
|
|
|
tmp[pos] = c;
|
|
|
|
pos++;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
2011-04-20 22:20:11 +02:00
|
|
|
tmp[min(pos, sizeof(tmp) - 1)] = 0;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
buf->buffer += pos;
|
|
|
|
buf->rem_bytes -= pos;
|
|
|
|
|
|
|
|
sscanf(tmp, "%f", &decimal);
|
|
|
|
|
|
|
|
TRACE("Found float %s - %f\n", tmp, decimal);
|
|
|
|
|
|
|
|
*(float*)buf->value = decimal;
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_integer(parse_buffer* buf)
|
|
|
|
{
|
2011-04-20 22:20:11 +02:00
|
|
|
char tmp[512];
|
2008-12-31 12:27:24 +01:00
|
|
|
DWORD pos = 0;
|
|
|
|
char c;
|
|
|
|
DWORD integer;
|
|
|
|
|
2011-04-20 22:20:01 +02:00
|
|
|
while (pos < buf->rem_bytes && !is_separator(c = *(buf->buffer+pos)))
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
if (!((c >= '0') && (c <= '9')))
|
|
|
|
return FALSE;
|
2011-04-20 22:20:11 +02:00
|
|
|
if (pos < sizeof(tmp))
|
|
|
|
tmp[pos] = c;
|
|
|
|
pos++;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
2011-04-20 22:20:11 +02:00
|
|
|
tmp[min(pos, sizeof(tmp) - 1)] = 0;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
buf->buffer += pos;
|
|
|
|
buf->rem_bytes -= pos;
|
|
|
|
|
|
|
|
sscanf(tmp, "%d", &integer);
|
|
|
|
|
|
|
|
TRACE("Found integer %s - %d\n", tmp, integer);
|
|
|
|
|
|
|
|
*(DWORD*)buf->value = integer;
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL is_string(parse_buffer* buf)
|
|
|
|
{
|
2011-04-20 22:20:11 +02:00
|
|
|
char tmp[512];
|
2008-12-31 12:27:24 +01:00
|
|
|
DWORD pos = 0;
|
|
|
|
char c;
|
2013-10-04 23:32:16 +02:00
|
|
|
BOOL ok = FALSE;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
if (*buf->buffer != '"')
|
|
|
|
return FALSE;
|
|
|
|
|
2012-03-09 11:54:00 +01:00
|
|
|
while ((pos+1) < buf->rem_bytes)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2012-03-09 11:54:00 +01:00
|
|
|
c = *(buf->buffer+pos+1);
|
2008-12-31 12:27:24 +01:00
|
|
|
if (c == '"')
|
|
|
|
{
|
2013-10-04 23:32:16 +02:00
|
|
|
ok = TRUE;
|
2008-12-31 12:27:24 +01:00
|
|
|
break;
|
|
|
|
}
|
2011-04-20 22:20:11 +02:00
|
|
|
if (pos < sizeof(tmp))
|
|
|
|
tmp[pos] = c;
|
|
|
|
pos++;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
2011-04-20 22:20:11 +02:00
|
|
|
tmp[min(pos, sizeof(tmp) - 1)] = 0;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
if (!ok)
|
|
|
|
{
|
|
|
|
TRACE("Wrong string %s\n", tmp);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
buf->buffer += pos + 2;
|
|
|
|
buf->rem_bytes -= pos + 2;
|
|
|
|
|
|
|
|
TRACE("Found string %s\n", tmp);
|
|
|
|
strcpy((char*)buf->value, tmp);
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static WORD parse_TOKEN(parse_buffer * buf)
|
|
|
|
{
|
|
|
|
WORD token;
|
|
|
|
|
|
|
|
if (buf->txt)
|
|
|
|
{
|
|
|
|
while(1)
|
|
|
|
{
|
|
|
|
char c;
|
|
|
|
if (!read_bytes(buf, &c, 1))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_NONE;
|
2008-12-31 12:27:24 +01:00
|
|
|
if ((c == '#') || (c == '/'))
|
|
|
|
{
|
|
|
|
/* Handle comment (# or //) */
|
|
|
|
if (c == '/')
|
|
|
|
{
|
|
|
|
if (!read_bytes(buf, &c, 1))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
if (c != '/')
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
c = 0;
|
|
|
|
while (c != 0x0A)
|
|
|
|
{
|
|
|
|
if (!read_bytes(buf, &c, 1))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_NONE;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (is_space(c))
|
|
|
|
continue;
|
|
|
|
if (is_operator(c) && (c != '<'))
|
|
|
|
{
|
|
|
|
token = get_operator_token(c);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
else if (c == '.')
|
|
|
|
{
|
|
|
|
token = TOKEN_DOT;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2009-01-11 16:41:56 +01:00
|
|
|
rewind_bytes(buf, 1);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
if ((token = get_keyword_token(buf)))
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (is_guid(buf))
|
|
|
|
{
|
|
|
|
token = TOKEN_GUID;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (is_integer(buf))
|
|
|
|
{
|
|
|
|
token = TOKEN_INTEGER;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (is_float(buf))
|
|
|
|
{
|
|
|
|
token = TOKEN_FLOAT;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (is_string(buf))
|
|
|
|
{
|
|
|
|
token = TOKEN_LPSTR;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (is_name(buf))
|
|
|
|
{
|
|
|
|
token = TOKEN_NAME;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
FIXME("Unrecognize element\n");
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2013-05-30 23:05:39 +02:00
|
|
|
if (!buf->list_nb_elements)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
if (!read_bytes(buf, &token, 2))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_NONE;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
/* Convert integer and float list into separate elements */
|
|
|
|
if (token == TOKEN_INTEGER_LIST)
|
|
|
|
{
|
2013-05-30 23:05:39 +02:00
|
|
|
if (!read_bytes(buf, &buf->list_nb_elements, 4))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
token = TOKEN_INTEGER;
|
2013-05-30 23:05:39 +02:00
|
|
|
buf->list_type_float = FALSE;
|
|
|
|
TRACE("Integer list (TOKEN_INTEGER_LIST) of size %d\n", buf->list_nb_elements);
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else if (token == TOKEN_FLOAT_LIST)
|
|
|
|
{
|
2013-05-30 23:05:39 +02:00
|
|
|
if (!read_bytes(buf, &buf->list_nb_elements, 4))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
token = TOKEN_FLOAT;
|
2013-05-30 23:05:39 +02:00
|
|
|
buf->list_type_float = TRUE;
|
|
|
|
TRACE("Float list (TOKEN_FLOAT_LIST) of size %d\n", buf->list_nb_elements);
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-05-30 23:05:39 +02:00
|
|
|
if (buf->list_nb_elements)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2013-05-30 23:05:47 +02:00
|
|
|
if (buf->list_separator)
|
|
|
|
{
|
|
|
|
buf->list_nb_elements--;
|
|
|
|
buf->list_separator = FALSE;
|
2016-05-29 12:42:56 +02:00
|
|
|
/* Insert separator between each value, and since list does not accept separator at the end
|
2013-05-30 23:05:47 +02:00
|
|
|
use a comma so any extra separator will generate an error */
|
|
|
|
token = TOKEN_COMMA;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
DWORD value;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
2013-05-30 23:05:47 +02:00
|
|
|
if (!read_bytes(buf, &value, 4))
|
|
|
|
return TOKEN_ERROR;
|
|
|
|
*(DWORD*)buf->value = value;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
2013-05-30 23:05:47 +02:00
|
|
|
buf->list_separator = TRUE;
|
2016-05-29 12:42:56 +02:00
|
|
|
/* Convert list into a series of their basic type counterpart */
|
2013-05-30 23:05:47 +02:00
|
|
|
token = buf->list_type_float ? TOKEN_FLOAT : TOKEN_INTEGER;
|
|
|
|
}
|
2008-12-31 12:27:24 +01:00
|
|
|
dump_TOKEN(token);
|
|
|
|
return token;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (token)
|
|
|
|
{
|
|
|
|
case TOKEN_NAME:
|
|
|
|
{
|
|
|
|
DWORD count;
|
2013-05-30 23:05:31 +02:00
|
|
|
char *name = (char*)buf->value;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
if (!read_bytes(buf, &count, 4))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2013-05-30 23:05:31 +02:00
|
|
|
if (!read_bytes(buf, name, count))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2013-05-30 23:05:31 +02:00
|
|
|
name[count] = 0;
|
|
|
|
TRACE("name = %s\n", name);
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
case TOKEN_INTEGER:
|
|
|
|
{
|
|
|
|
DWORD integer;
|
|
|
|
|
|
|
|
if (!read_bytes(buf, &integer, 4))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2013-05-30 23:05:31 +02:00
|
|
|
TRACE("integer = %u\n", integer);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
*(DWORD*)buf->value = integer;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case TOKEN_GUID:
|
|
|
|
{
|
|
|
|
char strguid[39];
|
|
|
|
GUID class_id;
|
|
|
|
|
|
|
|
if (!read_bytes(buf, &class_id, 16))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
sprintf(strguid, CLSIDFMT, class_id.Data1, class_id.Data2, class_id.Data3, class_id.Data4[0],
|
|
|
|
class_id.Data4[1], class_id.Data4[2], class_id.Data4[3], class_id.Data4[4], class_id.Data4[5],
|
|
|
|
class_id.Data4[6], class_id.Data4[7]);
|
2013-05-30 23:05:31 +02:00
|
|
|
TRACE("guid = %s\n", strguid);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
*(GUID*)buf->value = class_id;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case TOKEN_STRING:
|
|
|
|
{
|
|
|
|
DWORD count;
|
2013-05-30 23:05:31 +02:00
|
|
|
char *string = (char*)buf->value;
|
|
|
|
|
2008-12-31 12:27:24 +01:00
|
|
|
if (!read_bytes(buf, &count, 4))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2013-05-30 23:05:31 +02:00
|
|
|
if (!read_bytes(buf, string, count))
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2013-05-30 23:05:31 +02:00
|
|
|
string[count] = 0;
|
|
|
|
TRACE("string = %s\n", string);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
token = TOKEN_LPSTR;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case TOKEN_OBRACE:
|
|
|
|
case TOKEN_CBRACE:
|
|
|
|
case TOKEN_OPAREN:
|
|
|
|
case TOKEN_CPAREN:
|
|
|
|
case TOKEN_OBRACKET:
|
|
|
|
case TOKEN_CBRACKET:
|
|
|
|
case TOKEN_OANGLE:
|
|
|
|
case TOKEN_CANGLE:
|
|
|
|
case TOKEN_DOT:
|
|
|
|
case TOKEN_COMMA:
|
|
|
|
case TOKEN_SEMICOLON:
|
|
|
|
case TOKEN_TEMPLATE:
|
|
|
|
case TOKEN_WORD:
|
|
|
|
case TOKEN_DWORD:
|
|
|
|
case TOKEN_FLOAT:
|
|
|
|
case TOKEN_DOUBLE:
|
|
|
|
case TOKEN_CHAR:
|
|
|
|
case TOKEN_UCHAR:
|
|
|
|
case TOKEN_SWORD:
|
|
|
|
case TOKEN_SDWORD:
|
|
|
|
case TOKEN_VOID:
|
|
|
|
case TOKEN_LPSTR:
|
|
|
|
case TOKEN_UNICODE:
|
|
|
|
case TOKEN_CSTRING:
|
|
|
|
case TOKEN_ARRAY:
|
|
|
|
break;
|
|
|
|
default:
|
2013-04-28 22:46:09 +02:00
|
|
|
return TOKEN_ERROR;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
dump_TOKEN(token);
|
|
|
|
|
|
|
|
return token;
|
|
|
|
}
|
|
|
|
|
|
|
|
static WORD get_TOKEN(parse_buffer * buf)
|
|
|
|
{
|
|
|
|
if (buf->token_present)
|
|
|
|
{
|
|
|
|
buf->token_present = FALSE;
|
|
|
|
return buf->current_token;
|
|
|
|
}
|
|
|
|
|
|
|
|
buf->current_token = parse_TOKEN(buf);
|
|
|
|
|
|
|
|
return buf->current_token;
|
|
|
|
}
|
|
|
|
|
|
|
|
static WORD check_TOKEN(parse_buffer * buf)
|
|
|
|
{
|
|
|
|
if (buf->token_present)
|
|
|
|
return buf->current_token;
|
|
|
|
|
|
|
|
buf->current_token = parse_TOKEN(buf);
|
|
|
|
buf->token_present = TRUE;
|
|
|
|
|
|
|
|
return buf->current_token;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline BOOL is_primitive_type(WORD token)
|
|
|
|
{
|
|
|
|
BOOL ret;
|
|
|
|
switch(token)
|
|
|
|
{
|
|
|
|
case TOKEN_WORD:
|
|
|
|
case TOKEN_DWORD:
|
|
|
|
case TOKEN_FLOAT:
|
|
|
|
case TOKEN_DOUBLE:
|
|
|
|
case TOKEN_CHAR:
|
|
|
|
case TOKEN_UCHAR:
|
|
|
|
case TOKEN_SWORD:
|
|
|
|
case TOKEN_SDWORD:
|
|
|
|
case TOKEN_LPSTR:
|
|
|
|
case TOKEN_UNICODE:
|
|
|
|
case TOKEN_CSTRING:
|
2013-10-18 12:35:08 +02:00
|
|
|
ret = TRUE;
|
2008-12-31 12:27:24 +01:00
|
|
|
break;
|
|
|
|
default:
|
2013-10-18 12:35:08 +02:00
|
|
|
ret = FALSE;
|
2008-12-31 12:27:24 +01:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-01-08 22:43:24 +01:00
|
|
|
static BOOL parse_template_option_info(parse_buffer * buf)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
xtemplate* cur_template = &buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates];
|
|
|
|
|
|
|
|
if (check_TOKEN(buf) == TOKEN_DOT)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
if (get_TOKEN(buf) != TOKEN_DOT)
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_DOT)
|
|
|
|
return FALSE;
|
|
|
|
cur_template->open = TRUE;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
while (1)
|
|
|
|
{
|
|
|
|
if (get_TOKEN(buf) != TOKEN_NAME)
|
|
|
|
return FALSE;
|
2011-08-03 11:39:17 +02:00
|
|
|
strcpy(cur_template->children[cur_template->nb_children], (char*)buf->value);
|
2008-12-31 12:27:24 +01:00
|
|
|
if (check_TOKEN(buf) == TOKEN_GUID)
|
|
|
|
get_TOKEN(buf);
|
2011-08-03 11:39:17 +02:00
|
|
|
cur_template->nb_children++;
|
2008-12-31 12:27:24 +01:00
|
|
|
if (check_TOKEN(buf) != TOKEN_COMMA)
|
|
|
|
break;
|
|
|
|
get_TOKEN(buf);
|
|
|
|
}
|
|
|
|
cur_template->open = FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL parse_template_members_list(parse_buffer * buf)
|
|
|
|
{
|
|
|
|
int idx_member = 0;
|
|
|
|
member* cur_member;
|
|
|
|
|
|
|
|
while (1)
|
|
|
|
{
|
2013-10-18 12:35:08 +02:00
|
|
|
BOOL array = FALSE;
|
2008-12-31 12:27:24 +01:00
|
|
|
int nb_dims = 0;
|
|
|
|
cur_member = &buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].members[idx_member];
|
|
|
|
|
|
|
|
if (check_TOKEN(buf) == TOKEN_ARRAY)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
2013-10-18 12:35:08 +02:00
|
|
|
array = TRUE;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (check_TOKEN(buf) == TOKEN_NAME)
|
|
|
|
{
|
|
|
|
cur_member->type = get_TOKEN(buf);
|
2013-06-23 21:56:39 +02:00
|
|
|
if (!strcmp((char*)buf->value, "indexColor"))
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2013-06-23 21:56:39 +02:00
|
|
|
/* Case sensitive legacy type indexColor is described in the first template */
|
|
|
|
cur_member->idx_template = 0;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
2013-06-23 21:56:39 +02:00
|
|
|
else
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2013-06-23 21:56:39 +02:00
|
|
|
cur_member->idx_template = 1;
|
|
|
|
while (cur_member->idx_template < buf->pdxf->nb_xtemplates)
|
|
|
|
{
|
|
|
|
if (!strcasecmp((char*)buf->value, buf->pdxf->xtemplates[cur_member->idx_template].name))
|
|
|
|
break;
|
|
|
|
cur_member->idx_template++;
|
|
|
|
}
|
|
|
|
if (cur_member->idx_template == buf->pdxf->nb_xtemplates)
|
|
|
|
{
|
|
|
|
WARN("Reference to a nonexistent template '%s'\n", (char*)buf->value);
|
|
|
|
return FALSE;
|
|
|
|
}
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (is_primitive_type(check_TOKEN(buf)))
|
|
|
|
cur_member->type = get_TOKEN(buf);
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (get_TOKEN(buf) != TOKEN_NAME)
|
|
|
|
return FALSE;
|
|
|
|
strcpy(cur_member->name, (char*)buf->value);
|
|
|
|
|
|
|
|
if (array)
|
|
|
|
{
|
|
|
|
while (check_TOKEN(buf) == TOKEN_OBRACKET)
|
|
|
|
{
|
|
|
|
if (nb_dims >= MAX_ARRAY_DIM)
|
|
|
|
{
|
|
|
|
FIXME("Too many dimensions (%d) for multi-dimensional array\n", nb_dims + 1);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
get_TOKEN(buf);
|
|
|
|
if (check_TOKEN(buf) == TOKEN_INTEGER)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
cur_member->dim_fixed[nb_dims] = TRUE;
|
|
|
|
cur_member->dim_value[nb_dims] = *(DWORD*)buf->value;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_NAME)
|
|
|
|
return FALSE;
|
|
|
|
for (i = 0; i < idx_member; i++)
|
|
|
|
{
|
|
|
|
if (!strcmp((char*)buf->value, buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].members[i].name))
|
|
|
|
{
|
|
|
|
if (buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].members[i].nb_dims)
|
|
|
|
{
|
|
|
|
ERR("Array cannot be used to specify variable array size\n");
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
if (buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].members[i].type != TOKEN_DWORD)
|
|
|
|
{
|
|
|
|
FIXME("Only DWORD supported to specify variable array size\n");
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (i == idx_member)
|
|
|
|
{
|
|
|
|
ERR("Reference to unknown member %s\n", (char*)buf->value);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
cur_member->dim_fixed[nb_dims] = FALSE;
|
|
|
|
cur_member->dim_value[nb_dims] = i;
|
|
|
|
}
|
|
|
|
if (get_TOKEN(buf) != TOKEN_CBRACKET)
|
|
|
|
return FALSE;
|
|
|
|
nb_dims++;
|
|
|
|
}
|
|
|
|
if (!nb_dims)
|
|
|
|
return FALSE;
|
|
|
|
cur_member->nb_dims = nb_dims;
|
|
|
|
}
|
|
|
|
if (get_TOKEN(buf) != TOKEN_SEMICOLON)
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
idx_member++;
|
|
|
|
}
|
|
|
|
|
|
|
|
buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].nb_members = idx_member;
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL parse_template_parts(parse_buffer * buf)
|
|
|
|
{
|
|
|
|
if (!parse_template_members_list(buf))
|
|
|
|
return FALSE;
|
|
|
|
if (check_TOKEN(buf) == TOKEN_OBRACKET)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
if (!parse_template_option_info(buf))
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_CBRACKET)
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
2011-06-08 21:39:05 +02:00
|
|
|
static BOOL parse_template(parse_buffer * buf)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
if (get_TOKEN(buf) != TOKEN_TEMPLATE)
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_NAME)
|
|
|
|
return FALSE;
|
|
|
|
strcpy(buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].name, (char*)buf->value);
|
|
|
|
if (get_TOKEN(buf) != TOKEN_OBRACE)
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_GUID)
|
|
|
|
return FALSE;
|
|
|
|
buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].class_id = *(GUID*)buf->value;
|
|
|
|
if (!parse_template_parts(buf))
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_CBRACE)
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
TRACE("%d - %s - %s\n", buf->pdxf->nb_xtemplates, buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].name, debugstr_guid(&buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates].class_id));
|
|
|
|
buf->pdxf->nb_xtemplates++;
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
2013-04-29 21:42:55 +02:00
|
|
|
BOOL parse_templates(parse_buffer * buf, BOOL templates_only)
|
2011-06-08 21:39:05 +02:00
|
|
|
{
|
2013-04-29 21:42:55 +02:00
|
|
|
while (check_TOKEN(buf) != TOKEN_NONE)
|
2011-06-08 21:39:05 +02:00
|
|
|
{
|
2013-04-29 21:42:55 +02:00
|
|
|
if (templates_only && (check_TOKEN(buf) != TOKEN_TEMPLATE))
|
|
|
|
return TRUE;
|
2011-06-08 21:39:05 +02:00
|
|
|
if (!parse_template(buf))
|
|
|
|
{
|
|
|
|
WARN("Template is not correct\n");
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
TRACE("Template successfully parsed:\n");
|
|
|
|
if (TRACE_ON(d3dxof_parsing))
|
|
|
|
dump_template(buf->pdxf->xtemplates, &buf->pdxf->xtemplates[buf->pdxf->nb_xtemplates - 1]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
2009-03-24 23:59:37 +01:00
|
|
|
static BOOL check_buffer(parse_buffer * buf, ULONG size)
|
|
|
|
{
|
|
|
|
if ((buf->cur_pos_data + size) > buf->capacity)
|
|
|
|
{
|
|
|
|
LPBYTE pdata;
|
|
|
|
ULONG new_capacity = buf->capacity ? 2 * buf->capacity : 100000;
|
|
|
|
|
|
|
|
pdata = HeapAlloc(GetProcessHeap(), 0, new_capacity);
|
|
|
|
if (!pdata)
|
|
|
|
return FALSE;
|
|
|
|
memcpy(pdata, buf->pdata, buf->cur_pos_data);
|
|
|
|
HeapFree(GetProcessHeap(), 0, buf->pdata);
|
|
|
|
buf->capacity = new_capacity;
|
|
|
|
buf->pdata = pdata;
|
|
|
|
buf->pxo->root->pdata = pdata;
|
|
|
|
}
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
2009-01-07 12:35:08 +01:00
|
|
|
static BOOL parse_object_parts(parse_buffer * buf, BOOL allow_optional);
|
2008-12-31 12:27:24 +01:00
|
|
|
static BOOL parse_object_members_list(parse_buffer * buf)
|
|
|
|
{
|
|
|
|
DWORD token;
|
2013-01-03 23:46:51 +01:00
|
|
|
ULONG i;
|
2008-12-31 12:27:24 +01:00
|
|
|
xtemplate* pt = buf->pxt[buf->level];
|
|
|
|
|
2012-02-15 09:53:19 +01:00
|
|
|
buf->pxo->nb_members = pt->nb_members;
|
|
|
|
|
2008-12-31 12:27:24 +01:00
|
|
|
for (i = 0; i < pt->nb_members; i++)
|
|
|
|
{
|
2013-01-03 23:46:51 +01:00
|
|
|
ULONG k;
|
|
|
|
ULONG nb_elems = 1;
|
2013-05-21 20:02:03 +02:00
|
|
|
BOOL basic_type = TRUE;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
buf->pxo->members[i].name = pt->members[i].name;
|
2009-03-12 09:06:16 +01:00
|
|
|
buf->pxo->members[i].start = buf->cur_pos_data;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
for (k = 0; k < pt->members[i].nb_dims; k++)
|
|
|
|
{
|
|
|
|
if (pt->members[i].dim_fixed[k])
|
|
|
|
nb_elems *= pt->members[i].dim_value[k];
|
|
|
|
else
|
2009-03-12 09:06:16 +01:00
|
|
|
nb_elems *= *(DWORD*)(buf->pxo->root->pdata + buf->pxo->members[pt->members[i].dim_value[k]].start);
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
|
2013-01-03 23:46:51 +01:00
|
|
|
TRACE("Elements to consider: %u\n", nb_elems);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
for (k = 0; k < nb_elems; k++)
|
|
|
|
{
|
|
|
|
if (pt->members[i].type == TOKEN_NAME)
|
|
|
|
{
|
2013-01-03 23:46:51 +01:00
|
|
|
ULONG j;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
TRACE("Found sub-object %s\n", buf->pdxf->xtemplates[pt->members[i].idx_template].name);
|
2013-05-21 20:02:03 +02:00
|
|
|
basic_type = FALSE;
|
2008-12-31 12:27:24 +01:00
|
|
|
buf->level++;
|
|
|
|
/* To do template lookup */
|
|
|
|
for (j = 0; j < buf->pdxf->nb_xtemplates; j++)
|
|
|
|
{
|
|
|
|
if (!strcasecmp(buf->pdxf->xtemplates[pt->members[i].idx_template].name, buf->pdxf->xtemplates[j].name))
|
|
|
|
{
|
|
|
|
buf->pxt[buf->level] = &buf->pdxf->xtemplates[j];
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (j == buf->pdxf->nb_xtemplates)
|
|
|
|
{
|
|
|
|
ERR("Unknown template %s\n", (char*)buf->value);
|
|
|
|
buf->level--;
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
TRACE("Enter %s\n", buf->pdxf->xtemplates[pt->members[i].idx_template].name);
|
|
|
|
if (!parse_object_parts(buf, FALSE))
|
|
|
|
{
|
|
|
|
buf->level--;
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
buf->level--;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
token = check_TOKEN(buf);
|
|
|
|
if (token == TOKEN_INTEGER)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
TRACE("%s = %d\n", pt->members[i].name, *(DWORD*)buf->value);
|
|
|
|
/* Assume larger size */
|
2009-03-24 23:59:37 +01:00
|
|
|
if (!check_buffer(buf, 4))
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
if (pt->members[i].type == TOKEN_WORD)
|
|
|
|
{
|
2019-02-12 18:29:26 +01:00
|
|
|
*(WORD *)(buf->pdata + buf->cur_pos_data) = *(DWORD *)buf->value;
|
|
|
|
buf->cur_pos_data += 2;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else if (pt->members[i].type == TOKEN_DWORD)
|
|
|
|
{
|
2019-02-12 18:29:26 +01:00
|
|
|
*(DWORD *)(buf->pdata + buf->cur_pos_data) = *(DWORD *)buf->value;
|
|
|
|
buf->cur_pos_data += 4;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
FIXME("Token %d not supported\n", pt->members[i].type);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (token == TOKEN_FLOAT)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
TRACE("%s = %f\n", pt->members[i].name, *(float*)buf->value);
|
2009-03-24 23:59:37 +01:00
|
|
|
if (!check_buffer(buf, 4))
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
if (pt->members[i].type == TOKEN_FLOAT)
|
|
|
|
{
|
2019-02-12 18:29:26 +01:00
|
|
|
*(float *)(buf->pdata + buf->cur_pos_data) = *(float *)buf->value;
|
|
|
|
buf->cur_pos_data += 4;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
FIXME("Token %d not supported\n", pt->members[i].type);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (token == TOKEN_LPSTR)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
TRACE("%s = %s\n", pt->members[i].name, (char*)buf->value);
|
2011-05-20 00:35:17 +02:00
|
|
|
if (!check_buffer(buf, sizeof(LPSTR)))
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
if (pt->members[i].type == TOKEN_LPSTR)
|
|
|
|
{
|
|
|
|
int len = strlen((char*)buf->value) + 1;
|
|
|
|
if ((buf->cur_pstrings - buf->pstrings + len) > MAX_STRINGS_BUFFER)
|
|
|
|
{
|
|
|
|
FIXME("Buffer too small %p %p %d\n", buf->cur_pstrings, buf->pstrings, len);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
strcpy((char*)buf->cur_pstrings, (char*)buf->value);
|
2012-01-23 21:37:48 +01:00
|
|
|
*(((LPCSTR*)(buf->pdata + buf->cur_pos_data))) = (char*)buf->cur_pstrings;
|
2008-12-31 12:27:24 +01:00
|
|
|
buf->cur_pstrings += len;
|
2011-05-20 00:35:17 +02:00
|
|
|
buf->cur_pos_data += sizeof(LPSTR);
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
FIXME("Token %d not supported\n", pt->members[i].type);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
2013-05-21 20:02:03 +02:00
|
|
|
{
|
2013-05-30 23:05:54 +02:00
|
|
|
WARN("Unexpected token %d\n", token);
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
2012-02-25 12:36:51 +01:00
|
|
|
|
2013-05-21 20:02:03 +02:00
|
|
|
if (basic_type)
|
|
|
|
{
|
|
|
|
/* Handle separator only for basic types */
|
|
|
|
token = check_TOKEN(buf);
|
|
|
|
if ((token != TOKEN_COMMA) && (token != TOKEN_SEMICOLON))
|
|
|
|
return FALSE;
|
|
|
|
/* Allow multi-semicolons + single comma separator */
|
|
|
|
while (check_TOKEN(buf) == TOKEN_SEMICOLON)
|
|
|
|
get_TOKEN(buf);
|
|
|
|
if (check_TOKEN(buf) == TOKEN_COMMA)
|
|
|
|
get_TOKEN(buf);
|
|
|
|
}
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
2013-05-21 20:02:03 +02:00
|
|
|
|
2012-02-15 09:53:19 +01:00
|
|
|
buf->pxo->members[i].size = buf->cur_pos_data - buf->pxo->members[i].start;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static BOOL parse_object_parts(parse_buffer * buf, BOOL allow_optional)
|
|
|
|
{
|
2011-08-03 11:39:17 +02:00
|
|
|
buf->pxo->nb_children = 0;
|
2009-10-23 08:40:48 +02:00
|
|
|
|
2008-12-31 12:27:24 +01:00
|
|
|
if (!parse_object_members_list(buf))
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
if (allow_optional)
|
|
|
|
{
|
2009-03-12 09:06:16 +01:00
|
|
|
buf->pxo->size = buf->cur_pos_data - buf->pxo->pos_data;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
while (1)
|
|
|
|
{
|
|
|
|
if (check_TOKEN(buf) == TOKEN_OBRACE)
|
|
|
|
{
|
2013-01-03 23:46:51 +01:00
|
|
|
ULONG i, j;
|
2008-12-31 12:27:24 +01:00
|
|
|
get_TOKEN(buf);
|
|
|
|
if (get_TOKEN(buf) != TOKEN_NAME)
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_CBRACE)
|
|
|
|
return FALSE;
|
|
|
|
TRACE("Found optional reference %s\n", (char*)buf->value);
|
2009-12-30 22:06:18 +01:00
|
|
|
for (i = 0; i < (buf->nb_pxo_globals+1); i++)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
for (j = 0; j < (buf->pxo_globals[i])[0].nb_subobjects; j++)
|
|
|
|
{
|
|
|
|
if (!strcmp((buf->pxo_globals[i])[j].name, (char*)buf->value))
|
|
|
|
goto _exit;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
_exit:
|
2009-12-30 22:06:18 +01:00
|
|
|
if (i == (buf->nb_pxo_globals+1))
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
|
|
|
ERR("Reference to unknown object %s\n", (char*)buf->value);
|
|
|
|
return FALSE;
|
|
|
|
}
|
2010-03-23 08:32:20 +01:00
|
|
|
|
|
|
|
if (buf->pxo->root->nb_subobjects >= MAX_SUBOBJECTS)
|
|
|
|
{
|
|
|
|
FIXME("Too many sub-objects\n");
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
2011-08-03 11:39:17 +02:00
|
|
|
buf->pxo->children[buf->pxo->nb_children] = &buf->pxo_tab[buf->pxo->root->nb_subobjects++];
|
|
|
|
buf->pxo->children[buf->pxo->nb_children]->ptarget = &(buf->pxo_globals[i])[j];
|
|
|
|
buf->pxo->children[buf->pxo->nb_children]->binary = FALSE;
|
|
|
|
buf->pxo->nb_children++;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else if (check_TOKEN(buf) == TOKEN_NAME)
|
|
|
|
{
|
|
|
|
xobject* pxo = buf->pxo;
|
2010-03-23 08:32:20 +01:00
|
|
|
|
|
|
|
if (buf->pxo->root->nb_subobjects >= MAX_SUBOBJECTS)
|
|
|
|
{
|
|
|
|
FIXME("Too many sub-objects\n");
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
2015-11-22 21:37:20 +01:00
|
|
|
buf->pxo = buf->pxo->children[buf->pxo->nb_children] = &buf->pxo_tab[buf->pxo->root->nb_subobjects];
|
|
|
|
pxo->root->nb_subobjects++;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
TRACE("Enter optional %s\n", (char*)buf->value);
|
|
|
|
buf->level++;
|
|
|
|
if (!parse_object(buf))
|
|
|
|
{
|
|
|
|
buf->level--;
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
buf->level--;
|
|
|
|
buf->pxo = pxo;
|
2011-08-03 11:39:17 +02:00
|
|
|
buf->pxo->nb_children++;
|
2008-12-31 12:27:24 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-08-03 11:39:17 +02:00
|
|
|
if (buf->pxo->nb_children > MAX_CHILDREN)
|
2008-12-31 12:27:24 +01:00
|
|
|
{
|
2011-08-03 11:39:17 +02:00
|
|
|
FIXME("Too many children %d\n", buf->pxo->nb_children);
|
2008-12-31 12:27:24 +01:00
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
BOOL parse_object(parse_buffer * buf)
|
|
|
|
{
|
2013-01-03 23:46:51 +01:00
|
|
|
ULONG i;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
2009-03-12 09:06:16 +01:00
|
|
|
buf->pxo->pos_data = buf->cur_pos_data;
|
2008-12-31 12:27:24 +01:00
|
|
|
buf->pxo->ptarget = NULL;
|
2011-04-20 22:21:06 +02:00
|
|
|
buf->pxo->binary = FALSE;
|
2009-03-12 09:06:16 +01:00
|
|
|
buf->pxo->root = buf->pxo_tab;
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
if (get_TOKEN(buf) != TOKEN_NAME)
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
/* To do template lookup */
|
|
|
|
for (i = 0; i < buf->pdxf->nb_xtemplates; i++)
|
|
|
|
{
|
|
|
|
if (!strcasecmp((char*)buf->value, buf->pdxf->xtemplates[i].name))
|
|
|
|
{
|
|
|
|
buf->pxt[buf->level] = &buf->pdxf->xtemplates[i];
|
|
|
|
memcpy(&buf->pxo->type, &buf->pdxf->xtemplates[i].class_id, 16);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (i == buf->pdxf->nb_xtemplates)
|
|
|
|
{
|
|
|
|
ERR("Unknown template %s\n", (char*)buf->value);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (check_TOKEN(buf) == TOKEN_NAME)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
strcpy(buf->pxo->name, (char*)buf->value);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
buf->pxo->name[0] = 0;
|
|
|
|
|
|
|
|
if (get_TOKEN(buf) != TOKEN_OBRACE)
|
|
|
|
return FALSE;
|
|
|
|
if (check_TOKEN(buf) == TOKEN_GUID)
|
|
|
|
{
|
|
|
|
get_TOKEN(buf);
|
|
|
|
memcpy(&buf->pxo->class_id, buf->value, 16);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
memset(&buf->pxo->class_id, 0, 16);
|
|
|
|
|
|
|
|
if (!parse_object_parts(buf, TRUE))
|
|
|
|
return FALSE;
|
|
|
|
if (get_TOKEN(buf) != TOKEN_CBRACE)
|
|
|
|
return FALSE;
|
|
|
|
|
2013-04-29 21:43:03 +02:00
|
|
|
/* For seeking to a possibly eof to avoid parsing another object next time */
|
|
|
|
check_TOKEN(buf);
|
2008-12-31 12:27:24 +01:00
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|