0
============================================================================
THE UNOFFICIAL NEED FOR SPEED - SE FILE FORMAT SPECIFICATIONS - Version 0.2
============================================================================
E.1 GAMEDATA\CONFIG\PATHS.DAT
-------------------------
This file is essentially the same as for TNFS except for one addition. The
Track FAM files in NFSSE now come in 2 varieties, a normal file in
\simdata\ntrackfam and an internationalised file \simdata\etrackfam (English)
or \simdata\gtrackfam (German). The pattern of entries, then, is now as follows:
000 gamedata/config/ 2D0 F:\simdata\misc\
050 gamedata/savegame/ 320 F:\simdata\etrackfam\
0A0 F:\frontend\speech\ 370 F:\simdata\gtrackfam\
0F0 F:\simdata\soundbnk\ 3C0 F:\simdata\slides\
140 F:\frontend\music\ 410 F:\simdata\carfams\
190 F:\frontend\art\ 460 F:\simdata\soundbnk\
1E0 gamedata/modem/ 4B0 F:\simdata\carspecs\
230 F:\frontend\movielow\ 500 F:\simdata\dash\
280 gamedata/replay/ 550 F:\frontend\misc\
5A0 F:\frontend\show\
E.2 SIMDATA\MISC\*.TRI
------------------
These files describe the track itself, including the shape of the road
and the position of the scenery items. The objects are referenced by numbers
which correspond to entries in the corresponding files in SIMDATA\ETRACKFAM,
and SIMDATA\NTRACKFAM.
In order to understand the structure of TRI files you have to know that a
track is the superposition of three structures :
- first, a 'virtual road' : this is a sequence of points in space,
which will be called 'nodes'. These points correspond to successive
positions along the track, and all the cars have to pass near each of
these positions. During the game, the virtual road is invisible, but
you have to stay close to it.
- second, the scenery : this is a collection of points in space, which
will be called 'vertices', together with textures which are mapped onto
the polygons defined by consecutive vertices. These textures are used
to draw the road, the roadside and part of the landscape. Thus it is
of course preferable that the scenery remain close to the virtual road !
- and last, the objects : points in space together with bitmaps, which
are used for road signs, buildings, trees, etc... Some of them are
plain 2D bitmaps, but others have a more sophisticated polygonal 3D
structure.
The 3D coordinates x,y,z will always be used with the following meaning :
- x is an axis which is transversal to the starting line. Positive x values
correspond to points which are on the right of the starting position.
(i.e. if you start looking to the north, x points to the east)
- y is an axis which is parallel to the starting line. Positive y values
correspond to points which are ahead of the starting position.
(i.e. y points to the north)
- z is a vertical axis. Positive z values correspond to points higher than
the starting position.
a) TRI files begin with 98Ch bytes of headers and index tables. These are
as follows :
offset len data
------ --- ----
00 4 ?
04 8 ?
0C 4 x coordinate of the first node
10 4 z coordinate of the first node
14 4 y coordinate of the first node
18 4 ?
1C 4 ?
20 4 ?
24 4 length of the scenery data (in bytes : 120h per record)
28 4 ?
2C 960h first index table
The first index table is a succession of 32-bit offsets. It follows an
arithmetic progression by 120h as a general rule. This means the first
value is 0, followed by 548h, A90h, etc... On closed tracks, when the
end is reached, the offsets start again with 0, 548h, A90h, etc...
(so that the end of a lap is connected with the beginning of the following
one !). The table is filled to 960h bytes (600 entries) with zero values.
These offsets are added to the base address of the scenery data to get an
absolute offset into the file.
b) The virtual road data follows, and is constituted of 36-byte records
(one for each node). The first record is at offset 98Ch, and the last
allowed record is at offset 15B0Ch (this leaves room for 2400 records, and
since a scenery block corresponds to four nodes, both capacities are equal).
The unused records (after the last node) are filled with zero values.
The structure of each record is the following :
offset len data
------ --- ----
00 1 a0
01 1 a1
02 1 a2
03 1 a3
04 1 b0
05 1 b1
06 1 b2
07 1 b3
08 4 x coordinate
0C 4 z coordinate
10 4 y coordinate
14 2 slope
16 2 slant-A
18 2 orientation
1A 2 0
1C 2 y-orientation
1E 2 slant-B
20 2 x-orientation
22 2 0
- a0,a1,a2,a3 are 8-bit values which specify the width of the main road,
and the width of the finging area.
a0 = distance from the virtual road to the left hand verge
a1 = distance from the virtual road to the right hand verge
a2 = distance from virtual road to the absolute left hand edge
a3 = distanec from virtual road to the absolute right hand edge
- b0,b1,b2,b3 are 8-bit values of unknown purpose.
- x, z and y coordinates are signed long values.
- slope is a value indicating the slope at the current node, i.e. the
difference between the z coordinates of two consecutive nodes.
A good approximation is : slope(i) = (z(i+1) - z(i))/152.
However, to complicate things, it is stored as a signed 14-bit value,
complemented to 4000h. This means -1 is stored as 3FFFh, -2 as 3FFEh, ...
So in fact you must perform a logical and with 3FFFh before storing !
- slant-A is a value indicating how the road is slanted to the left or
to the right (as in the turns in Autumn Valley or Lost Vegas). It is
a signed 14-bit value, like slope. The value is positive if the road
is slanted to the right, negative if it is slanted to the left.
- slant-B has the same purpose, but is a standard signed 16-bit value.
Its value is positive for the left, negative for the right.
The approximative relation between slant-A and slant-B is
slant-B = -12.3 slant-A (remember that slant-A is 14-bit, though)
- orientation is a 14-bit value, and is equal to 0 for north (increasing y),
1000h for east (increasing x), 2000h for south (decreasing y), 3000h for
west (decreasing x), and back to 3FFFh for north.
- y-orientation is a signed 16-bit value, which is proportional to the
y coordinate variation. Meanwhile, x-orientation is proportional to the
*opposite* of the x coordinate variation. This means that the couple
(-xorientation,yorientation) gives the orientation of the track.
The norm (square root of xorient^2+yorient^2) is usually around 32000
(a little less than 8000h, to avoid numeric overflows), but can fluctuate
with the only condition that (-xor,yor) gives the correct orientation.
c) The objects data comes next. There are first several distinct zones,
many of which seem to be unused (?) :
offset len data
------ --- ----
15B0C 708h 3-byte records (there are 600... as many as scenery blocks ?)
16214 4 40h (?)
16218 4 3E8h = 1000 (size of the main block in records)
1621C 4 'SJOB'
16220 4 428Ch (total length of the remaining blocks)
16224 400h 16-byte records (unknown purpose)
16624 4 ?
16628 3E80h object data : 1000 records of 16 bytes (one per object)
The object data itself consists of a 16-byte record per object. The record
structure is the following :
offset len data
------ --- ----
00 4 reference node
04 1 bitmap number
05 1 flip
06 4 flags (unknown purpose)
0A 2 relative x coordinate
0C 2 relative z coordinate
0E 2 relative y coordinate
Each object is related to a reference node in the virtual road. The x,z,y
coordinates are then expressed as signed 16-bit values relative to the
coordinates of the reference node. Beware that the axes are simply translated
but not rotated ! (i.e. the x and y axes are still pointing east and north)
The objects are sorted in the order of increasing reference nodes.
A reference node value of -1 indicates that the record is unused (i.e. after
the end of the used records). Also note that the coordinates are not
expressed in the same unit as the 32-bit absolute coordinates seen above
(the units are much larger, so that the value fits in 16 bits).
The 8-bit flip value is equal to 0 for an object that is perfectly perpendi-
cular to the track (e.g. a road sign), larger values for objects that are
slightly turned, until 64 for an object that is mapped along the track
(e.g. an ad on the side of the road), then up to 128 which is the perfectly
reversed position (since the objects have no "back", this is the common
way of reversing a road sign for a turn in the other direction), then
up to 192 which is again longitudinal mapping (the other way) and until
255 which is back to the normal position.
The bitmap number corresponds to a texture in the corresponding .FAM file
(see B.8). The relevant bitmaps are in the second chunk of the .FAM file.
Two cases can occur :
- closed tracks : the second chunk is a 'wwww' structure containing a
single subchunk which is in turn a SHPI directory where the entry corres-
ponding to bitmap #n is called "nn00" where nn is n written in decimal.
(e.g. bitmap #18 is "1800"). Furthermore the object called "!pal" or "!PAL",
when it exists, is the corresponding palette (256 3-byte entries) ; FFh is
transparent.
- open roads : the second chunk contains a subchunk per bitmap, and each
subchunk is a SHPI containing at least the object "0000" (the bitmap), and
possibly a palette ("!pal" or "!PAL"). Object #n is then the bitmap "0000"
in the subchunk #n (the first subchunk is #0).
One must add to these 2D objects (plain bitmaps) the 3D objects described
in the fourth chunk of the .FAM file. They usually correspond to numbers
above the last 2D object ; however it happens, in closed tracks, that some
of the 3D objects are given numbers inside the range used by 2D objects.
In that case, the numbers describing 2D objects are shifted upwards.
(i.e. the bitmap "4400" corresponds to object #45 or #46). This phenomenon
apparently does not occur for open roads, where the 3D objects always follow
the 2D objects.
Furthermore, certain consecutive bitmaps represent successive states of an
animated object. In that case, the game will display successively the
relevant bitmaps. Note that if the second bitmap is given instead of the
first, the animation does not occur.
d) The scenery data starts at offset 1A4A8h. It is made up of records of
size 120h, each corresponding to four nodes in the virtual road. The
records are consecutive and the last record ends the .TRI file.
Each record has the following structure :
offset len data
------ --- ----
000 4 'TRKD'
004 4 114h = length of the record contents
008 4 00000000h
00C 2 ?
00E 10 textures
018 12 reference point
024 6 point A0
02A 6 point A1
... .. ........
05C 6 point A9
060 6 point A10
066 6 point B0
06C 6 point B1
... .. ........
09E 6 point B9
0A2 6 point B10
0A8 6 point C0
0AE 6 point C1
... .. ........
0DE 6 point C9
0E4 6 point C10
0EA 6 point D0
0F0 6 point D1
... .. ........
120 6 point D9
126 6 point D10
12C 6 point E0
130 6 point E1
... .. ........
162 6 point E9
168 6 point E10
Each point is given by three signed 16-bit relative coordinates (x,z,y as
usual). The coordinates are in the same reference frame as in the virtual
road data. The coordinates are relative to the virtual track point as given
in the virtual track data.
The points A0,...,A10 in record #n (starting with 0) correspond to the node
#4n (starting with 0) in the virtual road data. B0,...,B10 correspond to
node #4n+1, C0...C10 to node #4n+2, D0...D10 to node #4n+3 and E0...E10 to
node #4n+4. Thus the points E0...E10 are identical to the points A0...A10
of the following record.
The eleven point series (0 to 10) are arranged as follows : A0-E0 are near
the middle of the road, and thus close to the corresponding nodes.
A1-E1 are a little to the right, A2-E2 further right, ... until A5-E5.
A6-E6 are a little to the left, A7-E7 further left, ... until A10-E10.
(In tunnels, the points A5-E5 and A10-E10 get back to the center, consti-
tuting the ceiling).
To each record correspond ten textures (coded at the beginning), each given
by a 8-bit value, T1,T2,...,T10. T1 is used between A0-E0 and A1-E1,
T2 between A1-E1 and A2-E2, ..., T5 between A4-E4 and A5-E5 ; meanwhile,
T6 is used between A0-E0 and A6-E6, T7 between A6-E6 and A7-E7, ...,
T10 between A9-E9 and A10-E10. This is summarized on the following diagram :
E10---E9---E8---E7---E6---E0---E1---E2---E3---E4---E5 node 4n+4
| | | | | || | | | | |
| | | | | || | | | | |
D10 D9 D8 D7 D6 D0 D1 D2 D3 D4 D5 node 4n+3
| T | T | T | T | T || T | T | T | T | T |
| | | | | || | | | | |
C10 C9 C8 C7 C6 C0 C1 C2 C3 C4 C5 node 4n+2
| 10 | 9 | 8 | 7 | 6 || 1 | 2 | 3 | 4 | 5 |
| | | | | || | | | | |
B10 B9 B8 B7 B6 B0 B1 B2 B3 B4 B5 node 4n+1
| | | | | || | | | | |
| | | | | || | | | | |
A10---A9---A8---A7---A6---A0---A1---A2---A3---A4---A5 node 4n
^
the nodes are here |
The texture numbers are converted to bitmaps in the first chunk of the
.FAM file (see B.8). There are two different cases :
- closed tracks : the first chunk is a 'wwww' structure which contains a
single subchunk which is in turn a SHPI bitmap directory, possibly with a
palette '!PAL' or '!pal'. There is also often a bitmap called 'ga00' or
'GA00' (unknown interpretation). The names have the structure "xxls", where
xx is a decimal value indicating the texture group, l is 'A', 'B' or 'C',
and s indicates a scale ('0' is the largest, while '3'&'4' are very small).
The various scales are here to speed up the texture-mapping algorithm,
anyway the only texture that is always present is with s=0.
The xx and l values correspond to a texture number n in the following way :
n=3xx if l='A', 3xx+1 if l='B' and 3xx+2 if l='C'. Note that there are holes
in the numbering : many numbers do not have a bitmap.
Examples : bitmap "03C0" corresponds to texture #11 (3x3+2) at the largest
scale; bitmap "14A1" corresponds to texture #42 (3x14) at the second scale
available.
- open roads : the first chunk contains a subchunk per texture group (i.e.
the xx value is now the number of the subchunk, starting with 0). Each
subchunk is a SHPI directory containing potentially a palette, and bitmaps
labelled "l00s", where l is 'A','B' or 'C' and s is the scale.
As before, n=3xx if l='A', 3xx+1 if l='B', 3xx+2 if l='C', and there are
holes in the numbering.
Examples : texture #11 at scale '0' is now the bitmap "C000" in subchunk #3.
Texture #42 at scale '1' is now the bitmap called "A001" in subchunk #14.
F - APPENDIX -- Compression Formats
================================
Compressed data files start with a 5 byte header.
offset len data
------ --- ----
000 1 Pack Code Hi byte
001 1 Pack Code Lo byte (== FBh or 32h)
002 1 Expanded length Hi byte
003 1 Expanded length Mid byte
004 1 Expanded length Lo byte
If bit 0 of the pack code Hi byte (ie Offset 0 & 0x01) is set, then there
are now 3 padding bytes to allow the data to begin on a 32 bit boundary.
Otherwise, the data starts at offset 5. The interpretation of the data is
different for different pack codes. In the description that follows, it is
assumed that bit zero of the pack code Hi byte has been cleared.
F.1 Pack code == 10FBh, or 1032h
----------------------------
This is the pack code used by the .QFS files, and it indicates LZ77
compression has been used.
To decode, we read and decode chunks as per the following C code:
/*************************************************************************
Function: ReadPackFile (CFile&)
Purpose: Reads in and unpacks the specified file
Returns: A pointer to the unpacked data if successful.
Comments: A null pointer is returned on error
*************************************************************************/
#include "stdafx.h"
#include "pack.h"
// Note that we rely here on the characteristics of an
// overlapping 'copy up' of bytes. Hence we cannot use
// the memcpy library function.
unsigned char* ReadPackFile(CFile& file)
{
int filesize = file.GetLength();
unsigned char *pSourceData = (unsigned char *)malloc(filesize);
if (NULL == pSourceData) return(NULL);
int bytesread = file.Read(pSourceData, filesize);
if (bytesread != filesize){
free(pSourceData);
return(NULL);
}
int PackCode = (pSourceData[0]&0xfe)*256 + pSourceData[1];
if (PackCode != 0x10fb){
free(pSourceData);
return(NULL); // Invalid pack code
}
int ExpandedLength = (pSourceData[2] << 16) + (pSourceData[3] << 8) +
pSourceData[4];
int filepos = 5;
int TargetOffset = 0;
if (pSourceData[0] & 0x01) filepos = 8; // align if necessary.
unsigned char *pExpandedData = (unsigned char *)malloc(ExpandedLength);
if (pExpandedData){
while(filepos < filesize && pSourceData[filepos] < 0xfc){
unsigned char pack_byte = pSourceData[filepos];
int a = pSourceData[filepos+1];
int b = pSourceData[filepos+2];
if (!(pack_byte & 0x80)){
int len = pack_byte&0x03;
unsigned char *pDest =
pExpandedData+TargetOffset;
unsigned char *pSrc = pSourceData+filepos+2;
TargetOffset += len;
filepos += (len+2);
while (len--) *pDest++ = *pSrc++;
len = ((pack_byte & 0x1c)>>2) + 3;
int offset = (pack_byte >> 5) + a + 1;
pDest = pExpandedData + TargetOffset;
pSrc = pDest-offset;
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
else if (!(pack_byte & 0x40)){
int len = (a >> 6) &0x03;
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+3;
filepos += (len+3);
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
int offset = (a&0x3f)*256 + b + 1;
pDest = pExpandedData + TargetOffset;
pSrc = pDest-offset;
len = (pack_byte & 0x3f)+4;
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
else if (!(pack_byte & 0x20)){
int c = pSourceData[filepos+3];
int len = (pack_byte & 0x03);
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+4;
filepos += (len+4);
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
int offset = ((pack_byte & 0x10)<<0x0c) + 256*a
+ b + 1;
pDest = pExpandedData + TargetOffset;
pSrc = pDest-offset;
len = ((pack_byte >> 2)&0x03)*256+c+5;
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
else {
int len = (pack_byte&0x1f)*4+4;
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+1;
filepos += (len+1);
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
}
if (filepos < filesize && TargetOffset < ExpandedLength){
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+1;
int len = pSourceData[filepos]&0x03;
while (len--) *pDest++ = *pSrc++;
}
}
free(pSourceData);
return(pExpandedData);
}
THE UNOFFICIAL NEED FOR SPEED - SE FILE FORMAT SPECIFICATIONS - Version 0.2
============================================================================
E.1 GAMEDATA\CONFIG\PATHS.DAT
-------------------------
This file is essentially the same as for TNFS except for one addition. The
Track FAM files in NFSSE now come in 2 varieties, a normal file in
\simdata\ntrackfam and an internationalised file \simdata\etrackfam (English)
or \simdata\gtrackfam (German). The pattern of entries, then, is now as follows:
000 gamedata/config/ 2D0 F:\simdata\misc\
050 gamedata/savegame/ 320 F:\simdata\etrackfam\
0A0 F:\frontend\speech\ 370 F:\simdata\gtrackfam\
0F0 F:\simdata\soundbnk\ 3C0 F:\simdata\slides\
140 F:\frontend\music\ 410 F:\simdata\carfams\
190 F:\frontend\art\ 460 F:\simdata\soundbnk\
1E0 gamedata/modem/ 4B0 F:\simdata\carspecs\
230 F:\frontend\movielow\ 500 F:\simdata\dash\
280 gamedata/replay/ 550 F:\frontend\misc\
5A0 F:\frontend\show\
E.2 SIMDATA\MISC\*.TRI
------------------
These files describe the track itself, including the shape of the road
and the position of the scenery items. The objects are referenced by numbers
which correspond to entries in the corresponding files in SIMDATA\ETRACKFAM,
and SIMDATA\NTRACKFAM.
In order to understand the structure of TRI files you have to know that a
track is the superposition of three structures :
- first, a 'virtual road' : this is a sequence of points in space,
which will be called 'nodes'. These points correspond to successive
positions along the track, and all the cars have to pass near each of
these positions. During the game, the virtual road is invisible, but
you have to stay close to it.
- second, the scenery : this is a collection of points in space, which
will be called 'vertices', together with textures which are mapped onto
the polygons defined by consecutive vertices. These textures are used
to draw the road, the roadside and part of the landscape. Thus it is
of course preferable that the scenery remain close to the virtual road !
- and last, the objects : points in space together with bitmaps, which
are used for road signs, buildings, trees, etc... Some of them are
plain 2D bitmaps, but others have a more sophisticated polygonal 3D
structure.
The 3D coordinates x,y,z will always be used with the following meaning :
- x is an axis which is transversal to the starting line. Positive x values
correspond to points which are on the right of the starting position.
(i.e. if you start looking to the north, x points to the east)
- y is an axis which is parallel to the starting line. Positive y values
correspond to points which are ahead of the starting position.
(i.e. y points to the north)
- z is a vertical axis. Positive z values correspond to points higher than
the starting position.
a) TRI files begin with 98Ch bytes of headers and index tables. These are
as follows :
offset len data
------ --- ----
00 4 ?
04 8 ?
0C 4 x coordinate of the first node
10 4 z coordinate of the first node
14 4 y coordinate of the first node
18 4 ?
1C 4 ?
20 4 ?
24 4 length of the scenery data (in bytes : 120h per record)
28 4 ?
2C 960h first index table
The first index table is a succession of 32-bit offsets. It follows an
arithmetic progression by 120h as a general rule. This means the first
value is 0, followed by 548h, A90h, etc... On closed tracks, when the
end is reached, the offsets start again with 0, 548h, A90h, etc...
(so that the end of a lap is connected with the beginning of the following
one !). The table is filled to 960h bytes (600 entries) with zero values.
These offsets are added to the base address of the scenery data to get an
absolute offset into the file.
b) The virtual road data follows, and is constituted of 36-byte records
(one for each node). The first record is at offset 98Ch, and the last
allowed record is at offset 15B0Ch (this leaves room for 2400 records, and
since a scenery block corresponds to four nodes, both capacities are equal).
The unused records (after the last node) are filled with zero values.
The structure of each record is the following :
offset len data
------ --- ----
00 1 a0
01 1 a1
02 1 a2
03 1 a3
04 1 b0
05 1 b1
06 1 b2
07 1 b3
08 4 x coordinate
0C 4 z coordinate
10 4 y coordinate
14 2 slope
16 2 slant-A
18 2 orientation
1A 2 0
1C 2 y-orientation
1E 2 slant-B
20 2 x-orientation
22 2 0
- a0,a1,a2,a3 are 8-bit values which specify the width of the main road,
and the width of the finging area.
a0 = distance from the virtual road to the left hand verge
a1 = distance from the virtual road to the right hand verge
a2 = distance from virtual road to the absolute left hand edge
a3 = distanec from virtual road to the absolute right hand edge
- b0,b1,b2,b3 are 8-bit values of unknown purpose.
- x, z and y coordinates are signed long values.
- slope is a value indicating the slope at the current node, i.e. the
difference between the z coordinates of two consecutive nodes.
A good approximation is : slope(i) = (z(i+1) - z(i))/152.
However, to complicate things, it is stored as a signed 14-bit value,
complemented to 4000h. This means -1 is stored as 3FFFh, -2 as 3FFEh, ...
So in fact you must perform a logical and with 3FFFh before storing !
- slant-A is a value indicating how the road is slanted to the left or
to the right (as in the turns in Autumn Valley or Lost Vegas). It is
a signed 14-bit value, like slope. The value is positive if the road
is slanted to the right, negative if it is slanted to the left.
- slant-B has the same purpose, but is a standard signed 16-bit value.
Its value is positive for the left, negative for the right.
The approximative relation between slant-A and slant-B is
slant-B = -12.3 slant-A (remember that slant-A is 14-bit, though)
- orientation is a 14-bit value, and is equal to 0 for north (increasing y),
1000h for east (increasing x), 2000h for south (decreasing y), 3000h for
west (decreasing x), and back to 3FFFh for north.
- y-orientation is a signed 16-bit value, which is proportional to the
y coordinate variation. Meanwhile, x-orientation is proportional to the
*opposite* of the x coordinate variation. This means that the couple
(-xorientation,yorientation) gives the orientation of the track.
The norm (square root of xorient^2+yorient^2) is usually around 32000
(a little less than 8000h, to avoid numeric overflows), but can fluctuate
with the only condition that (-xor,yor) gives the correct orientation.
c) The objects data comes next. There are first several distinct zones,
many of which seem to be unused (?) :
offset len data
------ --- ----
15B0C 708h 3-byte records (there are 600... as many as scenery blocks ?)
16214 4 40h (?)
16218 4 3E8h = 1000 (size of the main block in records)
1621C 4 'SJOB'
16220 4 428Ch (total length of the remaining blocks)
16224 400h 16-byte records (unknown purpose)
16624 4 ?
16628 3E80h object data : 1000 records of 16 bytes (one per object)
The object data itself consists of a 16-byte record per object. The record
structure is the following :
offset len data
------ --- ----
00 4 reference node
04 1 bitmap number
05 1 flip
06 4 flags (unknown purpose)
0A 2 relative x coordinate
0C 2 relative z coordinate
0E 2 relative y coordinate
Each object is related to a reference node in the virtual road. The x,z,y
coordinates are then expressed as signed 16-bit values relative to the
coordinates of the reference node. Beware that the axes are simply translated
but not rotated ! (i.e. the x and y axes are still pointing east and north)
The objects are sorted in the order of increasing reference nodes.
A reference node value of -1 indicates that the record is unused (i.e. after
the end of the used records). Also note that the coordinates are not
expressed in the same unit as the 32-bit absolute coordinates seen above
(the units are much larger, so that the value fits in 16 bits).
The 8-bit flip value is equal to 0 for an object that is perfectly perpendi-
cular to the track (e.g. a road sign), larger values for objects that are
slightly turned, until 64 for an object that is mapped along the track
(e.g. an ad on the side of the road), then up to 128 which is the perfectly
reversed position (since the objects have no "back", this is the common
way of reversing a road sign for a turn in the other direction), then
up to 192 which is again longitudinal mapping (the other way) and until
255 which is back to the normal position.
The bitmap number corresponds to a texture in the corresponding .FAM file
(see B.8). The relevant bitmaps are in the second chunk of the .FAM file.
Two cases can occur :
- closed tracks : the second chunk is a 'wwww' structure containing a
single subchunk which is in turn a SHPI directory where the entry corres-
ponding to bitmap #n is called "nn00" where nn is n written in decimal.
(e.g. bitmap #18 is "1800"). Furthermore the object called "!pal" or "!PAL",
when it exists, is the corresponding palette (256 3-byte entries) ; FFh is
transparent.
- open roads : the second chunk contains a subchunk per bitmap, and each
subchunk is a SHPI containing at least the object "0000" (the bitmap), and
possibly a palette ("!pal" or "!PAL"). Object #n is then the bitmap "0000"
in the subchunk #n (the first subchunk is #0).
One must add to these 2D objects (plain bitmaps) the 3D objects described
in the fourth chunk of the .FAM file. They usually correspond to numbers
above the last 2D object ; however it happens, in closed tracks, that some
of the 3D objects are given numbers inside the range used by 2D objects.
In that case, the numbers describing 2D objects are shifted upwards.
(i.e. the bitmap "4400" corresponds to object #45 or #46). This phenomenon
apparently does not occur for open roads, where the 3D objects always follow
the 2D objects.
Furthermore, certain consecutive bitmaps represent successive states of an
animated object. In that case, the game will display successively the
relevant bitmaps. Note that if the second bitmap is given instead of the
first, the animation does not occur.
d) The scenery data starts at offset 1A4A8h. It is made up of records of
size 120h, each corresponding to four nodes in the virtual road. The
records are consecutive and the last record ends the .TRI file.
Each record has the following structure :
offset len data
------ --- ----
000 4 'TRKD'
004 4 114h = length of the record contents
008 4 00000000h
00C 2 ?
00E 10 textures
018 12 reference point
024 6 point A0
02A 6 point A1
... .. ........
05C 6 point A9
060 6 point A10
066 6 point B0
06C 6 point B1
... .. ........
09E 6 point B9
0A2 6 point B10
0A8 6 point C0
0AE 6 point C1
... .. ........
0DE 6 point C9
0E4 6 point C10
0EA 6 point D0
0F0 6 point D1
... .. ........
120 6 point D9
126 6 point D10
12C 6 point E0
130 6 point E1
... .. ........
162 6 point E9
168 6 point E10
Each point is given by three signed 16-bit relative coordinates (x,z,y as
usual). The coordinates are in the same reference frame as in the virtual
road data. The coordinates are relative to the virtual track point as given
in the virtual track data.
The points A0,...,A10 in record #n (starting with 0) correspond to the node
#4n (starting with 0) in the virtual road data. B0,...,B10 correspond to
node #4n+1, C0...C10 to node #4n+2, D0...D10 to node #4n+3 and E0...E10 to
node #4n+4. Thus the points E0...E10 are identical to the points A0...A10
of the following record.
The eleven point series (0 to 10) are arranged as follows : A0-E0 are near
the middle of the road, and thus close to the corresponding nodes.
A1-E1 are a little to the right, A2-E2 further right, ... until A5-E5.
A6-E6 are a little to the left, A7-E7 further left, ... until A10-E10.
(In tunnels, the points A5-E5 and A10-E10 get back to the center, consti-
tuting the ceiling).
To each record correspond ten textures (coded at the beginning), each given
by a 8-bit value, T1,T2,...,T10. T1 is used between A0-E0 and A1-E1,
T2 between A1-E1 and A2-E2, ..., T5 between A4-E4 and A5-E5 ; meanwhile,
T6 is used between A0-E0 and A6-E6, T7 between A6-E6 and A7-E7, ...,
T10 between A9-E9 and A10-E10. This is summarized on the following diagram :
E10---E9---E8---E7---E6---E0---E1---E2---E3---E4---E5 node 4n+4
| | | | | || | | | | |
| | | | | || | | | | |
D10 D9 D8 D7 D6 D0 D1 D2 D3 D4 D5 node 4n+3
| T | T | T | T | T || T | T | T | T | T |
| | | | | || | | | | |
C10 C9 C8 C7 C6 C0 C1 C2 C3 C4 C5 node 4n+2
| 10 | 9 | 8 | 7 | 6 || 1 | 2 | 3 | 4 | 5 |
| | | | | || | | | | |
B10 B9 B8 B7 B6 B0 B1 B2 B3 B4 B5 node 4n+1
| | | | | || | | | | |
| | | | | || | | | | |
A10---A9---A8---A7---A6---A0---A1---A2---A3---A4---A5 node 4n
^
the nodes are here |
The texture numbers are converted to bitmaps in the first chunk of the
.FAM file (see B.8). There are two different cases :
- closed tracks : the first chunk is a 'wwww' structure which contains a
single subchunk which is in turn a SHPI bitmap directory, possibly with a
palette '!PAL' or '!pal'. There is also often a bitmap called 'ga00' or
'GA00' (unknown interpretation). The names have the structure "xxls", where
xx is a decimal value indicating the texture group, l is 'A', 'B' or 'C',
and s indicates a scale ('0' is the largest, while '3'&'4' are very small).
The various scales are here to speed up the texture-mapping algorithm,
anyway the only texture that is always present is with s=0.
The xx and l values correspond to a texture number n in the following way :
n=3xx if l='A', 3xx+1 if l='B' and 3xx+2 if l='C'. Note that there are holes
in the numbering : many numbers do not have a bitmap.
Examples : bitmap "03C0" corresponds to texture #11 (3x3+2) at the largest
scale; bitmap "14A1" corresponds to texture #42 (3x14) at the second scale
available.
- open roads : the first chunk contains a subchunk per texture group (i.e.
the xx value is now the number of the subchunk, starting with 0). Each
subchunk is a SHPI directory containing potentially a palette, and bitmaps
labelled "l00s", where l is 'A','B' or 'C' and s is the scale.
As before, n=3xx if l='A', 3xx+1 if l='B', 3xx+2 if l='C', and there are
holes in the numbering.
Examples : texture #11 at scale '0' is now the bitmap "C000" in subchunk #3.
Texture #42 at scale '1' is now the bitmap called "A001" in subchunk #14.
F - APPENDIX -- Compression Formats
================================
Compressed data files start with a 5 byte header.
offset len data
------ --- ----
000 1 Pack Code Hi byte
001 1 Pack Code Lo byte (== FBh or 32h)
002 1 Expanded length Hi byte
003 1 Expanded length Mid byte
004 1 Expanded length Lo byte
If bit 0 of the pack code Hi byte (ie Offset 0 & 0x01) is set, then there
are now 3 padding bytes to allow the data to begin on a 32 bit boundary.
Otherwise, the data starts at offset 5. The interpretation of the data is
different for different pack codes. In the description that follows, it is
assumed that bit zero of the pack code Hi byte has been cleared.
F.1 Pack code == 10FBh, or 1032h
----------------------------
This is the pack code used by the .QFS files, and it indicates LZ77
compression has been used.
To decode, we read and decode chunks as per the following C code:
/*************************************************************************
Function: ReadPackFile (CFile&)
Purpose: Reads in and unpacks the specified file
Returns: A pointer to the unpacked data if successful.
Comments: A null pointer is returned on error
*************************************************************************/
#include "stdafx.h"
#include "pack.h"
// Note that we rely here on the characteristics of an
// overlapping 'copy up' of bytes. Hence we cannot use
// the memcpy library function.
unsigned char* ReadPackFile(CFile& file)
{
int filesize = file.GetLength();
unsigned char *pSourceData = (unsigned char *)malloc(filesize);
if (NULL == pSourceData) return(NULL);
int bytesread = file.Read(pSourceData, filesize);
if (bytesread != filesize){
free(pSourceData);
return(NULL);
}
int PackCode = (pSourceData[0]&0xfe)*256 + pSourceData[1];
if (PackCode != 0x10fb){
free(pSourceData);
return(NULL); // Invalid pack code
}
int ExpandedLength = (pSourceData[2] << 16) + (pSourceData[3] << 8) +
pSourceData[4];
int filepos = 5;
int TargetOffset = 0;
if (pSourceData[0] & 0x01) filepos = 8; // align if necessary.
unsigned char *pExpandedData = (unsigned char *)malloc(ExpandedLength);
if (pExpandedData){
while(filepos < filesize && pSourceData[filepos] < 0xfc){
unsigned char pack_byte = pSourceData[filepos];
int a = pSourceData[filepos+1];
int b = pSourceData[filepos+2];
if (!(pack_byte & 0x80)){
int len = pack_byte&0x03;
unsigned char *pDest =
pExpandedData+TargetOffset;
unsigned char *pSrc = pSourceData+filepos+2;
TargetOffset += len;
filepos += (len+2);
while (len--) *pDest++ = *pSrc++;
len = ((pack_byte & 0x1c)>>2) + 3;
int offset = (pack_byte >> 5) + a + 1;
pDest = pExpandedData + TargetOffset;
pSrc = pDest-offset;
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
else if (!(pack_byte & 0x40)){
int len = (a >> 6) &0x03;
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+3;
filepos += (len+3);
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
int offset = (a&0x3f)*256 + b + 1;
pDest = pExpandedData + TargetOffset;
pSrc = pDest-offset;
len = (pack_byte & 0x3f)+4;
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
else if (!(pack_byte & 0x20)){
int c = pSourceData[filepos+3];
int len = (pack_byte & 0x03);
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+4;
filepos += (len+4);
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
int offset = ((pack_byte & 0x10)<<0x0c) + 256*a
+ b + 1;
pDest = pExpandedData + TargetOffset;
pSrc = pDest-offset;
len = ((pack_byte >> 2)&0x03)*256+c+5;
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
else {
int len = (pack_byte&0x1f)*4+4;
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+1;
filepos += (len+1);
TargetOffset += len;
while (len--) *pDest++ = *pSrc++;
}
}
if (filepos < filesize && TargetOffset < ExpandedLength){
unsigned char *pDest = pExpandedData +
TargetOffset;
unsigned char *pSrc = pSourceData+filepos+1;
int len = pSourceData[filepos]&0x03;
while (len--) *pDest++ = *pSrc++;
}
}
free(pSourceData);
return(pExpandedData);
}