.DWG File Format – Bit codes and data types

Data type codes

A seeker is an RL holding an absolute file address or an offset from a known address. Some prescriptions also use EC (for vertex flags), S, L and D, which chapter 2 does not define; see Not confirmed.

3B: bit triplet

Read bits until a 0 bit is read or three bits have been read, whichever comes first. Each time a bit is read, shift the bits already read one place left. The result is 0 to 7.

BS: bitshort

The one-byte form is for positive values below 256; negative values use the short form, so −1 is 00 11111111 11111111, not 01 11111111.

0000000001000000011011010000111110

00 00000001 00000001   short 257
10                     0
11                     256
01 00001111            15
10                     0

BL: bitlong

As with BS, negative values use the full form.

000000000100000001000000000000000010010000111110

00 00000001 00000001 00000000 00000000   long 257
10                                       0
01 00001111                              15
10                                       0

BLL: bitlonglong

A 3B gives a length l, then l bytes follow, least significant first.

BD: bitdouble

MC: modular char

A run of bytes ending at the first byte whose high bit is 0. In each byte the high bit means “another byte follows” and the other seven bits are data, least significant group first. If the final byte also has bit 0x40 set, that bit is cleared and the number is negative. In the object map, modular chars hold handle offsets, which are never negative.

MS: modular short

Like a modular char, but in 16-bit little-endian units: the high bit of each short means another follows, and the other 15 bits are data. Modular shorts hold object sizes, so in practice there are one or two units.

BE: bit extrusion

In R13 and R14, a 3BD. From R2000, one bit: 1 means the extrusion is (0, 0, 1) and nothing more is stored; 0 means a 3BD follows.

DD: bitdouble with default

A 2-bit code, then data, applied to a default value that the field’s description names (often the previous value):

BT: bit thickness

In R13 and R14, a BD. From R2000, one bit: 1 means the thickness is 0.0; 0 means a BD follows.

Colours: CMC, ENC and TC

In R2000 and earlier a colour is a BS colour index. From R2004 there are two forms. CMC, used by objects:

ENC, used by entities, may refer to a DBCOLOR object and carry a transparency. It starts with a BS whose high byte holds flags and whose low bits hold the colour index:

TC (true colour) is the same as CMC in R2004 and later.

OT: object type

Up to R2007 the object type is a BS. From R2010 it is a 2-bit code followed by one or two bytes:

The type numbers are listed on Objects and entities.

H: handle references

Every object in an R13 or later file has a handle, and objects refer to each other by handle. A reference is stored as |CODE (4 bits)|COUNTER (4 bits)|HANDLE or OFFSET|: COUNTER is the number of handle bytes that follow, most significant first. The specification writes a reference as CODE.COUNTER.BYTE…, for example 5.1.0F, the usual reference to layer 0 (handle F) in drawings made by R13.

Where the reference expresses ownership, CODE gives its kind (TYPEDOBJHANDLE):

Elsewhere the handle is an offset from a reference handle, always a soft pointer (OFFSETOBJHANDLE):

5        2        05       E7
0101     0010     00000101 11100111

Strings: T, TU and TV

Up to R2004, text (T) is a BS length followed by that many bytes in the drawing’s code page. From R2007, text is Unicode (TU): a BS character count and UTF-16 characters, 2 bytes each. Field lists write TV for “T before R2007, TU from R2007”.

In R2007 and later, an object’s Unicode strings are not where the field list puts them: they are gathered in a string stream at the end of the object’s data, and the handles in a separate handle stream after that. How to find the string stream is described on Objects and entities. The header variables and classes sections use the same arrangement in R2007 and later.

Extended entity data and XRECORD use other string forms: in R13 to R2004 a 1-byte length, a 2-byte code page and the bytes; in R2007 and later a 2-byte length and that many UTF-16 characters (see Extended entity data).

8-bit CRC

Despite its name this CRC is 2 bytes long: a byte-at-a-time CRC with a 256-entry table of 16-bit values. CRCs are always on byte boundaries and never bit coded, so an object may end with unused bits before its CRC. AutoCAD uses a non-zero starting value in places, and sometimes XORs the result with a “magic number” (in R13 and later, only in the file header).

unsigned short crc8(unsigned short dx, const unsigned char *p, long n)
{
    while (n-- > 0) {
        unsigned char al = (unsigned char)(*p ^ (dx & 0xFF));
        dx = (dx >> 8) & 0xFF;
        dx = dx ^ crctable[al];
        p++;
    }
    return dx;
}

To accumulate a CRC, call it with 0 (or the starting value for that kind of data) and then with the last result. Starting values used: 0xC0C1 for the header variables; 0 with an XOR for the file header.

01000000 11100000 01010101 01010101

01         bitshort with one byte
00000011   value 3
100000     unused bits
01010101 01010101   the CRC

32-bit CRC

From R2004 a 32-bit CRC is also used, with a 256-entry table of 32-bit values that starts 0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA:

uint32_t crc32(const uint8_t *p, uint32_t n, uint32_t seed)
{
    uint32_t inverted = ~seed;
    while (n--) {
        uint8_t byte = *p++;
        inverted = (inverted >> 8) ^ crc32Table[(inverted ^ byte) & 0xff];
    }
    return ~inverted;
}

R2007 adds two 64-bit CRCs, described on File structure.

What changes by version

Not confirmed

• The codes EC, S, L and D appear in some prescriptions (for example the vertex flags and the DICTIONARY item count) but are not defined in chapter 2.
• Section 2.11 says colours are a BS index in R15 (R2000) and earlier, but the common entity format in section 20.2 lists an ENC entity colour for “R2000+”. The table above follows section 2.11.

Sources

• Open Design Specification for .dwg files, version 5.4.1 – Open Design Alliance; chapter 2 (bit codes and data definitions), and the string forms in chapters 20 and 28