.DXF File Format – Structure and group codes

Group code and value pairs

• An ASCII DXF file is a sequence of pairs, each on two lines: an integer group code, then its value. Leading spaces before the code are normal ( 0).
• The code’s range fixes the value’s type (below); the code and the object it belongs to fix its meaning.
• Code 0 begins every object, entity, table, table entry and section marker; an object ends where the next 0 begins.
• Strings are limited to 256 characters per line in the file (2049 single-byte characters for extended symbol names since AutoCAD 2000). Control characters are written as a caret and a letter (^G for BEL); a real caret is written ^ .
• Code 999 is a comment, ignored when reading and never written by AutoCAD.
• Angles are in degrees in DXF files (radians in AutoLISP and ObjectARX).

Sections

Each section starts with 0 SECTION, 2 name and ends with 0 ENDSEC. The file ends with 0 EOF.

The reference’s rules for files written by other programs: HEADER may be left out; any table, and the whole TABLES section, may be left out; LTYPE must come before LAYER if both are given; BLOCKS may be left out if no blocks are used, and must come before ENTITIES; layers named in entities but not defined are created with colour 7 and linetype CONTINUOUS; EOF must be present. A file of just an ENTITIES section and EOF is therefore valid.

Versions

The header variable $ACADVER (group 1) names the drawing database version:

From AC1021 on, text in a DXF file is UTF-8; earlier files use the code page named by $DWGCODEPAGE.

Group code value types

Group codes in numerical order

“Fixed” codes always mean the same thing; the others depend on the object. Negative codes appear only in AutoLISP and ObjectARX, never in files.

Tables and handles

  0
TABLE
  2
LAYER
  5
2
100
AcDbSymbolTable
 70
1
  0
LAYER
  5
10
100
AcDbSymbolTableRecord
100
AcDbLayerTableRecord
  2
0
 70
0
 62
7
  6
CONTINUOUS
  0
ENDTAB

Every object has a handle, a hexadecimal number unique in the drawing; the header variable $HANDSEED is larger than any handle in use. Group 330 gives the owner’s handle, and 102 groups {ACAD_REACTORS and {ACAD_XDICTIONARY list reactors and the extension dictionary. The DIMSTYLE table is an exception: its entries use group 105 for their handle.

Binary DXF

A binary DXF file, written by AutoCAD with the same .dxf extension, starts with a 22-byte sentinel:

41 75 74 6F 43 41 44 20 42 69 6E 61 72 79 20 44 58 46 0D 0A 1A 00
A  u  t  o  C  A  D     B  i  n  a  r  y     D  X  F  CR LF SUB NUL

Pairs follow in binary form. The group code is 2 bytes, little-endian (1 byte before Release 14, when extended data codes were written as a 255 byte followed by the 2-byte code). The value’s form follows from the code’s type:

The 999 comment is not used. Binary DXF keeps full double precision and is about a quarter smaller than ASCII. The sizes of the 32-bit, 64-bit and boolean values are not spelt out in the 2012 reference; they are those used by the ezdxf library’s binary reader.

Object coordinate systems

Planar entities (circles, arcs, 2D polylines, text, inserts and others) store their points in an object coordinate system defined only by the extrusion direction (210/220/230). Its axes come from the “arbitrary axis algorithm”: if the extrusion direction N has |N_x| < 1/64 and |N_y| < 1/64, the X axis is (0, 1, 0) × N, otherwise (0, 0, 1) × N, normalised; the Y axis is N × X. LINE, POINT, 3DFACE and 3D polylines use world coordinates.

Sections in detail

• Header variables – every HEADER variable, its group code and values
• Classes, tables and blocks – class records and default classes, every symbol table, BLOCK and ENDBLK
• Entities – the common entity codes and every entity
• Objects – the common object codes, every object, the thumbnail and extended data

Those pages follow the AutoCAD 2014 DXF Reference (March 2013), the newest edition Autodesk offers as a PDF; its structure chapters match the 2012 edition used here. Later editions are online help only; the 2018 one adds AC1032.