AutoCAD DXF

OGR supports reading most versions of AutoCAD DXF and writing AutoCAD 2000 version files. DXF is an ASCII format used for interchanging AutoCAD drawings between different software packages. The entire contents of the file is represented as a single layer named "entities".

DXF files are considered to have no georeferencing information through OGR. Features will all have the following generic attributes:

Layer: The name of the DXF layer. The default layer is "0".
SubClasses: Where available, a list of classes to which an element belongs.
ExtendedEntity: Where available, extended entity attributes all appended to form a single text attribute.
Linetype: Where available, the line type used for this entity.
EntityHandle: The hexadecimal entity handle. A sort of feature id.
Text: The text of labels.

Supported Elements

The following element types are supported:

POINT: Produces a simple point geometry feature.
MTEXT, TEXT: Produces a point feature with LABEL style information.
LINE, POLYLINE, LWPOLYLINE: translated as a LINESTRING. Rounded polylines (those with their vertices' budge attributes set) will be tessellated. Single-vertex polylines are translated to POINT.
CIRCLE, ELLIPSE, ARC: Translated as a LINESTRING, tessellating the arc into line segments.
INSERT: An attempt is made to insert the block definition as defined in the insert. Linework blocks are aggregated into a single feature with a geometry collection for the geometry. Text blocks are returned as one or more text features. To avoid merging blocks into a geometry collection the DXF_MERGE_BLOCK_GEOMETRIES config option may be set to FALSE.
DIMENSION: This element is exploded into a feature with arrows and leaders, and a feature with the dimension label.
HATCH: Line and arc boundaries are collected as a polygon geometry, but no effort is currently made to represent the fill style of HATCH entities.
3DFACE, SOLID: Translated as POLYGON.

A reasonable attempt is made to preserve line color, line width, text size and orientation via OGR feature styling information when translating elements. Currently no effort is made to preserve fill styles or complex line style attributes.

The approximation of arcs, ellipses, circles and rounded polylines as linestrings is done by splitting the arcs into subarcs of no more than a threshold angle. This angle is the OGR_ARC_STEPSIZE. This defaults to four degrees, but may be overridden by setting the configuration variable OGR_ARC_STEPSIZE.

DXF_INLINE_BLOCKS

The default behavior is for INSERT entities to be expanded with the geometry of the BLOCK they reference. However, if the DXF_INLINE_BLOCKS configuration option is set to the value FALSE, then the behavior is different as described here.

A new layer will be available called blocks. It will contain one or more features for each BLOCK defined in the file. In addition to the usual attributes, they will also have a BlockName attribute indicate what block they are part of.
The entities layer will have new attributes BlockName, BlockScale, and BlockAngle.
INSERT entities will populate these new fields with the corresponding information (they are null for all other entities).
INSERT entities will not have block geometry inlined - instead they will have a point geometry for the insertion point.

The intention is that with DXF_INLINE_BLOCKS disabled, the block references will remain as references and the original block definitions will be available via the blocks layer. On export this configuration will result in the creation of similar blocks.

Character Encodings

Normally DXF files are in the ANSI_1252 / Win1252 encoding. GDAL/OGR attempts to translate this to UTF-8 when reading and back into ANSI_1252 when writing. DXF files can also have a header field ($DWGCODEPAGE) indicating the encoding of the file. In GDAL 1.8.x and earlier this was ignored but from GDAL 1.9.0 and later an attempt is made to use this to recode other code pages to UTF-8. Whether this works will depend on the code page naming and whether GDAL/OGR is built against the iconv library for character recoding.

In some cases the $DWGCODEPAGE setting in a DXF file will be wrong, or unrecognised by OGR. It could be edited manually, or the DXF_ENCODING configuration variable can be used to override what id will be used by OGR in transcoding. The value of DXF_ENCODING should be an encoding name supported by CPLRecode() (i.e. an iconv name), not a DXF $DWGCODEPAGE name. Using a DXF_ENCODING name of "UTF-8" will avoid any attempt to recode the text as it is read.

Creation Issues

DXF files are written in AutoCAD 2000 format. A standard header (everything up to the ENTITIES keyword) is written from the $GDAL_DATA/header.dxf file, and the $GDAL_DATA/trailer.dxf file is added after the entities. Only one layer can be created on the output file.

Point features with LABEL styling are written as MTEXT entities based on the styling information.

Point features without LABEL styling are written as POINT entities.

LineString and MultiLineString features are written as one or more LWPOLYLINE entities, closed in the case of polygon rings. An effort is made to preserve line width and color.

Polygon and MultiPolygon features are written as HATCH entities. The dataset creation supports the following dataset creation options:

HEADER=filename: Override the header file used - in place of header.dxf located in the GDAL_DATA directory.
TRAILER=filename: Override the trailer file used - in place of trailer.dxf located in the GDAL_DATA directory.

Note that in GDAL 1.8 and later, the header and trailer templates can be complete DXF files. The driver will scan them and only extract the needed portions (portion before or after the ENTITIES section).

Block References

It is possible to export a "blocks" layer to DXF in addition to the "entities" layer in order to produce actual DXF BLOCKs definitions in the output file. It is also possible to write INSERT entities if a block name is provided for an entity. To make this work the follow conditions apply.

A "blocks" layer may be created, and it must be created before the entities layer.
The entities in the blocks layer should have the BlockName field populated.
Objects to be written as INSERTs in the entities layer should have a POINT geometry, and the BlockName field set.
If a block (name) is already defined in the template header, that will be used regardless of whether a new definition was provided in the blocks layer.

The intention is that a simple translation from DXF to with DXF_INLINE_BLOCKS set to FALSE will approximately reproduce the original blocks and keep INSERT entities as INSERT entities rather than exploding them.

Layer Definitions

When writing entities, if populated the Layer field is used to set the written entities layer. If the layer is not already defined in the template header then a new layer definition will be introduced, copied from the definition of the default layer ("0").

Line Type Definitions

When writing LWPOLYLINE entities the following rules apply with regard to Linetype definitions.

If the Linetype field is set on the written features, and that Linetype is already defined in the template header then it will be referenced from the entities regardless of whether an OGR style string existed.
If the Linetype is set, but the Linetype is not predefined in the header template, then a definition will be added if the feature has an OGR style string with a PEN tool and a "p" pattern setting.
If the feature has no Linetype field set, but it does have an OGR style string with a PEN tool with a "p" pattern set then an automatically named Linetype will be created in the output file.
It is assumed that patterns are using "g" (georeferenced) units for defining the line pattern. If not the scaling of the DXF patterns is likely to be wrong - potentially very wrong.

The intention is that "dot dash" style patterns will be preserved when written to DXF and that specific linetypes can be predefined in the header template, and referenced using the Linetype field if desired.

Units

At the moment GDAL writes DXF to report the measurement units as "English - Inches". For changing the units edit $MEASUREMENT and $INSUNITS variables in the header template.