
================================================================================
2.2.0
================================================================================

Changes of Note
===============

-  The Zstandard compression library is no longer required in order to
   build PDAL.
-  The Ubuntu conda image now installs all of PDAL under ``/usr``.
-  The default type of the ClusterID dimension has changed from
   ``uint64_t`` to ``int64_t``.
-  Most operations now store data using ``ColumnPointTable``, which
   stores data in arrays of dimensions, rather than arrays of point
   data. This can improve the performance of some operations
   substantially. Generally, this change should not be noticeable to
   users.
-  Normal refinement (``refine`` option) in ``filters.normal`` is now
   off by default. (#3133)
-  All filters now accept a ``where`` option to allow points to skip the
   filters. (#3118)
-  PDAL now searches for plugins in the same directory containing the
   PDAL library. (#3181)

API User Notes
==============

-  ``ColumnPointTable`` is now the default PDAL point table. If you have
   written code that expects row-major data rather than column-major
   data, be sure to use ``RowPointTable`` instead.

New Stages
==========

-  ``filters.lloydkmeans`` labels points with a ``ClusterID`` based on
   Lloyd’s algorithm. (#3075)
-  ``filters.litree`` provides segmentation of points into trees and
   non-trees. (#3111)
-  ``filters.faceraster`` creates a raster from an existing
   triangulation. (#3152)
-  ``filters.optimalneighborhood`` computes the optimal number of
   neighbors and corresponding radii (#3180)
-  ``reader.obj`` allows reading of OBJ files.

Removed Stages
==============

-  ``filters.hag`` has been removed. Use ``filters.hag_dem``,
   ``filter.hag_delaunay`` or ``filters.hag_nn`` instead.

Deprecated Stages
=================

-  ``filters.ferry``. Use the ``value`` option of ``filters.assign``
   instead.
-  ``writers.oci``. OCI support will be removed in a future release.
-  ``readers.oci``. OCI support will be removed in a future release.

Enhancements
============

-  ``filters.colorinterp`` now provides the ``clamp`` option to force
   values to be in the range [min, max]. (#2729 - Thanks to Matt
   Csencsits)
-  Long options can now be specified on the command line without and
   equal sign (‘=’) separating the option name and value. (#3016)
-  Added ``seed`` option ``readers.faux`` in order to allow consistent
   generation of random data. (#3014)
-  Improvements have been made to cmake installation to better support
   downstream projects. (#2262)
-  Added ``color_vertices`` option to ``writers.gltf`` to allow writing
   of RGB and normals. (#3043 - Thanks to Emma Krantz)
-  Added ``classbits`` option to ``filters.smrf`` to specify points to
   be ignored by the SMRF algorithm. (#3055)
-  Delaunay triangulation performance has been improved on systems using
   libstdc++. (#3076)
-  ``filters.covariancefeatures`` now provides the options ``min_k`` and
   ``radius`` to support radius search of neighbors in addition to
   k-nearest. (#3083)
-  ``filters.cluster`` now provide the option ``is3d`` to support
   2-dimensional clustering. (#3084)
-  ``filters.icp`` now emits a fully composed transformation matrix as
   metadata that can be provided to ``filters.transformation`` to modify
   point locations using a single operation. (#2939, #2962)
-  ``filters.eigenvalues`` added the options ``stride``, ``radius`` and
   ``min_k`` to allow radius and strided neighbor searching. (#3132)
-  ``writers.las`` now accepts the ``filename`` key in the ``vlrs``
   option to import VLR data from files. (#3146)
-  ``filters.transformation`` now provides the ``invert`` option to
   apply the inverse of the provided ``matrix``. (#3126)
-  ``readers.ept`` now returns a clipped boundary in the summary to
   match the behavior of the point count. (#3168)
-  ``filters.head`` is now streamable.
-  ``readers.i3s`` is now streamable.
-  ``readers.slpk`` no longer expands the source file into a temporary
   directory.
-  ``filters.assign`` added the ``value`` option to allow standard
   expression-based conditional assignment of values.(#1468)
- ``readers.tiledb`` and ``writers.tiledb`` now support TileDB version
  2 and duplicate points when using version 2 of TileDB. (#3030, #3047)

Bug Fixes
=========

-  A problem where the classifications of voting points were being
   changed in ``filters.neighborclassifier`` has been fixed. (#3210)
-  Fixed an issue where NITF-wrapped LAS files may fail to open (#3010)
-  Reading and writing of E57 files over 2GB in size on Windows has been
   fixed. (#3000)
-  Fixed various compilations issues in ``filters.matlab``. (#3013)
-  Added better handling for some cases of coordinate transformation
   failure. (#3009)
-  An error in calculating position for some data in ``filters.smrf``
   has been fixed. (#3069)
-  An error in calculating neighbors for isolated points in
   ``filters.smrf`` has been fixed. (#3071)
-  ``readers.las`` now creates dimensions ``ScanChannel`` and
   ``ClassFlags`` only when reading point formats 6-10. It used to
   create them for any version 1.4 file. (#3092)
-  ``filters.poisson`` now throws exceptions instead of calling
   ``exit()`` when it encounters a fatal error. (#3094)
-  Errors that could occur in ``gdal::Raster::statistics()`` are now
   returned to the caller, rather than resulting in a crash. (#3097)
-  ``pdal tindex`` now ensures that traditional GIS axis order is used.
   (#3099)
-  ``readers.tindex`` now uses the file SRS rather than the tile index
   column SRS unless overridden by the ``srs_column`` option. (#3012)
-  An error in ``readers.ept`` that could create bad data when both the
   ``bounds`` and ``addons`` options were specified has been fixed.
   (#3054)
-  ``filters.hag_nn`` and ``filter.hag_delaunay`` now log an error
   rather than terminate if no ground points are found. (#3112)
-  An error in reporting point count was when the number of points
   exceeded the maximum 32-bit value in ``readers.ept`` has been fixed.
   (#3120)
-  ``pdal tindex`` now stores WKT instead of EPSG codes to handle
   vertical SRS instances. (#3100)
-  A bug in ``filters.crop`` that didn’t properly handle 2D cropping has
   been fixed. (#3115)
-  ``readers.text`` now properly handles quoted dimension names in the
   header. (#2201)
-  Windows-style directory separators are now properly handled in cases
   using relative filenames. (#2936)
-  An exception reading E57 points with a transform has been fixed.
   (#3163)
-  Fix a failure in ``readers.pcd`` when preview() was called before
   execute(). (#3169)
-  The compression level for ``writers.tiledb`` is now properly set to
   7. (#3198)
-  An issue where ``filters.neighborclassifier`` would modify
   classifications as it calculated them has been fixed. (#3210)
-  An extra comma was written using ``readers.text`` when creating
   GEOJson output. This has been fixed. (#3204)

================================================================================
2.2.0
================================================================================

Changes of Note
===============

-  The Zstandard compression library is no longer required in order to
   build PDAL.
-  The Ubuntu conda image now installs all of PDAL under ``/usr``.
-  The default type of the ClusterID dimension has changed from
   ``uint64_t`` to ``int64_t``.
-  Most operations now store data using ``ColumnPointTable``, which
   stores data in arrays of dimensions, rather than arrays of point
   data. This can improve the performance of some operations
   substantially. Generally, this change should not be noticeable to
   users.
-  Normal refinement (``refine`` option) in ``filters.normal`` is now
   off by default. (#3133)
-  All filters now accept a ``where`` option to allow points to skip the
   filters. (#3118)
-  PDAL now searches for plugins in the same directory containing the
   PDAL library. (#3181)

API User Notes
==============

-  ``ColumnPointTable`` is now the default PDAL point table. If you have
   written code that expects row-major data rather than column-major
   data, be sure to use ``RowPointTable`` instead.

New Stages
==========

-  ``filters.lloydkmeans`` labels points with a ``ClusterID`` based on
   Lloyd’s algorithm. (#3075)
-  ``filters.litree`` provides segmentation of points into trees and
   non-trees. (#3111)
-  ``filters.faceraster`` creates a raster from an existing
   triangulation. (#3152)
-  ``filters.optimalneighborhood`` computes the optimal number of
   neighbors and corresponding radii (#3180)
-  ``reader.obj`` allows reading of OBJ files.

Removed Stages
==============

-  ``filters.hag`` has been removed. Use ``filters.hag_dem``,
   ``filter.hag_delaunay`` or ``filters.hag_nn`` instead.

Deprecated Stages
=================

-  ``filters.ferry``. Use the ``value`` option of ``filters.assign``
   instead.

Enhancements
============

-  ``filters.colorinterp`` now provides the ``clamp`` option to force
   values to be in the range [min, max]. (#2729 - Thanks to Matt
   Csencsits)
-  Long options can now be specified on the command line without and
   equal sign (‘=’) separating the option name and value. (#3016)
-  Added ``seed`` option ``readers.faux`` in order to allow consistent
   generation of random data. (#3014)
-  Improvements have been made to cmake installation to better support
   downstream projects. (#2262)
-  Added ``color_vertices`` option to ``writers.gltf`` to allow writing
   of RGB and normals. (#3043 - Thanks to Emma Krantz)
-  Added ``classbits`` option to ``filters.smrf`` to specify points to
   be ignored by the SMRF algorithm. (#3055)
-  Delaunay triangulation performance has been improved on systems using
   libstdc++. (#3076)
-  ``filters.covariancefeatures`` now provides the options ``min_k`` and
   ``radius`` to support radius search of neighbors in addition to
   k-nearest. (#3083)
-  ``filters.cluster`` now provide the option ``is3d`` to support
   2-dimensional clustering. (#3084)
-  ``filters.icp`` now emits a fully composed transformation matrix as
   metadata that can be provided to ``filters.transformation`` to modify
   point locations using a single operation. (#2939, #2962)
-  ``filters.eigenvalues`` added the options ``stride``, ``radius`` and
   ``min_k`` to allow radius and strided neighbor searching. (#3132)
-  ``writers.las`` now accepts the ``filename`` key in the ``vlrs``
   option to import VLR data from files. (#3146)
-  ``filters.transformation`` now provides the ``invert`` option to
   apply the inverse of the provided ``matrix``. (#3126)
-  ``readers.ept`` now returns a clipped boundary in the summary to
   match the behavior of the point count. (#3168)
-  ``filters.head`` is now streamable.
-  ``readers.i3s`` is now streamable.
-  ``readers.slpk`` no longer expands the source file into a temporary
   directory.
-  ``filters.assign`` added the ``value`` option to allow standard
   expression-based conditional assignment of values.(#1468)

Bug Fixes
=========

-  A problem where the classifications of voting points were being
   changed in ``filters.neighborclassifier`` has been fixed. (#3210)
-  Fixed an issue where NITF-wrapped LAS files may fail to open (#3010)
-  Reading and writing of E57 files over 2GB in size on Windows has been
   fixed. (#3000)
-  Fixed various compilations issues in ``filters.matlab``. (#3013)
-  Added better handling for some cases of coordinate transformation
   failure. (#3009)
-  An error in calculating position for some data in ``filters.smrf``
   has been fixed. (#3069)
-  An error in calculating neighbors for isolated points in
   ``filters.smrf`` has been fixed. (#3071)
-  ``readers.las`` now creates dimensions ``ScanChannel`` and
   ``ClassFlags`` only when reading point formats 6-10. It used to
   create them for any version 1.4 file. (#3092)
-  ``filters.poisson`` now throws exceptions instead of calling
   ``exit()`` when it encounters a fatal error. (#3094)
-  Errors that could occur in ``gdal::Raster::statistics()`` are now
   returned to the caller, rather than resulting in a crash. (#3097)
-  ``pdal tindex`` now ensures that traditional GIS axis order is used.
   (#3099)
-  ``readers.tindex`` now uses the file SRS rather than the tile index
   column SRS unless overridden by the ``srs_column`` option. (#3012)
-  An error in ``readers.ept`` that could create bad data when both the
   ``bounds`` and ``addons`` options were specified has been fixed.
   (#3054)
-  ``filters.hag_nn`` and ``filter.hag_delaunay`` now log an error
   rather than terminate if no ground points are found. (#3112)
-  An error in reporting point count was when the number of points
   exceeded the maximum 32-bit value in ``readers.ept`` has been fixed.
   (#3120)
-  ``pdal tindex`` now stores WKT instead of EPSG codes to handle
   vertical SRS instances. (#3100)
-  A bug in ``filters.crop`` that didn’t properly handle 2D cropping has
   been fixed. (#3115)
-  ``readers.text`` now properly handles quoted dimension names in the
   header. (#2201)
-  Windows-style directory separators are now properly handled in cases
   using relative filenames. (#2936)
-  An exception reading E57 points with a transform has been fixed.
   (#3163)
-  Fix a failure in ``readers.pcd`` when preview() was called before
   execute(). (#3169)
-  The compression level for ``writers.tiledb`` is now properly set to
   7. (#3198)
-  An issue where ``filters.neighborclassifier`` would modify
   classifications as it calculated them has been fixed. (#3210)

================================================================================
2.1.0
================================================================================

Changes of Note
===============

-  The X, Y and Z dimensions are now always stored internally as
   double-precision floating point. This was normally case previously,
   but it is now enforced. (#2892)
-  The interface include file directory is now set for downstream
   projects using cmake, eliminating the need to explicitly set the
   location of PDAL header files. (#2262)
-  PDAL is now always built using the arbiter library, which is included
   in the source distribution.
-  ``filters.python`` has been removed from the primary PDAL library and
   is part of the `PDAL python <https://pdal.io/python.html>`__ library.
-  When the ``bounds`` option of ``filters.crop`` represents a 3D
   volume, the filter now does 3D cropping, rather than 2D cropping as
   in previous releases. (#2904)
-  ``writers.ply`` no longer accepts ``default`` as a valid value for
   the ``storage_mode`` option. (#2385)

API User Notes
==============

-  Support has been added for range-based iteration of PointView
   objects. The dereferenced iterator yields a PointRef object.
-  We are now providing private header files as part of the
   distribution. The API exposed by these files may change or be
   eliminated in a future release. Use of these files may require the
   installation of the header files from other packages. The symbols for
   some of these functions may not be public, especially on Windows.
-  There are now enumerations for classifications (pdal::ClassLabel).
-  N-dimensional KD-tree support has been added.
-  There are some small API changes (const changes, for example), but
   almost all code using the PDAL library should build without
   modification.
-  We do not guarantee ABI compatibility, though in most cases this
   release should be backward compatible.

New Stages
==========

-  ``filters.voxeldownsize`` provides streamed support for voxel
   filtering. (#2699 - Thanks
   `Helix.re <http://https://helixre.com/>`__!)
-  ``filters.separatescanline`` splits input into separate output point
   views based on associated scan lines. (#2724 - Thanks Guilhem
   Villemin)
-  ``filters.farthestpointsampling`` add points from the input to the
   output one-at-a-time by selecting the point in the input that is the
   furthest from any point in the output. (#2761)
-  ``filters.skewnessbalancing`` classifies ground points using the
   method of Bartels. (#2740)
-  ``filters.dbscan`` provides density-based clustering. (#2748)
-  ``filters.csf`` provides ground classification of points using the
   cloth simulation method of Zhang. (#2765)
-  ``readers.hdf`` provides generic support for many HDF files. (#2934)
-  ``filters.hag_nn``, ``filters.hag_delaunay`` and ``filters.hag_dem``
   have been created to replace ``filters.hag``. (#2886)

Deprecated Stages
=================

-  ``filters.hag`` is deprecated and will be removed in a future
   release. Use ``filters.hag_nn``, ``filters.hag_delaunay`` and
   ``filters.hag_dem`` instead.

Removed Stages
==============

-  ``readers.sqlite`` and ``writers.sqlite``

Enhancements
============

-  ``readers.ept`` now supports the ``headers`` and ``query`` options to
   be forwarded to remote endpoints. (#2690)
-  ``readers.ept`` now supports multipolygons for point filtering.
   (#2758)
-  ``readers.ept`` now supports asynchronous operation when streaming.
   (#2772)
-  ``readers.ept`` now supports filtering via OGR queries. (#2295)
-  ``writers.ept`` now supports Zstandard compression. (#2561)
-  ``writers.gdal`` now supports a ``power`` option that can be used to
   impact the output of inverse distance weighting mode. (#2550)
-  Eliminated an error as exception in ``readers.e57`` in order to
   improve performance. (#2721)
-  ``readers.rxp`` now supports reading the ``EdgeOfFlightLine``
   dimension. (#2367)
-  ``filters.randomize`` now has a ``seed`` option to initialize the
   random number generator. (#2736)
-  ``filters.rdb`` and ``filters.rxp`` now support fetching data from
   remote sources.
-  ``filters.hexbin`` now has the option ``smooth``, to smooth generated
   geometry, and ``preserve_topology`` that forces creation of valid
   geometry during the smoothing operation.
-  ``filters.transformation`` now has a ``spatialreference`` option to
   set the spatial reference of output data. (#2796)
-  ``filters.overlay`` now uses the same point-in-polygon algorithm used
   by ``filters.crop``. (#2812)
-  ``writers.tiledb`` now supports TileDB attribute filters. (#2828)
-  Output from ``pdal translate`` can now be directed to the null device
   by using either ``devnull`` or ``/dev/null`` as the output filename.
   (#2835)
-  ``filters.transformation`` now supports providing a filename as the
   ``matrix`` option if the referenced file contains a matrix that
   matches the format supported inline. (#2856)
-  ``writers.las`` now allows specification of a ``filesource_id`` as an
   option. (#2862)
-  ``readers.text`` now handles quoted dimension names in a header line.
   (#2201)
-  ``pdal info`` now reports the reader used for a file, the current
   time and the size of the file. (#2900)
-  ``filters.reprojection`` now supports the options
   ``in_axis_ordering`` and ``out_axis_ordering`` to allow the matching
   feature in GDAL 3+ be accessed. (#2916)
-  ``filters.normal`` added the ``refine`` option to re-orient normal
   vectors using a minimum spanning tree. (#2860)

Bug Fixes
=========

-  ``readers.e57`` and ``writers.e57`` now build cleanly on Windows.
-  ``writers.pgpointcloud`` no longer uses the wrong schema when
   creating a new database and there is an existing database using the
   same SRS. (#2692)
-  Fixed a potential crash when destroying logs on shutdown. (#2702)
-  Fixed a case where ``filters.voxelcentroidnearestneighbor`` could
   return points for a voxel that weren’t actually in the voxel. (#2703)
-  Fixed a typo in the alternate names of the ``NormalY`` dimension.
   (#2723)
-  ``filters.divider`` now handles empty input. (#2727)
-  An error in calculating variance, standard deviation, skewness and
   kurtosis has been fixed in ``filters.stats``. (#2790)
-  An error where a list of dimensions would be expanded improperly in
   ``writers.ply`` under certain API uses. (#2791)
-  Crashes in filters using the delaunay triangulation code have been
   fixed. (#2805)
-  ``writers.pcd`` now always writes dimension names as lowercase.
   (#2789)
-  The extended VLR offset was not being written to LAS files. This has
   been fixed. (#2823)
-  Fixed some small memory leaks.
-  ``filters.elm`` now works properly when input is an empty point view.
   (#2921)
-  ``filters.pmf`` now classifies all points either as “Ground” or
   “Unclassified”. Points are no longer marked as “Created, Never
   Classified”. (#2898)
-  LAS VLRs specified in a pipeline are now properly handled by
   ``writers.las``. (#2937)
-  When not building with LASzip or LAZperf, ``writers.las`` would fail
   to properly set a spatial reference in some cases. This has been
   fixed.
-  ``writers.las`` now always uses WKT version 1. In the previous
   release it may have used WKT version 2 in some cases. (#2943)
-  A bug in ``pdal tile`` where the input spatial reference wasn’t set
   on the output has been fixed. (#2976)

================================================================================
2.0.0
================================================================================

Changes of Note
===============

-  Python support now requires at least version 3.5+ of Python and
   version 1.13 of Numpy.
-  ``filters.icp`` is no longer a plugin and provides options not
   previously available.
-  Linear algebra that was using single-precision math is now using
   double-precision math.
-  Support was added for GDAL version 3+.
-  PCL support has been removed. PCD read/write and ``filters.icp`` is
   now natively supported. The PCL filters have been moved to the
   `unsupported-plugins <https://github.com/PDAL/unsuppported-plugins>`__
   repository for those that might need functionality they provided.
-  PDAL now uses a JSON library that doesn’t support comments. If your
   pipelines contains non-standard JSON, they will no longer be
   accepted.
-  PDAL now supports GCC 9.2, clang 8.0.1 and Visual Studio 17

New Stages
==========

-  ``filters.covariancefeatures`` computes statistics about features of
   points in a neighborhood. (#2455 - Thanks
   `Helix.re <http://https://helixre.com/>`__!)
-  ``readers.e57`` reads the E57 format. (#2442 - Thanks
   `Helix.re <http://https://helixre.com/>`__!)
-  ``writers.e57`` writes the E57 format. (#2442 - Thanks
   `Helix.re <http://https://helixre.com/>`__!)
-  ``writers.gltf`` provides simple glTF mesh output support. (#2643)
-  ``filters.reciprocity`` computes the percentage of points that are
   considered uni-directional. (#2504 - @chambbj)
-  ``filters.miniball`` computes a criterion for neighbors based on the
   miniball algorithm. (#2504 - @chambbj)
-  ``filters.planefit`` computes a deviation of a point from a manifold
   approximating its neighbors. (#2504 - @chambbj)
-  ``readers.memory`` provides an API interface for direct reading of
   row-oriented point data. (#2620)
-  ``filters.projpipeline`` processes point through a `PROJ
   pipeline <https://proj.org/operations/transformations/deformation.html>`__.
   (#2626 - Thanks @vilaa!)

Deprecated Stages
=================

-  readers.sqlite (will be moved to
   `unsupported-plugins <https://github.com/PDAL/unsuppported-plugins>`__
   in a future release)
-  writers.sqlite (will be moved to
   `unsupported-plugins <https://github.com/PDAL/unsuppported-plugins>`__
   in a future release)
-  readers.geowave (will be moved to
   `unsupported-plugins <https://github.com/PDAL/unsuppported-plugins>`__
   in a future release)
-  writers.geowave (will be moved to
   `unsupported-plugins <https://github.com/PDAL/unsuppported-plugins>`__
   in a future release)

Removed Stages
==============

-  ``filters.mongus``
-  ``filters.kdistance``
-  ``readers.greyhound``

Enhancements
============

-  ``readers.mrsid`` now supports more recent versions of the Mr. Sid
   SDK. (#2460)
-  ``readers.tiledb`` and ``writers.tiledb`` now support stream mode
   processing. It also supports an ``append`` option.
-  ``readers.ept`` now accepts an optional ``\<spatial reference>``
   after the bounds specification in the ``bounds`` option to indicate
   the SRS of the provided boundary. (#2295)
-  ``filters.hag`` now supports interpolation from a TIN (triangulated
   network) of points. (#2520 - Thanks `James
   McClain <https://github.com/jamesmcclain>`__!)
-  ``filters.poisson`` now properly applies an inverse transform to
   points it creates. (#1840)
-  ``pdal --drivers`` now shows the file extensions associated with each
   driver. It also shows when stages are streamable. (#2571, #2528)
-  Some threaded execution has been added to ``readers.ept``.
-  Alternate dimension names have been added for “Red”, “Green” and
   “Blue” to better support PLY input. (#2570)

Bug Fixes
=========

-  Fixed a problem where data in a forwarded LAS VLR would be lost.
   (#2542)
-  Clear dimension list in ``filters.colorization`` to eliminate the
   possibility of writing invalid data. (#2487)
-  Delete code that removed control characters when converting to JSON.
   (#2355)
-  Fixed an issue where nested geometries were being written as invalid
   WKT in ``filter.hexbin``. (#2507)
-  Fixed a segmentation fault in ``readers.bpf`` if the specified file
   didn’t exist. (#2515)
-  Fixed an issue in ``writers.gdal`` where interpolation would
   terminate if the process hit the raster edge. (#2545)
-  Fixed an issue writing JSON when characters needed to be escaped.
   (#2558)
-  Delaunay triangulation no longer segfaults with fewer than three
   points. (#2513)
-  Issues with reading binary data with ``filters.ept`` have been fixed.
   (#2585)
-  ``pdal`` now errors if an option is provided that wasn’t expected.
   (#2574)
-  Forwarding for scale with multiple LAS inputs was fixed. (#2663)

================================================================================
1.9.1
================================================================================

Version 1.9.1 is a bug fix release.

Bug Fixes
---------

-  When writing numerous LAS files with the PDAL API, an internal list
   of fields to forward could grow excessively. This has been fixed.
   (#2452)
-  A computation in ``writers.gdal`` that could overflow with very large
   rasters has been fixed. (#2454)
-  A bug that could result in the same pipeline tag name being used for
   multiple stages of the same type has been fixed. (#2458)
-  A potential segmentation fault in ``pdal density`` has been fixed.
   (#2461)
-  A segmentation fault in ``filters.chipper`` when processing multiple
   inputs has been fixed. (#2479)
-  An issue with ``readers.ept`` overwriting temporary files before they
   weren’t needed has been fixed. (#2474)
-  Fixed an issue where ``filters.icp`` would create data that appeared
   invalid due to the precision of data when positions weren’t located
   near the origin. (#2492)

================================================================================
1.9.0
================================================================================

Changes of Note
===============

-  PDAL now requires at least version 2.2 of GDAL.
-  Pipelines can optionally be specified as an array of stages (without
   an encapsulating “pipeline” object).
-  The location of installed cmake files has been changed to match the
   cmake convention.
-  ``filters.delaunay`` now uses the delaunator library instead of
   geogram.
-  PDAL no longer depends on the GEOS library. However, some polygon
   functions will not work if PDAL isn’t built with a version of the
   GDAL library that has GEOS support.
-  PDAL now uses the standard cmake script to locate the necessary GDAL
   files when building.
-  ``readers.sbet`` now reads angles as degrees rather than radians.
   (#2354)
-  The ``tindex`` command now requires one of the subcommands ``create``
   or ``merge``. It no longer accepts the ``merge`` option, which has
   been replaced by the ``merge`` subcommand.
-  The argument for the function ``PipelineManager::executeStream`` is
   now of type ``StreamPointTable&`` instead of ``FixedPointTable&``.

Enhancements
============

-  All readers now accept the ``default_srs`` and ``override_srs``
   options. ``override_srs`` is an alias for the existing
   ``spatialreference`` option. ``default_srs`` applies only if no
   spatial reference exists in the source file. (#2194, #2195)
-  Added some dimension detection capability to ``readers.pts``. (#2317)
-  Hardened API use of LASzip in ``writers.las`` for points with data
   format 6 and above and set the extended_point_type variable. (#2320,
   #2329)
-  Support the ``libgeotiff`` path suffix when searching for the geotiff
   library with cmake.
-  Added a ``resolution`` option to ``readers.ept`` to limit the number
   of points read. (#2331)
-  ``readers.rxp`` now supports stream mode.
-  Added the ``invert`` option to ``filters.head`` and ``filters.tail``.
-  Added support for subcommands in kernels. (#2293)

New stages
==========

-  ``filters.mongo`` - Filtering of points using the mongo expression
   language.
-  ``readers.tiledb`` - Reads points from the tileDB database.
-  ``writers.tiledb`` - Writes points to the tileDB database.
-  ``writers.ept_addon`` - Adds sidecar data to an existing EPT dataset.

Deprecated stages
=================

-  ``filters.kdistance`` - Replaced by ``filters.nndistance``.
   ``filters.kdistance`` will be removed in the next PDAL release.
-  ``filters.mongus`` - Replaces by ``filters.pmf`` and ``filters.smrf``

Bug fixes
=========

-  The tindex command now uses sufficient precision when using the
   ``fast_boundary`` option (#2271)
-  ``filters.smrf`` and ``filters.pmf`` now process all points if the
   ``NumberOfReturns`` or ``ReturnNumber`` dimension is missing. This
   fixes a regression introduced with PDAL 1.8. (#2275)
-  The density command is now found by pdal. This fixes a regression
   introduced with PDAL 1.8. (#2378)
-  The ``scale=auto`` and ``offset=auto`` options now work with
   ``writers.bpf`` in stream mode. (#1983)
-  Pipelines with diamond shapes now properly re-execute ancestor
   stages. This fixes a regression introduced with PDAL 1.8. (#2290)
-  ``filters.crop`` now uses the correct logic when the ``outside``
   option is used in standard mode. (#2305)
-  A bug in ``writers.gdal`` with streaming mode that caused an improper
   location shift in some instances has been fixed. (#2292)
-  Added an explicit dependency on the arbiter library for curl. This
   fixes build issues on some Unix platforms. (#1822)
-  Fixed a bug in ``filters.icp`` introduced by a change in PCL (see:
   https://github.com/PointCloudLibrary/pcl/issues/2724) that improperly
   removed a function call. (#2319)
-  ``readers.tindex`` now provides a default of “EPSG:4326” for the
   ``filter_srs`` option. (#2316)
-  Points in buffer corners are now properly placed in tiles in
   ``filters.splitter``. (#2372)
-  Fixed reference counting in ``filters.python`` that might have caused
   a crash in some circumstances.
-  Generalized the python library re-loaded by PDAL when being used as a
   python extension. This is only relevant on some Unix distributions.

================================================================================
1.8.0
================================================================================

Changes of Note
===============

-  PointTableRef is now publicly accessible from PointView (#1926)
-  Minimum CMake version is now 3.5
-  ``filters.hexbin`` is now a built-in stage, rather than a plugin.
   (#2001)
-  Removed support for ``ght`` compression in ``writers.pgpointcloud``.
   (#2148)
-  On OSX, plugins are now installed with ID of ``@rpath`` rather than
   ``@loader_path/../lib``

Enhancements
============

-  Added ``denoise`` and ``reset`` options to ``pdal ground``. (#1579)
-  ``readers.gdal`` now supports stream mode and provides the ``header``
   option to map dimensions. It also supports fetching bounds without
   reading the entire file. (#1819)
-  ``readers.mbio`` added ``datatype`` option to support reading
   sidescan data. (#1852)
-  ``filters.stats`` was computing expensive kurtosis and skewness
   statistics by default. These statistics are now available with the
   ``advanced`` option. (#1878)
-  Added backtrace support for alpine linux-based Docker containers.
   (#1904)
-  Added a ``condition`` option for ``filters.assign`` to limit
   assignment. (#1956)
-  Add access to artifact manager keys. (#2026)
-  Added support for LAZ compression in ``writers.pgpointcloud`` (#2050)
-  Replaced ``last`` option with ``returns`` to support more flexible
   segmentation in ``filters.smrf`` and ``filters.pmf``. (#2053)
-  ``writers.text`` now supports stream mode. (#2064)
-  Added ``pdal tile`` kernel with streaming support to conveniently
   tile data. (#2065)
-  A KD-tree used in one filter will now be reused in subsequent filters
   when possible. (#2123)
-  ``writers.ply`` now has a ``precision`` option to specify output
   precision. (#2144)
-  ``filters.smrf`` and ``filters.pmf`` supports complete range syntax
   for the ``ignore`` option. (#2157)
-  ``filters.hexbin`` now supports stream mode. (#2170)
-  ``readers.numpy`` now has the ``order`` option, which replaces the
   previous ``x``, ``y`` and ``z`` options. It also supports structured
   numpy arrays and maps values to the X, Y and Z dimensions
   accordingly.
-  All readers now support setting a spatial reference to override any
   in the data with the ``spatialreference`` option.
-  Add support for unicode filenames in pipelines on Windows platforms.
-  Added NumpyReader::setArray() to support direct setting of a numpy
   array into ``readers.numpy``.
-  Added StreamPointTable::setNumPoints() and support in
   Streamable::execute() allowing custom point tables to know the number
   of points in each pass through the point table.
-  Added SpatialReference::isProjected() to allow callers to determine
   if a spatial reference specifies a projection. Also added
   SpatialReference::identifyHorizontalEPSG() and
   SpatialReference::identifyVerticalEPSG() to return an EPSG code from
   a spatial reference if possible.
-  Added support for reading BPF files stored remotely.

New stages
==========

-  ``readers.rdb`` - Support reading RIEGL RDB data.
-  ``readers.i3s`` - Support reading of web service-style Esri I3S point
   clouds.
-  ``readers.slpk`` - Support reading of file-based I3S-style point
   clouds.
-  ``writers.fbx`` - Experimental Unity engine (SDK required) support.
   (#2127)
-  ``filters.nndistance`` - Replaces ``filters.kdistance`` and adds
   average distance support. (#2071)
-  ``filters.dem`` - Filter data based on bounds relative to a raster
   value. (#2090)
-  ``filters.delaunay`` - Create a delauany triangulation of a point
   cloud. (#1855)
-  ``filters.info`` - Generate metadata about an input point set. Used
   by ``pdal info``.

Deprecated stages
=================

-  ``filters.kdistance`` - Replaced by ``filters.nndistance``.

Bug fixes
=========

-  Fixed an incorrect error message suggesting there were multiple SRSs
   in some cases when reading multiple inputs. (#2009)
-  Fixed a problem in ``filters.reprojection`` in stream mode that would
   improperly reproject when there were multiple input sources with
   differing SRSs. (#2058)
-  Fixed a problem in stream mode where a stage with no SRS would
   override the active SRS during processing. (#2069)
-  Fixed a problem in ``writers.gdal`` where output would be aggregated
   if multiple inputs were provided. (#2074)
-  The ``count`` option was not respected in stream mode. It now
   properly limits the number of points read. (#2086)
-  Fixed an off-by-one error that could create improper values in
   ``writers.gdal``. Most data differences were small and not usually
   problematic. (#2095)
-  Multiple option values can be specified on the command line by
   repeating the option assignment. (#2114)
-  Added a missing initialization in ``filters.returns`` that could
   cause more point views to be returned than requested. (#2115)
-  Emit an error if the ``count`` option isn’t set for ``readers.faux``.
   (#2128)
-  PipelineManager::getStage() now returns a proper leaf node. (#2149)
-  Fixed logic for ``filters.crop`` in streaming mode with multiple crop
   areas that could return too few points. (#2198)
-  Added the ``minimal`` option for ``readers.rxp`` that was documented
   but not fully implemented. (#2225)
-  Fixed an error in failing to read all points in ``readers.rxp``
   exposed with a newer SDK. (#2226)
-  Fixed an error in fetching floating point data from a PointContainer
   when the value was NaN. (#2239)

================================================================================
1.7.0
================================================================================

Changes of Note
===============

-  ``filter.ferry`` now creates output dimensions with the same type as
   the input dimension. It also takes an arrow ‘=>’ in addition to ‘=’
   in the ``--dimension`` specification.
-  ``filters.hexbin`` now falls back to slow boundary creation if no
   bounds information exists to do fast boundary creation.
-  Dimension names can now contain the forward slash (‘/’) character.
-  ``readers.gdal`` and ``filters.colorization`` now attempt to create
   dimensions with the type of the associated raster.
-  The Python PDAL extension code was removed from the PDAL source tree
   to its own `repository <https://github.com/PDAL/python>`__.
-  The JAVA PDAL interface was remove from the PDAL source tree to its
   own `repository <https://github.com/PDAL/java>`__.
-  ``pdal pipeline``\ and ``pdal translate`` now use stream mode if the
   operation being performed is streamable. An option ``--nostream`` has
   been added to both commands to prevent the use of stream mode. The
   ``--stream`` option of ``pdal pipeline`` is now obsolete and ignored.
-  A new interface has been provided for the creation of plugins
   requiring less boilerplate code. There has been no API change.
-  Stages and pipelines can now be tested to check whether they are
   streamable.

Enhancements
============

-  Added options ``--timegap`` and ``--speedmin`` to ``readers.mbio`` to
   allow configuration of which points should be read.
-  Added support for compression schemes (xz, lzma, zstd) and created a
   standardized interface (#1722).
-  ``writers.bpf`` now supports the option ``auto`` for the ``coord_id``
   option to allow the UTM zone to be set from the spatial reference if
   possible (#1723).
-  Added the ability read stage-specific options from a file with the
   ``--option_file`` option (#1641).
-  Replace the GDAL point-in-polygon with a much faster implementation
   in ``filters.crop``.
-  Add a ``--reverse`` option to ``filters.mortonorder`` to provide a
   point ordering for good dispersal.
-  ``readers.bpf`` now supports the TCR (ECEF - earth centered, earth
   fixed) coordinate system.
-  Added option ``--use_eb_vlr`` to allow ``readers.las`` to interpret
   an extra bytes VLR as if the file were version 1.4 even if it’s using
   an earlier LAS version.
-  ``readers.text`` added options ``--header`` and ``--skip`` to provide
   an alternative header line and to allow skipping lines before reading
   the header, respectively.
-  ``writers.text`` now supports the ability to specify individual
   dimension precision using a colon (‘:’) and integer following the
   dimension name in the ``--order`` option.
-  ``readers.numpy`` adds support for reading from Numpy (.npy) save
   files.
-  ``pdal info`` now provides the ``--enumerate`` option. See the
   documentation for
   `filters.stats <https://pdal.io/stages/filters.stats.html>`__ for
   details.
-  Added a general option ``--logtiming`` to cause log output to contain
   the elapsed time from the start of the program. (#1882)

Documentation
=============

-  Added a description of the Alpine Linux environment used for Travis
   and Docker.
-  Updated the documentation for building PDAL on Windows.
-  Added an example of how to loop files with PowerShell.
-  Corrected output shown in the documentation for ``filters.hexbin``.
-  Reorganized the stages landing page to make it easier to find.
   (#1880)

New stages
==========

-  ``readers.greyhound`` - Allows reading points from a source using the
   `greyhound <https://github.com/hobu/greyhound>`__ protocol.
-  ``filters.neighborclassifier`` - Re-classifies points based on the
   classification of neighboring points.

Removed stages
==============

-  ``filters.computerange`` - Use ``filters.python`` to simulate the
   functionality.

Bug fixes
=========

-  ``filters.range`` now always rejects NaN values as out of range.
-  Changed the default ``--timegap`` value in ``readers.mbio`` from 0 to
   1 second.
-  Fixed a bug when reading pointers from metadata on some OSX systems.
-  Fixed a problem in ``pdal translate`` where overriding a reader or
   writer from a pipeline wouldn’t create the proper pipeline.
-  Fixed a problem where multiple LASzip VLRs would get written to LAS
   files (#1726).
-  Fixed an installation problem with the Python extension.
-  Fixed a bug in ``writers.tindex`` that could cause a crash if the
   output file wasn’t LAS.
-  Fixed JSON output from ``filters.hexbin`` when density/area can’t be
   calculated.
-  Fixed a problem where output might be tagged with the wrong SRS when
   using ``filters.reprojection`` in stream mode. (#1877)
-  PDAL_DRIVER_PATH would be improperly parsed on Windows systems when
   the path contained colons. Windows builds now use the semicolon as
   the path separator. (#1889)
-  Convert NaN and infinite double values to proper strings for output
   as JSON.
-  Synthetic, keypoint and withheld flags are now added to the
   classification dimension for version 1.0 - 1.3 files in
   ``readers.las``.
-  Support missed case supporting points with color but no density in
   ``filters.poisson``.
-  Throw an error if the input file can’t be opened in ``readers.ply``.
-  The ``--stdin`` option for ``kernels.tindex`` didn’t work. Now it
   does. Various other fixes were made.
-  ``writers.gdal`` now throws an error if an attempt is made to write
   an output file with no points available.
-  A build error that would be generated if lazsip was not found, even
   if it was not requested, has been resolved.
-  A build error that would be generated if Python was found but not
   requested has been resolved.
-  PDAL defaults to using `normal CMake
   interface <https://cmake.org/cmake/help/v3.11/policy/CMP0022.html>`__
   linking (#1890)
-  Fixed an issue where dimensions from ``readers.pcd`` and
   ``writers.pcd`` could get confused with dimensions from
   ``readers.sbet`` and ``writers.sbet``.
-  Fixed index computation in ``filters.voxelcentroidnearestneighbor``
   and ``filters.voxelcenternearestneighbor`` #1901
-  Fixed libdl linking #1900

================================================================================
1.6.0
================================================================================

Changes of Note
===============

-  PDAL's Travis CI configuration is now based on Alpine Linux.
-  PDAL is now built into containers with Alpine linux in addition to
   Ubuntu linux. `Tags <https://hub.docker.com/r/pdal/pdal/tags/>`__
   exist for each release, starting with 1.4, as well as the master
   branch.
-  Pipeline tag names can now contain capital letters. They can also
   contain underscores after the first character.
-  Replace ``filters.programmable`` and ``filters.predicate`` with the
   more general ``filters.python``.
-  Add support for Matlab with ``filters.matlab``\ (#1661).
-  Remove the ``approximate`` option from ``filters.pmf`` and add an
   ``exponential`` option.p
-  Placed base64 encoded VLR data in a subnode of the VLR itself with
   the key "data" rather than duplicate the VLR node itself (#1648).
-  XML pipelines are no longer supported (#1666).
-  The number of proprietary dimensions in ``readers.text`` was expanded
   from 255 to 4095 (#1657).
-  API hooks have been added to support the use of PDAL with JVM
   languages such as Java or Scala.
-  Added support for LASzip 1.4 and switch to use the new LASzip API.
   (#1205). LASzip support in PDAL will require LASzip.org release 3.1.0
   or greater.
-  The cpd kernel has been replaced with ``filters.cpd``.
-  No more warnings about ReturnNumber or NumberOfReturns for LAS
   permuations (#1682).
-  The KernelFactory class has been removed. Its functionality has been
   moved to StageFactory.
-  Built-in eigen support has changed from version 3.2.8 to 3.3.4
   (#1681).

Enhancements
============

-  API users can now create synonyms for existing arguments to a stage.
-  ``filters.splitter`` can now create buffered tiles with the
   ``buffer`` option.
-  ``writers.ply``\ can now be made to write faces of an existing mesh
   (created with ``filters.greedyprojection`` or ``filters.poisson``) if
   the ``faces`` option is used. An option ``dims`` has also been added
   that allows specification of the dimensions to be written as PLY
   elements. The writer also now supports streaming mode.
-  ``readers.text`` is now automatically invoked for .csv files.
-  PDAL\_PLUGIN\_INSTALL\_PATH can now be set via override when building
   PDAL from source.
-  Changed the use of null devices to eliminate potentially running out
   of file descriptors on Windows.
-  ``filters.randomize`` can now be created by the stage factory
   (#1598).
-  Provide the ability to specify a viewpoint and normal orientation in
   ``filters.normal`` (#1638).
-  ``readers.las`` now provides the ``ignore_vlr`` option to allow named
   VLRs to be dropped when read (#1651).
-  Allow ``writers.gdal`` to write output rasters of type other than
   double (#1497).
-  ``filters.sqlite`` is now invoked automatically for .gpkg files.
-  ``filters.colorinterp`` can now be used in streaming mode in some
   cases (#1675).
-  Pointers can now be stored as metadata.
-  ``filters.ferry`` can now create new dimensions without copying data
   (#1694).

Documentation
-------------

-  Remove some leftover references to the ``classify`` and ``extract``
   options that were removed from ``filters.ground`` in the last
   release.
-  Add a note about running pgpointcloud tests.
-  Added a tutorial on filtering data with python.
-  Remove lingering XML pipeline examples and replace with JSON.
-  Many updates and corrections to the workshop.
-  Added to the FAQs and entry about why a stage might not be found.
-  Added information to stage docs to indicate whether or not they were
   buit-in rather than plugins (#1612).
-  Added information to stage docs to indicate when they are streamable
   (#1606).

New filters
===========

-  ``filters.greedyprojection`` - Performs triangulation of points
   (surface reconstruction) based on the greedy projection algorithm.
-  ``filters.poisson`` - Performs triangulation of points (surface
   reconstruction) based on the algorithm of Kazhdan.
-  ``filters.head`` - Passes through only the first N points.
-  ``filters.tail`` - Passes through only the last N points.
-  ``filters.cpd`` - Calculates and applies a transformation to align
   two datasets using the `Coherent Point
   Drift <https://sites.google.com/site/myronenko/research/cpd>`__
   registration algorithm.
-  ``filters.icp`` - Calculates and applies a transformation to align
   two datasets using the `Iterative Closest
   Point <http://docs.pointclouds.org/trunk/classpcl_1_1_iterative_closest_point.html>`__
   registration algorithm.
-  ``filters.voxelcenternearestneighbor`` - Finds points closest to the
   center of a voxel (#1597).
-  ``filters.voxelcentroidnearestneighbor`` - Finds points closest to
   the controid of points in a voxel (#1597).
-  ``filters.python`` - Replaces ``filters.predicate`` and
   ``filters.programmable``.
-  ``filters.matlab`` - Provides support for matlab manipulation of PDAL
   points and metadata (#1661).

New readers
===========

-  Add ``readers.osg`` to support Open Scene Graph format.
-  Add ``readers.matlab`` to support reading data from a user-defined
   Matlab array struct. The same structure is written by
   ``writers.matlab``.

Bug fixes
=========

-  Fixed a case where\ ``kernels.tindex`` would unconditionally set the
   spatial reference on a feature from the ``a_srs`` option. The spatial
   reference stored in ``a_srs`` is now only used if explicitly set or
   no spatial reference was present.
-  Fixed a case where 'writers.gdal\` could fail to check for an
   out-of-bounds point, potentially leading to a crash.
-  Fix an error in 'filters.cluster' where the points wouldn't properly
   be placed in the first cluster because the starting cluster number
   was incorrect.
-  Fixed an error in freeing OGR features that could cause a crash when
   running "pdal density".
-  Fix potential memory leaks when creating OGRSpatialReference objects.
-  Make sure the ``global_encoding`` option is initialized to 0 in
   ``writers.las`` (#1595).
-  Fix eigen::computeCovariance to compute the correct sample
   covariance.
-  In some cases, the ``filters.crop`` would attempt to treat a 2D
   bounding box as 3D, yeilding a NULL bounding box and an error in
   behavior (#1626).
-  Fixed potential crash when using PDAL with multiple threads by
   providing locking for gdal::ErrorHandler (#1637)
-  Made sure that an uncompressed LAS file would be properly read even
   if the ``compression`` option was provided.
-  Throw an exception instead of crash when attempting to access a
   non-existent color ramp. (#1688)

================================================================================
1.5.0
================================================================================
Changes of Note
===============

-  PCL ``--visualize`` capability of the ``pdal`` command line
   application has been removed.
-  ``writer.derivative`` has been removed. Use
   `gdaldem <http://www.gdal.org/gdaldem.html>`__ for faster and more
   featureful equivalent functionality.
-  GeoTIFF and Proj.4 are now required dependencies.
-  ``writers.p2g`` has been removed. It was replaced by ``writers.gdal``
   in 1.4, but the P2G writer was essentially unmaintained and we will
   be using the GDAL one going forward.
-  ``filters.attribute`` was split into ``filters.assign`` and
   ``filters.overlay`` to separate their functionalities
-  ``filters.pmf`` and ``filters.outlier`` have dropped the ``classify``
   and ``extract`` options. They now only classify points and leave it
   downstream filters to ignore/extract classifications as needed.
-  ``filters.outlier`` has changed the default classification for noise
   points from ``18`` to ``7`` to match the LAS classification code for
   "Low point (noise)".

Enhancements
============

-  ``pdal pipeline`` now supports a ``--stream`` option which will
   default to one-at-a-time or chunk-at-a-time point processing when all
   stages in the pipeline support it. You can use this option to control
   memory consumption -- for example when interpolating a very large
   file with ``writers.gdal``
-  ``filters.crop`` was enhanced to support transformed filter polygons,
   streaming, and radius cropping.
-  ``readers.greyhound`` updated to support greyhound.io 1.0 release,
   with the most significant enhancement being support for passing
   downstream JSON filters.
-  ``user_data`` JSON object can be applied to any PDAL pipeline object
   and it will be carried through processing. You can use this mechanism
   for carrying your own information in PDAL pipelines without having to
   sidecar data. #1427
-  ``writers.las`` now can write ``pdal_metadata`` and ``pdal_pipeline``
   VLRs for processing history tracking. #1509 #1525
-  ``metadata``, ``schema``, and ``spatialreference`` objects added to
   global module for ``filters.programmable`` and ``filters.predicate``
   Python filters.
-  ``pdalargs`` option for ``filters.programmable`` and
   ``filters.predicate`` allow you to pass in a JSON dictionary to your
   Python module for override or modification of your script
-  Stage tags can be used in pipeline override scenarios
-  User-settable VLRs in ``writers.las`` #1542
-  ``filters.sort`` now supports descending order and uses
   ``std::stable_sort`` #1530 (Thanks to new contributor @wrenoud )
-  ``pdal tindex`` will now use data bounds if ``filters.hexbin`` cannot
   be loaded for boundaries #1533
-  ``filters.pmf`` and ``filters.smrf`` improved performance #1531 and
   #1541
-  ``filters.assign`` now supports
   `Range <https://pdal.io/stages/filters.range.html>`__-based
   filters
-  ``filters.outlier`` now accepts a user-specified ``class`` to
   override the default value of ``7`` for points deemed outliers. #1545
-  ``filters.pmf`` and ``filters.smrf`` now accept a
   `Range <https://pdal.io/stages/ranges.html#ranges>`__ via the
   ``ignore`` option to specify values that should be excluded from
   ground segmentation. #1545
-  ``filters.pmf`` and ``filters.smrf`` now consider only last returns
   (when return information is available) as the default behavior. The
   ``last`` option can be set to ``false`` to consider all returns.
   #1545

Documentation
-------------

-  New `About page <https://pdal.io/about.html>`__ adapted from
   `workshop <https://pdal.io/workshop/>`__
-  New `LAS reading and writing <https://pdal.io/tutorial/las.html>`__
   tutorial
-  Consolidation of `Python <https://pdal.io/python.html>`__ information

New filters
-----------

-  ``filters.cluster`` - Perform Euclidean cluster extraction, and label
   each point by its cluster ID. By @chambbj.
-  ``filters.groupby`` - Split incoming PointView into individual
   PointViews categorically, e.g., by Classification. By @chambbj.
-  ``filters.locate`` - Locate and return the point with the minimum or
   maximum value for a given dimension. By @chambbj.
-  ``filters.emf`` - Extended Local Maximum filter. By @chambbj.

New readers
-----------

-  ``readers.mbio`` Bathymetric point cloud support for formats
   supported by the
   `MB-System <https://www.ldeo.columbia.edu/res/pi/MB-System/>`__
   software library

Bug fixes
---------

-  ``writers.pgpointcloud`` needed to treat table schema correctly
   https://github.com/PDAL/PDAL/pull/1540 (thanks @elemoine)
-  ``pdal density`` kernel now supports overriding ``filters.hexbin``
   options #1487
-  Arbiter embedded library updated to support setting Curl options
   (certificate settings, etc).
-  Provided a default value for ``radius`` in ``writers.gdal`` #1475
-  ``writers.ply`` broken for non-standard dimensions #1556
-  No EVLRs for ``writers.las`` for files < LAS 1.4 #1551
-  LAS extra dims handling for standard PDAL dimension names #1555
-  LASzip defines #1549


================================================================================
1.4.0
================================================================================
Changes of Note
===============

-  GeoTIFF is now required to compile PDAL
-  ``--scale`` and ``--offset`` kernel options are no longer supported.
   Specify using stage-specific options as needed.
-  The ``--validate`` option of the ``pdal pipeline`` command now
   invokes the preparation portion of the pipeline to force validation
   of options.
-  The ``--verbose`` option to ``pdal`` now accepts log level names
   ("Error", "Warning", "Info", "Debug", "Debug1", "Debug2", "Debug3",
   "Debug4" and "Debug5") in addition to the corresponding numeric
   values (0 - 8).

Enhancements
============

New filters
-----------

-  ```filters.colorinterp`` <http://pdal.io/stages/filters.colorinterp.html>`__
   - Ramp RGB colors based on a specified dimension. By @hobu
-  ```filters.mad`` <http://pdal.io/stages/filters.mad.html>`__ - Filter
   outliers in a given dimension by computing Median Absolute Deviation
   (MAD). By @chambbj
-  ```filters.lof`` <http://pdal.io/stages/filters.lof.html>`__ -Filters
   outliers by Local Outlier Factor (LOF). By @chambbj
-  ```filters.estimaterank`` <http://pdal.io/stages/filters.estimaterank.html>`__
   - Estimate rank of each neighborhood of k-nearest neighbors. By
   @chambbj
-  ```filters.eigenvalues`` <http://pdal.io/stages/filters.eigenvalues.html>`__
   - Compute pointwise Eigenvalues. By @chambbj
-  ```filters.iqr`` <http://pdal/io/stages/filters.iqr.html>`__ - Filter
   outliers in a given dimension by computing Interquartile Range (IQR).
   By @chambbj
-  ```filters.kdistance`` <http://pdal.io/stages/filters.kdistance.html>`__
   - Compute pointwise K-distance. By @chambbj
-  ```filters.radialdensity`` <http://pdal.io/stages/filters.radialdensity.html>`__
   - Compute pointwise radial density. By @chambbj
-  ```filters.outlier`` <http://pdal.io/stages/filters.outlier.html>`__
   - Radius and statistical outliers. By @chambbj

New writers
-----------

-  ```writers.gdal`` <http://pdal.io/stages/writers.gdal.html>`__ -
   `points2grid <http://github.com/crrel/points2grid>`__ replacement. By
   @abellgithub

New kernels
-----------

-  ```kernels.hausdorff`` <http://pdal.io/apps/hausdorff.html>`__ -
   Compute `Hausdorff
   distance <https://en.wikipedia.org/wiki/Hausdorff_distance>`__
   between two point clouds. By @chambbj

Improvements
------------

-  `Filename
   globbing <http://man7.org/linux/man-pages/man7/glob.7.html>`__ is now
   supported in the JSON pipeline specification of reader input files.
   Note that tilde expansion is NOT supported.
-  Source tree reorganization
   https://lists.osgeo.org/pipermail/pdal/2016-December/001099.html
-  CMake updates to utilize ``target_include_directory`` and
   ``target_link_libraries``.
-  JSON output for ``pdal --showjson --drivers`` and
   ``pdal --showjson --options`` to support application builders being
   able to fetch active lists of stages, kernels, and options.
   https://github.com/PDAL/PDAL/issues/1315
-  Stacktrace logging to stderr on Unix systems
   https://github.com/PDAL/PDAL/pull/1329
-  Geometry ingestion enhancements now support using
   `GeoJSON <http://geojson.org>`__ or WKT in pipeline options
   https://github.com/PDAL/PDAL/pull/1339.
-  Significant Python extension refactor
   https://github.com/PDAL/PDAL/pull/1367 including ability to fetch
   data schema, log, and pipeline information. Common utility classes to
   support the Python extension were refactored in support of the Java
   extension.
-  Java extension by `Azavea <https://www.azavea.com/>`__ to support
   using PDAL in `Spark <http://spark.apache.org/>`__ and friends.
   https://github.com/PDAL/PDAL/pull/1371
-  ```kernels.density`` <http://pdal.io/stages/kernels.density.html>`__
   - Density kernel now supports writing into an existing OGR datasource
   https://github.com/PDAL/PDAL/pull/1396
-  ```readers.greyhound`` <http://pdal.io/stages/readers.greyhound.html>`__
   - Greyhound reader refactor.
-  Multi-threaded read support
-  Server-side filtering pass-through
-  ```writers.derivative`` <http://pdal.io/stages/writers.derivative.html>`__
   - Derivative writer refactor.
-  ``slope_d8``
-  ``slope_fd``
-  ``aspect_d8``
-  ``aspect_fd``
-  ``contour_curvature``
-  ``profile_curvature``
-  ``tangential_curvature``
-  ``hillshade``
-  ``total_curvature``
-  Output to any GDAL-writable format
   https://github.com/PDAL/PDAL/issues/1146
-  ```filters.crop`` <http://pdal.io/stages/filters.crop.html>`__ -
   Radial cropping https://github.com/PDAL/PDAL/issues/1387
-  ```filters.stats`` <http://pdal.io/stages/filters.stats.html>`__ -
   Optional per-dimension median and MAD computation
-  Support was added for the recently added cartesian coordinate in BPF
   files.
-  ```writers.p2g`` <http://pdal.io/stages/writers.p2g.html>`__ now uses
   the InCoreInterp method of the points2grid code. This uses more
   memory but runs faster and doesn't crash.
-  The application now provides better error feedback on command-line
   errors by indicating the invoked kernel when an error is detected.
-  PDAL now searches by default in the following locations for plugins:
   ``"." "./lib", "../lib", "./bin", "../bin"``. Use
   ``PDAL_DRIVER_PATH`` to explicitly override the plugin search
   location.
-  Vector-based command-line arguments now accept default values in the
   API.
-  JSON parsing errors of pipeline files now provide more detailed
   messages.
-  Writers now add output filenames to metadata.
-  Stage names provided as input to other stages in pipelines can now be
   specified as strings or arrays of strings. The previous version
   required single input stage names to be placed in an array.
-  Added ``--smooth`` option to
   ```filters.hexbin`` <http://pdal.io/stages/filters.hexbin.html>`__ to
   allow user control of boundary smoothing.

Bug fixes
---------

-  Well-known text provided as a spatial reference isn't interpreted by
   GDAL unless necessary.
-  ```filters.hexbin`` <http://pdal.io/stages/filters.hexbin.html>`__
   now returns ``MULTIPOLYGON EMPTY`` when it is unable to compute a
   boundary.
-  Reading a not a number (nan) value from a text file now works
   properly.
-  The ``--compression`` option for
   ```writers.pcd`` <http://pdal.io/stages/writers.pcd.html>`__ has been
   fixed so that the writer actually compresses as requested.
-  The stage manager (and hence, pipelines) now properly recognizes the
   text reader as
   ```readers.text`` <http://pdal.io/stages/readers.text.html>`__.
-  ```readers.text`` <http://pdal.io/stages/readers.text.html>`__ now
   detects the case when a dimension has been specified more than once
   in an input file.
-  Fixed a problem where
   ```filters.splitter`` <http://pdal.io/stages/filters.splitter.html>`__
   could create cells larger than requested about the X and Y axes.
-  ```writers.nitf`` <http://pdal.io/stages/writers.nitf.html>`__ now
   errors if it attempts to write an FTITLE field that exceeds the
   allowable length.
-  If PDAL is build with LAZperf but without LASzip, the program now
   properly defaults to using LAZperf.
-  Fixed a problem where
   ```filters.sort`` <http://pdal.io/stages/filters.sort.html>`__ could
   fail to properly order points depending on the implementation of the
   C++ sort algorithm.
-  Fixed a problem in pgpostgres readers and writers where a failure in
   a query could lead to a crash.


================================================================================
1.3.0
================================================================================

Changes of Note
================================================================================

- Command line parsing has been reworked to cause invalid options to emit
  an error message.  Stage options specified in pipelines and on the command
  line are handled similarly.
- The dimension PlatformHeading has been renamed to Azimuth.  When looking
  up a dimension by name, the string "platformheading" is still accepted and
  returns Azimuth.
- Errors detected by GDAL are no longer thrown as exceptions.  A log message
  is emitted instead.
- Well-known dimensions are now added to PDAL by editing a JSON file,
  Dimension.json.
- Linking with PDAL using CMake no longer requires explicit linking with
  curl, jsoncpp or arbiter libraries.
- PDAL now searches for plugins in the following locations and order by
  default: ./lib, ../lib, ../bin, the location where PDAL was installed.
- The '--debug' and '--verbose' options are no longer supported as stage
  options.  The '--verbose' option is accepted on the PDAL command line.  The
  '--debug' option is deprecated, and if specified on the command line is
  equivalent to '--verbose=3'.  One can enable logging programmatically by
  calling setLog() on a PipelineManager or a specific stage.
- pdal::Dimension types are now C++11 enumeration classes. The change may
  require editing any Stage implementations you might have and removing the
  extraneous ::Enum type specification.

Enhancements
================================================================================

- Pipelines can now be read directly from standard input.
- Files can now be read from Amazon S3 buckets by providing an appropriate
  URL.
- Many new filters have been added: filters.approximatecoplanar,
  filters.eigenvalues, filters.estimaterank, filters.hag, filters.normal,
  filters.outlier, filters.pmf, filters.sample.  Most of these are algorithm
  extractions from the PCL library, with the hope of eliminating the need
  for PCL in some future PDAL release.
- The PLY reader now loads dimensions that aren't predefined PDAL dimensions.
- A '--driver' option has been added to allow a specific driver to be loaded
  for a file without regard to its extension.
- The PDAL_DRIVER_PATH environment variable now accepts a list of locations
  to search for drivers.
- Beta release quality drivers improvements in readers.greyhound
- Beta quality implementation of Mongus and Zalik ground filter
- Experimental implementation of Pingel et al. ground filter
- writers.pcd enhancements by Logan Byers (binary, compression) -- requires
  PCL
- Docker images upgraded to Ubuntu Xenial
- Cyclone PTS reader -- readers.pts
- skewness, kurtosis, stddev, and variance added to filters.stats output
- Python API now available https://pypi.python.org/pypi/pdal

Fixes
================================================================================

- A failure that may have resulted when using filters.hexbin to calculate
  density in the southern hemisphere has been corrected.
- A failure to create the index file with 'pdal tindex' and GDAL 2.X has
  been fixed.
- The '--tindex' option for the 'pdal tindex' command is now a positional
  option as specified in the documentation.
- The icebridge reader now reads the X dimension as longitude and forces
  the value in the range (-180, 180].  It also properly uses the dimension
  Azimuth instead of ScanAngleRank.
- An error in writers.pgpointcloud where it ignored SQL to be run at the end
  of the stage has been fixed.
- An error that might incorrectly write values stored internally as bytes
  when written as a different data type has been fixed.
- A problem where 'pdal info' wouldn't properly report dimension names not
  predefined by PDAL has been fixed.
- A bug in filters.crop that wouldn't properly transform coordinates when
  provided the '--a_srs' option has been fixed.

================================================================================
1.2.0
================================================================================

Changes of Note
================================================================================

- The GEOS library is now required to build PDAL.  In earlier versions it was
  an optional component.
- Boost is no longer a required component.  Unless you are building plugins
  that require boost (notably PCL and Geowave), you no longer will need
  boost installed on your system to build or run PDAL.
- PDAL now builds on Microsoft Visual Studio 2015.
- The PipelineReader class has been removed and its functionality has been
  merged into PipelineManager.
- Plugin libraries now support Linux versioning.
- Naming changes have been made to allow packaging with the Debian release.
- filters.height now uses the dimension 'HeightAboveGround' instead of a
  dimension named 'Height' to be compatible with the filters.heightaboveground.
- Option names no longer contain lowercase characters.
- PDAL now works with GDAL version 1.9 and later.
- Stages created with the StageFactory are now owned by the factory.
- filters.dartthrowing has been renamed filters.dartsample
- 'pipeline-serialization' now produces JSON output instead of XML.

Enhancements
================================================================================

- Pipelines may now be specified using a JSON syntax.  XML syntax is still
  supported but users should switch to JSON when possible as the XML support
  will be removed in a future version.
- PDAL now can be built into a Docker container.
- Many stages now support "streaming," which allows control of the number
  of points stored in memory during processing.  See
  Stage::execute(StreamPointTable&) for more information.
- A basic text reader has been added.
- Added support for the dimension 'ClassFlags' in readers.las.
- The derivative writer can now produce output for multiple primitive types
  with a single execution.
- 'pdal info' now provides bounding box output instead of a more refined
  boundary when the hexbin plugin isn't found.
- Added 'pdal density' to provide a command-line interface to the
  filters.hexbin density calcuations.
- The icebridge reader can now load an associated metadata file.  The reader
  also now marks the associated coordinate system as WGS84.
- The stats filter now emits bounding box information in native and WGS84
  projections.
- PDAL command-line programs now (generally) check their argument lists for
  correctness and report syntax errors.
- 'pdal info' now provides spatial reference attributes in addition to
  the actual well-known text.
- Geometry can now be specified as GeoJSON as well as well-known-text in
  most contexts.  Geometry optionally provides Z-dimension output.
- Stage and plugin creation is now thread-safe (NOTE: Most of PDAL is
  NOT thread-safe, so tread carefully).
- Many, many documentation enhancements.

Fixes
================================================================================

- A bug in generating PCIDs when with multiple simultaneous PDAL executions
  to the same Postgres database has been fixed.
- Fixed a bug in generated SQL delete statements when certain table names
  were used in the writers.postgres driver.
- Properly escape quotes when generating JSON output.
- Fix an off-by-one error when writing data with the derivative writer that
  could lead to a crash.
- Fixed a dependency error during builds that could lead to a failure to
  properly load Python extensions on Linux.
- Fixed a bug where passing certain options to 'pdal info' could be handled
  in ambiguous ways.
- Fixed bugs in the reading of raster data using readers.gdal.
- Fixed population of the AIMIDB and ACFTB attributes in writers.nitf.
- Corrected the parsing of some dimension names in filters.colorization.
- Fixed a potential truncation in the GlobalEncoding dimension of readers.las.

================================================================================
1.1.0
================================================================================

Enhancements
================================================================================

- Add support for the LAZperf LAS compressor in decoding/encoding LAS files.
  LAZperf can be enabled with the 'compression' option in readers.las and
  writers.las.
- Add PCL functionality as filters (filters.greedyprojection,
  filters.gridprojection, filters.ground filters.movingleastsquares,
  filters.poisson, filters.radiusoutlier, filters.statisticaloutlier,
  filters.voxelgrid, filters.height, filters.dartsample)
- Add readers.gdal to support reading raster sets as point clouds
- Update writers.geowave and readers.geowave to work with the latest version
  of GeoWave software.
- Add readers.ilvis2 to support the Icebridge ILVIS2 format.
- Disallow nested options.  Check stage documentation for changes in option
  names and handling. (filters.ferry, filters.colorization, filters.attribute,
  filters.crop).  Change filters.attribute to handle only a single dimension.
- Add 'output_dims' options in writers.bpf to allow control of the dimensions
  that should be written.
- Add 'all' keyword in 'extra_dims' options of writers.las to cause all
  dimensions to be written to either the standard or extra dimensions of
  a LAS point.
- Add filters.randomize to allow randomized order of points.
- Add filters.divider to split a set of points into subsets of a fixed number
  or into subsets containing a specific number of points.
- Update to version 1.1.4 of rply in readers.rply.
- Change the logic of the range filter to allow multiple ranges for a single
  dimension and support a simple boolean logic.
- Change the default scaling on writer.bpf to 'auto'.
- Add support for smoothing boundaries generated by filters.hexbin.
- Add readers.tindex to allow vector-filtered input of point cloud files.
- Allow merging of datasets with non-matching spatial references.
- Many, many documentation enhancements.

Fixes
================================================================================

- Handle error with Pgpointcloud when pointcloud extension is not installed
  on postgres server.  Skip tests if extention is missing.
- Set precision on output of doubles to metadata.
- Fix a divide-by-zero error in readers.faux when the point count was 1.
  (https://github.com/PDAL/PDAL/issues/1015)
- Fix fatal error loading numpy library that occurred when running
  filters.predicate or filters.programmable.
  (https://github.com/PDAL/PDAL/issues/1010)
- Correct readers.las to properly check WKT bit when choosing spatial
  reference VLR.
  (https://github.com/PDAL/PDAL/issues/1040)
- Correct writer.las to emit only WKT or GeoTiff VLR, not both.
  (https://github.com/PDAL/PDAL/issues/1040)
- Check object ID against table column id (attrelid) to ensure correct PCID
  retrieval in readers.pgpointcloud.
  (https://github.com/PDAL/PDAL/pull/1051)

