FastROCS TK 1.6.0
================================================================================

New features
--------------------------------------------------------------------------------

* The performance of **FastROCS** when loading a database into memory has
  been significantly improved when the appropriate caching has
  been done to the input OEB file.
  :oe:func:`OEFastROCS::OEPrepareFastROCSMol` function has been added
  to allow preparing a molecule to be written to OEB so that it can be
  read very quickly by :oe:meth:`OEFastROCS::OEShapeDatabase::Open`.
  It is highly recommended to pre-process input OEB
  files with this function to achieve faster load times. This works
  especially well in combination with :oe:func:`OEChem::OEPRECompress`,
  resulting in uncompressed OEB files that are comparable to their
  gzip-compressed counterparts, even with extra cached information
  inside them.

  .. warning::

     :oe:func:`OEFastROCS::OEPrepareFastROCSMol` strips out
     information that **OMEGA** generates (i.e., hydrogen coordinates and
     conformer energies). Care must be taken when using
     these OEB files for alternative use cases. Please refer to the
     :oe:func:`OEFastROCS::OEPrepareFastROCSMol` documentation for a
     full accounting of how the molecules are changed.

* ``OEFastROCSCacheStuff`` was never officially supported and has
  been replaced by the :oe:func:`OEFastROCS::OEPrepareFastROCSMol`
  function. It now throws a deprecation warning and will be
  removed in a future release.

* **FastROCS TK** should be binary compatible with more NVidia driver
  versions as of this release. Previously, **FastROCS TK** could only
  target a single NVidia driver version. Now, several can be targeted
  in a single toolkit build. This version of **FastROCS TK** will work
  on the following NVidia driver versions: 340.x, 346.x, and 352.x. This
  should significantly reduce **FastROCS TK** installation headaches.

* :oe:meth:`OEFastROCS::OEShapeDatabase::SetNumOpenThreads` method
  has been added to control how many CPU operating-system threads are created
  when calling :oe:meth:`OEFastROCS::OEShapeDatabase::Open` on an
  :ref:`OEMolDatabase <OEChem_OEMolDatabase>` object.

* All :ref:`OEShapeDatabase <OEFastROCS_OEShapeDatabase>` instances now share the
  same GPU context. Previously, creating many
  :ref:`OEShapeDatabase <OEFastROCS_OEShapeDatabase>` instances would result in many
  operating system threads created per instance. This would
  often hit resource limits around the number of threads
  allowed in a single process.

Minor bug fixes
--------------------------------------------------------------------------------

* :oe:meth:`OEFastROCS::OEShapeDatabase::Open` now returns ``false``
  if no molecules are read from the :ref:`OEMolDatabase <OEChem_OEMolDatabase>`
  object.

Python-specific changes
--------------------------------------------------------------------------------

* All **FastROCS TK** Python examples now use ``/usr/bin/env python``
  to determine the appropriate Python interpreter to use. Previously,
  some examples were erroneously hardcoded to use ``/usr/bin/python``.

* ``karma.py`` shape clustering example had a minor bug that affected how invalid
  file names were handled. This has been fixed.

Documentation changes
--------------------------------------------------------------------------------

* The **FastROCS TK** :ref:`fastrocs_part_installation` section is now
  more general and defers to the NVidia website for installation
  instructions. A list of supported NVidia drivers for **FastROCS TK**
  can also be found in the :ref:`fastrocs_part_installation` section.
