Dlisio

-------------------
* dlisio can now read files wrapped in Tape Image Format (tif).
* dlisio can now read files that do not contain a Storage Unit Label.
* The numpy array returned by ``frame.curves()`` can now be indexed with
fingerprints in addition to the normal mnemonic indexing. Fingerprints are a
more reliable indexing method as these are required to be unique by the
standard, unlike mnemonics. This should mainly be of interest to automation
pipelines where reliable indexing is key.
* dlisio can now read frames with duplicated channels. This behavior is
explicitly forbidden by the spec. However, it is frequently violated. By
default, ``frame.curves()`` still fails, but this can now be bypassed with
``strict=False``.
* dlisio no longer accepts files where the last Visible Record is truncated, but
the last Logical Record is intact. Support for such truncated files was
never intended in the first place, but happened to work.
* ``Channel.curves()`` fails more gracefully when there is no recorded curve
data.
* The documentation has been revamped and new sections focusing on
understanding the content and structure of dlis-files are added.
* Fixes a bug that caused ``channel.curves()`` to use too much memory.
* Fixes a bug that causes ``dlisio.load()`` to fail if the file contained
encrypted fdata record(s).
* Fixes a bug that caused ``dlisio.load()`` to fail if the obname of a fdata
record spanned multiple Visible Records.
* Fixes a bug that re-read unknown objects from disk even if they were cached
from previous reads.

--------------------
* Fixes a bug were ``dlisio.load()`` did not properly close the memory mapping to
the file when loading failed.
* Fixes a bug where ``dlis.match()`` and ``dlis.object()`` returned the same object
multiple times under certain circumstances.
* ``dlis.describe()`` again includes the object-count of each object-type.
* ``dlisio.load()`` now warns if a file contains ``Update``-objects. The current lack
of support for such objects means that dlisio may wrongfully present data in
files with ``Update``-objects.
* There is now a list of organization codes on readthedocs
* Fixes a bug in the Process-docs

--------------------
* Metadata objects are now parsed and loaded when needed, rather than all at
once in ``dlisio.load()``. This is not directly observable for the user, other
than it improves performance for ``dlisio.load()``. For files with a lot of
metadata, the performance gain is huge.
* dlisio can now read even more curve-data. Specifically, where multiple FDATA
(rows) are stored in the same IFLR.
* The array from ``Frame.curves()`` now includes FRAMENO as the first column.
FRAMENO are the row numbers as represented in the file. It might happen that
there are missing rows or that they are out-of-order in the file, that is now
observable by inspecting FRAMENO.
* Better support for non-ascii strings. It is now possible to tell dlisio which
string encodings to try if decoding with 'utf-8' fails. Supply a list of
encodings o ``set_encodings()`` and dlisio will try them in order.
* ``Frame.index`` now returns the Channel mnemonic, not the ``Channel``-object.
* ``Channel.index`` is removed.
* Validated types are now represented as tuples, not lists.
* Fixes a bug were microseconds in datetime objects were interpreted as
milliseconds.
* Better error message when incomplete Channels objects cause parsing of curves
to fail as a result.

--------------------
* dlisio has learned to read curves with variable length data types. Thus,
every data-type that the standard allows for curves is now supported by
dlisio.
* ``Frame``- and ``Channel``-objects now have an index-property. ``index`` returns the
``Channel``-object that serves as the index-channel for the given Frame/Channel.

-------------------
* The sphinx documentation on readthedocs_ has a few new sections: About the
project, an introduction to some dlis-concepts and a quick guide to help new
users to get started with dlisio.
* API documentation has seen some improvements as well. The ``dlis``-class
documentation is revamped to better help users to work with logical files and
accessing objects. ``Frame`` and ``Channel`` are more thoroughly documented, and
more examples on how to work with curve data are provided.
* Direct access to specific objects has been made more convenient with
``dlis.object()``.
* ``dlis.match()`` is no longer case sensitive.
* ``dlis.fileheader`` now returns the ``Fileheader``-object directly, not wrapped as
dict_values.
* ``dlis.objects`` has been removed
* CircleCI is added to the ci-pipeline for building and testing on linux
* Python test suite has seen some refactoring
* It is now possible to build the python module with ``setup.py``, provided the
core library is already installed on the system.

--------------------
* Output a readable summary of any metadata-object, logical file or batch-object
with ``.describe()``.
* Access to curves directly through ``Frame``- and ``Channel``-objects.
* dlisio has learned to read the following metadata-objects: ``Process``, ``Path``,
``Splice``, ``Well reference point``, ``Group``, ``Message``, ``Comment``.
* ``dlis.match()`` lets you search for objects with a regular expression.
* dlisio now reads even more files. Restrictions such as number-of-objects in an
object_set and missing representation codes in templates have been lifted.
* The parsing routine has seen some improvements. This includes giving the user
more freedom to customize object-parsing.
* Multidimensional metadata attributes are handled correctly.
* ``BasicObject.update_stash`` has been removed.
* ``dlis.getobjects()`` has been removed.
* ``dlis.object_set`` has been renamed to ``dlis.indexedobjects``.
* ``Computation.source`` is now a scalar, not vector.
* ``BasicObject``'s ``type`` and ``attic`` is now attributes, not properties.
* Objects are allowed to have empty ids (name/mnemonic).
* The API documentation has seen some minor updates.
* dlisio uses endianness.h rather than its own implementation.
* Some of the binary test files have been simplified.
* core functionality such as ``findfdata``, ``findsul``, ``findvrl``, ``findoffsets`` and
``stream.at`` are more thoroughly tested.
* Parts of the Python test suite have been refactored.
* Fixed a bug were long obnames were allocated insufficient memory.
* Fixed a bug were multi-dimensional fdata were interpreted incorrectly.
* Fixed a bug that caused incorrectly partitioning from physical- to logical
file(s).
* Fixed a bug that caused parsing of a encrypted logical record to fail.