All notable changes to this project will be documented in this file, for a complete overview of changes, please refer to the git log.
The format is based on Keep a Changelog, but most notably, without sectioning changes into type-of-change.
- fixes a bug in
dl::findoffsetsthat caused an infinite loop for certain broken files.
- Fixes a bug in
dlisio.load()that caused it to leak open file handles when load failed.
- Added official support and distributed wheels for python 3.8.
- Better error message is reported when attempting to load files which do not exist.
- dlisio can now read files which contain empty logical records.
- The cli tool describe.cpp is removed as it has not been maintained and used.
- Fixes a bug in the build script that creates the macos wheels. The lfp library was not properly included, resulting in an import error when importing 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
frame.curves()still fails, but this can now be bypassed with
- 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.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
- 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
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
set_encodings()and dlisio will try them in order.
Frame.indexnow returns the Channel mnemonic, not the
- 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.
Channel-objects now have an index-property.
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.
Channelare 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.match()is no longer case sensitive.
dlis.fileheadernow returns the
Fileheader-object directly, not wrapped as dict_values.
dlis.objectshas 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
- Access to curves directly through
- dlisio has learned to read the following metadata-objects:
Well reference point,
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_stashhas been removed.
dlis.getobjects()has been removed.
dlis.object_sethas been renamed to
Computation.sourceis now a scalar, not vector.
atticis 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
stream.atare 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.