1.2.0

New features:

• Enhance DataChunkIterator to support h5 datasets better, skip empty blocks, and work on arbitrary iteration axes #132 (@oruebel, @rly)
• Drop Python 2.7 support #126 (@rly)

Minor enhancements:

• Always use parent names in autogenerated child spec names #130 (@rly)
• Improve unit testing #131 (@rly)
• Improve coverage testing #128 (@rly)
• Implement better support for floating point data for Python 3.5 on Windows #127 (@rly)

1.1.2

Minor enhancements:

• Add LabelledDict class from PyNWB (@rly)
• Add ability to resolve link to links (@ajtritt)
• Support Scalar HDF5 dataspaces in H5SpecReader (@t-b)
• Determine dtype for empty numpy datasets (@oruebel)

Bug fixes:

• Fix shape enforcement in docval when argument does not have len (@rly)

1.1.1

Bug fixes:

• Fix error when writing new data to existing file without object IDs (@rly)
• Fix errors when methods called on H5DataIO linked dataset that is closed (@rly)
• Fix I/O error when optional link is not specified (@rly)
• Make ObjectMapper properly use object attribute override method (@ajtritt)
• Improve warning and error messages (@rly)
• Improve unit tests (@rly)

1.1.0

Major features:

• Add unique ID to Container objects (@ajtritt, @rly)
• Note that as of 1.0.4, the spec is cached by default when writing a Container to disk

Enhancements:

• Move setting of parent and container_source of Container objects out of __init__ (@ajtritt, @rly)
• Update all requirements (@rly)
• Have TypeMap return dict instead of tuple (@bendichter)
• Added dtype and astype functions to DataChunk (@oruebel)
• Overload get_docval to find arg with matching name (@rly)
• Remove warning for empty datasets (@rly)
• Improve unit tests (@rly)

Fixes:

• get_class now uses default_name argument (@bendichter)
• Fix on determining the correct dtype based on the spec or AbstractDataChunkIterator (@oruebel)
• Honor dtype from builder when writing data iteratively (@oruebel)
• Fix checking of docval shape on types without len (@rly)
• Prevent __pycache__ from being written in release tarballs (@rly)

1.0.5

Bug fixes:

• Fixed critical bug on creating links #87

1.0.4

Enhancements:

• Cache the spec by default when writing a container to disk
• Update the default order of keys of specs written to disk
• Add warnings when an IO operation is not supported in the current mode (e.g. writing in 'r' mode)
• Store the cached spec in UTF-8
• Handle zero-length lists and tuples
• Improve links in documentation
• Passively read datasets that are vectors of strings

Bug fixes:

• Allow boolean attributes in custom classes / get_class
• Write lists of strings in their original encoding instead of ASCII
• Allow a dynamically created container within another dynamically created container to be written to file
• Validate file against cached namespace
• Validate arrays of datetime types correctly
• Properly populate Builder source when reading from HDF5
• Use supplied name for custom classes properly in get_class
• Fix type inference

1.0.3

hdmf 1.0.3

1.0.2

hdmf 1.0.2

1.0.1

hdmf 1.0.1

1.0

hdmf 1.0