Critical bug-fixes

This release fixes two critical bugs:

• Artifacts for empty blocks (thanks for reporting @aschampion)
• Gzip compression for zarr

In addition, several issues and inconsistencies for [] of Dataset were fixed (thanks to @clbarnes who implemented most of this) and the behavior of require_dataset and visititems was improved.

This will be the last release supporting python 2.7.

Fix multi-threading bug

This release fixes a critical bug in multi-threaded chunk I/O, which led to occasional seg-faults due to a thread-safety issue.
You can use multi-threaded I/O from python by setting the n_threads attribute of a dataset:

ds.n_threads = 8

Also adds support for vector instructions via xsimd.

Support ROI in converters

Supports Region of Interests (ROI) in the hdf5 converters and utility functions.
All Changes:

• Change positional order of z5py.File constructor arguments s.t. mode can be passed as second positional argument (this makes it closer to h5py API).
• Rename z5py.util.rechunk to z5py.util.copy_dataset
• Rename z5py.util.copy_dataset, z5py.converter.convert_to_n5, z5py.converter.convert_from_h5 arguments out_chunks to chunks and out_blocks to block_shape; chunks now has default value None. In this case the chunk shape of input dataset is used.
• z5py.util.copy_dataset, z5py.converter.convert_to_n5, z5py.converter.convert_from_h5 now support arguments roi and fit_to_roi. roi is optional slice or tuple of slices which restricts operation to region of interest defined by the slice(s). If 'fit_to_roi' is True shape of output dataset is set to roi shape if roi is given (default is False).

More h5py compatibility

Introduce more compatibility with h5py:

• Implement read_direct / write_direct, thanks to @paulhfu
• Add compression attribute to Datasetand rename compression_options to compression_opts

Additional changes:

• Add unix environment files for python 2.7, 3.5, 3.6, 3.7 and use for travis build, thanks to @clbarnes
• Fix issue in c++ implementation of blocking
• Remove json submodule

Support N5 varlength

Varlength chunks store chunk data with arbitrary number of elements.
In z5py, varlength chunks cannot be read/written by [].

Instead, this release also exposes access to chunks via
Dataset.write_chunk / Dataset.read_chunk.
Chunks with variable length can be written via:

ds.write_chunk(chunk_id, var_len_data, varlen=True)

and chunks in varlength mode will be correctly read by read_chunk.

Note that the zarr format does not support varlength chunks.

Fix Conda build

This release fixes the build for the new conda-forge gcc7 toolchain.
Thanks to @isuruf for this contribution.

Additional changes:

• __getitem__ does not squeeze return value
• json encoder can be obtained via get_json_encoder (thanks to @paulhfu )

Custom json, visititems, tif converter

This release includes:

• support for custom json serializers via z5py.set_json_encoder / z5py.set_json_decoder
• converters for tif images using imageio: z5py.converter.convert_from_tif
• visititems functionality in z5py.Group; thanks to @paulhfu for this contribution.

First Major Release

First major release:

• N5 specification on filesystem is supported except for var-size chunks.
• Zarr specification on filesystem is suported except for F-order and variable endianess.
• Python API is mostly compatible with h5py and should be stable.
• Conda-forge packages for relevant systems and python versions.

Update build

0.5.5

Update build

Fix build issues

0.5.4

Update dataToFormat implementation