• Subassign syntax to allow masks sized for the assignment dimensions rather than the full output dimensions
• input_mask allows masks to apply to the full input object rather than the extracted dimensions
• Call recorder to auto-generate C code from grblas code
• to_values() returns numpy arrays rather than tuples
• fast_import/fast_export suitesparse extensions are exposed under the .ss namespace

• Much nicer reprs (plain and html) for objects (both data and expressions), which replaces .show()
• Add “pygraphblas” as a backend choice. This add ability to convert between grblas and pygraphblas.
• Data objects now have names! (passed in at construction time)
• Moved tests/ inside the grblas package so we can test it after install.
• Use versioneer.py to manage version.
• Apply black code formatter.
• More extensive testing and a few minor bug fixes.

Update to SuiteSparse::GraphBLAS 3.3.1

• Apply binary
• Kronecker
• Remove element
• Resize
• Complex numbers
• Additional builtin unary, binary, monoid, semiring

This release has breaking changes to the API:

• Matrix.new_from_type became Matrix.new
• Matrix.new_from_existing was deprecated. Use A.dup() instead.
• Matrix.new_from_values became Matrix.from_values
• Masks now require explicit indication of values M.V or structure M.S
• ewise_add with binary ops will raise unless require_monoid is set to False in the function call
• binary.div was renamed to binary.cdiv. Two additional division operators were added: binary.truediv and binary.floordiv
• Comparison via == was removed. New comparison methods M.isequal and M.isclose were added.

The version of GraphBLAS was updated to 3.2.2 to gain access to structural masks.
Most numpy operators are also available under unary.numpy or binary.numpy.

This release has breaking changes to the API:

• Matrix.new_from_type became Matrix.new
• Matrix.new_from_existing was deprecated. Use A.dup() instead.
• Matrix.new_from_values became Matrix.from_values
• Masks now require explicit indication of values M.V or structure M.S
• ewise_add with binary ops will raise unless require_monoid is set to False in the function call
• binary.div was renamed to binary.cdiv. Two additional division operators were added: binary.truediv and binary.floordiv
• Comparison via == was removed. New comparison methods M.isequal and M.isclose were added.

The version of GraphBLAS was updated to 3.2.2 to gain access to structural masks.
Most numpy operators are also available under unary.numpy or binary.numpy.

Update to SuiteSparse::GraphBLAS v3.2.2

Major overhaul of syntax – replace C[mask] = ... with C(mask) << .... This allows keyword arguments and separation of indexes for assign.

Operators have also changed. From BinaryOp.PLUS to binary.plus