• ReleaseNotes
• Release_2.2.1rc1

Changes from 2.2 to 2.2.1rc1

• When using ObjectAtom objects in VLArrays the HIGHEST_PROTOCOL is used for pickling objects. For NumPy arrays, this simple change leads to space savings up to 3x and time improvements up to 30x. Closes #301.
• tables.Expr can perform operations on scalars now. Thanks to Gaëtan de Menten for providing a patch for this. Closes #287.
• Fixed a problem with indexes larger than 32-bit on leaf objects on 32-bit machines. Fixes #283.
• Merged in Blosc 1.1 for fixing a problem with large datatypes and subprocess issues. Closes #288 and #295.
• Due to the adoption of Blosc 1.1, the pthreads-win32 library dependency is dropped on Windows platforms.
• Fixed a problem with tables.Expr and operands with vary large rowsizes. Closes #300.
• leaf[numpy.array[scalar]] idiom returns a NumPy array instead of an scalar. This has been done for compatibility with NumPy. Closes #303.
• Optimization for Table.copy() so that FIELD_* attrs are not overwritten during the copy. This can lead to speed-ups up to 100x for short tables that have hundreds of columns. Closes #304.
• For external links, its relative paths are resolved now with respect to the directory of the main HDF5 file, rather than with respect to the current directory. Closes #306.
• Expr.setInputsRange() and Expr.setOutputRange() do support numpy.integer types now. Closes #285.
• Column names in tables can start with '__' now. Closes #291.
• Unicode empty strings are supported now as atributes. Addresses #307.
• Cython 0.13 and higher is supported now. Fixes #293.
• PyTables should be more 'easy_install'-able now. Addresses #298.