Release notes for PyTables 2.1 series

Author:Francesc Alted i Abad
Contact:faltet@pytables.org

Changes from 2.1.1 to 2.1.2

Bug fixes

  • Solved problems with Table.modifyColumn() when the column(s) is multidimensional. Fixes #228.
  • The row attribute of a table seems stalled after a table move or rename. Fixes #224.
  • Fixed a problem with len(array) in 32-bit platforms when array is large enough (> 2**31).
  • Added missing _c_classId attribute to the UnImplemented class. ptrepack no longer chokes while copying Unimplemented classes.
  • The FIELD_* sys attrs are no longer copied when the PYTABLES_SYS_ATTRS parameter is set to false.
  • The FILTERS attribute is not added anymore when PYTABLES_SYS_ATTR parameter is set to false.
  • Disable the printing of Unicode characters that cannot be printed on win32 platform. Fixes #235.

Other changes

  • When retrieving a row of a 1-dimensional array, a 0-dim array was returned instead of a numpy scalar. Now, an actuall numpy scalar is returned. Closes #222.
  • LZO and bzip2 filters adapted to an API fix introduced in HDF5 1.8.3. Closes #225.
  • Unsupported HDF5 types in attributes are no longer transferred during copies. A new _v_unimplemented list have been added in AttributeSet class so as to keep track of such attributes. Closes #240.
  • LZO binaries have disappeared from the GnuWin32 repository. Until they come eventually back, they have been put at http://www.pytables.org/download/lzo-win. This has been documented in the install chapter.

Changes from 2.1 to 2.1.1

Bug fixes

  • Fixed a memory leak when a lot of queries were made. Closes #203 and #207.
  • The chunkshape=”auto” parameter value of Leaf.copy() is honored now, even when the (start, stop, step) parameters are specified. Closes #204.
  • Due to a flaw in its design, the File class was not able to be subclassed. This has been fixed. Closes #205.
  • Default values were not correctly retrieved when opening already created CArray/EArray objects. Fixed. Closes #212.
  • Fixed a problem with the installation of the nctoh5 script that prevented it from being executed. Closes #215.
  • [Pro] The iterseq cache ignored non-indexed conditions, giving wrong results when those appeared in condition expressions. This has been fixed. Closes #206.

Other changes

  • openFile(), isHDF5File() and isPyTablesFile() functions accept Unicode filenames now. Closes #202 and #214.
  • When creating large type sizes (exceeding 64 KB), HDF5 complained and refused to do so. The HDF5 team has logged the issue as a bug, but meanwhile it has been implemented a workaround in PyTables that allows to create such large datatypes for situations that does not require defaults other than zero. Addresses #211.
  • In order to be consistent with how are stored the other data types, Unicode attributes are retrieved now as NumPy scalars instead of Python Unicode strings or NumPy arrays. For the moment, I’ve fixed this through pickling the Unicode strings. In the future, when HDF5 1.8.x series would be a requirement, that should be done via a HDF5 native Unicode type. Closes #213.

Enjoy data!

—The PyTables Team