What’s new in PyTables 1.0

Author:

Francesc Altet

Contact:

faltet@carabos.com

Author:

Ivan Vilata i Balaguer

Contact:

ivilata@carabos.com

This document details the modifications to PyTables since version 0.9.1. Its main purpose is help you ensure that your programs will be runnable when you switch from PyTables 0.9.1 to PyTables 1.0.

API additions

  • The new Table.col() method can be used to get a column from a table as a NumArray or CharArray object. This is preferred over the syntax table['colname'].

  • The new Table.readCoordinates() method reads a set of rows given their indexes into an in-memory object.

  • The new Table.readAppend() method Append rows fulfilling the condition to a destination table.

Backward-incompatible changes

  • Trying to open a nonexistent file or a file of unknown type raises IOError instead of RuntimeError. Using an invalid mode raises ValueError instead of RuntimeError.

  • Getting a child node from a closed group raises ValueError instead of RuntimeError.

  • Running an action on the wrong type of node now (i.e. using file.listNodes() on a leaf) raises a TypeError instead of a NodeError.

  • Removing a non-existing child now raises a NoSuchNodeError, instead of doing nothing.

  • Removing a non-empty child group using del group.child fails with a NodeError instead of recursively doing the removal. This is because of the potential damage it may cause when used inadvertently. If a recursive behavior is needed, use the _f_remove() method of the child node.

  • The recursive flag of Group._f_walkNodes() is True by default now. Before it was False.

  • Now, deleting and getting a non-existing attribute raises an AttributeError instead of a RuntimeError.

  • Swapped last two arguments of File.copyAttrs() to match the other methods. Please use File.copyNodeAttrs() anyway.

  • Failing to infer the size of a string column raises ValueError instead of RuntimeError.

  • Excessive table column name length and number of columns now raise ValueError instead of IndexError and NameError.

  • Excessive table row length now raises ValueError instead of RuntimeError.

  • table[integer] returns a numarray.records.Record object instead of a tuple. This was the original behavior before PyTables 0.9 and proved to be more consistent than the last one (tables do not have an explicit ordering of columns).

  • Specifying a nonexistent column in Table.read() raises a ValueError instead of a LookupError.

  • When start >= stop an empty iterator is returned by Table.iterrows() instead of an empty RecArray. Thanks to Ashley Walsh for noting this.

  • The interface of isHDF5File() and isPyTablesFile() file has been unified so that they both return true or false values on success and raise HDF5ExtError or errors. The true value in isPyTablesFile() is the format version string of the file.

  • Table.whereIndexed() and Table.whereInRange() are now private methods, since the Table.where() method is able to choose the most adequate option.

  • The global variables ExtVersion and HDF5Version have been renamed to extVersion and hdf5Version, respectively.

  • whichLibVersion() returns None on querying unavailable libraries, and raises ValueError on unknown ones.

The following modifications, though being (strictly speaking) modifications of the API, will most probably not cause compatibility problems (but your mileage may vary):

  • The default values for name and classname arguments in File.getNode() are now None, although the empty string is still allowed for backwards compatibility. File hierarchy manipulation and attribute handling operations using those arguments have changed to reflect this.

  • Copy operations (Group._f_copyChildren(), File.copyChildren(), File.copyNode()…) do no longer return a tuple with the new node and statistics. Instead, they only return the new node, and statistics are collected via an optional keyword argument.

  • The copyFile() function in File.py has changed its signature from:

    copyFile(srcfilename=None, dstfilename=None, title=None, filters=None,
             copyuserattrs=True, overwrite=False, stats=None)
    

    to:

    copyFile(srcfilename, dstfilename, overwrite=False, **kwargs)
    

    Thus, the function allows the same options as File.copyFile().

  • The File.copyFile() method has changed its signature from:

    copyFile(self, dstfilename=None, title=None, filters=None,
             copyuserattrs=1, overwrite=0, stats=None):
    

    to:

    copyFile(self, dstfilename, overwrite=False, **kwargs)
    

    This enables this method to pass on arbitrary flags and options supported by copying methods of inner nodes in the hierarchy.

  • The File.copyChildren() method has changed its signature from:

    copyChildren(self, wheresrc, wheredst, recursive=False, filters=None,
                 copyuserattrs=True, start=0, stop=None, step=1,
                 overwrite=False, stats=None)
    

    to:

    copyChildren(self, srcgroup, dstgroup, overwrite=False, recursive=False,
                 **kwargs):
    

    Thus, the function allows the same options as Group._f_copyChildren().

  • The Group._f_copyChildren() method has changed its signature from:

    _f_copyChildren(self, where, recursive=False, filters=None,
                    copyuserattrs=True, start=0, stop=None, step=1,
                    overwrite=False, stats=None)
    

    to:

    _f_copyChildren(self, dstgroup, overwrite=False, recursive=False,
                    **kwargs)
    

    This enables this method to pass on arbitrary flags and options supported by copying methods of inner nodes in the group.

  • Renamed srcFilename and dstFilename arguments in copyFile() and File.copyFile() to srcfilename and dstfilename, respectively. Renamed whereSrc and whereDst arguments in File.copyChildren() to wheresrc and wheredst, respectively. Renamed dstNode argument in File.copyAttrs() to dstnode. Tose arguments should be easier to type in interactive sessions (although 99% of the time it is not necessary to specify them).

  • Renamed object argument in EArray.append() to sequence.

  • The rows argument in Table.append() is now compulsory.

  • The start argument in Table.removeRows() is now compulsory.

API refinements

  • The isHDF5() function has been deprecated in favor of isHDF5File().

  • Node attribute-handling methods in File have been renamed for a better coherence and understanding of their purpose:

    • getAttrNode() is now called getNodeAttr()

    • setAttrNode() is now called setNodeAttr()

    • delAttrNode() is now called delNodeAttr()

    • copyAttrs() is now called copyNodeAttrs()

    They keep their respective signatures, and the old versions still exist for backwards compatibility, though they issue a DeprecationWarning.

  • Using VLArray.append() with multiple arguments is now deprecated for its ambiguity. You should put the arguments in a single sequence object (list, tuple, array…) and pass it as the only argument.

  • Using table['colname'] is deprecated. Using table.col('colname') (with the new col() method) is preferred.

Bug fixes (affecting API)

  • Table.iterrows() returns an empty iterator when no rows are selected, instead of returning None.


Enjoy data!

—The PyTables Team