Comment 3 for bug 1314544

Reviewed: https://review.openstack.org/96514
Committed: https://git.openstack.org/cgit/openstack/python-ceilometerclient/commit/?id=2a2e2dd2732240a08d7f0e1288229a39ae380d1c
Submitter: Jenkins
Branch: master

commit 2a2e2dd2732240a08d7f0e1288229a39ae380d1c
Author: Chris Dent <email address hidden>
Date: Fri May 30 11:40:11 2014 +0100

    Refactor split_by_op and split_by_datatype

    split_by_op was returning data against which further error checking was
    then required. This change moves that error handling inside the method
    and simplifies the regular expression so that it splits (greedily) on
    the first operator it finds. If the split is not possible, it is
    a ValueError. If the field or value are empty, that is a ValueError.

    Both split_by_op and split_by_datatype were doing a findall() where
    a match() and split() do the right job and more efficiently.

    Regular expression compilation has been moved to the module level
    to insure they need only be compiled once. Operator keys must be
    sorted by length to ensure the point at which the split happens is
    most greedy. Using a split keeps the regex short and removes any
    statements about the left and right hand sides of the operator.

    Tests added to cover the method more completely, including testing
    for corner cases such as single character field or values or
    operators showing up in unexpected locations. 'string' variable
    renamed to 'query' and 'query_value' to avoid confusion. Named parameters
    on string substitution for clarity.

    Note that the tests which do self.assertRaises could more explicitly
    check the exception with self.assertRaisesRegexp but that would
    break compatibility with Python 2.6.

    Change-Id: Icd815ff65aba9eae3f76afee3bb33e85d85bea72
    Closes-Bug: #1314544