It appears that Python's __getslice__(self, a, b) overload has a surprising
behavior. From
http://docs.python.org/reference/datamodel.html#object.__getslice__:
"If negative indexes are used in the slice, the length of the sequence is
added to that index."
So when you evaluate rows[-20:], Python evaluates len(rows) (which is 13 in
the above example), and calls:
rows.__getslice__(-20+13, 2147483647)
rather than:
rows.__getslice__(-20, None)
or rows.__getslice__(-20, 2147483647)
as one might initially suspect. The Rows object just passes the received
indices on to self.records, so you're getting a new Rows back containing
records[-7:] - clearly not what you want!!
SOLUTION:
It turns out that __getslice__ is deprecated since Python 2.0, and that
__getitem__ should be used instead. If __getslice__ is missing and a a slice
is requested, __getitem__ receives a "slice" object. A proposed fix:
- remove __getslice__ from Rows
- replace __getitem__ with:
def __getitem__(self, i):
if isinstance(i, slice):
return Rows(self.db, self.records[i], self.colnames)
else:
row = self.records[i]
keys = row.keys()
if self.compact and len(keys) == 1 and keys[0] != '_extra':
return row[row.keys()[0]]
return row
I've done a bit of testing with this. It fixes the problem noted by Gene,
and does not appear to break anything obvious.
Cheers,
Kevin
