def test_int8_missing_range_encoding():
# Also test with negative numbers!
for offset in (0, -100):
s = pd.Series((1+offset, None, 2**8+offset-1))
c = select_codec('column', s, None)
assert type(c) == codec.Int8Missing
assert c.min == 1+offset
_check_encode(c, s, b'\x00\xff\xfe')
def test_int16_range_encoding_maximal():
# Also test with negative numbers!
for offset in (0, -10000):
s = pd.Series((1+offset, 2**8+offset, 2**16+offset))
c = select_codec('column', s, None)
assert type(c) == codec.Int16
assert c.min == 1+offset
_check_encode(c, s, b'\x00\x00\xff\x00\xff\xff')
def test_int32_range_encoding():
"""
n.b. the Int32 codec is a bit crappy. It does _not_ include an offset value
--> It only encodes the legit values of a SIGNED 32bit integer
--> 64bit integers are todo (but break some fortran compatibility, as not all
64bit integers can be represented as doubles).
--> Can include missing values
"""
s = pd.Series((-2**31, None, 2**31-2))
c = select_codec('column', s, None)
assert isinstance(c, codec.Int32)
assert c.min == -2**31
_check_encode(c, s, b'\x00\x00\x00\x80\xff\xff\xff\x7f\xfe\xff\xff\x7f')
def test_int16_range_encoding_minimal():
"""
A span of integers that _just_ requires int16
"""
# Also test with negative numbers!
for offset in (0, -10000):
s = pd.Series((1+offset, 2**8+offset+1))
c = select_codec('column', s, None)
assert type(c) == codec.Int16
assert c.min == 1+offset
_check_encode(c, s, b'\x00\x00\x00\x01')
def test_wider_range_unsupported():
s = pd.Series((-2**31, 2**31-1))
with pytest.raises(NotImplementedError):
select_codec('column', s, None)