Skip to content

Instantly share code, notes, and snippets.

@honno
Last active November 2, 2022 19:20
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save honno/98371048b4d5b9fc4a003377a714e827 to your computer and use it in GitHub Desktop.
Save honno/98371048b4d5b9fc4a003377a714e827 to your computer and use it in GitHub Desktop.
import math
import sys
from itertools import product
from operator import index
import numpy as np
import pytest
from hypothesis import assume, given, note
from hypothesis import strategies as st
from hypothesis.extra import numpy as nps
assert sys.maxsize == 2**63 - 1 # assume 64-bit Python
builtin_to_dtype = {
bool: np.bool_,
int: np.int64,
float: np.float64,
complex: np.complex128,
}
@pytest.mark.parametrize("use_np", [False, True])
@pytest.mark.parametrize(
"builtin, arg_kind", product([bool, int, float, complex], repeat=2)
)
@given(data=st.data())
def test_builtin_casting(builtin, arg_kind, use_np, data):
if use_np:
_math = np
dtype = builtin_to_dtype[arg_kind]
# We forcibly map examples with np.asarray to ensure we don't get NumPy
# scalars. This is because they have slightly different behaviour, e.g.
#
# >>> np_array = np.asarray(1+1j, dtype=np.complex128)
# >>> int(np_array)
# TypeError: can't convert complex to int
# >>> np_scalar = np.complex128(1+1j)
# >>> int(np_scalar)
# 1
#
arg_strat = nps.from_dtype(np.dtype(dtype)).map(np.asarray)
else:
_math = math
arg_strat = st.from_type(arg_kind)
arg = data.draw(arg_strat, label="arg")
f_expr = f"{builtin.__name__}({arg})"
if builtin == int and arg_kind == float and not _math.isfinite(arg):
if _math.isinf(arg):
errclass = OverflowError
else:
assert _math.isnan(arg) # sanity check
errclass = ValueError
with pytest.raises(errclass):
builtin(arg)
note(f"{f_expr}=<{errclass.__name__}>")
elif builtin in [int, float] and arg_kind == complex:
with pytest.raises(TypeError):
builtin(arg)
note(f"{f_expr}=<TypeError>")
else:
res = builtin(arg)
note(f"{f_expr}={res}")
assert type(res) == builtin
if builtin == bool:
assert not res if arg == 0 else res
elif arg_kind == bool:
assert res == 1 if arg else res == 0
elif builtin == int:
assert arg_kind in [int, float] # sanity check
if arg_kind == float:
assume(abs(arg) < 2**63)
assert res == _math.trunc(arg)
elif builtin == float:
assert arg_kind in [int, float] # sanity check
if _math.isnan(arg):
assert _math.isnan(res)
else:
if arg_kind == int:
assume(abs(arg) <= 2**53)
assert res == arg
else:
assert builtin == complex # sanity check
if arg_kind in [int, float]:
assert res.imag == 0
if _math.isnan(arg):
assert _math.isnan(res.real)
else:
if arg_kind == int:
assume(abs(arg) <= 2**53)
assert res.real == arg
else:
assert arg_kind == complex # sanity check
for component, arg_component in [
(res.real, arg.real),
(res.imag, arg.imag),
]:
if _math.isnan(arg_component):
assert _math.isnan(component)
else:
assert component == arg_component
@pytest.mark.parametrize("use_np", [False, True])
@pytest.mark.parametrize("arg_kind", [bool, int, float, complex])
@given(data=st.data())
def test_index(arg_kind, use_np, data):
if use_np:
_math = np
dtype = builtin_to_dtype[arg_kind]
# See comment in test_builtin_casting
arg_strat = nps.from_dtype(np.dtype(dtype)).map(np.asarray)
else:
_math = math
arg_strat = st.from_type(arg_kind)
arg = data.draw(arg_strat, label="arg")
f_expr = f"index({arg})"
if arg_kind in [float, complex] or (arg_kind == bool and use_np):
with pytest.raises(TypeError):
index(arg)
note(f"{f_expr}=<TypeError>")
else:
res = index(arg)
note(f"{f_expr}={res}")
assert type(res) == bool if arg_kind == bool else int
if arg_kind == bool:
assert res is True if arg else res is False
else:
assert arg_kind == int # sanity check
assert res == _math.trunc(arg)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment