tests: Update float2int tests for new range classifications

- Tests vary based on build configuration (32/64-bit and internal int type).
- Added tests for exceptions raised on overflow of int type.
This commit is contained in:
David Steinberg 2015-01-23 17:52:22 +00:00 committed by Damien George
parent 0fb17f6ef4
commit 4c1a7e0d6a
2 changed files with 134 additions and 19 deletions

View File

@ -1,12 +1,33 @@
# check cases converting float to int, relying only on single precision float # check cases converting float to int, relying only on single precision float
import struct
# work out configuration values
is_64bit = struct.calcsize("P") == 8
# 0 = none, 1 = long long, 2 = mpz
try:
dummy = 0x7fffffffffffffff
try:
if (0xffffffffffffffff + 1) > 0:
ll_type = 2
else:
ll_type = 1
except:
# in case the sum in the if statement above changes to raising an exception on overflow
ll_type = 1
except:
ll_type = 0
# basic conversion
print(int(14187745.)) print(int(14187745.))
print(int(2.**100))
print("%d" % 14187745.) print("%d" % 14187745.)
print("%d" % 2.**100) if ll_type == 2:
print(int(2.**100))
print("%d" % 2.**100)
testpass = True testpass = True
for i in range(0,128): p2_rng = ((30,63,127),(62,63,127))[is_64bit][ll_type]
for i in range(0,p2_rng):
bitcnt = len(bin(int(2.**i))) - 3; bitcnt = len(bin(int(2.**i))) - 3;
if i != bitcnt: if i != bitcnt:
print('fail: 2**%u was %u bits long' % (i, bitcnt)); print('fail: 2**%u was %u bits long' % (i, bitcnt));
@ -15,25 +36,51 @@ print("power of 2 test: %s" % (testpass and 'passed' or 'failed'))
# TODO why does 10**12 fail this test for single precision float? # TODO why does 10**12 fail this test for single precision float?
testpass = True testpass = True
for i in range(0,12): p10_rng = 9 if (ll_type == 0 and ~is_64bit) else 11
for i in range(0,p10_rng):
digcnt = len(str(int(10.**i))) - 1; digcnt = len(str(int(10.**i))) - 1;
if i != digcnt: if i != digcnt:
print('fail: 10**%u was %u digits long' % (i, digcnt)); print('fail: 10**%u was %u digits long' % (i, digcnt));
testpass = False testpass = False
print("power of 10 test: %s" % (testpass and 'passed' or 'failed')) print("power of 10 test: %s" % (testpass and 'passed' or 'failed'))
# test inf conversion def fp2int_test(num, name, should_fail):
try: try:
int(float('inf')) x = int(num)
except OverflowError: passed = ~should_fail
print("OverflowError") except:
passed = should_fail
print('%s: %s' % (name, passed and 'passed' or 'failed'))
# test nan conversion if ll_type != 2:
try: if ll_type == 0:
int(float('nan')) if is_64bit:
except ValueError: neg_bad_fp = -1.00000005*2.**62.
print("ValueError") pos_bad_fp = 2.**62.
neg_good_fp = -2.**62.
pos_good_fp = 0.99999993*2.**62.
else:
neg_bad_fp = -1.00000005*2.**30.
pos_bad_fp = 2.**30.
neg_good_fp = -2.**30.
pos_good_fp = 0.9999999499*2.**30.
else:
neg_bad_fp = -0.51*2.**64.
pos_bad_fp = 2.**63.
neg_good_fp = -2.**63.
pos_good_fp = 1.9999998*2.**62.
fp2int_test(neg_bad_fp, 'neg bad', True)
fp2int_test(pos_bad_fp, 'pos bad', True)
fp2int_test(neg_good_fp, 'neg good', False)
fp2int_test(pos_good_fp, 'pos good', False)
else:
fp2int_test(-1.999999879*2.**127., 'large neg', False)
fp2int_test(1.999999879*2.**127., 'large pos', False)
fp2int_test(float('inf'), 'inf test', True)
fp2int_test(float('nan'), 'NaN test', True)
# test numbers < 1 (this used to fail; see issue #1044) # test numbers < 1 (this used to fail; see issue #1044)
import struct fp2int_test(0.0001, 'small num', False)
struct.pack('I', int(1/2)) struct.pack('I', int(1/2))

View File

@ -1,11 +1,37 @@
# check cases converting float to int, requiring double precision float # check cases converting float to int, requiring double precision float
import struct
# work out configuration values
is_64bit = struct.calcsize("P") == 8
# 0 = none, 1 = long long, 2 = mpz
try:
dummy = 0x7fffffffffffffff
try:
if (0xffffffffffffffff + 1) > 0:
ll_type = 2
else:
ll_type = 1
except:
# in case the sum in the if statement above changes to raising an exception on overflow
ll_type = 1
except:
ll_type = 0
# This case occurs with time.time() values # This case occurs with time.time() values
print(int(1418774543.)) if ll_type != 0:
print("%d" % 1418774543.) print(int(1418774543.))
print("%d" % 1418774543.)
if ll_type == 3:
print(int(2.**100))
print("%d" % 2.**100)
else:
print(int(1073741823.))
print("%d" % 1073741823.)
testpass = True testpass = True
for i in range(0,1024): p2_rng = ((30,63,1024),(62,63,1024))[is_64bit][ll_type]
for i in range(0,p2_rng):
bitcnt = len(bin(int(2.**i))) - 3; bitcnt = len(bin(int(2.**i))) - 3;
if i != bitcnt: if i != bitcnt:
print('fail: 2**%u was %u bits long' % (i, bitcnt)); print('fail: 2**%u was %u bits long' % (i, bitcnt));
@ -13,9 +39,51 @@ for i in range(0,1024):
print("power of 2 test: %s" % (testpass and 'passed' or 'failed')) print("power of 2 test: %s" % (testpass and 'passed' or 'failed'))
testpass = True testpass = True
for i in range(0,23): p10_rng = ((9,18,23),(18,18,23))[is_64bit][ll_type]
for i in range(0,p10_rng):
digcnt = len(str(int(10.**i))) - 1; digcnt = len(str(int(10.**i))) - 1;
if i != digcnt: if i != digcnt:
print('fail: 10**%u was %u digits long' % (i, digcnt)); print('fail: 10**%u was %u digits long' % (i, digcnt));
testpass = False testpass = False
print("power of 10 test: %s" % (testpass and 'passed' or 'failed')) print("power of 10 test: %s" % (testpass and 'passed' or 'failed'))
def fp2int_test(num, name, should_fail):
try:
x = int(num)
passed = ~should_fail
except:
passed = should_fail
print('%s: %s' % (name, passed and 'passed' or 'failed'))
if ll_type != 2:
if ll_type == 0:
if is_64bit:
neg_bad_fp = -1.00000005*2.**62.
pos_bad_fp = 2.**62.
neg_good_fp = -2.**62.
pos_good_fp = 0.99999993*2.**62.
else:
neg_bad_fp = -1.00000005*2.**30.
pos_bad_fp = 2.**30.
neg_good_fp = -2.**30.
pos_good_fp = 0.9999999499*2.**30.
else:
neg_bad_fp = -0.51*2.**64.
pos_bad_fp = 2.**63.
neg_good_fp = -2.**63.
pos_good_fp = 1.9999998*2.**62.
fp2int_test(neg_bad_fp, 'neg bad', True)
fp2int_test(pos_bad_fp, 'pos bad', True)
fp2int_test(neg_good_fp, 'neg good', False)
fp2int_test(pos_good_fp, 'pos good', False)
else:
fp2int_test(-1.9999999999999981*2.**1023., 'large neg', False)
fp2int_test(1.9999999999999981*2.**1023., 'large pos', False)
fp2int_test(float('inf'), 'inf test', True)
fp2int_test(float('nan'), 'NaN test', True)
# test numbers < 1 (this used to fail; see issue #1044)
fp2int_test(0.0001, 'small num', False)
struct.pack('I', int(1/2))