7d414a1b52
Previous to this patch, a big-int, float or imag constant was interned (made into a qstr) and then parsed at runtime to create an object each time it was needed. This is wasteful in RAM and not efficient. Now, these constants are parsed straight away in the parser and turned into objects. This allows constants with large numbers of digits (so addresses issue #1103) and takes us a step closer to #722.
84 lines
1.4 KiB
Python
84 lines
1.4 KiB
Python
print(int(False))
|
|
print(int(True))
|
|
|
|
print(int(0))
|
|
print(int(1))
|
|
print(int(+1))
|
|
print(int(-1))
|
|
|
|
print(int('0'))
|
|
print(int('+0'))
|
|
print(int('-0'))
|
|
print(int('1'))
|
|
print(int('+1'))
|
|
print(int('-1'))
|
|
print(int('01'))
|
|
print(int('9'))
|
|
print(int('10'))
|
|
print(int('+10'))
|
|
print(int('-10'))
|
|
print(int('12'))
|
|
print(int('-12'))
|
|
print(int('99'))
|
|
print(int('100'))
|
|
print(int('314'))
|
|
print(int(' 314'))
|
|
print(int('314 '))
|
|
print(int(' \t\t 314 \t\t '))
|
|
print(int(' 1 '))
|
|
print(int(' -3 '))
|
|
|
|
print(int('0', 10))
|
|
print(int('1', 10))
|
|
print(int(' \t 1 \t ', 10))
|
|
print(int('11', 10))
|
|
print(int('11', 16))
|
|
print(int('11', 8))
|
|
print(int('11', 2))
|
|
print(int('11', 36))
|
|
print(int('0o123', 0))
|
|
print(int('8388607'))
|
|
print(int('0x123', 16))
|
|
print(int('0X123', 16))
|
|
print(int('0A', 16))
|
|
print(int('0o123', 8))
|
|
print(int('0O123', 8))
|
|
print(int('0123', 8))
|
|
print(int('0b100', 2))
|
|
print(int('0B100', 2))
|
|
print(int('0100', 2))
|
|
print(int(' \t 0o12', 8))
|
|
print(int('0o12 \t ', 8))
|
|
print(int(b"12", 10))
|
|
print(int(b"12"))
|
|
|
|
|
|
def test(value, base):
|
|
try:
|
|
print(int(value, base))
|
|
except ValueError:
|
|
print('ValueError')
|
|
|
|
|
|
test('x', 0)
|
|
test('1x', 0)
|
|
test(' 1x', 0)
|
|
test(' 1' + chr(2) + ' ', 0)
|
|
test('', 0)
|
|
test(' ', 0)
|
|
test(' \t\t ', 0)
|
|
test("\u0200", 0)
|
|
test('0x', 16)
|
|
test('0x', 0)
|
|
test('0o', 8)
|
|
test('0o', 0)
|
|
test('0b', 2)
|
|
test('0b', 0)
|
|
test('0b2', 2)
|
|
test('0o8', 8)
|
|
test('0xg', 16)
|
|
test('1 1', 16)
|
|
|
|
# check that we don't parse this as a floating point number
|
|
print(0x1e+1)
|