499 lines
18 KiB
Python
499 lines
18 KiB
Python
import unittest
|
|
import sys
|
|
import _ast
|
|
from test import test_support
|
|
import textwrap
|
|
|
|
class TestSpecifics(unittest.TestCase):
|
|
|
|
def test_no_ending_newline(self):
|
|
compile("hi", "<test>", "exec")
|
|
compile("hi\r", "<test>", "exec")
|
|
|
|
def test_empty(self):
|
|
compile("", "<test>", "exec")
|
|
|
|
def test_other_newlines(self):
|
|
compile("\r\n", "<test>", "exec")
|
|
compile("\r", "<test>", "exec")
|
|
compile("hi\r\nstuff\r\ndef f():\n pass\r", "<test>", "exec")
|
|
compile("this_is\rreally_old_mac\rdef f():\n pass", "<test>", "exec")
|
|
|
|
def test_debug_assignment(self):
|
|
# catch assignments to __debug__
|
|
self.assertRaises(SyntaxError, compile, '__debug__ = 1', '?', 'single')
|
|
import __builtin__
|
|
prev = __builtin__.__debug__
|
|
setattr(__builtin__, '__debug__', 'sure')
|
|
setattr(__builtin__, '__debug__', prev)
|
|
|
|
def test_argument_handling(self):
|
|
# detect duplicate positional and keyword arguments
|
|
self.assertRaises(SyntaxError, eval, 'lambda a,a:0')
|
|
self.assertRaises(SyntaxError, eval, 'lambda a,a=1:0')
|
|
self.assertRaises(SyntaxError, eval, 'lambda a=1,a=1:0')
|
|
try:
|
|
exec 'def f(a, a): pass'
|
|
self.fail("duplicate arguments")
|
|
except SyntaxError:
|
|
pass
|
|
try:
|
|
exec 'def f(a = 0, a = 1): pass'
|
|
self.fail("duplicate keyword arguments")
|
|
except SyntaxError:
|
|
pass
|
|
try:
|
|
exec 'def f(a): global a; a = 1'
|
|
self.fail("variable is global and local")
|
|
except SyntaxError:
|
|
pass
|
|
|
|
def test_syntax_error(self):
|
|
self.assertRaises(SyntaxError, compile, "1+*3", "filename", "exec")
|
|
|
|
def test_none_keyword_arg(self):
|
|
self.assertRaises(SyntaxError, compile, "f(None=1)", "<string>", "exec")
|
|
|
|
def test_duplicate_global_local(self):
|
|
try:
|
|
exec 'def f(a): global a; a = 1'
|
|
self.fail("variable is global and local")
|
|
except SyntaxError:
|
|
pass
|
|
|
|
def test_exec_with_general_mapping_for_locals(self):
|
|
|
|
class M:
|
|
"Test mapping interface versus possible calls from eval()."
|
|
def __getitem__(self, key):
|
|
if key == 'a':
|
|
return 12
|
|
raise KeyError
|
|
def __setitem__(self, key, value):
|
|
self.results = (key, value)
|
|
def keys(self):
|
|
return list('xyz')
|
|
|
|
m = M()
|
|
g = globals()
|
|
exec 'z = a' in g, m
|
|
self.assertEqual(m.results, ('z', 12))
|
|
try:
|
|
exec 'z = b' in g, m
|
|
except NameError:
|
|
pass
|
|
else:
|
|
self.fail('Did not detect a KeyError')
|
|
exec 'z = dir()' in g, m
|
|
self.assertEqual(m.results, ('z', list('xyz')))
|
|
exec 'z = globals()' in g, m
|
|
self.assertEqual(m.results, ('z', g))
|
|
exec 'z = locals()' in g, m
|
|
self.assertEqual(m.results, ('z', m))
|
|
try:
|
|
exec 'z = b' in m
|
|
except TypeError:
|
|
pass
|
|
else:
|
|
self.fail('Did not validate globals as a real dict')
|
|
|
|
class A:
|
|
"Non-mapping"
|
|
pass
|
|
m = A()
|
|
try:
|
|
exec 'z = a' in g, m
|
|
except TypeError:
|
|
pass
|
|
else:
|
|
self.fail('Did not validate locals as a mapping')
|
|
|
|
# Verify that dict subclasses work as well
|
|
class D(dict):
|
|
def __getitem__(self, key):
|
|
if key == 'a':
|
|
return 12
|
|
return dict.__getitem__(self, key)
|
|
d = D()
|
|
exec 'z = a' in g, d
|
|
self.assertEqual(d['z'], 12)
|
|
|
|
def test_extended_arg(self):
|
|
longexpr = 'x = x or ' + '-x' * 2500
|
|
code = '''
|
|
def f(x):
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
%s
|
|
# the expressions above have no effect, x == argument
|
|
while x:
|
|
x -= 1
|
|
# EXTENDED_ARG/JUMP_ABSOLUTE here
|
|
return x
|
|
''' % ((longexpr,)*10)
|
|
exec code
|
|
self.assertEqual(f(5), 0)
|
|
|
|
def test_complex_args(self):
|
|
|
|
with test_support.check_py3k_warnings(
|
|
("tuple parameter unpacking has been removed", SyntaxWarning)):
|
|
exec textwrap.dedent('''
|
|
def comp_args((a, b)):
|
|
return a,b
|
|
self.assertEqual(comp_args((1, 2)), (1, 2))
|
|
|
|
def comp_args((a, b)=(3, 4)):
|
|
return a, b
|
|
self.assertEqual(comp_args((1, 2)), (1, 2))
|
|
self.assertEqual(comp_args(), (3, 4))
|
|
|
|
def comp_args(a, (b, c)):
|
|
return a, b, c
|
|
self.assertEqual(comp_args(1, (2, 3)), (1, 2, 3))
|
|
|
|
def comp_args(a=2, (b, c)=(3, 4)):
|
|
return a, b, c
|
|
self.assertEqual(comp_args(1, (2, 3)), (1, 2, 3))
|
|
self.assertEqual(comp_args(), (2, 3, 4))
|
|
''')
|
|
|
|
def test_argument_order(self):
|
|
try:
|
|
exec 'def f(a=1, (b, c)): pass'
|
|
self.fail("non-default args after default")
|
|
except SyntaxError:
|
|
pass
|
|
|
|
def test_float_literals(self):
|
|
# testing bad float literals
|
|
self.assertRaises(SyntaxError, eval, "2e")
|
|
self.assertRaises(SyntaxError, eval, "2.0e+")
|
|
self.assertRaises(SyntaxError, eval, "1e-")
|
|
self.assertRaises(SyntaxError, eval, "3-4e/21")
|
|
|
|
def test_indentation(self):
|
|
# testing compile() of indented block w/o trailing newline"
|
|
s = """
|
|
if 1:
|
|
if 2:
|
|
pass"""
|
|
compile(s, "<string>", "exec")
|
|
|
|
# This test is probably specific to CPython and may not generalize
|
|
# to other implementations. We are trying to ensure that when
|
|
# the first line of code starts after 256, correct line numbers
|
|
# in tracebacks are still produced.
|
|
def test_leading_newlines(self):
|
|
s256 = "".join(["\n"] * 256 + ["spam"])
|
|
co = compile(s256, 'fn', 'exec')
|
|
self.assertEqual(co.co_firstlineno, 257)
|
|
self.assertEqual(co.co_lnotab, '')
|
|
|
|
def test_literals_with_leading_zeroes(self):
|
|
for arg in ["077787", "0xj", "0x.", "0e", "090000000000000",
|
|
"080000000000000", "000000000000009", "000000000000008",
|
|
"0b42", "0BADCAFE", "0o123456789", "0b1.1", "0o4.2",
|
|
"0b101j2", "0o153j2", "0b100e1", "0o777e1", "0o8", "0o78"]:
|
|
self.assertRaises(SyntaxError, eval, arg)
|
|
|
|
self.assertEqual(eval("0777"), 511)
|
|
self.assertEqual(eval("0777L"), 511)
|
|
self.assertEqual(eval("000777"), 511)
|
|
self.assertEqual(eval("0xff"), 255)
|
|
self.assertEqual(eval("0xffL"), 255)
|
|
self.assertEqual(eval("0XfF"), 255)
|
|
self.assertEqual(eval("0777."), 777)
|
|
self.assertEqual(eval("0777.0"), 777)
|
|
self.assertEqual(eval("000000000000000000000000000000000000000000000000000777e0"), 777)
|
|
self.assertEqual(eval("0777e1"), 7770)
|
|
self.assertEqual(eval("0e0"), 0)
|
|
self.assertEqual(eval("0000E-012"), 0)
|
|
self.assertEqual(eval("09.5"), 9.5)
|
|
self.assertEqual(eval("0777j"), 777j)
|
|
self.assertEqual(eval("00j"), 0j)
|
|
self.assertEqual(eval("00.0"), 0)
|
|
self.assertEqual(eval("0e3"), 0)
|
|
self.assertEqual(eval("090000000000000."), 90000000000000.)
|
|
self.assertEqual(eval("090000000000000.0000000000000000000000"), 90000000000000.)
|
|
self.assertEqual(eval("090000000000000e0"), 90000000000000.)
|
|
self.assertEqual(eval("090000000000000e-0"), 90000000000000.)
|
|
self.assertEqual(eval("090000000000000j"), 90000000000000j)
|
|
self.assertEqual(eval("000000000000007"), 7)
|
|
self.assertEqual(eval("000000000000008."), 8.)
|
|
self.assertEqual(eval("000000000000009."), 9.)
|
|
self.assertEqual(eval("0b101010"), 42)
|
|
self.assertEqual(eval("-0b000000000010"), -2)
|
|
self.assertEqual(eval("0o777"), 511)
|
|
self.assertEqual(eval("-0o0000010"), -8)
|
|
self.assertEqual(eval("020000000000.0"), 20000000000.0)
|
|
self.assertEqual(eval("037777777777e0"), 37777777777.0)
|
|
self.assertEqual(eval("01000000000000000000000.0"),
|
|
1000000000000000000000.0)
|
|
|
|
def test_unary_minus(self):
|
|
# Verify treatment of unary minus on negative numbers SF bug #660455
|
|
if sys.maxint == 2147483647:
|
|
# 32-bit machine
|
|
all_one_bits = '0xffffffff'
|
|
self.assertEqual(eval(all_one_bits), 4294967295L)
|
|
self.assertEqual(eval("-" + all_one_bits), -4294967295L)
|
|
elif sys.maxint == 9223372036854775807:
|
|
# 64-bit machine
|
|
all_one_bits = '0xffffffffffffffff'
|
|
self.assertEqual(eval(all_one_bits), 18446744073709551615L)
|
|
self.assertEqual(eval("-" + all_one_bits), -18446744073709551615L)
|
|
else:
|
|
self.fail("How many bits *does* this machine have???")
|
|
# Verify treatment of constant folding on -(sys.maxint+1)
|
|
# i.e. -2147483648 on 32 bit platforms. Should return int, not long.
|
|
self.assertIsInstance(eval("%s" % (-sys.maxint - 1)), int)
|
|
self.assertIsInstance(eval("%s" % (-sys.maxint - 2)), long)
|
|
|
|
if sys.maxint == 9223372036854775807:
|
|
def test_32_63_bit_values(self):
|
|
a = +4294967296 # 1 << 32
|
|
b = -4294967296 # 1 << 32
|
|
c = +281474976710656 # 1 << 48
|
|
d = -281474976710656 # 1 << 48
|
|
e = +4611686018427387904 # 1 << 62
|
|
f = -4611686018427387904 # 1 << 62
|
|
g = +9223372036854775807 # 1 << 63 - 1
|
|
h = -9223372036854775807 # 1 << 63 - 1
|
|
|
|
for variable in self.test_32_63_bit_values.func_code.co_consts:
|
|
if variable is not None:
|
|
self.assertIsInstance(variable, int)
|
|
|
|
def test_sequence_unpacking_error(self):
|
|
# Verify sequence packing/unpacking with "or". SF bug #757818
|
|
i,j = (1, -1) or (-1, 1)
|
|
self.assertEqual(i, 1)
|
|
self.assertEqual(j, -1)
|
|
|
|
def test_none_assignment(self):
|
|
stmts = [
|
|
'None = 0',
|
|
'None += 0',
|
|
'__builtins__.None = 0',
|
|
'def None(): pass',
|
|
'class None: pass',
|
|
'(a, None) = 0, 0',
|
|
'for None in range(10): pass',
|
|
'def f(None): pass',
|
|
'import None',
|
|
'import x as None',
|
|
'from x import None',
|
|
'from x import y as None'
|
|
]
|
|
for stmt in stmts:
|
|
stmt += "\n"
|
|
self.assertRaises(SyntaxError, compile, stmt, 'tmp', 'single')
|
|
self.assertRaises(SyntaxError, compile, stmt, 'tmp', 'exec')
|
|
# This is ok.
|
|
compile("from None import x", "tmp", "exec")
|
|
compile("from x import None as y", "tmp", "exec")
|
|
compile("import None as x", "tmp", "exec")
|
|
|
|
def test_import(self):
|
|
succeed = [
|
|
'import sys',
|
|
'import os, sys',
|
|
'import os as bar',
|
|
'import os.path as bar',
|
|
'from __future__ import nested_scopes, generators',
|
|
'from __future__ import (nested_scopes,\ngenerators)',
|
|
'from __future__ import (nested_scopes,\ngenerators,)',
|
|
'from sys import stdin, stderr, stdout',
|
|
'from sys import (stdin, stderr,\nstdout)',
|
|
'from sys import (stdin, stderr,\nstdout,)',
|
|
'from sys import (stdin\n, stderr, stdout)',
|
|
'from sys import (stdin\n, stderr, stdout,)',
|
|
'from sys import stdin as si, stdout as so, stderr as se',
|
|
'from sys import (stdin as si, stdout as so, stderr as se)',
|
|
'from sys import (stdin as si, stdout as so, stderr as se,)',
|
|
]
|
|
fail = [
|
|
'import (os, sys)',
|
|
'import (os), (sys)',
|
|
'import ((os), (sys))',
|
|
'import (sys',
|
|
'import sys)',
|
|
'import (os,)',
|
|
'import os As bar',
|
|
'import os.path a bar',
|
|
'from sys import stdin As stdout',
|
|
'from sys import stdin a stdout',
|
|
'from (sys) import stdin',
|
|
'from __future__ import (nested_scopes',
|
|
'from __future__ import nested_scopes)',
|
|
'from __future__ import nested_scopes,\ngenerators',
|
|
'from sys import (stdin',
|
|
'from sys import stdin)',
|
|
'from sys import stdin, stdout,\nstderr',
|
|
'from sys import stdin si',
|
|
'from sys import stdin,'
|
|
'from sys import (*)',
|
|
'from sys import (stdin,, stdout, stderr)',
|
|
'from sys import (stdin, stdout),',
|
|
]
|
|
for stmt in succeed:
|
|
compile(stmt, 'tmp', 'exec')
|
|
for stmt in fail:
|
|
self.assertRaises(SyntaxError, compile, stmt, 'tmp', 'exec')
|
|
|
|
def test_for_distinct_code_objects(self):
|
|
# SF bug 1048870
|
|
def f():
|
|
f1 = lambda x=1: x
|
|
f2 = lambda x=2: x
|
|
return f1, f2
|
|
f1, f2 = f()
|
|
self.assertNotEqual(id(f1.func_code), id(f2.func_code))
|
|
|
|
def test_lambda_doc(self):
|
|
l = lambda: "foo"
|
|
self.assertIsNone(l.__doc__)
|
|
|
|
def test_unicode_encoding(self):
|
|
code = u"# -*- coding: utf-8 -*-\npass\n"
|
|
self.assertRaises(SyntaxError, compile, code, "tmp", "exec")
|
|
|
|
def test_subscripts(self):
|
|
# SF bug 1448804
|
|
# Class to make testing subscript results easy
|
|
class str_map(object):
|
|
def __init__(self):
|
|
self.data = {}
|
|
def __getitem__(self, key):
|
|
return self.data[str(key)]
|
|
def __setitem__(self, key, value):
|
|
self.data[str(key)] = value
|
|
def __delitem__(self, key):
|
|
del self.data[str(key)]
|
|
def __contains__(self, key):
|
|
return str(key) in self.data
|
|
d = str_map()
|
|
# Index
|
|
d[1] = 1
|
|
self.assertEqual(d[1], 1)
|
|
d[1] += 1
|
|
self.assertEqual(d[1], 2)
|
|
del d[1]
|
|
self.assertNotIn(1, d)
|
|
# Tuple of indices
|
|
d[1, 1] = 1
|
|
self.assertEqual(d[1, 1], 1)
|
|
d[1, 1] += 1
|
|
self.assertEqual(d[1, 1], 2)
|
|
del d[1, 1]
|
|
self.assertNotIn((1, 1), d)
|
|
# Simple slice
|
|
d[1:2] = 1
|
|
self.assertEqual(d[1:2], 1)
|
|
d[1:2] += 1
|
|
self.assertEqual(d[1:2], 2)
|
|
del d[1:2]
|
|
self.assertNotIn(slice(1, 2), d)
|
|
# Tuple of simple slices
|
|
d[1:2, 1:2] = 1
|
|
self.assertEqual(d[1:2, 1:2], 1)
|
|
d[1:2, 1:2] += 1
|
|
self.assertEqual(d[1:2, 1:2], 2)
|
|
del d[1:2, 1:2]
|
|
self.assertNotIn((slice(1, 2), slice(1, 2)), d)
|
|
# Extended slice
|
|
d[1:2:3] = 1
|
|
self.assertEqual(d[1:2:3], 1)
|
|
d[1:2:3] += 1
|
|
self.assertEqual(d[1:2:3], 2)
|
|
del d[1:2:3]
|
|
self.assertNotIn(slice(1, 2, 3), d)
|
|
# Tuple of extended slices
|
|
d[1:2:3, 1:2:3] = 1
|
|
self.assertEqual(d[1:2:3, 1:2:3], 1)
|
|
d[1:2:3, 1:2:3] += 1
|
|
self.assertEqual(d[1:2:3, 1:2:3], 2)
|
|
del d[1:2:3, 1:2:3]
|
|
self.assertNotIn((slice(1, 2, 3), slice(1, 2, 3)), d)
|
|
# Ellipsis
|
|
d[...] = 1
|
|
self.assertEqual(d[...], 1)
|
|
d[...] += 1
|
|
self.assertEqual(d[...], 2)
|
|
del d[...]
|
|
self.assertNotIn(Ellipsis, d)
|
|
# Tuple of Ellipses
|
|
d[..., ...] = 1
|
|
self.assertEqual(d[..., ...], 1)
|
|
d[..., ...] += 1
|
|
self.assertEqual(d[..., ...], 2)
|
|
del d[..., ...]
|
|
self.assertNotIn((Ellipsis, Ellipsis), d)
|
|
|
|
def test_mangling(self):
|
|
class A:
|
|
def f():
|
|
__mangled = 1
|
|
__not_mangled__ = 2
|
|
import __mangled_mod
|
|
import __package__.module
|
|
|
|
self.assertIn("_A__mangled", A.f.func_code.co_varnames)
|
|
self.assertIn("__not_mangled__", A.f.func_code.co_varnames)
|
|
self.assertIn("_A__mangled_mod", A.f.func_code.co_varnames)
|
|
self.assertIn("__package__", A.f.func_code.co_varnames)
|
|
|
|
def test_compile_ast(self):
|
|
fname = __file__
|
|
if fname.lower().endswith(('pyc', 'pyo')):
|
|
fname = fname[:-1]
|
|
with open(fname, 'r') as f:
|
|
fcontents = f.read()
|
|
sample_code = [
|
|
['<assign>', 'x = 5'],
|
|
['<print1>', 'print 1'],
|
|
['<printv>', 'print v'],
|
|
['<printTrue>', 'print True'],
|
|
['<printList>', 'print []'],
|
|
['<ifblock>', """if True:\n pass\n"""],
|
|
['<forblock>', """for n in [1, 2, 3]:\n print n\n"""],
|
|
['<deffunc>', """def foo():\n pass\nfoo()\n"""],
|
|
[fname, fcontents],
|
|
]
|
|
|
|
for fname, code in sample_code:
|
|
co1 = compile(code, '%s1' % fname, 'exec')
|
|
ast = compile(code, '%s2' % fname, 'exec', _ast.PyCF_ONLY_AST)
|
|
self.assertTrue(type(ast) == _ast.Module)
|
|
co2 = compile(ast, '%s3' % fname, 'exec')
|
|
self.assertEqual(co1, co2)
|
|
# the code object's filename comes from the second compilation step
|
|
self.assertEqual(co2.co_filename, '%s3' % fname)
|
|
|
|
# raise exception when node type doesn't match with compile mode
|
|
co1 = compile('print 1', '<string>', 'exec', _ast.PyCF_ONLY_AST)
|
|
self.assertRaises(TypeError, compile, co1, '<ast>', 'eval')
|
|
|
|
# raise exception when node type is no start node
|
|
self.assertRaises(TypeError, compile, _ast.If(), '<ast>', 'exec')
|
|
|
|
# raise exception when node has invalid children
|
|
ast = _ast.Module()
|
|
ast.body = [_ast.BoolOp()]
|
|
self.assertRaises(TypeError, compile, ast, '<ast>', 'exec')
|
|
|
|
|
|
def test_main():
|
|
test_support.run_unittest(TestSpecifics)
|
|
|
|
if __name__ == "__main__":
|
|
test_main()
|