cpython/Lib/test/test_shlex.py

199 lines
5.7 KiB
Python
Raw Normal View History

import io
import shlex
import string
import unittest
# The original test data set was from shellwords, by Hartmut Goebel.
data = r"""x|x|
foo bar|foo|bar|
foo bar|foo|bar|
foo bar |foo|bar|
foo bar bla fasel|foo|bar|bla|fasel|
x y z xxxx|x|y|z|xxxx|
\x bar|\|x|bar|
\ x bar|\|x|bar|
\ bar|\|bar|
foo \x bar|foo|\|x|bar|
foo \ x bar|foo|\|x|bar|
foo \ bar|foo|\|bar|
foo "bar" bla|foo|"bar"|bla|
"foo" "bar" "bla"|"foo"|"bar"|"bla"|
"foo" bar "bla"|"foo"|bar|"bla"|
"foo" bar bla|"foo"|bar|bla|
foo 'bar' bla|foo|'bar'|bla|
'foo' 'bar' 'bla'|'foo'|'bar'|'bla'|
'foo' bar 'bla'|'foo'|bar|'bla'|
'foo' bar bla|'foo'|bar|bla|
blurb foo"bar"bar"fasel" baz|blurb|foo"bar"bar"fasel"|baz|
blurb foo'bar'bar'fasel' baz|blurb|foo'bar'bar'fasel'|baz|
""|""|
''|''|
foo "" bar|foo|""|bar|
foo '' bar|foo|''|bar|
foo "" "" "" bar|foo|""|""|""|bar|
foo '' '' '' bar|foo|''|''|''|bar|
\""|\|""|
"\"|"\"|
"foo\ bar"|"foo\ bar"|
"foo\\ bar"|"foo\\ bar"|
"foo\\ bar\"|"foo\\ bar\"|
"foo\\" bar\""|"foo\\"|bar|\|""|
"foo\\ bar\" dfadf"|"foo\\ bar\"|dfadf"|
"foo\\\ bar\" dfadf"|"foo\\\ bar\"|dfadf"|
"foo\\\x bar\" dfadf"|"foo\\\x bar\"|dfadf"|
"foo\x bar\" dfadf"|"foo\x bar\"|dfadf"|
\''|\|''|
'foo\ bar'|'foo\ bar'|
'foo\\ bar'|'foo\\ bar'|
"foo\\\x bar\" df'a\ 'df'|"foo\\\x bar\"|df'a|\|'df'|
\"foo"|\|"foo"|
\"foo"\x|\|"foo"|\|x|
"foo\x"|"foo\x"|
"foo\ "|"foo\ "|
foo\ xx|foo|\|xx|
foo\ x\x|foo|\|x|\|x|
foo\ x\x\""|foo|\|x|\|x|\|""|
"foo\ x\x"|"foo\ x\x"|
"foo\ x\x\\"|"foo\ x\x\\"|
"foo\ x\x\\""foobar"|"foo\ x\x\\"|"foobar"|
"foo\ x\x\\"\''"foobar"|"foo\ x\x\\"|\|''|"foobar"|
"foo\ x\x\\"\'"fo'obar"|"foo\ x\x\\"|\|'"fo'|obar"|
"foo\ x\x\\"\'"fo'obar" 'don'\''t'|"foo\ x\x\\"|\|'"fo'|obar"|'don'|\|''|t'|
'foo\ bar'|'foo\ bar'|
'foo\\ bar'|'foo\\ bar'|
foo\ bar|foo|\|bar|
foo#bar\nbaz|foobaz|
:-) ;-)|:|-|)|;|-|)|
2010-10-27 15:52:48 -03:00
áéíóú|á|é|í|ó|ú|
"""
posix_data = r"""x|x|
foo bar|foo|bar|
foo bar|foo|bar|
foo bar |foo|bar|
foo bar bla fasel|foo|bar|bla|fasel|
x y z xxxx|x|y|z|xxxx|
\x bar|x|bar|
\ x bar| x|bar|
\ bar| bar|
foo \x bar|foo|x|bar|
foo \ x bar|foo| x|bar|
foo \ bar|foo| bar|
foo "bar" bla|foo|bar|bla|
"foo" "bar" "bla"|foo|bar|bla|
"foo" bar "bla"|foo|bar|bla|
"foo" bar bla|foo|bar|bla|
foo 'bar' bla|foo|bar|bla|
'foo' 'bar' 'bla'|foo|bar|bla|
'foo' bar 'bla'|foo|bar|bla|
'foo' bar bla|foo|bar|bla|
blurb foo"bar"bar"fasel" baz|blurb|foobarbarfasel|baz|
blurb foo'bar'bar'fasel' baz|blurb|foobarbarfasel|baz|
""||
''||
foo "" bar|foo||bar|
foo '' bar|foo||bar|
foo "" "" "" bar|foo||||bar|
foo '' '' '' bar|foo||||bar|
\"|"|
"\""|"|
"foo\ bar"|foo\ bar|
"foo\\ bar"|foo\ bar|
"foo\\ bar\""|foo\ bar"|
"foo\\" bar\"|foo\|bar"|
"foo\\ bar\" dfadf"|foo\ bar" dfadf|
"foo\\\ bar\" dfadf"|foo\\ bar" dfadf|
"foo\\\x bar\" dfadf"|foo\\x bar" dfadf|
"foo\x bar\" dfadf"|foo\x bar" dfadf|
\'|'|
'foo\ bar'|foo\ bar|
'foo\\ bar'|foo\\ bar|
"foo\\\x bar\" df'a\ 'df"|foo\\x bar" df'a\ 'df|
\"foo|"foo|
\"foo\x|"foox|
"foo\x"|foo\x|
"foo\ "|foo\ |
foo\ xx|foo xx|
foo\ x\x|foo xx|
foo\ x\x\"|foo xx"|
"foo\ x\x"|foo\ x\x|
"foo\ x\x\\"|foo\ x\x\|
"foo\ x\x\\""foobar"|foo\ x\x\foobar|
"foo\ x\x\\"\'"foobar"|foo\ x\x\'foobar|
"foo\ x\x\\"\'"fo'obar"|foo\ x\x\'fo'obar|
"foo\ x\x\\"\'"fo'obar" 'don'\''t'|foo\ x\x\'fo'obar|don't|
"foo\ x\x\\"\'"fo'obar" 'don'\''t' \\|foo\ x\x\'fo'obar|don't|\|
'foo\ bar'|foo\ bar|
'foo\\ bar'|foo\\ bar|
foo\ bar|foo bar|
foo#bar\nbaz|foo|baz|
:-) ;-)|:-)|;-)|
2010-10-27 15:52:48 -03:00
áéíóú|áéíóú|
"""
class ShlexTest(unittest.TestCase):
def setUp(self):
self.data = [x.split("|")[:-1]
for x in data.splitlines()]
self.posix_data = [x.split("|")[:-1]
for x in posix_data.splitlines()]
for item in self.data:
item[0] = item[0].replace(r"\n", "\n")
for item in self.posix_data:
item[0] = item[0].replace(r"\n", "\n")
def splitTest(self, data, comments):
for i in range(len(data)):
l = shlex.split(data[i][0], comments=comments)
self.assertEqual(l, data[i][1:],
"%s: %s != %s" %
(data[i][0], l, data[i][1:]))
def oldSplit(self, s):
ret = []
2007-06-12 14:43:43 -03:00
lex = shlex.shlex(io.StringIO(s))
tok = lex.get_token()
while tok:
ret.append(tok)
tok = lex.get_token()
return ret
2003-04-24 13:02:54 -03:00
def testSplitPosix(self):
"""Test data splitting with posix parser"""
2003-04-24 13:02:54 -03:00
self.splitTest(self.posix_data, comments=True)
def testCompat(self):
"""Test compatibility interface"""
for i in range(len(self.data)):
l = self.oldSplit(self.data[i][0])
self.assertEqual(l, self.data[i][1:],
"%s: %s != %s" %
(self.data[i][0], l, self.data[i][1:]))
def testQuote(self):
safeunquoted = string.ascii_letters + string.digits + '@%_-+=:,./'
unicode_sample = '\xe9\xe0\xdf' # e + acute accent, a + grave, sharp s
unsafe = '"`$\\!' + unicode_sample
self.assertEqual(shlex.quote(''), "''")
self.assertEqual(shlex.quote(safeunquoted), safeunquoted)
self.assertEqual(shlex.quote('test file name'), "'test file name'")
for u in unsafe:
self.assertEqual(shlex.quote('test%sname' % u),
"'test%sname'" % u)
for u in unsafe:
self.assertEqual(shlex.quote("test%s'name'" % u),
"'test%s'\"'\"'name'\"'\"''" % u)
# Allow this test to be used with old shlex.py
if not getattr(shlex, "split", None):
for methname in dir(ShlexTest):
if methname.startswith("test") and methname != "testCompat":
delattr(ShlexTest, methname)
if __name__ == "__main__":
unittest.main()