Show More
@@ -1,119 +1,120 b'' | |||||
1 | from __future__ import absolute_import |
|
1 | from __future__ import absolute_import | |
2 |
|
2 | |||
3 | import glob |
|
3 | import glob | |
4 | import os |
|
4 | import os | |
5 | import shutil |
|
5 | import shutil | |
6 | import tempfile |
|
6 | import tempfile | |
7 | import unittest |
|
7 | import unittest | |
8 |
|
8 | |||
9 | from mercurial import ( |
|
9 | from mercurial import ( | |
10 | util, |
|
10 | util, | |
11 | ) |
|
11 | ) | |
12 | atomictempfile = util.atomictempfile |
|
12 | atomictempfile = util.atomictempfile | |
13 |
|
13 | |||
14 | class testatomictempfile(unittest.TestCase): |
|
14 | class testatomictempfile(unittest.TestCase): | |
15 | def setUp(self): |
|
15 | def setUp(self): | |
16 | self._testdir = tempfile.mkdtemp('atomictempfiletest') |
|
16 | self._testdir = tempfile.mkdtemp('atomictempfiletest') | |
17 | self._filename = os.path.join(self._testdir, 'testfilename') |
|
17 | self._filename = os.path.join(self._testdir, 'testfilename') | |
18 |
|
18 | |||
19 | def tearDown(self): |
|
19 | def tearDown(self): | |
20 | shutil.rmtree(self._testdir, True) |
|
20 | shutil.rmtree(self._testdir, True) | |
21 |
|
21 | |||
22 | def testsimple(self): |
|
22 | def testsimple(self): | |
23 | file = atomictempfile(self._filename) |
|
23 | file = atomictempfile(self._filename) | |
24 | self.assertFalse(os.path.isfile(self._filename)) |
|
24 | self.assertFalse(os.path.isfile(self._filename)) | |
25 | tempfilename = file._tempname |
|
25 | tempfilename = file._tempname | |
26 | self.assertTrue(tempfilename in glob.glob( |
|
26 | self.assertTrue(tempfilename in glob.glob( | |
27 | os.path.join(self._testdir, '.testfilename-*'))) |
|
27 | os.path.join(self._testdir, '.testfilename-*'))) | |
28 |
|
28 | |||
29 | file.write(b'argh\n') |
|
29 | file.write(b'argh\n') | |
30 | file.close() |
|
30 | file.close() | |
31 |
|
31 | |||
32 | self.assertTrue(os.path.isfile(self._filename)) |
|
32 | self.assertTrue(os.path.isfile(self._filename)) | |
33 | self.assertTrue(tempfilename not in glob.glob( |
|
33 | self.assertTrue(tempfilename not in glob.glob( | |
34 | os.path.join(self._testdir, '.testfilename-*'))) |
|
34 | os.path.join(self._testdir, '.testfilename-*'))) | |
35 |
|
35 | |||
36 | # discard() removes the temp file without making the write permanent |
|
36 | # discard() removes the temp file without making the write permanent | |
37 | def testdiscard(self): |
|
37 | def testdiscard(self): | |
38 | file = atomictempfile(self._filename) |
|
38 | file = atomictempfile(self._filename) | |
39 | (dir, basename) = os.path.split(file._tempname) |
|
39 | (dir, basename) = os.path.split(file._tempname) | |
40 |
|
40 | |||
41 | file.write(b'yo\n') |
|
41 | file.write(b'yo\n') | |
42 | file.discard() |
|
42 | file.discard() | |
43 |
|
43 | |||
44 | self.assertFalse(os.path.isfile(self._filename)) |
|
44 | self.assertFalse(os.path.isfile(self._filename)) | |
45 | self.assertTrue(basename not in os.listdir('.')) |
|
45 | self.assertTrue(basename not in os.listdir('.')) | |
46 |
|
46 | |||
47 | # if a programmer screws up and passes bad args to atomictempfile, they |
|
47 | # if a programmer screws up and passes bad args to atomictempfile, they | |
48 | # get a plain ordinary TypeError, not infinite recursion |
|
48 | # get a plain ordinary TypeError, not infinite recursion | |
49 | def testoops(self): |
|
49 | def testoops(self): | |
50 |
self.assertRaises(TypeError |
|
50 | with self.assertRaises(TypeError): | |
|
51 | atomictempfile() | |||
51 |
|
52 | |||
52 | # checkambig=True avoids ambiguity of timestamp |
|
53 | # checkambig=True avoids ambiguity of timestamp | |
53 | def testcheckambig(self): |
|
54 | def testcheckambig(self): | |
54 | def atomicwrite(checkambig): |
|
55 | def atomicwrite(checkambig): | |
55 | f = atomictempfile(self._filename, checkambig=checkambig) |
|
56 | f = atomictempfile(self._filename, checkambig=checkambig) | |
56 | f.write('FOO') |
|
57 | f.write('FOO') | |
57 | f.close() |
|
58 | f.close() | |
58 |
|
59 | |||
59 | # try some times, because reproduction of ambiguity depends on |
|
60 | # try some times, because reproduction of ambiguity depends on | |
60 | # "filesystem time" |
|
61 | # "filesystem time" | |
61 | for i in xrange(5): |
|
62 | for i in xrange(5): | |
62 | atomicwrite(False) |
|
63 | atomicwrite(False) | |
63 | oldstat = os.stat(self._filename) |
|
64 | oldstat = os.stat(self._filename) | |
64 | if oldstat.st_ctime != oldstat.st_mtime: |
|
65 | if oldstat.st_ctime != oldstat.st_mtime: | |
65 | # subsequent changing never causes ambiguity |
|
66 | # subsequent changing never causes ambiguity | |
66 | continue |
|
67 | continue | |
67 |
|
68 | |||
68 | repetition = 3 |
|
69 | repetition = 3 | |
69 |
|
70 | |||
70 | # repeat atomic write with checkambig=True, to examine |
|
71 | # repeat atomic write with checkambig=True, to examine | |
71 | # whether st_mtime is advanced multiple times as expected |
|
72 | # whether st_mtime is advanced multiple times as expected | |
72 | for j in xrange(repetition): |
|
73 | for j in xrange(repetition): | |
73 | atomicwrite(True) |
|
74 | atomicwrite(True) | |
74 | newstat = os.stat(self._filename) |
|
75 | newstat = os.stat(self._filename) | |
75 | if oldstat.st_ctime != newstat.st_ctime: |
|
76 | if oldstat.st_ctime != newstat.st_ctime: | |
76 | # timestamp ambiguity was naturally avoided while repetition |
|
77 | # timestamp ambiguity was naturally avoided while repetition | |
77 | continue |
|
78 | continue | |
78 |
|
79 | |||
79 | # st_mtime should be advanced "repetition" times, because |
|
80 | # st_mtime should be advanced "repetition" times, because | |
80 | # all atomicwrite() occurred at same time (in sec) |
|
81 | # all atomicwrite() occurred at same time (in sec) | |
81 | self.assertTrue(newstat.st_mtime == |
|
82 | self.assertTrue(newstat.st_mtime == | |
82 | ((oldstat.st_mtime + repetition) & 0x7fffffff)) |
|
83 | ((oldstat.st_mtime + repetition) & 0x7fffffff)) | |
83 | # no more examination is needed, if assumption above is true |
|
84 | # no more examination is needed, if assumption above is true | |
84 | break |
|
85 | break | |
85 | else: |
|
86 | else: | |
86 | # This platform seems too slow to examine anti-ambiguity |
|
87 | # This platform seems too slow to examine anti-ambiguity | |
87 | # of file timestamp (or test happened to be executed at |
|
88 | # of file timestamp (or test happened to be executed at | |
88 | # bad timing). Exit silently in this case, because running |
|
89 | # bad timing). Exit silently in this case, because running | |
89 | # on other faster platforms can detect problems |
|
90 | # on other faster platforms can detect problems | |
90 | pass |
|
91 | pass | |
91 |
|
92 | |||
92 | def testread(self): |
|
93 | def testread(self): | |
93 | with open(self._filename, 'wb') as f: |
|
94 | with open(self._filename, 'wb') as f: | |
94 | f.write(b'foobar\n') |
|
95 | f.write(b'foobar\n') | |
95 | file = atomictempfile(self._filename, mode='rb') |
|
96 | file = atomictempfile(self._filename, mode='rb') | |
96 | self.assertTrue(file.read(), b'foobar\n') |
|
97 | self.assertTrue(file.read(), b'foobar\n') | |
97 | file.discard() |
|
98 | file.discard() | |
98 |
|
99 | |||
99 | def testcontextmanagersuccess(self): |
|
100 | def testcontextmanagersuccess(self): | |
100 | """When the context closes, the file is closed""" |
|
101 | """When the context closes, the file is closed""" | |
101 | with atomictempfile('foo') as f: |
|
102 | with atomictempfile('foo') as f: | |
102 | self.assertFalse(os.path.isfile('foo')) |
|
103 | self.assertFalse(os.path.isfile('foo')) | |
103 | f.write(b'argh\n') |
|
104 | f.write(b'argh\n') | |
104 | self.assertTrue(os.path.isfile('foo')) |
|
105 | self.assertTrue(os.path.isfile('foo')) | |
105 |
|
106 | |||
106 | def testcontextmanagerfailure(self): |
|
107 | def testcontextmanagerfailure(self): | |
107 | """On exception, the file is discarded""" |
|
108 | """On exception, the file is discarded""" | |
108 | try: |
|
109 | try: | |
109 | with atomictempfile('foo') as f: |
|
110 | with atomictempfile('foo') as f: | |
110 | self.assertFalse(os.path.isfile('foo')) |
|
111 | self.assertFalse(os.path.isfile('foo')) | |
111 | f.write(b'argh\n') |
|
112 | f.write(b'argh\n') | |
112 | raise ValueError |
|
113 | raise ValueError | |
113 | except ValueError: |
|
114 | except ValueError: | |
114 | pass |
|
115 | pass | |
115 | self.assertFalse(os.path.isfile('foo')) |
|
116 | self.assertFalse(os.path.isfile('foo')) | |
116 |
|
117 | |||
117 | if __name__ == '__main__': |
|
118 | if __name__ == '__main__': | |
118 | import silenttestrunner |
|
119 | import silenttestrunner | |
119 | silenttestrunner.main(__name__) |
|
120 | silenttestrunner.main(__name__) |
@@ -1,79 +1,77 b'' | |||||
1 | from __future__ import absolute_import |
|
1 | from __future__ import absolute_import | |
2 |
|
2 | |||
3 | import silenttestrunner |
|
3 | import silenttestrunner | |
4 | import unittest |
|
4 | import unittest | |
5 |
|
5 | |||
6 | from mercurial import util |
|
6 | from mercurial import util | |
7 |
|
7 | |||
8 | class contextmanager(object): |
|
8 | class contextmanager(object): | |
9 | def __init__(self, name, trace): |
|
9 | def __init__(self, name, trace): | |
10 | self.name = name |
|
10 | self.name = name | |
11 | self.entered = False |
|
11 | self.entered = False | |
12 | self.exited = False |
|
12 | self.exited = False | |
13 | self.trace = trace |
|
13 | self.trace = trace | |
14 |
|
14 | |||
15 | def __enter__(self): |
|
15 | def __enter__(self): | |
16 | self.entered = True |
|
16 | self.entered = True | |
17 | self.trace(('enter', self.name)) |
|
17 | self.trace(('enter', self.name)) | |
18 | return self |
|
18 | return self | |
19 |
|
19 | |||
20 | def __exit__(self, exc_type, exc_val, exc_tb): |
|
20 | def __exit__(self, exc_type, exc_val, exc_tb): | |
21 | self.exited = exc_type, exc_val, exc_tb |
|
21 | self.exited = exc_type, exc_val, exc_tb | |
22 | self.trace(('exit', self.name)) |
|
22 | self.trace(('exit', self.name)) | |
23 |
|
23 | |||
24 | def __repr__(self): |
|
24 | def __repr__(self): | |
25 | return '<ctx %r>' % self.name |
|
25 | return '<ctx %r>' % self.name | |
26 |
|
26 | |||
27 | class ctxerror(Exception): |
|
27 | class ctxerror(Exception): | |
28 | pass |
|
28 | pass | |
29 |
|
29 | |||
30 | class raise_on_enter(contextmanager): |
|
30 | class raise_on_enter(contextmanager): | |
31 | def __enter__(self): |
|
31 | def __enter__(self): | |
32 | self.trace(('raise', self.name)) |
|
32 | self.trace(('raise', self.name)) | |
33 | raise ctxerror(self.name) |
|
33 | raise ctxerror(self.name) | |
34 |
|
34 | |||
35 | class raise_on_exit(contextmanager): |
|
35 | class raise_on_exit(contextmanager): | |
36 | def __exit__(self, exc_type, exc_val, exc_tb): |
|
36 | def __exit__(self, exc_type, exc_val, exc_tb): | |
37 | self.trace(('raise', self.name)) |
|
37 | self.trace(('raise', self.name)) | |
38 | raise ctxerror(self.name) |
|
38 | raise ctxerror(self.name) | |
39 |
|
39 | |||
40 | def ctxmgr(name, trace): |
|
40 | def ctxmgr(name, trace): | |
41 | return lambda: contextmanager(name, trace) |
|
41 | return lambda: contextmanager(name, trace) | |
42 |
|
42 | |||
43 | class test_ctxmanager(unittest.TestCase): |
|
43 | class test_ctxmanager(unittest.TestCase): | |
44 | def test_basics(self): |
|
44 | def test_basics(self): | |
45 | trace = [] |
|
45 | trace = [] | |
46 | addtrace = trace.append |
|
46 | addtrace = trace.append | |
47 | with util.ctxmanager(ctxmgr('a', addtrace), ctxmgr('b', addtrace)) as c: |
|
47 | with util.ctxmanager(ctxmgr('a', addtrace), ctxmgr('b', addtrace)) as c: | |
48 | a, b = c.enter() |
|
48 | a, b = c.enter() | |
49 | c.atexit(addtrace, ('atexit', 'x')) |
|
49 | c.atexit(addtrace, ('atexit', 'x')) | |
50 | c.atexit(addtrace, ('atexit', 'y')) |
|
50 | c.atexit(addtrace, ('atexit', 'y')) | |
51 | self.assertEqual(trace, [('enter', 'a'), ('enter', 'b'), |
|
51 | self.assertEqual(trace, [('enter', 'a'), ('enter', 'b'), | |
52 | ('atexit', 'y'), ('atexit', 'x'), |
|
52 | ('atexit', 'y'), ('atexit', 'x'), | |
53 | ('exit', 'b'), ('exit', 'a')]) |
|
53 | ('exit', 'b'), ('exit', 'a')]) | |
54 |
|
54 | |||
55 | def test_raise_on_enter(self): |
|
55 | def test_raise_on_enter(self): | |
56 | trace = [] |
|
56 | trace = [] | |
57 | addtrace = trace.append |
|
57 | addtrace = trace.append | |
58 | def go(): |
|
58 | with self.assertRaises(ctxerror): | |
59 | with util.ctxmanager(ctxmgr('a', addtrace), |
|
59 | with util.ctxmanager(ctxmgr('a', addtrace), | |
60 | lambda: raise_on_enter('b', addtrace)) as c: |
|
60 | lambda: raise_on_enter('b', addtrace)) as c: | |
61 | c.enter() |
|
61 | c.enter() | |
62 | addtrace('unreachable') |
|
62 | addtrace('unreachable') | |
63 | self.assertRaises(ctxerror, go) |
|
|||
64 | self.assertEqual(trace, [('enter', 'a'), ('raise', 'b'), ('exit', 'a')]) |
|
63 | self.assertEqual(trace, [('enter', 'a'), ('raise', 'b'), ('exit', 'a')]) | |
65 |
|
64 | |||
66 | def test_raise_on_exit(self): |
|
65 | def test_raise_on_exit(self): | |
67 | trace = [] |
|
66 | trace = [] | |
68 | addtrace = trace.append |
|
67 | addtrace = trace.append | |
69 | def go(): |
|
68 | with self.assertRaises(ctxerror): | |
70 | with util.ctxmanager(ctxmgr('a', addtrace), |
|
69 | with util.ctxmanager(ctxmgr('a', addtrace), | |
71 | lambda: raise_on_exit('b', addtrace)) as c: |
|
70 | lambda: raise_on_exit('b', addtrace)) as c: | |
72 | c.enter() |
|
71 | c.enter() | |
73 | addtrace('running') |
|
72 | addtrace('running') | |
74 | self.assertRaises(ctxerror, go) |
|
|||
75 | self.assertEqual(trace, [('enter', 'a'), ('enter', 'b'), 'running', |
|
73 | self.assertEqual(trace, [('enter', 'a'), ('enter', 'b'), 'running', | |
76 | ('raise', 'b'), ('exit', 'a')]) |
|
74 | ('raise', 'b'), ('exit', 'a')]) | |
77 |
|
75 | |||
78 | if __name__ == '__main__': |
|
76 | if __name__ == '__main__': | |
79 | silenttestrunner.main(__name__) |
|
77 | silenttestrunner.main(__name__) |
@@ -1,298 +1,296 b'' | |||||
1 | from __future__ import absolute_import |
|
1 | from __future__ import absolute_import | |
2 |
|
2 | |||
3 | import copy |
|
3 | import copy | |
4 | import errno |
|
4 | import errno | |
5 | import os |
|
5 | import os | |
6 | import silenttestrunner |
|
6 | import silenttestrunner | |
7 | import tempfile |
|
7 | import tempfile | |
8 | import types |
|
8 | import types | |
9 | import unittest |
|
9 | import unittest | |
10 |
|
10 | |||
11 | from mercurial import ( |
|
11 | from mercurial import ( | |
12 | error, |
|
12 | error, | |
13 | lock, |
|
13 | lock, | |
14 | vfs as vfsmod, |
|
14 | vfs as vfsmod, | |
15 | ) |
|
15 | ) | |
16 |
|
16 | |||
17 | testlockname = 'testlock' |
|
17 | testlockname = 'testlock' | |
18 |
|
18 | |||
19 | # work around http://bugs.python.org/issue1515 |
|
19 | # work around http://bugs.python.org/issue1515 | |
20 | if types.MethodType not in copy._deepcopy_dispatch: |
|
20 | if types.MethodType not in copy._deepcopy_dispatch: | |
21 | def _deepcopy_method(x, memo): |
|
21 | def _deepcopy_method(x, memo): | |
22 | return type(x)(x.im_func, copy.deepcopy(x.im_self, memo), x.im_class) |
|
22 | return type(x)(x.im_func, copy.deepcopy(x.im_self, memo), x.im_class) | |
23 | copy._deepcopy_dispatch[types.MethodType] = _deepcopy_method |
|
23 | copy._deepcopy_dispatch[types.MethodType] = _deepcopy_method | |
24 |
|
24 | |||
25 | class lockwrapper(lock.lock): |
|
25 | class lockwrapper(lock.lock): | |
26 | def __init__(self, pidoffset, *args, **kwargs): |
|
26 | def __init__(self, pidoffset, *args, **kwargs): | |
27 | # lock.lock.__init__() calls lock(), so the pidoffset assignment needs |
|
27 | # lock.lock.__init__() calls lock(), so the pidoffset assignment needs | |
28 | # to be earlier |
|
28 | # to be earlier | |
29 | self._pidoffset = pidoffset |
|
29 | self._pidoffset = pidoffset | |
30 | super(lockwrapper, self).__init__(*args, **kwargs) |
|
30 | super(lockwrapper, self).__init__(*args, **kwargs) | |
31 | def _getpid(self): |
|
31 | def _getpid(self): | |
32 | return super(lockwrapper, self)._getpid() + self._pidoffset |
|
32 | return super(lockwrapper, self)._getpid() + self._pidoffset | |
33 |
|
33 | |||
34 | class teststate(object): |
|
34 | class teststate(object): | |
35 | def __init__(self, testcase, dir, pidoffset=0): |
|
35 | def __init__(self, testcase, dir, pidoffset=0): | |
36 | self._testcase = testcase |
|
36 | self._testcase = testcase | |
37 | self._acquirecalled = False |
|
37 | self._acquirecalled = False | |
38 | self._releasecalled = False |
|
38 | self._releasecalled = False | |
39 | self._postreleasecalled = False |
|
39 | self._postreleasecalled = False | |
40 | self.vfs = vfsmod.vfs(dir, audit=False) |
|
40 | self.vfs = vfsmod.vfs(dir, audit=False) | |
41 | self._pidoffset = pidoffset |
|
41 | self._pidoffset = pidoffset | |
42 |
|
42 | |||
43 | def makelock(self, *args, **kwargs): |
|
43 | def makelock(self, *args, **kwargs): | |
44 | l = lockwrapper(self._pidoffset, self.vfs, testlockname, |
|
44 | l = lockwrapper(self._pidoffset, self.vfs, testlockname, | |
45 | releasefn=self.releasefn, acquirefn=self.acquirefn, |
|
45 | releasefn=self.releasefn, acquirefn=self.acquirefn, | |
46 | *args, **kwargs) |
|
46 | *args, **kwargs) | |
47 | l.postrelease.append(self.postreleasefn) |
|
47 | l.postrelease.append(self.postreleasefn) | |
48 | return l |
|
48 | return l | |
49 |
|
49 | |||
50 | def acquirefn(self): |
|
50 | def acquirefn(self): | |
51 | self._acquirecalled = True |
|
51 | self._acquirecalled = True | |
52 |
|
52 | |||
53 | def releasefn(self): |
|
53 | def releasefn(self): | |
54 | self._releasecalled = True |
|
54 | self._releasecalled = True | |
55 |
|
55 | |||
56 | def postreleasefn(self): |
|
56 | def postreleasefn(self): | |
57 | self._postreleasecalled = True |
|
57 | self._postreleasecalled = True | |
58 |
|
58 | |||
59 | def assertacquirecalled(self, called): |
|
59 | def assertacquirecalled(self, called): | |
60 | self._testcase.assertEqual( |
|
60 | self._testcase.assertEqual( | |
61 | self._acquirecalled, called, |
|
61 | self._acquirecalled, called, | |
62 | 'expected acquire to be %s but was actually %s' % ( |
|
62 | 'expected acquire to be %s but was actually %s' % ( | |
63 | self._tocalled(called), |
|
63 | self._tocalled(called), | |
64 | self._tocalled(self._acquirecalled), |
|
64 | self._tocalled(self._acquirecalled), | |
65 | )) |
|
65 | )) | |
66 |
|
66 | |||
67 | def resetacquirefn(self): |
|
67 | def resetacquirefn(self): | |
68 | self._acquirecalled = False |
|
68 | self._acquirecalled = False | |
69 |
|
69 | |||
70 | def assertreleasecalled(self, called): |
|
70 | def assertreleasecalled(self, called): | |
71 | self._testcase.assertEqual( |
|
71 | self._testcase.assertEqual( | |
72 | self._releasecalled, called, |
|
72 | self._releasecalled, called, | |
73 | 'expected release to be %s but was actually %s' % ( |
|
73 | 'expected release to be %s but was actually %s' % ( | |
74 | self._tocalled(called), |
|
74 | self._tocalled(called), | |
75 | self._tocalled(self._releasecalled), |
|
75 | self._tocalled(self._releasecalled), | |
76 | )) |
|
76 | )) | |
77 |
|
77 | |||
78 | def assertpostreleasecalled(self, called): |
|
78 | def assertpostreleasecalled(self, called): | |
79 | self._testcase.assertEqual( |
|
79 | self._testcase.assertEqual( | |
80 | self._postreleasecalled, called, |
|
80 | self._postreleasecalled, called, | |
81 | 'expected postrelease to be %s but was actually %s' % ( |
|
81 | 'expected postrelease to be %s but was actually %s' % ( | |
82 | self._tocalled(called), |
|
82 | self._tocalled(called), | |
83 | self._tocalled(self._postreleasecalled), |
|
83 | self._tocalled(self._postreleasecalled), | |
84 | )) |
|
84 | )) | |
85 |
|
85 | |||
86 | def assertlockexists(self, exists): |
|
86 | def assertlockexists(self, exists): | |
87 | actual = self.vfs.lexists(testlockname) |
|
87 | actual = self.vfs.lexists(testlockname) | |
88 | self._testcase.assertEqual( |
|
88 | self._testcase.assertEqual( | |
89 | actual, exists, |
|
89 | actual, exists, | |
90 | 'expected lock to %s but actually did %s' % ( |
|
90 | 'expected lock to %s but actually did %s' % ( | |
91 | self._toexists(exists), |
|
91 | self._toexists(exists), | |
92 | self._toexists(actual), |
|
92 | self._toexists(actual), | |
93 | )) |
|
93 | )) | |
94 |
|
94 | |||
95 | def _tocalled(self, called): |
|
95 | def _tocalled(self, called): | |
96 | if called: |
|
96 | if called: | |
97 | return 'called' |
|
97 | return 'called' | |
98 | else: |
|
98 | else: | |
99 | return 'not called' |
|
99 | return 'not called' | |
100 |
|
100 | |||
101 | def _toexists(self, exists): |
|
101 | def _toexists(self, exists): | |
102 | if exists: |
|
102 | if exists: | |
103 | return 'exist' |
|
103 | return 'exist' | |
104 | else: |
|
104 | else: | |
105 | return 'not exist' |
|
105 | return 'not exist' | |
106 |
|
106 | |||
107 | class testlock(unittest.TestCase): |
|
107 | class testlock(unittest.TestCase): | |
108 | def testlock(self): |
|
108 | def testlock(self): | |
109 | state = teststate(self, tempfile.mkdtemp(dir=os.getcwd())) |
|
109 | state = teststate(self, tempfile.mkdtemp(dir=os.getcwd())) | |
110 | lock = state.makelock() |
|
110 | lock = state.makelock() | |
111 | state.assertacquirecalled(True) |
|
111 | state.assertacquirecalled(True) | |
112 | lock.release() |
|
112 | lock.release() | |
113 | state.assertreleasecalled(True) |
|
113 | state.assertreleasecalled(True) | |
114 | state.assertpostreleasecalled(True) |
|
114 | state.assertpostreleasecalled(True) | |
115 | state.assertlockexists(False) |
|
115 | state.assertlockexists(False) | |
116 |
|
116 | |||
117 | def testrecursivelock(self): |
|
117 | def testrecursivelock(self): | |
118 | state = teststate(self, tempfile.mkdtemp(dir=os.getcwd())) |
|
118 | state = teststate(self, tempfile.mkdtemp(dir=os.getcwd())) | |
119 | lock = state.makelock() |
|
119 | lock = state.makelock() | |
120 | state.assertacquirecalled(True) |
|
120 | state.assertacquirecalled(True) | |
121 |
|
121 | |||
122 | state.resetacquirefn() |
|
122 | state.resetacquirefn() | |
123 | lock.lock() |
|
123 | lock.lock() | |
124 | # recursive lock should not call acquirefn again |
|
124 | # recursive lock should not call acquirefn again | |
125 | state.assertacquirecalled(False) |
|
125 | state.assertacquirecalled(False) | |
126 |
|
126 | |||
127 | lock.release() # brings lock refcount down from 2 to 1 |
|
127 | lock.release() # brings lock refcount down from 2 to 1 | |
128 | state.assertreleasecalled(False) |
|
128 | state.assertreleasecalled(False) | |
129 | state.assertpostreleasecalled(False) |
|
129 | state.assertpostreleasecalled(False) | |
130 | state.assertlockexists(True) |
|
130 | state.assertlockexists(True) | |
131 |
|
131 | |||
132 | lock.release() # releases the lock |
|
132 | lock.release() # releases the lock | |
133 | state.assertreleasecalled(True) |
|
133 | state.assertreleasecalled(True) | |
134 | state.assertpostreleasecalled(True) |
|
134 | state.assertpostreleasecalled(True) | |
135 | state.assertlockexists(False) |
|
135 | state.assertlockexists(False) | |
136 |
|
136 | |||
137 | def testlockfork(self): |
|
137 | def testlockfork(self): | |
138 | state = teststate(self, tempfile.mkdtemp(dir=os.getcwd())) |
|
138 | state = teststate(self, tempfile.mkdtemp(dir=os.getcwd())) | |
139 | lock = state.makelock() |
|
139 | lock = state.makelock() | |
140 | state.assertacquirecalled(True) |
|
140 | state.assertacquirecalled(True) | |
141 |
|
141 | |||
142 | # fake a fork |
|
142 | # fake a fork | |
143 | forklock = copy.deepcopy(lock) |
|
143 | forklock = copy.deepcopy(lock) | |
144 | forklock._pidoffset = 1 |
|
144 | forklock._pidoffset = 1 | |
145 | forklock.release() |
|
145 | forklock.release() | |
146 | state.assertreleasecalled(False) |
|
146 | state.assertreleasecalled(False) | |
147 | state.assertpostreleasecalled(False) |
|
147 | state.assertpostreleasecalled(False) | |
148 | state.assertlockexists(True) |
|
148 | state.assertlockexists(True) | |
149 |
|
149 | |||
150 | # release the actual lock |
|
150 | # release the actual lock | |
151 | lock.release() |
|
151 | lock.release() | |
152 | state.assertreleasecalled(True) |
|
152 | state.assertreleasecalled(True) | |
153 | state.assertpostreleasecalled(True) |
|
153 | state.assertpostreleasecalled(True) | |
154 | state.assertlockexists(False) |
|
154 | state.assertlockexists(False) | |
155 |
|
155 | |||
156 | def testinheritlock(self): |
|
156 | def testinheritlock(self): | |
157 | d = tempfile.mkdtemp(dir=os.getcwd()) |
|
157 | d = tempfile.mkdtemp(dir=os.getcwd()) | |
158 | parentstate = teststate(self, d) |
|
158 | parentstate = teststate(self, d) | |
159 | parentlock = parentstate.makelock() |
|
159 | parentlock = parentstate.makelock() | |
160 | parentstate.assertacquirecalled(True) |
|
160 | parentstate.assertacquirecalled(True) | |
161 |
|
161 | |||
162 | # set up lock inheritance |
|
162 | # set up lock inheritance | |
163 | with parentlock.inherit() as lockname: |
|
163 | with parentlock.inherit() as lockname: | |
164 | parentstate.assertreleasecalled(True) |
|
164 | parentstate.assertreleasecalled(True) | |
165 | parentstate.assertpostreleasecalled(False) |
|
165 | parentstate.assertpostreleasecalled(False) | |
166 | parentstate.assertlockexists(True) |
|
166 | parentstate.assertlockexists(True) | |
167 |
|
167 | |||
168 | childstate = teststate(self, d, pidoffset=1) |
|
168 | childstate = teststate(self, d, pidoffset=1) | |
169 | childlock = childstate.makelock(parentlock=lockname) |
|
169 | childlock = childstate.makelock(parentlock=lockname) | |
170 | childstate.assertacquirecalled(True) |
|
170 | childstate.assertacquirecalled(True) | |
171 |
|
171 | |||
172 | childlock.release() |
|
172 | childlock.release() | |
173 | childstate.assertreleasecalled(True) |
|
173 | childstate.assertreleasecalled(True) | |
174 | childstate.assertpostreleasecalled(False) |
|
174 | childstate.assertpostreleasecalled(False) | |
175 | childstate.assertlockexists(True) |
|
175 | childstate.assertlockexists(True) | |
176 |
|
176 | |||
177 | parentstate.resetacquirefn() |
|
177 | parentstate.resetacquirefn() | |
178 |
|
178 | |||
179 | parentstate.assertacquirecalled(True) |
|
179 | parentstate.assertacquirecalled(True) | |
180 |
|
180 | |||
181 | parentlock.release() |
|
181 | parentlock.release() | |
182 | parentstate.assertreleasecalled(True) |
|
182 | parentstate.assertreleasecalled(True) | |
183 | parentstate.assertpostreleasecalled(True) |
|
183 | parentstate.assertpostreleasecalled(True) | |
184 | parentstate.assertlockexists(False) |
|
184 | parentstate.assertlockexists(False) | |
185 |
|
185 | |||
186 | def testmultilock(self): |
|
186 | def testmultilock(self): | |
187 | d = tempfile.mkdtemp(dir=os.getcwd()) |
|
187 | d = tempfile.mkdtemp(dir=os.getcwd()) | |
188 | state0 = teststate(self, d) |
|
188 | state0 = teststate(self, d) | |
189 | lock0 = state0.makelock() |
|
189 | lock0 = state0.makelock() | |
190 | state0.assertacquirecalled(True) |
|
190 | state0.assertacquirecalled(True) | |
191 |
|
191 | |||
192 | with lock0.inherit() as lock0name: |
|
192 | with lock0.inherit() as lock0name: | |
193 | state0.assertreleasecalled(True) |
|
193 | state0.assertreleasecalled(True) | |
194 | state0.assertpostreleasecalled(False) |
|
194 | state0.assertpostreleasecalled(False) | |
195 | state0.assertlockexists(True) |
|
195 | state0.assertlockexists(True) | |
196 |
|
196 | |||
197 | state1 = teststate(self, d, pidoffset=1) |
|
197 | state1 = teststate(self, d, pidoffset=1) | |
198 | lock1 = state1.makelock(parentlock=lock0name) |
|
198 | lock1 = state1.makelock(parentlock=lock0name) | |
199 | state1.assertacquirecalled(True) |
|
199 | state1.assertacquirecalled(True) | |
200 |
|
200 | |||
201 | # from within lock1, acquire another lock |
|
201 | # from within lock1, acquire another lock | |
202 | with lock1.inherit() as lock1name: |
|
202 | with lock1.inherit() as lock1name: | |
203 | # since the file on disk is lock0's this should have the same |
|
203 | # since the file on disk is lock0's this should have the same | |
204 | # name |
|
204 | # name | |
205 | self.assertEqual(lock0name, lock1name) |
|
205 | self.assertEqual(lock0name, lock1name) | |
206 |
|
206 | |||
207 | state2 = teststate(self, d, pidoffset=2) |
|
207 | state2 = teststate(self, d, pidoffset=2) | |
208 | lock2 = state2.makelock(parentlock=lock1name) |
|
208 | lock2 = state2.makelock(parentlock=lock1name) | |
209 | state2.assertacquirecalled(True) |
|
209 | state2.assertacquirecalled(True) | |
210 |
|
210 | |||
211 | lock2.release() |
|
211 | lock2.release() | |
212 | state2.assertreleasecalled(True) |
|
212 | state2.assertreleasecalled(True) | |
213 | state2.assertpostreleasecalled(False) |
|
213 | state2.assertpostreleasecalled(False) | |
214 | state2.assertlockexists(True) |
|
214 | state2.assertlockexists(True) | |
215 |
|
215 | |||
216 | state1.resetacquirefn() |
|
216 | state1.resetacquirefn() | |
217 |
|
217 | |||
218 | state1.assertacquirecalled(True) |
|
218 | state1.assertacquirecalled(True) | |
219 |
|
219 | |||
220 | lock1.release() |
|
220 | lock1.release() | |
221 | state1.assertreleasecalled(True) |
|
221 | state1.assertreleasecalled(True) | |
222 | state1.assertpostreleasecalled(False) |
|
222 | state1.assertpostreleasecalled(False) | |
223 | state1.assertlockexists(True) |
|
223 | state1.assertlockexists(True) | |
224 |
|
224 | |||
225 | lock0.release() |
|
225 | lock0.release() | |
226 |
|
226 | |||
227 | def testinheritlockfork(self): |
|
227 | def testinheritlockfork(self): | |
228 | d = tempfile.mkdtemp(dir=os.getcwd()) |
|
228 | d = tempfile.mkdtemp(dir=os.getcwd()) | |
229 | parentstate = teststate(self, d) |
|
229 | parentstate = teststate(self, d) | |
230 | parentlock = parentstate.makelock() |
|
230 | parentlock = parentstate.makelock() | |
231 | parentstate.assertacquirecalled(True) |
|
231 | parentstate.assertacquirecalled(True) | |
232 |
|
232 | |||
233 | # set up lock inheritance |
|
233 | # set up lock inheritance | |
234 | with parentlock.inherit() as lockname: |
|
234 | with parentlock.inherit() as lockname: | |
235 | childstate = teststate(self, d, pidoffset=1) |
|
235 | childstate = teststate(self, d, pidoffset=1) | |
236 | childlock = childstate.makelock(parentlock=lockname) |
|
236 | childlock = childstate.makelock(parentlock=lockname) | |
237 | childstate.assertacquirecalled(True) |
|
237 | childstate.assertacquirecalled(True) | |
238 |
|
238 | |||
239 | # fork the child lock |
|
239 | # fork the child lock | |
240 | forkchildlock = copy.deepcopy(childlock) |
|
240 | forkchildlock = copy.deepcopy(childlock) | |
241 | forkchildlock._pidoffset += 1 |
|
241 | forkchildlock._pidoffset += 1 | |
242 | forkchildlock.release() |
|
242 | forkchildlock.release() | |
243 | childstate.assertreleasecalled(False) |
|
243 | childstate.assertreleasecalled(False) | |
244 | childstate.assertpostreleasecalled(False) |
|
244 | childstate.assertpostreleasecalled(False) | |
245 | childstate.assertlockexists(True) |
|
245 | childstate.assertlockexists(True) | |
246 |
|
246 | |||
247 | # release the child lock |
|
247 | # release the child lock | |
248 | childlock.release() |
|
248 | childlock.release() | |
249 | childstate.assertreleasecalled(True) |
|
249 | childstate.assertreleasecalled(True) | |
250 | childstate.assertpostreleasecalled(False) |
|
250 | childstate.assertpostreleasecalled(False) | |
251 | childstate.assertlockexists(True) |
|
251 | childstate.assertlockexists(True) | |
252 |
|
252 | |||
253 | parentlock.release() |
|
253 | parentlock.release() | |
254 |
|
254 | |||
255 | def testinheritcheck(self): |
|
255 | def testinheritcheck(self): | |
256 | d = tempfile.mkdtemp(dir=os.getcwd()) |
|
256 | d = tempfile.mkdtemp(dir=os.getcwd()) | |
257 | state = teststate(self, d) |
|
257 | state = teststate(self, d) | |
258 | def check(): |
|
258 | def check(): | |
259 | raise error.LockInheritanceContractViolation('check failed') |
|
259 | raise error.LockInheritanceContractViolation('check failed') | |
260 | lock = state.makelock(inheritchecker=check) |
|
260 | lock = state.makelock(inheritchecker=check) | |
261 | state.assertacquirecalled(True) |
|
261 | state.assertacquirecalled(True) | |
262 |
|
262 | |||
263 | def tryinherit(): |
|
263 | with self.assertRaises(error.LockInheritanceContractViolation): | |
264 | with lock.inherit(): |
|
264 | with lock.inherit(): | |
265 | pass |
|
265 | pass | |
266 |
|
266 | |||
267 | self.assertRaises(error.LockInheritanceContractViolation, tryinherit) |
|
|||
268 |
|
||||
269 | lock.release() |
|
267 | lock.release() | |
270 |
|
268 | |||
271 | def testfrequentlockunlock(self): |
|
269 | def testfrequentlockunlock(self): | |
272 | """This tests whether lock acquisition fails as expected, even if |
|
270 | """This tests whether lock acquisition fails as expected, even if | |
273 | (1) lock can't be acquired (makelock fails by EEXIST), and |
|
271 | (1) lock can't be acquired (makelock fails by EEXIST), and | |
274 | (2) locker info can't be read in (readlock fails by ENOENT) while |
|
272 | (2) locker info can't be read in (readlock fails by ENOENT) while | |
275 | retrying 5 times. |
|
273 | retrying 5 times. | |
276 | """ |
|
274 | """ | |
277 |
|
275 | |||
278 | d = tempfile.mkdtemp(dir=os.getcwd()) |
|
276 | d = tempfile.mkdtemp(dir=os.getcwd()) | |
279 | state = teststate(self, d) |
|
277 | state = teststate(self, d) | |
280 |
|
278 | |||
281 | def emulatefrequentlock(*args): |
|
279 | def emulatefrequentlock(*args): | |
282 | raise OSError(errno.EEXIST, "File exists") |
|
280 | raise OSError(errno.EEXIST, "File exists") | |
283 | def emulatefrequentunlock(*args): |
|
281 | def emulatefrequentunlock(*args): | |
284 | raise OSError(errno.ENOENT, "No such file or directory") |
|
282 | raise OSError(errno.ENOENT, "No such file or directory") | |
285 |
|
283 | |||
286 | state.vfs.makelock = emulatefrequentlock |
|
284 | state.vfs.makelock = emulatefrequentlock | |
287 | state.vfs.readlock = emulatefrequentunlock |
|
285 | state.vfs.readlock = emulatefrequentunlock | |
288 |
|
286 | |||
289 | try: |
|
287 | try: | |
290 | state.makelock(timeout=0) |
|
288 | state.makelock(timeout=0) | |
291 | self.fail("unexpected lock acquisition") |
|
289 | self.fail("unexpected lock acquisition") | |
292 | except error.LockHeld as why: |
|
290 | except error.LockHeld as why: | |
293 | self.assertTrue(why.errno == errno.ETIMEDOUT) |
|
291 | self.assertTrue(why.errno == errno.ETIMEDOUT) | |
294 | self.assertTrue(why.locker == "") |
|
292 | self.assertTrue(why.locker == "") | |
295 | state.assertlockexists(False) |
|
293 | state.assertlockexists(False) | |
296 |
|
294 | |||
297 | if __name__ == '__main__': |
|
295 | if __name__ == '__main__': | |
298 | silenttestrunner.main(__name__) |
|
296 | silenttestrunner.main(__name__) |
@@ -1,487 +1,491 b'' | |||||
1 | from __future__ import absolute_import |
|
1 | from __future__ import absolute_import | |
2 |
|
2 | |||
3 | import binascii |
|
3 | import binascii | |
4 | import itertools |
|
4 | import itertools | |
5 | import silenttestrunner |
|
5 | import silenttestrunner | |
6 | import unittest |
|
6 | import unittest | |
7 |
|
7 | |||
8 | from mercurial import ( |
|
8 | from mercurial import ( | |
9 | manifest as manifestmod, |
|
9 | manifest as manifestmod, | |
10 | match as matchmod, |
|
10 | match as matchmod, | |
11 | ) |
|
11 | ) | |
12 |
|
12 | |||
13 | EMTPY_MANIFEST = '' |
|
13 | EMTPY_MANIFEST = '' | |
14 | EMTPY_MANIFEST_V2 = '\0\n' |
|
14 | EMTPY_MANIFEST_V2 = '\0\n' | |
15 |
|
15 | |||
16 | HASH_1 = '1' * 40 |
|
16 | HASH_1 = '1' * 40 | |
17 | BIN_HASH_1 = binascii.unhexlify(HASH_1) |
|
17 | BIN_HASH_1 = binascii.unhexlify(HASH_1) | |
18 | HASH_2 = 'f' * 40 |
|
18 | HASH_2 = 'f' * 40 | |
19 | BIN_HASH_2 = binascii.unhexlify(HASH_2) |
|
19 | BIN_HASH_2 = binascii.unhexlify(HASH_2) | |
20 | HASH_3 = '1234567890abcdef0987654321deadbeef0fcafe' |
|
20 | HASH_3 = '1234567890abcdef0987654321deadbeef0fcafe' | |
21 | BIN_HASH_3 = binascii.unhexlify(HASH_3) |
|
21 | BIN_HASH_3 = binascii.unhexlify(HASH_3) | |
22 | A_SHORT_MANIFEST = ( |
|
22 | A_SHORT_MANIFEST = ( | |
23 | 'bar/baz/qux.py\0%(hash2)s%(flag2)s\n' |
|
23 | 'bar/baz/qux.py\0%(hash2)s%(flag2)s\n' | |
24 | 'foo\0%(hash1)s%(flag1)s\n' |
|
24 | 'foo\0%(hash1)s%(flag1)s\n' | |
25 | ) % {'hash1': HASH_1, |
|
25 | ) % {'hash1': HASH_1, | |
26 | 'flag1': '', |
|
26 | 'flag1': '', | |
27 | 'hash2': HASH_2, |
|
27 | 'hash2': HASH_2, | |
28 | 'flag2': 'l', |
|
28 | 'flag2': 'l', | |
29 | } |
|
29 | } | |
30 |
|
30 | |||
31 | # Same data as A_SHORT_MANIFEST |
|
31 | # Same data as A_SHORT_MANIFEST | |
32 | A_SHORT_MANIFEST_V2 = ( |
|
32 | A_SHORT_MANIFEST_V2 = ( | |
33 | '\0\n' |
|
33 | '\0\n' | |
34 | '\x00bar/baz/qux.py\0%(flag2)s\n%(hash2)s\n' |
|
34 | '\x00bar/baz/qux.py\0%(flag2)s\n%(hash2)s\n' | |
35 | '\x00foo\0%(flag1)s\n%(hash1)s\n' |
|
35 | '\x00foo\0%(flag1)s\n%(hash1)s\n' | |
36 | ) % {'hash1': BIN_HASH_1, |
|
36 | ) % {'hash1': BIN_HASH_1, | |
37 | 'flag1': '', |
|
37 | 'flag1': '', | |
38 | 'hash2': BIN_HASH_2, |
|
38 | 'hash2': BIN_HASH_2, | |
39 | 'flag2': 'l', |
|
39 | 'flag2': 'l', | |
40 | } |
|
40 | } | |
41 |
|
41 | |||
42 | # Same data as A_SHORT_MANIFEST |
|
42 | # Same data as A_SHORT_MANIFEST | |
43 | A_METADATA_MANIFEST = ( |
|
43 | A_METADATA_MANIFEST = ( | |
44 | '\0foo\0bar\n' |
|
44 | '\0foo\0bar\n' | |
45 | '\x00bar/baz/qux.py\0%(flag2)s\0foo\0bar\n%(hash2)s\n' # flag and metadata |
|
45 | '\x00bar/baz/qux.py\0%(flag2)s\0foo\0bar\n%(hash2)s\n' # flag and metadata | |
46 | '\x00foo\0%(flag1)s\0foo\n%(hash1)s\n' # no flag, but metadata |
|
46 | '\x00foo\0%(flag1)s\0foo\n%(hash1)s\n' # no flag, but metadata | |
47 | ) % {'hash1': BIN_HASH_1, |
|
47 | ) % {'hash1': BIN_HASH_1, | |
48 | 'flag1': '', |
|
48 | 'flag1': '', | |
49 | 'hash2': BIN_HASH_2, |
|
49 | 'hash2': BIN_HASH_2, | |
50 | 'flag2': 'l', |
|
50 | 'flag2': 'l', | |
51 | } |
|
51 | } | |
52 |
|
52 | |||
53 | A_STEM_COMPRESSED_MANIFEST = ( |
|
53 | A_STEM_COMPRESSED_MANIFEST = ( | |
54 | '\0\n' |
|
54 | '\0\n' | |
55 | '\x00bar/baz/qux.py\0%(flag2)s\n%(hash2)s\n' |
|
55 | '\x00bar/baz/qux.py\0%(flag2)s\n%(hash2)s\n' | |
56 | '\x04qux/foo.py\0%(flag1)s\n%(hash1)s\n' # simple case of 4 stem chars |
|
56 | '\x04qux/foo.py\0%(flag1)s\n%(hash1)s\n' # simple case of 4 stem chars | |
57 | '\x0az.py\0%(flag1)s\n%(hash1)s\n' # tricky newline = 10 stem characters |
|
57 | '\x0az.py\0%(flag1)s\n%(hash1)s\n' # tricky newline = 10 stem characters | |
58 | '\x00%(verylongdir)sx/x\0\n%(hash1)s\n' |
|
58 | '\x00%(verylongdir)sx/x\0\n%(hash1)s\n' | |
59 | '\xffx/y\0\n%(hash2)s\n' # more than 255 stem chars |
|
59 | '\xffx/y\0\n%(hash2)s\n' # more than 255 stem chars | |
60 | ) % {'hash1': BIN_HASH_1, |
|
60 | ) % {'hash1': BIN_HASH_1, | |
61 | 'flag1': '', |
|
61 | 'flag1': '', | |
62 | 'hash2': BIN_HASH_2, |
|
62 | 'hash2': BIN_HASH_2, | |
63 | 'flag2': 'l', |
|
63 | 'flag2': 'l', | |
64 | 'verylongdir': 255 * 'x', |
|
64 | 'verylongdir': 255 * 'x', | |
65 | } |
|
65 | } | |
66 |
|
66 | |||
67 | A_DEEPER_MANIFEST = ( |
|
67 | A_DEEPER_MANIFEST = ( | |
68 | 'a/b/c/bar.py\0%(hash3)s%(flag1)s\n' |
|
68 | 'a/b/c/bar.py\0%(hash3)s%(flag1)s\n' | |
69 | 'a/b/c/bar.txt\0%(hash1)s%(flag1)s\n' |
|
69 | 'a/b/c/bar.txt\0%(hash1)s%(flag1)s\n' | |
70 | 'a/b/c/foo.py\0%(hash3)s%(flag1)s\n' |
|
70 | 'a/b/c/foo.py\0%(hash3)s%(flag1)s\n' | |
71 | 'a/b/c/foo.txt\0%(hash2)s%(flag2)s\n' |
|
71 | 'a/b/c/foo.txt\0%(hash2)s%(flag2)s\n' | |
72 | 'a/b/d/baz.py\0%(hash3)s%(flag1)s\n' |
|
72 | 'a/b/d/baz.py\0%(hash3)s%(flag1)s\n' | |
73 | 'a/b/d/qux.py\0%(hash1)s%(flag2)s\n' |
|
73 | 'a/b/d/qux.py\0%(hash1)s%(flag2)s\n' | |
74 | 'a/b/d/ten.txt\0%(hash3)s%(flag2)s\n' |
|
74 | 'a/b/d/ten.txt\0%(hash3)s%(flag2)s\n' | |
75 | 'a/b/dog.py\0%(hash3)s%(flag1)s\n' |
|
75 | 'a/b/dog.py\0%(hash3)s%(flag1)s\n' | |
76 | 'a/b/fish.py\0%(hash2)s%(flag1)s\n' |
|
76 | 'a/b/fish.py\0%(hash2)s%(flag1)s\n' | |
77 | 'a/c/london.py\0%(hash3)s%(flag2)s\n' |
|
77 | 'a/c/london.py\0%(hash3)s%(flag2)s\n' | |
78 | 'a/c/paper.txt\0%(hash2)s%(flag2)s\n' |
|
78 | 'a/c/paper.txt\0%(hash2)s%(flag2)s\n' | |
79 | 'a/c/paris.py\0%(hash2)s%(flag1)s\n' |
|
79 | 'a/c/paris.py\0%(hash2)s%(flag1)s\n' | |
80 | 'a/d/apple.py\0%(hash3)s%(flag1)s\n' |
|
80 | 'a/d/apple.py\0%(hash3)s%(flag1)s\n' | |
81 | 'a/d/pizza.py\0%(hash3)s%(flag2)s\n' |
|
81 | 'a/d/pizza.py\0%(hash3)s%(flag2)s\n' | |
82 | 'a/green.py\0%(hash1)s%(flag2)s\n' |
|
82 | 'a/green.py\0%(hash1)s%(flag2)s\n' | |
83 | 'a/purple.py\0%(hash2)s%(flag1)s\n' |
|
83 | 'a/purple.py\0%(hash2)s%(flag1)s\n' | |
84 | 'app.py\0%(hash3)s%(flag1)s\n' |
|
84 | 'app.py\0%(hash3)s%(flag1)s\n' | |
85 | 'readme.txt\0%(hash2)s%(flag1)s\n' |
|
85 | 'readme.txt\0%(hash2)s%(flag1)s\n' | |
86 | ) % {'hash1': HASH_1, |
|
86 | ) % {'hash1': HASH_1, | |
87 | 'flag1': '', |
|
87 | 'flag1': '', | |
88 | 'hash2': HASH_2, |
|
88 | 'hash2': HASH_2, | |
89 | 'flag2': 'l', |
|
89 | 'flag2': 'l', | |
90 | 'hash3': HASH_3, |
|
90 | 'hash3': HASH_3, | |
91 | } |
|
91 | } | |
92 |
|
92 | |||
93 | HUGE_MANIFEST_ENTRIES = 200001 |
|
93 | HUGE_MANIFEST_ENTRIES = 200001 | |
94 |
|
94 | |||
95 | A_HUGE_MANIFEST = ''.join(sorted( |
|
95 | A_HUGE_MANIFEST = ''.join(sorted( | |
96 | 'file%d\0%s%s\n' % (i, h, f) for i, h, f in |
|
96 | 'file%d\0%s%s\n' % (i, h, f) for i, h, f in | |
97 | itertools.izip(xrange(200001), |
|
97 | itertools.izip(xrange(200001), | |
98 | itertools.cycle((HASH_1, HASH_2)), |
|
98 | itertools.cycle((HASH_1, HASH_2)), | |
99 | itertools.cycle(('', 'x', 'l'))))) |
|
99 | itertools.cycle(('', 'x', 'l'))))) | |
100 |
|
100 | |||
101 | class basemanifesttests(object): |
|
101 | class basemanifesttests(object): | |
102 | def parsemanifest(self, text): |
|
102 | def parsemanifest(self, text): | |
103 | raise NotImplementedError('parsemanifest not implemented by test case') |
|
103 | raise NotImplementedError('parsemanifest not implemented by test case') | |
104 |
|
104 | |||
105 | def assertIn(self, thing, container, msg=None): |
|
105 | def assertIn(self, thing, container, msg=None): | |
106 | # assertIn new in 2.7, use it if available, otherwise polyfill |
|
106 | # assertIn new in 2.7, use it if available, otherwise polyfill | |
107 | sup = getattr(unittest.TestCase, 'assertIn', False) |
|
107 | sup = getattr(unittest.TestCase, 'assertIn', False) | |
108 | if sup: |
|
108 | if sup: | |
109 | return sup(self, thing, container, msg=msg) |
|
109 | return sup(self, thing, container, msg=msg) | |
110 | if not msg: |
|
110 | if not msg: | |
111 | msg = 'Expected %r in %r' % (thing, container) |
|
111 | msg = 'Expected %r in %r' % (thing, container) | |
112 | self.assert_(thing in container, msg) |
|
112 | self.assert_(thing in container, msg) | |
113 |
|
113 | |||
114 | def testEmptyManifest(self): |
|
114 | def testEmptyManifest(self): | |
115 | m = self.parsemanifest(EMTPY_MANIFEST) |
|
115 | m = self.parsemanifest(EMTPY_MANIFEST) | |
116 | self.assertEqual(0, len(m)) |
|
116 | self.assertEqual(0, len(m)) | |
117 | self.assertEqual([], list(m)) |
|
117 | self.assertEqual([], list(m)) | |
118 |
|
118 | |||
119 | def testEmptyManifestv2(self): |
|
119 | def testEmptyManifestv2(self): | |
120 | m = self.parsemanifest(EMTPY_MANIFEST_V2) |
|
120 | m = self.parsemanifest(EMTPY_MANIFEST_V2) | |
121 | self.assertEqual(0, len(m)) |
|
121 | self.assertEqual(0, len(m)) | |
122 | self.assertEqual([], list(m)) |
|
122 | self.assertEqual([], list(m)) | |
123 |
|
123 | |||
124 | def testManifest(self): |
|
124 | def testManifest(self): | |
125 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
125 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
126 | self.assertEqual(['bar/baz/qux.py', 'foo'], list(m)) |
|
126 | self.assertEqual(['bar/baz/qux.py', 'foo'], list(m)) | |
127 | self.assertEqual(BIN_HASH_2, m['bar/baz/qux.py']) |
|
127 | self.assertEqual(BIN_HASH_2, m['bar/baz/qux.py']) | |
128 | self.assertEqual('l', m.flags('bar/baz/qux.py')) |
|
128 | self.assertEqual('l', m.flags('bar/baz/qux.py')) | |
129 | self.assertEqual(BIN_HASH_1, m['foo']) |
|
129 | self.assertEqual(BIN_HASH_1, m['foo']) | |
130 | self.assertEqual('', m.flags('foo')) |
|
130 | self.assertEqual('', m.flags('foo')) | |
131 |
self.assertRaises(KeyError |
|
131 | with self.assertRaises(KeyError): | |
|
132 | m['wat'] | |||
132 |
|
133 | |||
133 | def testParseManifestV2(self): |
|
134 | def testParseManifestV2(self): | |
134 | m1 = self.parsemanifest(A_SHORT_MANIFEST) |
|
135 | m1 = self.parsemanifest(A_SHORT_MANIFEST) | |
135 | m2 = self.parsemanifest(A_SHORT_MANIFEST_V2) |
|
136 | m2 = self.parsemanifest(A_SHORT_MANIFEST_V2) | |
136 | # Should have same content as A_SHORT_MANIFEST |
|
137 | # Should have same content as A_SHORT_MANIFEST | |
137 | self.assertEqual(m1.text(), m2.text()) |
|
138 | self.assertEqual(m1.text(), m2.text()) | |
138 |
|
139 | |||
139 | def testParseManifestMetadata(self): |
|
140 | def testParseManifestMetadata(self): | |
140 | # Metadata is for future-proofing and should be accepted but ignored |
|
141 | # Metadata is for future-proofing and should be accepted but ignored | |
141 | m = self.parsemanifest(A_METADATA_MANIFEST) |
|
142 | m = self.parsemanifest(A_METADATA_MANIFEST) | |
142 | self.assertEqual(A_SHORT_MANIFEST, m.text()) |
|
143 | self.assertEqual(A_SHORT_MANIFEST, m.text()) | |
143 |
|
144 | |||
144 | def testParseManifestStemCompression(self): |
|
145 | def testParseManifestStemCompression(self): | |
145 | m = self.parsemanifest(A_STEM_COMPRESSED_MANIFEST) |
|
146 | m = self.parsemanifest(A_STEM_COMPRESSED_MANIFEST) | |
146 | self.assertIn('bar/baz/qux.py', m) |
|
147 | self.assertIn('bar/baz/qux.py', m) | |
147 | self.assertIn('bar/qux/foo.py', m) |
|
148 | self.assertIn('bar/qux/foo.py', m) | |
148 | self.assertIn('bar/qux/foz.py', m) |
|
149 | self.assertIn('bar/qux/foz.py', m) | |
149 | self.assertIn(256 * 'x' + '/x', m) |
|
150 | self.assertIn(256 * 'x' + '/x', m) | |
150 | self.assertIn(256 * 'x' + '/y', m) |
|
151 | self.assertIn(256 * 'x' + '/y', m) | |
151 | self.assertEqual(A_STEM_COMPRESSED_MANIFEST, m.text(usemanifestv2=True)) |
|
152 | self.assertEqual(A_STEM_COMPRESSED_MANIFEST, m.text(usemanifestv2=True)) | |
152 |
|
153 | |||
153 | def testTextV2(self): |
|
154 | def testTextV2(self): | |
154 | m1 = self.parsemanifest(A_SHORT_MANIFEST) |
|
155 | m1 = self.parsemanifest(A_SHORT_MANIFEST) | |
155 | v2text = m1.text(usemanifestv2=True) |
|
156 | v2text = m1.text(usemanifestv2=True) | |
156 | self.assertEqual(A_SHORT_MANIFEST_V2, v2text) |
|
157 | self.assertEqual(A_SHORT_MANIFEST_V2, v2text) | |
157 |
|
158 | |||
158 | def testSetItem(self): |
|
159 | def testSetItem(self): | |
159 | want = BIN_HASH_1 |
|
160 | want = BIN_HASH_1 | |
160 |
|
161 | |||
161 | m = self.parsemanifest(EMTPY_MANIFEST) |
|
162 | m = self.parsemanifest(EMTPY_MANIFEST) | |
162 | m['a'] = want |
|
163 | m['a'] = want | |
163 | self.assertIn('a', m) |
|
164 | self.assertIn('a', m) | |
164 | self.assertEqual(want, m['a']) |
|
165 | self.assertEqual(want, m['a']) | |
165 | self.assertEqual('a\0' + HASH_1 + '\n', m.text()) |
|
166 | self.assertEqual('a\0' + HASH_1 + '\n', m.text()) | |
166 |
|
167 | |||
167 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
168 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
168 | m['a'] = want |
|
169 | m['a'] = want | |
169 | self.assertEqual(want, m['a']) |
|
170 | self.assertEqual(want, m['a']) | |
170 | self.assertEqual('a\0' + HASH_1 + '\n' + A_SHORT_MANIFEST, |
|
171 | self.assertEqual('a\0' + HASH_1 + '\n' + A_SHORT_MANIFEST, | |
171 | m.text()) |
|
172 | m.text()) | |
172 |
|
173 | |||
173 | def testSetFlag(self): |
|
174 | def testSetFlag(self): | |
174 | want = 'x' |
|
175 | want = 'x' | |
175 |
|
176 | |||
176 | m = self.parsemanifest(EMTPY_MANIFEST) |
|
177 | m = self.parsemanifest(EMTPY_MANIFEST) | |
177 | # first add a file; a file-less flag makes no sense |
|
178 | # first add a file; a file-less flag makes no sense | |
178 | m['a'] = BIN_HASH_1 |
|
179 | m['a'] = BIN_HASH_1 | |
179 | m.setflag('a', want) |
|
180 | m.setflag('a', want) | |
180 | self.assertEqual(want, m.flags('a')) |
|
181 | self.assertEqual(want, m.flags('a')) | |
181 | self.assertEqual('a\0' + HASH_1 + want + '\n', m.text()) |
|
182 | self.assertEqual('a\0' + HASH_1 + want + '\n', m.text()) | |
182 |
|
183 | |||
183 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
184 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
184 | # first add a file; a file-less flag makes no sense |
|
185 | # first add a file; a file-less flag makes no sense | |
185 | m['a'] = BIN_HASH_1 |
|
186 | m['a'] = BIN_HASH_1 | |
186 | m.setflag('a', want) |
|
187 | m.setflag('a', want) | |
187 | self.assertEqual(want, m.flags('a')) |
|
188 | self.assertEqual(want, m.flags('a')) | |
188 | self.assertEqual('a\0' + HASH_1 + want + '\n' + A_SHORT_MANIFEST, |
|
189 | self.assertEqual('a\0' + HASH_1 + want + '\n' + A_SHORT_MANIFEST, | |
189 | m.text()) |
|
190 | m.text()) | |
190 |
|
191 | |||
191 | def testCopy(self): |
|
192 | def testCopy(self): | |
192 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
193 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
193 | m['a'] = BIN_HASH_1 |
|
194 | m['a'] = BIN_HASH_1 | |
194 | m2 = m.copy() |
|
195 | m2 = m.copy() | |
195 | del m |
|
196 | del m | |
196 | del m2 # make sure we don't double free() anything |
|
197 | del m2 # make sure we don't double free() anything | |
197 |
|
198 | |||
198 | def testCompaction(self): |
|
199 | def testCompaction(self): | |
199 | unhex = binascii.unhexlify |
|
200 | unhex = binascii.unhexlify | |
200 | h1, h2 = unhex(HASH_1), unhex(HASH_2) |
|
201 | h1, h2 = unhex(HASH_1), unhex(HASH_2) | |
201 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
202 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
202 | m['alpha'] = h1 |
|
203 | m['alpha'] = h1 | |
203 | m['beta'] = h2 |
|
204 | m['beta'] = h2 | |
204 | del m['foo'] |
|
205 | del m['foo'] | |
205 | want = 'alpha\0%s\nbar/baz/qux.py\0%sl\nbeta\0%s\n' % ( |
|
206 | want = 'alpha\0%s\nbar/baz/qux.py\0%sl\nbeta\0%s\n' % ( | |
206 | HASH_1, HASH_2, HASH_2) |
|
207 | HASH_1, HASH_2, HASH_2) | |
207 | self.assertEqual(want, m.text()) |
|
208 | self.assertEqual(want, m.text()) | |
208 | self.assertEqual(3, len(m)) |
|
209 | self.assertEqual(3, len(m)) | |
209 | self.assertEqual(['alpha', 'bar/baz/qux.py', 'beta'], list(m)) |
|
210 | self.assertEqual(['alpha', 'bar/baz/qux.py', 'beta'], list(m)) | |
210 | self.assertEqual(h1, m['alpha']) |
|
211 | self.assertEqual(h1, m['alpha']) | |
211 | self.assertEqual(h2, m['bar/baz/qux.py']) |
|
212 | self.assertEqual(h2, m['bar/baz/qux.py']) | |
212 | self.assertEqual(h2, m['beta']) |
|
213 | self.assertEqual(h2, m['beta']) | |
213 | self.assertEqual('', m.flags('alpha')) |
|
214 | self.assertEqual('', m.flags('alpha')) | |
214 | self.assertEqual('l', m.flags('bar/baz/qux.py')) |
|
215 | self.assertEqual('l', m.flags('bar/baz/qux.py')) | |
215 | self.assertEqual('', m.flags('beta')) |
|
216 | self.assertEqual('', m.flags('beta')) | |
216 |
self.assertRaises(KeyError |
|
217 | with self.assertRaises(KeyError): | |
|
218 | m['foo'] | |||
217 |
|
219 | |||
218 | def testSetGetNodeSuffix(self): |
|
220 | def testSetGetNodeSuffix(self): | |
219 | clean = self.parsemanifest(A_SHORT_MANIFEST) |
|
221 | clean = self.parsemanifest(A_SHORT_MANIFEST) | |
220 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
222 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
221 | h = m['foo'] |
|
223 | h = m['foo'] | |
222 | f = m.flags('foo') |
|
224 | f = m.flags('foo') | |
223 | want = h + 'a' |
|
225 | want = h + 'a' | |
224 | # Merge code wants to set 21-byte fake hashes at times |
|
226 | # Merge code wants to set 21-byte fake hashes at times | |
225 | m['foo'] = want |
|
227 | m['foo'] = want | |
226 | self.assertEqual(want, m['foo']) |
|
228 | self.assertEqual(want, m['foo']) | |
227 | self.assertEqual([('bar/baz/qux.py', BIN_HASH_2), |
|
229 | self.assertEqual([('bar/baz/qux.py', BIN_HASH_2), | |
228 | ('foo', BIN_HASH_1 + 'a')], |
|
230 | ('foo', BIN_HASH_1 + 'a')], | |
229 | list(m.iteritems())) |
|
231 | list(m.iteritems())) | |
230 | # Sometimes it even tries a 22-byte fake hash, but we can |
|
232 | # Sometimes it even tries a 22-byte fake hash, but we can | |
231 | # return 21 and it'll work out |
|
233 | # return 21 and it'll work out | |
232 | m['foo'] = want + '+' |
|
234 | m['foo'] = want + '+' | |
233 | self.assertEqual(want, m['foo']) |
|
235 | self.assertEqual(want, m['foo']) | |
234 | # make sure the suffix survives a copy |
|
236 | # make sure the suffix survives a copy | |
235 | match = matchmod.match('', '', ['re:foo']) |
|
237 | match = matchmod.match('', '', ['re:foo']) | |
236 | m2 = m.matches(match) |
|
238 | m2 = m.matches(match) | |
237 | self.assertEqual(want, m2['foo']) |
|
239 | self.assertEqual(want, m2['foo']) | |
238 | self.assertEqual(1, len(m2)) |
|
240 | self.assertEqual(1, len(m2)) | |
239 | m2 = m.copy() |
|
241 | m2 = m.copy() | |
240 | self.assertEqual(want, m2['foo']) |
|
242 | self.assertEqual(want, m2['foo']) | |
241 | # suffix with iteration |
|
243 | # suffix with iteration | |
242 | self.assertEqual([('bar/baz/qux.py', BIN_HASH_2), |
|
244 | self.assertEqual([('bar/baz/qux.py', BIN_HASH_2), | |
243 | ('foo', want)], |
|
245 | ('foo', want)], | |
244 | list(m.iteritems())) |
|
246 | list(m.iteritems())) | |
245 |
|
247 | |||
246 | # shows up in diff |
|
248 | # shows up in diff | |
247 | self.assertEqual({'foo': ((want, f), (h, ''))}, m.diff(clean)) |
|
249 | self.assertEqual({'foo': ((want, f), (h, ''))}, m.diff(clean)) | |
248 | self.assertEqual({'foo': ((h, ''), (want, f))}, clean.diff(m)) |
|
250 | self.assertEqual({'foo': ((h, ''), (want, f))}, clean.diff(m)) | |
249 |
|
251 | |||
250 | def testMatchException(self): |
|
252 | def testMatchException(self): | |
251 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
253 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
252 | match = matchmod.match('', '', ['re:.*']) |
|
254 | match = matchmod.match('', '', ['re:.*']) | |
253 | def filt(path): |
|
255 | def filt(path): | |
254 | if path == 'foo': |
|
256 | if path == 'foo': | |
255 | assert False |
|
257 | assert False | |
256 | return True |
|
258 | return True | |
257 | match.matchfn = filt |
|
259 | match.matchfn = filt | |
258 |
self.assertRaises(AssertionError |
|
260 | with self.assertRaises(AssertionError): | |
|
261 | m.matches(match) | |||
259 |
|
262 | |||
260 | def testRemoveItem(self): |
|
263 | def testRemoveItem(self): | |
261 | m = self.parsemanifest(A_SHORT_MANIFEST) |
|
264 | m = self.parsemanifest(A_SHORT_MANIFEST) | |
262 | del m['foo'] |
|
265 | del m['foo'] | |
263 |
self.assertRaises(KeyError |
|
266 | with self.assertRaises(KeyError): | |
|
267 | m['foo'] | |||
264 | self.assertEqual(1, len(m)) |
|
268 | self.assertEqual(1, len(m)) | |
265 | self.assertEqual(1, len(list(m))) |
|
269 | self.assertEqual(1, len(list(m))) | |
266 | # now restore and make sure everything works right |
|
270 | # now restore and make sure everything works right | |
267 | m['foo'] = 'a' * 20 |
|
271 | m['foo'] = 'a' * 20 | |
268 | self.assertEqual(2, len(m)) |
|
272 | self.assertEqual(2, len(m)) | |
269 | self.assertEqual(2, len(list(m))) |
|
273 | self.assertEqual(2, len(list(m))) | |
270 |
|
274 | |||
271 | def testManifestDiff(self): |
|
275 | def testManifestDiff(self): | |
272 | MISSING = (None, '') |
|
276 | MISSING = (None, '') | |
273 | addl = 'z-only-in-left\0' + HASH_1 + '\n' |
|
277 | addl = 'z-only-in-left\0' + HASH_1 + '\n' | |
274 | addr = 'z-only-in-right\0' + HASH_2 + 'x\n' |
|
278 | addr = 'z-only-in-right\0' + HASH_2 + 'x\n' | |
275 | left = self.parsemanifest( |
|
279 | left = self.parsemanifest( | |
276 | A_SHORT_MANIFEST.replace(HASH_1, HASH_3 + 'x') + addl) |
|
280 | A_SHORT_MANIFEST.replace(HASH_1, HASH_3 + 'x') + addl) | |
277 | right = self.parsemanifest(A_SHORT_MANIFEST + addr) |
|
281 | right = self.parsemanifest(A_SHORT_MANIFEST + addr) | |
278 | want = { |
|
282 | want = { | |
279 | 'foo': ((BIN_HASH_3, 'x'), |
|
283 | 'foo': ((BIN_HASH_3, 'x'), | |
280 | (BIN_HASH_1, '')), |
|
284 | (BIN_HASH_1, '')), | |
281 | 'z-only-in-left': ((BIN_HASH_1, ''), MISSING), |
|
285 | 'z-only-in-left': ((BIN_HASH_1, ''), MISSING), | |
282 | 'z-only-in-right': (MISSING, (BIN_HASH_2, 'x')), |
|
286 | 'z-only-in-right': (MISSING, (BIN_HASH_2, 'x')), | |
283 | } |
|
287 | } | |
284 | self.assertEqual(want, left.diff(right)) |
|
288 | self.assertEqual(want, left.diff(right)) | |
285 |
|
289 | |||
286 | want = { |
|
290 | want = { | |
287 | 'bar/baz/qux.py': (MISSING, (BIN_HASH_2, 'l')), |
|
291 | 'bar/baz/qux.py': (MISSING, (BIN_HASH_2, 'l')), | |
288 | 'foo': (MISSING, (BIN_HASH_3, 'x')), |
|
292 | 'foo': (MISSING, (BIN_HASH_3, 'x')), | |
289 | 'z-only-in-left': (MISSING, (BIN_HASH_1, '')), |
|
293 | 'z-only-in-left': (MISSING, (BIN_HASH_1, '')), | |
290 | } |
|
294 | } | |
291 | self.assertEqual(want, self.parsemanifest(EMTPY_MANIFEST).diff(left)) |
|
295 | self.assertEqual(want, self.parsemanifest(EMTPY_MANIFEST).diff(left)) | |
292 |
|
296 | |||
293 | want = { |
|
297 | want = { | |
294 | 'bar/baz/qux.py': ((BIN_HASH_2, 'l'), MISSING), |
|
298 | 'bar/baz/qux.py': ((BIN_HASH_2, 'l'), MISSING), | |
295 | 'foo': ((BIN_HASH_3, 'x'), MISSING), |
|
299 | 'foo': ((BIN_HASH_3, 'x'), MISSING), | |
296 | 'z-only-in-left': ((BIN_HASH_1, ''), MISSING), |
|
300 | 'z-only-in-left': ((BIN_HASH_1, ''), MISSING), | |
297 | } |
|
301 | } | |
298 | self.assertEqual(want, left.diff(self.parsemanifest(EMTPY_MANIFEST))) |
|
302 | self.assertEqual(want, left.diff(self.parsemanifest(EMTPY_MANIFEST))) | |
299 | copy = right.copy() |
|
303 | copy = right.copy() | |
300 | del copy['z-only-in-right'] |
|
304 | del copy['z-only-in-right'] | |
301 | del right['foo'] |
|
305 | del right['foo'] | |
302 | want = { |
|
306 | want = { | |
303 | 'foo': (MISSING, (BIN_HASH_1, '')), |
|
307 | 'foo': (MISSING, (BIN_HASH_1, '')), | |
304 | 'z-only-in-right': ((BIN_HASH_2, 'x'), MISSING), |
|
308 | 'z-only-in-right': ((BIN_HASH_2, 'x'), MISSING), | |
305 | } |
|
309 | } | |
306 | self.assertEqual(want, right.diff(copy)) |
|
310 | self.assertEqual(want, right.diff(copy)) | |
307 |
|
311 | |||
308 | short = self.parsemanifest(A_SHORT_MANIFEST) |
|
312 | short = self.parsemanifest(A_SHORT_MANIFEST) | |
309 | pruned = short.copy() |
|
313 | pruned = short.copy() | |
310 | del pruned['foo'] |
|
314 | del pruned['foo'] | |
311 | want = { |
|
315 | want = { | |
312 | 'foo': ((BIN_HASH_1, ''), MISSING), |
|
316 | 'foo': ((BIN_HASH_1, ''), MISSING), | |
313 | } |
|
317 | } | |
314 | self.assertEqual(want, short.diff(pruned)) |
|
318 | self.assertEqual(want, short.diff(pruned)) | |
315 | want = { |
|
319 | want = { | |
316 | 'foo': (MISSING, (BIN_HASH_1, '')), |
|
320 | 'foo': (MISSING, (BIN_HASH_1, '')), | |
317 | } |
|
321 | } | |
318 | self.assertEqual(want, pruned.diff(short)) |
|
322 | self.assertEqual(want, pruned.diff(short)) | |
319 | want = { |
|
323 | want = { | |
320 | 'bar/baz/qux.py': None, |
|
324 | 'bar/baz/qux.py': None, | |
321 | 'foo': (MISSING, (BIN_HASH_1, '')), |
|
325 | 'foo': (MISSING, (BIN_HASH_1, '')), | |
322 | } |
|
326 | } | |
323 | self.assertEqual(want, pruned.diff(short, clean=True)) |
|
327 | self.assertEqual(want, pruned.diff(short, clean=True)) | |
324 |
|
328 | |||
325 | def testReversedLines(self): |
|
329 | def testReversedLines(self): | |
326 | backwards = ''.join( |
|
330 | backwards = ''.join( | |
327 | l + '\n' for l in reversed(A_SHORT_MANIFEST.split('\n')) if l) |
|
331 | l + '\n' for l in reversed(A_SHORT_MANIFEST.split('\n')) if l) | |
328 | try: |
|
332 | try: | |
329 | self.parsemanifest(backwards) |
|
333 | self.parsemanifest(backwards) | |
330 | self.fail('Should have raised ValueError') |
|
334 | self.fail('Should have raised ValueError') | |
331 | except ValueError as v: |
|
335 | except ValueError as v: | |
332 | self.assertIn('Manifest lines not in sorted order.', str(v)) |
|
336 | self.assertIn('Manifest lines not in sorted order.', str(v)) | |
333 |
|
337 | |||
334 | def testNoTerminalNewline(self): |
|
338 | def testNoTerminalNewline(self): | |
335 | try: |
|
339 | try: | |
336 | self.parsemanifest(A_SHORT_MANIFEST + 'wat') |
|
340 | self.parsemanifest(A_SHORT_MANIFEST + 'wat') | |
337 | self.fail('Should have raised ValueError') |
|
341 | self.fail('Should have raised ValueError') | |
338 | except ValueError as v: |
|
342 | except ValueError as v: | |
339 | self.assertIn('Manifest did not end in a newline.', str(v)) |
|
343 | self.assertIn('Manifest did not end in a newline.', str(v)) | |
340 |
|
344 | |||
341 | def testNoNewLineAtAll(self): |
|
345 | def testNoNewLineAtAll(self): | |
342 | try: |
|
346 | try: | |
343 | self.parsemanifest('wat') |
|
347 | self.parsemanifest('wat') | |
344 | self.fail('Should have raised ValueError') |
|
348 | self.fail('Should have raised ValueError') | |
345 | except ValueError as v: |
|
349 | except ValueError as v: | |
346 | self.assertIn('Manifest did not end in a newline.', str(v)) |
|
350 | self.assertIn('Manifest did not end in a newline.', str(v)) | |
347 |
|
351 | |||
348 | def testHugeManifest(self): |
|
352 | def testHugeManifest(self): | |
349 | m = self.parsemanifest(A_HUGE_MANIFEST) |
|
353 | m = self.parsemanifest(A_HUGE_MANIFEST) | |
350 | self.assertEqual(HUGE_MANIFEST_ENTRIES, len(m)) |
|
354 | self.assertEqual(HUGE_MANIFEST_ENTRIES, len(m)) | |
351 | self.assertEqual(len(m), len(list(m))) |
|
355 | self.assertEqual(len(m), len(list(m))) | |
352 |
|
356 | |||
353 | def testMatchesMetadata(self): |
|
357 | def testMatchesMetadata(self): | |
354 | '''Tests matches() for a few specific files to make sure that both |
|
358 | '''Tests matches() for a few specific files to make sure that both | |
355 | the set of files as well as their flags and nodeids are correct in |
|
359 | the set of files as well as their flags and nodeids are correct in | |
356 | the resulting manifest.''' |
|
360 | the resulting manifest.''' | |
357 | m = self.parsemanifest(A_HUGE_MANIFEST) |
|
361 | m = self.parsemanifest(A_HUGE_MANIFEST) | |
358 |
|
362 | |||
359 | match = matchmod.match('/', '', |
|
363 | match = matchmod.match('/', '', | |
360 | ['file1', 'file200', 'file300'], exact=True) |
|
364 | ['file1', 'file200', 'file300'], exact=True) | |
361 | m2 = m.matches(match) |
|
365 | m2 = m.matches(match) | |
362 |
|
366 | |||
363 | w = ('file1\0%sx\n' |
|
367 | w = ('file1\0%sx\n' | |
364 | 'file200\0%sl\n' |
|
368 | 'file200\0%sl\n' | |
365 | 'file300\0%s\n') % (HASH_2, HASH_1, HASH_1) |
|
369 | 'file300\0%s\n') % (HASH_2, HASH_1, HASH_1) | |
366 | self.assertEqual(w, m2.text()) |
|
370 | self.assertEqual(w, m2.text()) | |
367 |
|
371 | |||
368 | def testMatchesNonexistentFile(self): |
|
372 | def testMatchesNonexistentFile(self): | |
369 | '''Tests matches() for a small set of specific files, including one |
|
373 | '''Tests matches() for a small set of specific files, including one | |
370 | nonexistent file to make sure in only matches against existing files. |
|
374 | nonexistent file to make sure in only matches against existing files. | |
371 | ''' |
|
375 | ''' | |
372 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
376 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
373 |
|
377 | |||
374 | match = matchmod.match('/', '', |
|
378 | match = matchmod.match('/', '', | |
375 | ['a/b/c/bar.txt', 'a/b/d/qux.py', 'readme.txt', 'nonexistent'], |
|
379 | ['a/b/c/bar.txt', 'a/b/d/qux.py', 'readme.txt', 'nonexistent'], | |
376 | exact=True) |
|
380 | exact=True) | |
377 | m2 = m.matches(match) |
|
381 | m2 = m.matches(match) | |
378 |
|
382 | |||
379 | self.assertEqual( |
|
383 | self.assertEqual( | |
380 | ['a/b/c/bar.txt', 'a/b/d/qux.py', 'readme.txt'], |
|
384 | ['a/b/c/bar.txt', 'a/b/d/qux.py', 'readme.txt'], | |
381 | m2.keys()) |
|
385 | m2.keys()) | |
382 |
|
386 | |||
383 | def testMatchesNonexistentDirectory(self): |
|
387 | def testMatchesNonexistentDirectory(self): | |
384 | '''Tests matches() for a relpath match on a directory that doesn't |
|
388 | '''Tests matches() for a relpath match on a directory that doesn't | |
385 | actually exist.''' |
|
389 | actually exist.''' | |
386 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
390 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
387 |
|
391 | |||
388 | match = matchmod.match('/', '', ['a/f'], default='relpath') |
|
392 | match = matchmod.match('/', '', ['a/f'], default='relpath') | |
389 | m2 = m.matches(match) |
|
393 | m2 = m.matches(match) | |
390 |
|
394 | |||
391 | self.assertEqual([], m2.keys()) |
|
395 | self.assertEqual([], m2.keys()) | |
392 |
|
396 | |||
393 | def testMatchesExactLarge(self): |
|
397 | def testMatchesExactLarge(self): | |
394 | '''Tests matches() for files matching a large list of exact files. |
|
398 | '''Tests matches() for files matching a large list of exact files. | |
395 | ''' |
|
399 | ''' | |
396 | m = self.parsemanifest(A_HUGE_MANIFEST) |
|
400 | m = self.parsemanifest(A_HUGE_MANIFEST) | |
397 |
|
401 | |||
398 | flist = m.keys()[80:300] |
|
402 | flist = m.keys()[80:300] | |
399 | match = matchmod.match('/', '', flist, exact=True) |
|
403 | match = matchmod.match('/', '', flist, exact=True) | |
400 | m2 = m.matches(match) |
|
404 | m2 = m.matches(match) | |
401 |
|
405 | |||
402 | self.assertEqual(flist, m2.keys()) |
|
406 | self.assertEqual(flist, m2.keys()) | |
403 |
|
407 | |||
404 | def testMatchesFull(self): |
|
408 | def testMatchesFull(self): | |
405 | '''Tests matches() for what should be a full match.''' |
|
409 | '''Tests matches() for what should be a full match.''' | |
406 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
410 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
407 |
|
411 | |||
408 | match = matchmod.match('/', '', ['']) |
|
412 | match = matchmod.match('/', '', ['']) | |
409 | m2 = m.matches(match) |
|
413 | m2 = m.matches(match) | |
410 |
|
414 | |||
411 | self.assertEqual(m.keys(), m2.keys()) |
|
415 | self.assertEqual(m.keys(), m2.keys()) | |
412 |
|
416 | |||
413 | def testMatchesDirectory(self): |
|
417 | def testMatchesDirectory(self): | |
414 | '''Tests matches() on a relpath match on a directory, which should |
|
418 | '''Tests matches() on a relpath match on a directory, which should | |
415 | match against all files within said directory.''' |
|
419 | match against all files within said directory.''' | |
416 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
420 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
417 |
|
421 | |||
418 | match = matchmod.match('/', '', ['a/b'], default='relpath') |
|
422 | match = matchmod.match('/', '', ['a/b'], default='relpath') | |
419 | m2 = m.matches(match) |
|
423 | m2 = m.matches(match) | |
420 |
|
424 | |||
421 | self.assertEqual([ |
|
425 | self.assertEqual([ | |
422 | 'a/b/c/bar.py', 'a/b/c/bar.txt', 'a/b/c/foo.py', 'a/b/c/foo.txt', |
|
426 | 'a/b/c/bar.py', 'a/b/c/bar.txt', 'a/b/c/foo.py', 'a/b/c/foo.txt', | |
423 | 'a/b/d/baz.py', 'a/b/d/qux.py', 'a/b/d/ten.txt', 'a/b/dog.py', |
|
427 | 'a/b/d/baz.py', 'a/b/d/qux.py', 'a/b/d/ten.txt', 'a/b/dog.py', | |
424 | 'a/b/fish.py'], m2.keys()) |
|
428 | 'a/b/fish.py'], m2.keys()) | |
425 |
|
429 | |||
426 | def testMatchesExactPath(self): |
|
430 | def testMatchesExactPath(self): | |
427 | '''Tests matches() on an exact match on a directory, which should |
|
431 | '''Tests matches() on an exact match on a directory, which should | |
428 | result in an empty manifest because you can't perform an exact match |
|
432 | result in an empty manifest because you can't perform an exact match | |
429 | against a directory.''' |
|
433 | against a directory.''' | |
430 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
434 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
431 |
|
435 | |||
432 | match = matchmod.match('/', '', ['a/b'], exact=True) |
|
436 | match = matchmod.match('/', '', ['a/b'], exact=True) | |
433 | m2 = m.matches(match) |
|
437 | m2 = m.matches(match) | |
434 |
|
438 | |||
435 | self.assertEqual([], m2.keys()) |
|
439 | self.assertEqual([], m2.keys()) | |
436 |
|
440 | |||
437 | def testMatchesCwd(self): |
|
441 | def testMatchesCwd(self): | |
438 | '''Tests matches() on a relpath match with the current directory ('.') |
|
442 | '''Tests matches() on a relpath match with the current directory ('.') | |
439 | when not in the root directory.''' |
|
443 | when not in the root directory.''' | |
440 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
444 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
441 |
|
445 | |||
442 | match = matchmod.match('/', 'a/b', ['.'], default='relpath') |
|
446 | match = matchmod.match('/', 'a/b', ['.'], default='relpath') | |
443 | m2 = m.matches(match) |
|
447 | m2 = m.matches(match) | |
444 |
|
448 | |||
445 | self.assertEqual([ |
|
449 | self.assertEqual([ | |
446 | 'a/b/c/bar.py', 'a/b/c/bar.txt', 'a/b/c/foo.py', 'a/b/c/foo.txt', |
|
450 | 'a/b/c/bar.py', 'a/b/c/bar.txt', 'a/b/c/foo.py', 'a/b/c/foo.txt', | |
447 | 'a/b/d/baz.py', 'a/b/d/qux.py', 'a/b/d/ten.txt', 'a/b/dog.py', |
|
451 | 'a/b/d/baz.py', 'a/b/d/qux.py', 'a/b/d/ten.txt', 'a/b/dog.py', | |
448 | 'a/b/fish.py'], m2.keys()) |
|
452 | 'a/b/fish.py'], m2.keys()) | |
449 |
|
453 | |||
450 | def testMatchesWithPattern(self): |
|
454 | def testMatchesWithPattern(self): | |
451 | '''Tests matches() for files matching a pattern that reside |
|
455 | '''Tests matches() for files matching a pattern that reside | |
452 | deeper than the specified directory.''' |
|
456 | deeper than the specified directory.''' | |
453 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
457 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
454 |
|
458 | |||
455 | match = matchmod.match('/', '', ['a/b/*/*.txt']) |
|
459 | match = matchmod.match('/', '', ['a/b/*/*.txt']) | |
456 | m2 = m.matches(match) |
|
460 | m2 = m.matches(match) | |
457 |
|
461 | |||
458 | self.assertEqual( |
|
462 | self.assertEqual( | |
459 | ['a/b/c/bar.txt', 'a/b/c/foo.txt', 'a/b/d/ten.txt'], |
|
463 | ['a/b/c/bar.txt', 'a/b/c/foo.txt', 'a/b/d/ten.txt'], | |
460 | m2.keys()) |
|
464 | m2.keys()) | |
461 |
|
465 | |||
462 | class testmanifestdict(unittest.TestCase, basemanifesttests): |
|
466 | class testmanifestdict(unittest.TestCase, basemanifesttests): | |
463 | def parsemanifest(self, text): |
|
467 | def parsemanifest(self, text): | |
464 | return manifestmod.manifestdict(text) |
|
468 | return manifestmod.manifestdict(text) | |
465 |
|
469 | |||
466 | class testtreemanifest(unittest.TestCase, basemanifesttests): |
|
470 | class testtreemanifest(unittest.TestCase, basemanifesttests): | |
467 | def parsemanifest(self, text): |
|
471 | def parsemanifest(self, text): | |
468 | return manifestmod.treemanifest('', text) |
|
472 | return manifestmod.treemanifest('', text) | |
469 |
|
473 | |||
470 | def testWalkSubtrees(self): |
|
474 | def testWalkSubtrees(self): | |
471 | m = self.parsemanifest(A_DEEPER_MANIFEST) |
|
475 | m = self.parsemanifest(A_DEEPER_MANIFEST) | |
472 |
|
476 | |||
473 | dirs = [s._dir for s in m.walksubtrees()] |
|
477 | dirs = [s._dir for s in m.walksubtrees()] | |
474 | self.assertEqual( |
|
478 | self.assertEqual( | |
475 | sorted(['', 'a/', 'a/c/', 'a/d/', 'a/b/', 'a/b/c/', 'a/b/d/']), |
|
479 | sorted(['', 'a/', 'a/c/', 'a/d/', 'a/b/', 'a/b/c/', 'a/b/d/']), | |
476 | sorted(dirs) |
|
480 | sorted(dirs) | |
477 | ) |
|
481 | ) | |
478 |
|
482 | |||
479 | match = matchmod.match('/', '', ['path:a/b/']) |
|
483 | match = matchmod.match('/', '', ['path:a/b/']) | |
480 | dirs = [s._dir for s in m.walksubtrees(matcher=match)] |
|
484 | dirs = [s._dir for s in m.walksubtrees(matcher=match)] | |
481 | self.assertEqual( |
|
485 | self.assertEqual( | |
482 | sorted(['a/b/', 'a/b/c/', 'a/b/d/']), |
|
486 | sorted(['a/b/', 'a/b/c/', 'a/b/d/']), | |
483 | sorted(dirs) |
|
487 | sorted(dirs) | |
484 | ) |
|
488 | ) | |
485 |
|
489 | |||
486 | if __name__ == '__main__': |
|
490 | if __name__ == '__main__': | |
487 | silenttestrunner.main(__name__) |
|
491 | silenttestrunner.main(__name__) |
@@ -1,84 +1,84 b'' | |||||
1 | from __future__ import absolute_import |
|
1 | from __future__ import absolute_import | |
2 |
|
2 | |||
3 | import unittest |
|
3 | import unittest | |
4 | import silenttestrunner |
|
4 | import silenttestrunner | |
5 |
|
5 | |||
6 | from mercurial import ( |
|
6 | from mercurial import ( | |
7 | error, |
|
7 | error, | |
8 | scmutil, |
|
8 | scmutil, | |
9 | ) |
|
9 | ) | |
10 |
|
10 | |||
11 | class mockfile(object): |
|
11 | class mockfile(object): | |
12 | def __init__(self, name, fs): |
|
12 | def __init__(self, name, fs): | |
13 | self.name = name |
|
13 | self.name = name | |
14 | self.fs = fs |
|
14 | self.fs = fs | |
15 |
|
15 | |||
16 | def __enter__(self): |
|
16 | def __enter__(self): | |
17 | return self |
|
17 | return self | |
18 |
|
18 | |||
19 | def __exit__(self, *args, **kwargs): |
|
19 | def __exit__(self, *args, **kwargs): | |
20 | pass |
|
20 | pass | |
21 |
|
21 | |||
22 | def write(self, text): |
|
22 | def write(self, text): | |
23 | self.fs.contents[self.name] = text |
|
23 | self.fs.contents[self.name] = text | |
24 |
|
24 | |||
25 | def read(self): |
|
25 | def read(self): | |
26 | return self.fs.contents[self.name] |
|
26 | return self.fs.contents[self.name] | |
27 |
|
27 | |||
28 | class mockvfs(object): |
|
28 | class mockvfs(object): | |
29 | def __init__(self): |
|
29 | def __init__(self): | |
30 | self.contents = {} |
|
30 | self.contents = {} | |
31 |
|
31 | |||
32 | def read(self, path): |
|
32 | def read(self, path): | |
33 | return mockfile(path, self).read() |
|
33 | return mockfile(path, self).read() | |
34 |
|
34 | |||
35 | def readlines(self, path): |
|
35 | def readlines(self, path): | |
36 | # lines need to contain the trailing '\n' to mock the real readlines |
|
36 | # lines need to contain the trailing '\n' to mock the real readlines | |
37 | return [l for l in mockfile(path, self).read().splitlines(True)] |
|
37 | return [l for l in mockfile(path, self).read().splitlines(True)] | |
38 |
|
38 | |||
39 | def __call__(self, path, mode, atomictemp): |
|
39 | def __call__(self, path, mode, atomictemp): | |
40 | return mockfile(path, self) |
|
40 | return mockfile(path, self) | |
41 |
|
41 | |||
42 | class testsimplekeyvaluefile(unittest.TestCase): |
|
42 | class testsimplekeyvaluefile(unittest.TestCase): | |
43 | def setUp(self): |
|
43 | def setUp(self): | |
44 | self.vfs = mockvfs() |
|
44 | self.vfs = mockvfs() | |
45 |
|
45 | |||
46 | def testbasicwritingiandreading(self): |
|
46 | def testbasicwritingiandreading(self): | |
47 | dw = {'key1': 'value1', 'Key2': 'value2'} |
|
47 | dw = {'key1': 'value1', 'Key2': 'value2'} | |
48 | scmutil.simplekeyvaluefile(self.vfs, 'kvfile').write(dw) |
|
48 | scmutil.simplekeyvaluefile(self.vfs, 'kvfile').write(dw) | |
49 | self.assertEqual(sorted(self.vfs.read('kvfile').split('\n')), |
|
49 | self.assertEqual(sorted(self.vfs.read('kvfile').split('\n')), | |
50 | ['', 'Key2=value2', 'key1=value1']) |
|
50 | ['', 'Key2=value2', 'key1=value1']) | |
51 | dr = scmutil.simplekeyvaluefile(self.vfs, 'kvfile').read() |
|
51 | dr = scmutil.simplekeyvaluefile(self.vfs, 'kvfile').read() | |
52 | self.assertEqual(dr, dw) |
|
52 | self.assertEqual(dr, dw) | |
53 |
|
53 | |||
54 | def testinvalidkeys(self): |
|
54 | def testinvalidkeys(self): | |
55 | d = {'0key1': 'value1', 'Key2': 'value2'} |
|
55 | d = {'0key1': 'value1', 'Key2': 'value2'} | |
56 | self.assertRaises(error.ProgrammingError, |
|
56 | with self.assertRaisesRegexp(error.ProgrammingError, | |
57 | scmutil.simplekeyvaluefile(self.vfs, 'kvfile').write, |
|
57 | 'keys must start with a letter.*'): | |
58 | d) |
|
58 | scmutil.simplekeyvaluefile(self.vfs, 'kvfile').write(d) | |
|
59 | ||||
59 | d = {'key1@': 'value1', 'Key2': 'value2'} |
|
60 | d = {'key1@': 'value1', 'Key2': 'value2'} | |
60 | self.assertRaises(error.ProgrammingError, |
|
61 | with self.assertRaisesRegexp(error.ProgrammingError, 'invalid key.*'): | |
61 |
|
|
62 | scmutil.simplekeyvaluefile(self.vfs, 'kvfile').write(d) | |
62 | d) |
|
|||
63 |
|
63 | |||
64 | def testinvalidvalues(self): |
|
64 | def testinvalidvalues(self): | |
65 | d = {'key1': 'value1', 'Key2': 'value2\n'} |
|
65 | d = {'key1': 'value1', 'Key2': 'value2\n'} | |
66 | self.assertRaises(error.ProgrammingError, |
|
66 | with self.assertRaisesRegexp(error.ProgrammingError, 'invalid val.*'): | |
67 |
|
|
67 | scmutil.simplekeyvaluefile(self.vfs, 'kvfile').write(d) | |
68 | d) |
|
|||
69 |
|
68 | |||
70 | def testcorruptedfile(self): |
|
69 | def testcorruptedfile(self): | |
71 | self.vfs.contents['badfile'] = 'ababagalamaga\n' |
|
70 | self.vfs.contents['badfile'] = 'ababagalamaga\n' | |
72 | self.assertRaises(error.CorruptedState, |
|
71 | with self.assertRaisesRegexp(error.CorruptedState, | |
73 | scmutil.simplekeyvaluefile(self.vfs, 'badfile').read) |
|
72 | 'dictionary.*element.*'): | |
|
73 | scmutil.simplekeyvaluefile(self.vfs, 'badfile').read() | |||
74 |
|
74 | |||
75 | def testfirstline(self): |
|
75 | def testfirstline(self): | |
76 | dw = {'key1': 'value1'} |
|
76 | dw = {'key1': 'value1'} | |
77 | scmutil.simplekeyvaluefile(self.vfs, 'fl').write(dw, firstline='1.0') |
|
77 | scmutil.simplekeyvaluefile(self.vfs, 'fl').write(dw, firstline='1.0') | |
78 | self.assertEqual(self.vfs.read('fl'), '1.0\nkey1=value1\n') |
|
78 | self.assertEqual(self.vfs.read('fl'), '1.0\nkey1=value1\n') | |
79 | dr = scmutil.simplekeyvaluefile(self.vfs, 'fl')\ |
|
79 | dr = scmutil.simplekeyvaluefile(self.vfs, 'fl')\ | |
80 | .read(firstlinenonkeyval=True) |
|
80 | .read(firstlinenonkeyval=True) | |
81 | self.assertEqual(dr, {'__firstline': '1.0', 'key1': 'value1'}) |
|
81 | self.assertEqual(dr, {'__firstline': '1.0', 'key1': 'value1'}) | |
82 |
|
82 | |||
83 | if __name__ == "__main__": |
|
83 | if __name__ == "__main__": | |
84 | silenttestrunner.main(__name__) |
|
84 | silenttestrunner.main(__name__) |
@@ -1,360 +1,361 b'' | |||||
1 | # Copyright (C) 2004, 2005 Canonical Ltd |
|
1 | # Copyright (C) 2004, 2005 Canonical Ltd | |
2 | # |
|
2 | # | |
3 | # This program is free software; you can redistribute it and/or modify |
|
3 | # This program is free software; you can redistribute it and/or modify | |
4 | # it under the terms of the GNU General Public License as published by |
|
4 | # it under the terms of the GNU General Public License as published by | |
5 | # the Free Software Foundation; either version 2 of the License, or |
|
5 | # the Free Software Foundation; either version 2 of the License, or | |
6 | # (at your option) any later version. |
|
6 | # (at your option) any later version. | |
7 | # |
|
7 | # | |
8 | # This program is distributed in the hope that it will be useful, |
|
8 | # This program is distributed in the hope that it will be useful, | |
9 | # but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
9 | # but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
|
10 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
11 | # GNU General Public License for more details. |
|
11 | # GNU General Public License for more details. | |
12 | # |
|
12 | # | |
13 | # You should have received a copy of the GNU General Public License |
|
13 | # You should have received a copy of the GNU General Public License | |
14 | # along with this program; if not, see <http://www.gnu.org/licenses/>. |
|
14 | # along with this program; if not, see <http://www.gnu.org/licenses/>. | |
15 |
|
15 | |||
16 | from __future__ import absolute_import |
|
16 | from __future__ import absolute_import | |
17 |
|
17 | |||
18 | import unittest |
|
18 | import unittest | |
19 | from mercurial import ( |
|
19 | from mercurial import ( | |
20 | error, |
|
20 | error, | |
21 | simplemerge, |
|
21 | simplemerge, | |
22 | util, |
|
22 | util, | |
23 | ) |
|
23 | ) | |
24 |
|
24 | |||
25 | TestCase = unittest.TestCase |
|
25 | TestCase = unittest.TestCase | |
26 | # bzr compatible interface, for the tests |
|
26 | # bzr compatible interface, for the tests | |
27 | class Merge3(simplemerge.Merge3Text): |
|
27 | class Merge3(simplemerge.Merge3Text): | |
28 | """3-way merge of texts. |
|
28 | """3-way merge of texts. | |
29 |
|
29 | |||
30 | Given BASE, OTHER, THIS, tries to produce a combined text |
|
30 | Given BASE, OTHER, THIS, tries to produce a combined text | |
31 | incorporating the changes from both BASE->OTHER and BASE->THIS. |
|
31 | incorporating the changes from both BASE->OTHER and BASE->THIS. | |
32 | All three will typically be sequences of lines.""" |
|
32 | All three will typically be sequences of lines.""" | |
33 | def __init__(self, base, a, b): |
|
33 | def __init__(self, base, a, b): | |
34 | basetext = '\n'.join([i.strip('\n') for i in base] + ['']) |
|
34 | basetext = '\n'.join([i.strip('\n') for i in base] + ['']) | |
35 | atext = '\n'.join([i.strip('\n') for i in a] + ['']) |
|
35 | atext = '\n'.join([i.strip('\n') for i in a] + ['']) | |
36 | btext = '\n'.join([i.strip('\n') for i in b] + ['']) |
|
36 | btext = '\n'.join([i.strip('\n') for i in b] + ['']) | |
37 | if util.binary(basetext) or util.binary(atext) or util.binary(btext): |
|
37 | if util.binary(basetext) or util.binary(atext) or util.binary(btext): | |
38 | raise error.Abort("don't know how to merge binary files") |
|
38 | raise error.Abort("don't know how to merge binary files") | |
39 | simplemerge.Merge3Text.__init__(self, basetext, atext, btext, |
|
39 | simplemerge.Merge3Text.__init__(self, basetext, atext, btext, | |
40 | base, a, b) |
|
40 | base, a, b) | |
41 |
|
41 | |||
42 | CantReprocessAndShowBase = simplemerge.CantReprocessAndShowBase |
|
42 | CantReprocessAndShowBase = simplemerge.CantReprocessAndShowBase | |
43 |
|
43 | |||
44 | def split_lines(t): |
|
44 | def split_lines(t): | |
45 | return util.stringio(t).readlines() |
|
45 | return util.stringio(t).readlines() | |
46 |
|
46 | |||
47 | ############################################################ |
|
47 | ############################################################ | |
48 | # test case data from the gnu diffutils manual |
|
48 | # test case data from the gnu diffutils manual | |
49 | # common base |
|
49 | # common base | |
50 | TZU = split_lines(""" The Nameless is the origin of Heaven and Earth; |
|
50 | TZU = split_lines(""" The Nameless is the origin of Heaven and Earth; | |
51 | The named is the mother of all things. |
|
51 | The named is the mother of all things. | |
52 |
|
52 | |||
53 | Therefore let there always be non-being, |
|
53 | Therefore let there always be non-being, | |
54 | so we may see their subtlety, |
|
54 | so we may see their subtlety, | |
55 | And let there always be being, |
|
55 | And let there always be being, | |
56 | so we may see their outcome. |
|
56 | so we may see their outcome. | |
57 | The two are the same, |
|
57 | The two are the same, | |
58 | But after they are produced, |
|
58 | But after they are produced, | |
59 | they have different names. |
|
59 | they have different names. | |
60 | They both may be called deep and profound. |
|
60 | They both may be called deep and profound. | |
61 | Deeper and more profound, |
|
61 | Deeper and more profound, | |
62 | The door of all subtleties! |
|
62 | The door of all subtleties! | |
63 | """) |
|
63 | """) | |
64 |
|
64 | |||
65 | LAO = split_lines(""" The Way that can be told of is not the eternal Way; |
|
65 | LAO = split_lines(""" The Way that can be told of is not the eternal Way; | |
66 | The name that can be named is not the eternal name. |
|
66 | The name that can be named is not the eternal name. | |
67 | The Nameless is the origin of Heaven and Earth; |
|
67 | The Nameless is the origin of Heaven and Earth; | |
68 | The Named is the mother of all things. |
|
68 | The Named is the mother of all things. | |
69 | Therefore let there always be non-being, |
|
69 | Therefore let there always be non-being, | |
70 | so we may see their subtlety, |
|
70 | so we may see their subtlety, | |
71 | And let there always be being, |
|
71 | And let there always be being, | |
72 | so we may see their outcome. |
|
72 | so we may see their outcome. | |
73 | The two are the same, |
|
73 | The two are the same, | |
74 | But after they are produced, |
|
74 | But after they are produced, | |
75 | they have different names. |
|
75 | they have different names. | |
76 | """) |
|
76 | """) | |
77 |
|
77 | |||
78 |
|
78 | |||
79 | TAO = split_lines(""" The Way that can be told of is not the eternal Way; |
|
79 | TAO = split_lines(""" The Way that can be told of is not the eternal Way; | |
80 | The name that can be named is not the eternal name. |
|
80 | The name that can be named is not the eternal name. | |
81 | The Nameless is the origin of Heaven and Earth; |
|
81 | The Nameless is the origin of Heaven and Earth; | |
82 | The named is the mother of all things. |
|
82 | The named is the mother of all things. | |
83 |
|
83 | |||
84 | Therefore let there always be non-being, |
|
84 | Therefore let there always be non-being, | |
85 | so we may see their subtlety, |
|
85 | so we may see their subtlety, | |
86 | And let there always be being, |
|
86 | And let there always be being, | |
87 | so we may see their result. |
|
87 | so we may see their result. | |
88 | The two are the same, |
|
88 | The two are the same, | |
89 | But after they are produced, |
|
89 | But after they are produced, | |
90 | they have different names. |
|
90 | they have different names. | |
91 |
|
91 | |||
92 | -- The Way of Lao-Tzu, tr. Wing-tsit Chan |
|
92 | -- The Way of Lao-Tzu, tr. Wing-tsit Chan | |
93 |
|
93 | |||
94 | """) |
|
94 | """) | |
95 |
|
95 | |||
96 | MERGED_RESULT = split_lines("""\ |
|
96 | MERGED_RESULT = split_lines("""\ | |
97 | The Way that can be told of is not the eternal Way; |
|
97 | The Way that can be told of is not the eternal Way; | |
98 | The name that can be named is not the eternal name. |
|
98 | The name that can be named is not the eternal name. | |
99 | The Nameless is the origin of Heaven and Earth; |
|
99 | The Nameless is the origin of Heaven and Earth; | |
100 | The Named is the mother of all things. |
|
100 | The Named is the mother of all things. | |
101 | Therefore let there always be non-being, |
|
101 | Therefore let there always be non-being, | |
102 | so we may see their subtlety, |
|
102 | so we may see their subtlety, | |
103 | And let there always be being, |
|
103 | And let there always be being, | |
104 | so we may see their result. |
|
104 | so we may see their result. | |
105 | The two are the same, |
|
105 | The two are the same, | |
106 | But after they are produced, |
|
106 | But after they are produced, | |
107 | they have different names. |
|
107 | they have different names. | |
108 | <<<<<<< LAO |
|
108 | <<<<<<< LAO | |
109 | ======= |
|
109 | ======= | |
110 |
|
110 | |||
111 | -- The Way of Lao-Tzu, tr. Wing-tsit Chan |
|
111 | -- The Way of Lao-Tzu, tr. Wing-tsit Chan | |
112 |
|
112 | |||
113 | >>>>>>> TAO |
|
113 | >>>>>>> TAO | |
114 | """) |
|
114 | """) | |
115 |
|
115 | |||
116 | class TestMerge3(TestCase): |
|
116 | class TestMerge3(TestCase): | |
117 | def log(self, msg): |
|
117 | def log(self, msg): | |
118 | pass |
|
118 | pass | |
119 |
|
119 | |||
120 | def test_no_changes(self): |
|
120 | def test_no_changes(self): | |
121 | """No conflicts because nothing changed""" |
|
121 | """No conflicts because nothing changed""" | |
122 | m3 = Merge3(['aaa', 'bbb'], |
|
122 | m3 = Merge3(['aaa', 'bbb'], | |
123 | ['aaa', 'bbb'], |
|
123 | ['aaa', 'bbb'], | |
124 | ['aaa', 'bbb']) |
|
124 | ['aaa', 'bbb']) | |
125 |
|
125 | |||
126 | self.assertEquals(m3.find_unconflicted(), |
|
126 | self.assertEquals(m3.find_unconflicted(), | |
127 | [(0, 2)]) |
|
127 | [(0, 2)]) | |
128 |
|
128 | |||
129 | self.assertEquals(list(m3.find_sync_regions()), |
|
129 | self.assertEquals(list(m3.find_sync_regions()), | |
130 | [(0, 2, |
|
130 | [(0, 2, | |
131 | 0, 2, |
|
131 | 0, 2, | |
132 | 0, 2), |
|
132 | 0, 2), | |
133 | (2, 2, 2, 2, 2, 2)]) |
|
133 | (2, 2, 2, 2, 2, 2)]) | |
134 |
|
134 | |||
135 | self.assertEquals(list(m3.merge_regions()), |
|
135 | self.assertEquals(list(m3.merge_regions()), | |
136 | [('unchanged', 0, 2)]) |
|
136 | [('unchanged', 0, 2)]) | |
137 |
|
137 | |||
138 | self.assertEquals(list(m3.merge_groups()), |
|
138 | self.assertEquals(list(m3.merge_groups()), | |
139 | [('unchanged', ['aaa', 'bbb'])]) |
|
139 | [('unchanged', ['aaa', 'bbb'])]) | |
140 |
|
140 | |||
141 | def test_front_insert(self): |
|
141 | def test_front_insert(self): | |
142 | m3 = Merge3(['zz'], |
|
142 | m3 = Merge3(['zz'], | |
143 | ['aaa', 'bbb', 'zz'], |
|
143 | ['aaa', 'bbb', 'zz'], | |
144 | ['zz']) |
|
144 | ['zz']) | |
145 |
|
145 | |||
146 | # todo: should use a sentinel at end as from get_matching_blocks |
|
146 | # todo: should use a sentinel at end as from get_matching_blocks | |
147 | # to match without zz |
|
147 | # to match without zz | |
148 | self.assertEquals(list(m3.find_sync_regions()), |
|
148 | self.assertEquals(list(m3.find_sync_regions()), | |
149 | [(0, 1, 2, 3, 0, 1), |
|
149 | [(0, 1, 2, 3, 0, 1), | |
150 | (1, 1, 3, 3, 1, 1)]) |
|
150 | (1, 1, 3, 3, 1, 1)]) | |
151 |
|
151 | |||
152 | self.assertEquals(list(m3.merge_regions()), |
|
152 | self.assertEquals(list(m3.merge_regions()), | |
153 | [('a', 0, 2), |
|
153 | [('a', 0, 2), | |
154 | ('unchanged', 0, 1)]) |
|
154 | ('unchanged', 0, 1)]) | |
155 |
|
155 | |||
156 | self.assertEquals(list(m3.merge_groups()), |
|
156 | self.assertEquals(list(m3.merge_groups()), | |
157 | [('a', ['aaa', 'bbb']), |
|
157 | [('a', ['aaa', 'bbb']), | |
158 | ('unchanged', ['zz'])]) |
|
158 | ('unchanged', ['zz'])]) | |
159 |
|
159 | |||
160 | def test_null_insert(self): |
|
160 | def test_null_insert(self): | |
161 | m3 = Merge3([], |
|
161 | m3 = Merge3([], | |
162 | ['aaa', 'bbb'], |
|
162 | ['aaa', 'bbb'], | |
163 | []) |
|
163 | []) | |
164 | # todo: should use a sentinel at end as from get_matching_blocks |
|
164 | # todo: should use a sentinel at end as from get_matching_blocks | |
165 | # to match without zz |
|
165 | # to match without zz | |
166 | self.assertEquals(list(m3.find_sync_regions()), |
|
166 | self.assertEquals(list(m3.find_sync_regions()), | |
167 | [(0, 0, 2, 2, 0, 0)]) |
|
167 | [(0, 0, 2, 2, 0, 0)]) | |
168 |
|
168 | |||
169 | self.assertEquals(list(m3.merge_regions()), |
|
169 | self.assertEquals(list(m3.merge_regions()), | |
170 | [('a', 0, 2)]) |
|
170 | [('a', 0, 2)]) | |
171 |
|
171 | |||
172 | self.assertEquals(list(m3.merge_lines()), |
|
172 | self.assertEquals(list(m3.merge_lines()), | |
173 | ['aaa', 'bbb']) |
|
173 | ['aaa', 'bbb']) | |
174 |
|
174 | |||
175 | def test_no_conflicts(self): |
|
175 | def test_no_conflicts(self): | |
176 | """No conflicts because only one side changed""" |
|
176 | """No conflicts because only one side changed""" | |
177 | m3 = Merge3(['aaa', 'bbb'], |
|
177 | m3 = Merge3(['aaa', 'bbb'], | |
178 | ['aaa', '111', 'bbb'], |
|
178 | ['aaa', '111', 'bbb'], | |
179 | ['aaa', 'bbb']) |
|
179 | ['aaa', 'bbb']) | |
180 |
|
180 | |||
181 | self.assertEquals(m3.find_unconflicted(), |
|
181 | self.assertEquals(m3.find_unconflicted(), | |
182 | [(0, 1), (1, 2)]) |
|
182 | [(0, 1), (1, 2)]) | |
183 |
|
183 | |||
184 | self.assertEquals(list(m3.find_sync_regions()), |
|
184 | self.assertEquals(list(m3.find_sync_regions()), | |
185 | [(0, 1, 0, 1, 0, 1), |
|
185 | [(0, 1, 0, 1, 0, 1), | |
186 | (1, 2, 2, 3, 1, 2), |
|
186 | (1, 2, 2, 3, 1, 2), | |
187 | (2, 2, 3, 3, 2, 2)]) |
|
187 | (2, 2, 3, 3, 2, 2)]) | |
188 |
|
188 | |||
189 | self.assertEquals(list(m3.merge_regions()), |
|
189 | self.assertEquals(list(m3.merge_regions()), | |
190 | [('unchanged', 0, 1), |
|
190 | [('unchanged', 0, 1), | |
191 | ('a', 1, 2), |
|
191 | ('a', 1, 2), | |
192 | ('unchanged', 1, 2)]) |
|
192 | ('unchanged', 1, 2)]) | |
193 |
|
193 | |||
194 | def test_append_a(self): |
|
194 | def test_append_a(self): | |
195 | m3 = Merge3(['aaa\n', 'bbb\n'], |
|
195 | m3 = Merge3(['aaa\n', 'bbb\n'], | |
196 | ['aaa\n', 'bbb\n', '222\n'], |
|
196 | ['aaa\n', 'bbb\n', '222\n'], | |
197 | ['aaa\n', 'bbb\n']) |
|
197 | ['aaa\n', 'bbb\n']) | |
198 |
|
198 | |||
199 | self.assertEquals(''.join(m3.merge_lines()), |
|
199 | self.assertEquals(''.join(m3.merge_lines()), | |
200 | 'aaa\nbbb\n222\n') |
|
200 | 'aaa\nbbb\n222\n') | |
201 |
|
201 | |||
202 | def test_append_b(self): |
|
202 | def test_append_b(self): | |
203 | m3 = Merge3(['aaa\n', 'bbb\n'], |
|
203 | m3 = Merge3(['aaa\n', 'bbb\n'], | |
204 | ['aaa\n', 'bbb\n'], |
|
204 | ['aaa\n', 'bbb\n'], | |
205 | ['aaa\n', 'bbb\n', '222\n']) |
|
205 | ['aaa\n', 'bbb\n', '222\n']) | |
206 |
|
206 | |||
207 | self.assertEquals(''.join(m3.merge_lines()), |
|
207 | self.assertEquals(''.join(m3.merge_lines()), | |
208 | 'aaa\nbbb\n222\n') |
|
208 | 'aaa\nbbb\n222\n') | |
209 |
|
209 | |||
210 | def test_append_agreement(self): |
|
210 | def test_append_agreement(self): | |
211 | m3 = Merge3(['aaa\n', 'bbb\n'], |
|
211 | m3 = Merge3(['aaa\n', 'bbb\n'], | |
212 | ['aaa\n', 'bbb\n', '222\n'], |
|
212 | ['aaa\n', 'bbb\n', '222\n'], | |
213 | ['aaa\n', 'bbb\n', '222\n']) |
|
213 | ['aaa\n', 'bbb\n', '222\n']) | |
214 |
|
214 | |||
215 | self.assertEquals(''.join(m3.merge_lines()), |
|
215 | self.assertEquals(''.join(m3.merge_lines()), | |
216 | 'aaa\nbbb\n222\n') |
|
216 | 'aaa\nbbb\n222\n') | |
217 |
|
217 | |||
218 | def test_append_clash(self): |
|
218 | def test_append_clash(self): | |
219 | m3 = Merge3(['aaa\n', 'bbb\n'], |
|
219 | m3 = Merge3(['aaa\n', 'bbb\n'], | |
220 | ['aaa\n', 'bbb\n', '222\n'], |
|
220 | ['aaa\n', 'bbb\n', '222\n'], | |
221 | ['aaa\n', 'bbb\n', '333\n']) |
|
221 | ['aaa\n', 'bbb\n', '333\n']) | |
222 |
|
222 | |||
223 | ml = m3.merge_lines(name_a='a', |
|
223 | ml = m3.merge_lines(name_a='a', | |
224 | name_b='b', |
|
224 | name_b='b', | |
225 | start_marker='<<', |
|
225 | start_marker='<<', | |
226 | mid_marker='--', |
|
226 | mid_marker='--', | |
227 | end_marker='>>') |
|
227 | end_marker='>>') | |
228 | self.assertEquals(''.join(ml), |
|
228 | self.assertEquals(''.join(ml), | |
229 | 'aaa\n' |
|
229 | 'aaa\n' | |
230 | 'bbb\n' |
|
230 | 'bbb\n' | |
231 | '<< a\n' |
|
231 | '<< a\n' | |
232 | '222\n' |
|
232 | '222\n' | |
233 | '--\n' |
|
233 | '--\n' | |
234 | '333\n' |
|
234 | '333\n' | |
235 | '>> b\n' |
|
235 | '>> b\n' | |
236 | ) |
|
236 | ) | |
237 |
|
237 | |||
238 | def test_insert_agreement(self): |
|
238 | def test_insert_agreement(self): | |
239 | m3 = Merge3(['aaa\n', 'bbb\n'], |
|
239 | m3 = Merge3(['aaa\n', 'bbb\n'], | |
240 | ['aaa\n', '222\n', 'bbb\n'], |
|
240 | ['aaa\n', '222\n', 'bbb\n'], | |
241 | ['aaa\n', '222\n', 'bbb\n']) |
|
241 | ['aaa\n', '222\n', 'bbb\n']) | |
242 |
|
242 | |||
243 | ml = m3.merge_lines(name_a='a', |
|
243 | ml = m3.merge_lines(name_a='a', | |
244 | name_b='b', |
|
244 | name_b='b', | |
245 | start_marker='<<', |
|
245 | start_marker='<<', | |
246 | mid_marker='--', |
|
246 | mid_marker='--', | |
247 | end_marker='>>') |
|
247 | end_marker='>>') | |
248 | self.assertEquals(''.join(ml), 'aaa\n222\nbbb\n') |
|
248 | self.assertEquals(''.join(ml), 'aaa\n222\nbbb\n') | |
249 |
|
249 | |||
250 |
|
250 | |||
251 | def test_insert_clash(self): |
|
251 | def test_insert_clash(self): | |
252 | """Both try to insert lines in the same place.""" |
|
252 | """Both try to insert lines in the same place.""" | |
253 | m3 = Merge3(['aaa\n', 'bbb\n'], |
|
253 | m3 = Merge3(['aaa\n', 'bbb\n'], | |
254 | ['aaa\n', '111\n', 'bbb\n'], |
|
254 | ['aaa\n', '111\n', 'bbb\n'], | |
255 | ['aaa\n', '222\n', 'bbb\n']) |
|
255 | ['aaa\n', '222\n', 'bbb\n']) | |
256 |
|
256 | |||
257 | self.assertEquals(m3.find_unconflicted(), |
|
257 | self.assertEquals(m3.find_unconflicted(), | |
258 | [(0, 1), (1, 2)]) |
|
258 | [(0, 1), (1, 2)]) | |
259 |
|
259 | |||
260 | self.assertEquals(list(m3.find_sync_regions()), |
|
260 | self.assertEquals(list(m3.find_sync_regions()), | |
261 | [(0, 1, 0, 1, 0, 1), |
|
261 | [(0, 1, 0, 1, 0, 1), | |
262 | (1, 2, 2, 3, 2, 3), |
|
262 | (1, 2, 2, 3, 2, 3), | |
263 | (2, 2, 3, 3, 3, 3)]) |
|
263 | (2, 2, 3, 3, 3, 3)]) | |
264 |
|
264 | |||
265 | self.assertEquals(list(m3.merge_regions()), |
|
265 | self.assertEquals(list(m3.merge_regions()), | |
266 | [('unchanged', 0, 1), |
|
266 | [('unchanged', 0, 1), | |
267 | ('conflict', 1, 1, 1, 2, 1, 2), |
|
267 | ('conflict', 1, 1, 1, 2, 1, 2), | |
268 | ('unchanged', 1, 2)]) |
|
268 | ('unchanged', 1, 2)]) | |
269 |
|
269 | |||
270 | self.assertEquals(list(m3.merge_groups()), |
|
270 | self.assertEquals(list(m3.merge_groups()), | |
271 | [('unchanged', ['aaa\n']), |
|
271 | [('unchanged', ['aaa\n']), | |
272 | ('conflict', [], ['111\n'], ['222\n']), |
|
272 | ('conflict', [], ['111\n'], ['222\n']), | |
273 | ('unchanged', ['bbb\n']), |
|
273 | ('unchanged', ['bbb\n']), | |
274 | ]) |
|
274 | ]) | |
275 |
|
275 | |||
276 | ml = m3.merge_lines(name_a='a', |
|
276 | ml = m3.merge_lines(name_a='a', | |
277 | name_b='b', |
|
277 | name_b='b', | |
278 | start_marker='<<', |
|
278 | start_marker='<<', | |
279 | mid_marker='--', |
|
279 | mid_marker='--', | |
280 | end_marker='>>') |
|
280 | end_marker='>>') | |
281 | self.assertEquals(''.join(ml), |
|
281 | self.assertEquals(''.join(ml), | |
282 | '''aaa |
|
282 | '''aaa | |
283 | << a |
|
283 | << a | |
284 | 111 |
|
284 | 111 | |
285 | -- |
|
285 | -- | |
286 | 222 |
|
286 | 222 | |
287 | >> b |
|
287 | >> b | |
288 | bbb |
|
288 | bbb | |
289 | ''') |
|
289 | ''') | |
290 |
|
290 | |||
291 | def test_replace_clash(self): |
|
291 | def test_replace_clash(self): | |
292 | """Both try to insert lines in the same place.""" |
|
292 | """Both try to insert lines in the same place.""" | |
293 | m3 = Merge3(['aaa', '000', 'bbb'], |
|
293 | m3 = Merge3(['aaa', '000', 'bbb'], | |
294 | ['aaa', '111', 'bbb'], |
|
294 | ['aaa', '111', 'bbb'], | |
295 | ['aaa', '222', 'bbb']) |
|
295 | ['aaa', '222', 'bbb']) | |
296 |
|
296 | |||
297 | self.assertEquals(m3.find_unconflicted(), |
|
297 | self.assertEquals(m3.find_unconflicted(), | |
298 | [(0, 1), (2, 3)]) |
|
298 | [(0, 1), (2, 3)]) | |
299 |
|
299 | |||
300 | self.assertEquals(list(m3.find_sync_regions()), |
|
300 | self.assertEquals(list(m3.find_sync_regions()), | |
301 | [(0, 1, 0, 1, 0, 1), |
|
301 | [(0, 1, 0, 1, 0, 1), | |
302 | (2, 3, 2, 3, 2, 3), |
|
302 | (2, 3, 2, 3, 2, 3), | |
303 | (3, 3, 3, 3, 3, 3)]) |
|
303 | (3, 3, 3, 3, 3, 3)]) | |
304 |
|
304 | |||
305 | def test_replace_multi(self): |
|
305 | def test_replace_multi(self): | |
306 | """Replacement with regions of different size.""" |
|
306 | """Replacement with regions of different size.""" | |
307 | m3 = Merge3(['aaa', '000', '000', 'bbb'], |
|
307 | m3 = Merge3(['aaa', '000', '000', 'bbb'], | |
308 | ['aaa', '111', '111', '111', 'bbb'], |
|
308 | ['aaa', '111', '111', '111', 'bbb'], | |
309 | ['aaa', '222', '222', '222', '222', 'bbb']) |
|
309 | ['aaa', '222', '222', '222', '222', 'bbb']) | |
310 |
|
310 | |||
311 | self.assertEquals(m3.find_unconflicted(), |
|
311 | self.assertEquals(m3.find_unconflicted(), | |
312 | [(0, 1), (3, 4)]) |
|
312 | [(0, 1), (3, 4)]) | |
313 |
|
313 | |||
314 |
|
314 | |||
315 | self.assertEquals(list(m3.find_sync_regions()), |
|
315 | self.assertEquals(list(m3.find_sync_regions()), | |
316 | [(0, 1, 0, 1, 0, 1), |
|
316 | [(0, 1, 0, 1, 0, 1), | |
317 | (3, 4, 4, 5, 5, 6), |
|
317 | (3, 4, 4, 5, 5, 6), | |
318 | (4, 4, 5, 5, 6, 6)]) |
|
318 | (4, 4, 5, 5, 6, 6)]) | |
319 |
|
319 | |||
320 | def test_merge_poem(self): |
|
320 | def test_merge_poem(self): | |
321 | """Test case from diff3 manual""" |
|
321 | """Test case from diff3 manual""" | |
322 | m3 = Merge3(TZU, LAO, TAO) |
|
322 | m3 = Merge3(TZU, LAO, TAO) | |
323 | ml = list(m3.merge_lines('LAO', 'TAO')) |
|
323 | ml = list(m3.merge_lines('LAO', 'TAO')) | |
324 | self.log('merge result:') |
|
324 | self.log('merge result:') | |
325 | self.log(''.join(ml)) |
|
325 | self.log(''.join(ml)) | |
326 | self.assertEquals(ml, MERGED_RESULT) |
|
326 | self.assertEquals(ml, MERGED_RESULT) | |
327 |
|
327 | |||
328 | def test_binary(self): |
|
328 | def test_binary(self): | |
329 |
self.assertRaises(error.Abort |
|
329 | with self.assertRaises(error.Abort): | |
|
330 | Merge3(['\x00'], ['a'], ['b']) | |||
330 |
|
331 | |||
331 | def test_dos_text(self): |
|
332 | def test_dos_text(self): | |
332 | base_text = 'a\r\n' |
|
333 | base_text = 'a\r\n' | |
333 | this_text = 'b\r\n' |
|
334 | this_text = 'b\r\n' | |
334 | other_text = 'c\r\n' |
|
335 | other_text = 'c\r\n' | |
335 | m3 = Merge3(base_text.splitlines(True), other_text.splitlines(True), |
|
336 | m3 = Merge3(base_text.splitlines(True), other_text.splitlines(True), | |
336 | this_text.splitlines(True)) |
|
337 | this_text.splitlines(True)) | |
337 | m_lines = m3.merge_lines('OTHER', 'THIS') |
|
338 | m_lines = m3.merge_lines('OTHER', 'THIS') | |
338 | self.assertEqual('<<<<<<< OTHER\r\nc\r\n=======\r\nb\r\n' |
|
339 | self.assertEqual('<<<<<<< OTHER\r\nc\r\n=======\r\nb\r\n' | |
339 | '>>>>>>> THIS\r\n'.splitlines(True), list(m_lines)) |
|
340 | '>>>>>>> THIS\r\n'.splitlines(True), list(m_lines)) | |
340 |
|
341 | |||
341 | def test_mac_text(self): |
|
342 | def test_mac_text(self): | |
342 | base_text = 'a\r' |
|
343 | base_text = 'a\r' | |
343 | this_text = 'b\r' |
|
344 | this_text = 'b\r' | |
344 | other_text = 'c\r' |
|
345 | other_text = 'c\r' | |
345 | m3 = Merge3(base_text.splitlines(True), other_text.splitlines(True), |
|
346 | m3 = Merge3(base_text.splitlines(True), other_text.splitlines(True), | |
346 | this_text.splitlines(True)) |
|
347 | this_text.splitlines(True)) | |
347 | m_lines = m3.merge_lines('OTHER', 'THIS') |
|
348 | m_lines = m3.merge_lines('OTHER', 'THIS') | |
348 | self.assertEqual('<<<<<<< OTHER\rc\r=======\rb\r' |
|
349 | self.assertEqual('<<<<<<< OTHER\rc\r=======\rb\r' | |
349 | '>>>>>>> THIS\r'.splitlines(True), list(m_lines)) |
|
350 | '>>>>>>> THIS\r'.splitlines(True), list(m_lines)) | |
350 |
|
351 | |||
351 | if __name__ == '__main__': |
|
352 | if __name__ == '__main__': | |
352 | # hide the timer |
|
353 | # hide the timer | |
353 | import time |
|
354 | import time | |
354 | orig = time.time |
|
355 | orig = time.time | |
355 | try: |
|
356 | try: | |
356 | time.time = lambda: 0 |
|
357 | time.time = lambda: 0 | |
357 | unittest.main() |
|
358 | unittest.main() | |
358 | finally: |
|
359 | finally: | |
359 | time.time = orig |
|
360 | time.time = orig | |
360 |
|
361 |
General Comments 0
You need to be logged in to leave comments.
Login now