The result of SF patch #1471578: big-memory tests for strings, lists and

tuples. Lots to be added, still, but this will give big-memory people
something to play with in 2.5 alpha 2, and hopefully get more people to
write these tests.

Lib/test/regrtest.py

@@ -25,6 +25,7 @@ Command line options:
 -N: nocoverdir -- Put coverage files alongside modules
 -L: runleaks   -- run the leaks(1) command just before exit
 -R: huntrleaks -- search for reference leaks (needs debug build, v. slow)
+-M: memlimit   -- run very large memory-consuming tests
 
 If non-option arguments are present, they are names for tests to run,
 unless -x is given, in which case they are names for tests not to run.
@@ -63,6 +64,19 @@ of times further it is run and 'fname' is the name of the file the
 reports are written to.  These parameters all have defaults (5, 4 and
 "reflog.txt" respectively), so the minimal invocation is '-R ::'.
 
+-M runs tests that require an exorbitant amount of memory. These tests
+typically try to ascertain containers keep working when containing more than
+2 bilion objects, and only work on 64-bit systems. The passed-in memlimit,
+which is a string in the form of '2.5Gb', determines howmuch memory the
+tests will limit themselves to (but they may go slightly over.) The number
+shouldn't be more memory than the machine has (including swap memory). You
+should also keep in mind that swap memory is generally much, much slower
+than RAM, and setting memlimit to all available RAM or higher will heavily
+tax the machine. On the other hand, it is no use running these tests with a
+limit of less than 2.5Gb, and many require more than 20Gb. Tests that expect
+to use more than memlimit memory will be skipped. The big-memory tests
+generally run very, very long.
+
 -u is used to specify which special resource intensive tests to run,
 such as those requiring large file support or network connectivity.
 The argument is a comma-separated list of words indicating the
@@ -180,12 +194,12 @@ def main(tests=None, testdir=None, verbose=0, quiet=False, generate=False,
 
     test_support.record_original_stdout(sys.stdout)
     try:
-        opts, args = getopt.getopt(sys.argv[1:], 'hvgqxsrf:lu:t:TD:NLR:w',
+        opts, args = getopt.getopt(sys.argv[1:], 'hvgqxsrf:lu:t:TD:NLR:wM:',
                                    ['help', 'verbose', 'quiet', 'generate',
                                     'exclude', 'single', 'random', 'fromfile',
                                     'findleaks', 'use=', 'threshold=', 'trace',
                                     'coverdir=', 'nocoverdir', 'runleaks',
-                                    'huntrleaks=', 'verbose2',
+                                    'huntrleaks=', 'verbose2', 'memlimit=',
                                     ])
     except getopt.error, msg:
         usage(2, msg)
@@ -241,6 +255,8 @@ def main(tests=None, testdir=None, verbose=0, quiet=False, generate=False,
                 huntrleaks[1] = int(huntrleaks[1])
             if len(huntrleaks[2]) == 0:
                 huntrleaks[2] = "reflog.txt"
+        elif o in ('-M', '--memlimit'):
+            test_support.set_memlimit(a)
         elif o in ('-u', '--use'):
             u = [x.lower() for x in a.split(',')]
             for r in u:

Lib/test/test_bigmem.py

+from test import test_support
+from test.test_support import bigmemtest, _1G, _2G
+
+import unittest
+import operator
+import string
+import sys
+
+# Bigmem testing houserules:
+#
+#  - Try not to allocate too many large objects. It's okay to rely on
+#    refcounting semantics, but don't forget that 's = create_largestring()'
+#    doesn't release the old 's' (if it exists) until well after its new
+#    value has been created. Use 'del s' before the create_largestring call.
+#
+#  - Do *not* compare large objects using assertEquals or similar. It's a
+#    lengty operation and the errormessage will be utterly useless due to
+#    its size. To make sure whether a result has the right contents, better
+#    to use the strip or count methods, or compare meaningful slices.
+#
+#  - Don't forget to test for large indices, offsets and results and such,
+#    in addition to large sizes.
+#
+#  - When repeating an object (say, a substring, or a small list) to create
+#    a large object, make the subobject of a length that is not a power of
+#    2. That way, int-wrapping problems are more easily detected.
+#
+#  - While the bigmemtest decorator speaks of 'minsize', all tests will
+#    actually be called with a much smaller number too, in the normal
+#    test run (5Kb currently.) This is so the tests themselves get frequent
+#    testing Consequently, always make all large allocations based on the
+#    passed-in 'size', and don't rely on the size being very large. Also,
+#    memuse-per-size should remain sane (less than a few thousand); if your
+#    test uses more, adjust 'size' upward, instead.
+
+class StrTest(unittest.TestCase):
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_capitalize(self, size):
+        SUBSTR = ' abc def ghi'
+        s = '-' * size + SUBSTR
+        caps = s.capitalize()
+        self.assertEquals(caps[-len(SUBSTR):],
+                         SUBSTR.capitalize())
+        self.assertEquals(caps.lstrip('-'), SUBSTR)
+
+    @bigmemtest(minsize=_2G + 10, memuse=1)
+    def test_center(self, size):
+        SUBSTR = ' abc def ghi'
+        s = SUBSTR.center(size)
+        self.assertEquals(len(s), size)
+        lpadsize = rpadsize = (len(s) - len(SUBSTR)) // 2
+        if len(s) % 2:
+            lpadsize += 1
+        self.assertEquals(s[lpadsize:-rpadsize], SUBSTR)
+        self.assertEquals(s.strip(), SUBSTR.strip())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_count(self, size):
+        SUBSTR = ' abc def ghi'
+        s = '.' * size + SUBSTR
+        self.assertEquals(s.count('.'), size)
+        s += '.'
+        self.assertEquals(s.count('.'), size + 1)
+        self.assertEquals(s.count(' '), 3)
+        self.assertEquals(s.count('i'), 1)
+        self.assertEquals(s.count('j'), 0)
+
+    @bigmemtest(minsize=0, memuse=1)
+    def test_decode(self, size):
+        pass
+
+    @bigmemtest(minsize=0, memuse=1)
+    def test_encode(self, size):
+        pass
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_endswith(self, size):
+        SUBSTR = ' abc def ghi'
+        s = '-' * size + SUBSTR
+        self.failUnless(s.endswith(SUBSTR))
+        self.failUnless(s.endswith(s))
+        s2 = '...' + s
+        self.failUnless(s2.endswith(s))
+        self.failIf(s.endswith('a' + SUBSTR))
+        self.failIf(SUBSTR.endswith(s))
+
+    @bigmemtest(minsize=_2G + 10, memuse=2)
+    def test_expandtabs(self, size):
+        s = '-' * size
+        tabsize = 8
+        self.assertEquals(s.expandtabs(), s)
+        del s
+        slen, remainder = divmod(size, tabsize)
+        s = '       \t' * slen
+        s = s.expandtabs(tabsize)
+        self.assertEquals(len(s), size - remainder)
+        self.assertEquals(len(s.strip(' ')), 0)
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_find(self, size):
+        SUBSTR = ' abc def ghi'
+        sublen = len(SUBSTR)
+        s = ''.join([SUBSTR, '-' * size, SUBSTR])
+        self.assertEquals(s.find(' '), 0)
+        self.assertEquals(s.find(SUBSTR), 0)
+        self.assertEquals(s.find(' ', sublen), sublen + size)
+        self.assertEquals(s.find(SUBSTR, len(SUBSTR)), sublen + size)
+        self.assertEquals(s.find('i'), SUBSTR.find('i'))
+        self.assertEquals(s.find('i', sublen),
+                         sublen + size + SUBSTR.find('i'))
+        self.assertEquals(s.find('i', size),
+                         sublen + size + SUBSTR.find('i'))
+        self.assertEquals(s.find('j'), -1)
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_index(self, size):
+        SUBSTR = ' abc def ghi'
+        sublen = len(SUBSTR)
+        s = ''.join([SUBSTR, '-' * size, SUBSTR])
+        self.assertEquals(s.index(' '), 0)
+        self.assertEquals(s.index(SUBSTR), 0)
+        self.assertEquals(s.index(' ', sublen), sublen + size)
+        self.assertEquals(s.index(SUBSTR, sublen), sublen + size)
+        self.assertEquals(s.index('i'), SUBSTR.index('i'))
+        self.assertEquals(s.index('i', sublen),
+                         sublen + size + SUBSTR.index('i'))
+        self.assertEquals(s.index('i', size),
+                         sublen + size + SUBSTR.index('i'))
+        self.assertRaises(ValueError, s.index, 'j')
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_isalnum(self, size):
+        SUBSTR = '123456'
+        s = 'a' * size + SUBSTR
+        self.failUnless(s.isalnum())
+        s += '.'
+        self.failIf(s.isalnum())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_isalpha(self, size):
+        SUBSTR = 'zzzzzzz'
+        s = 'a' * size + SUBSTR
+        self.failUnless(s.isalpha())
+        s += '.'
+        self.failIf(s.isalpha())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_isdigit(self, size):
+        SUBSTR = '123456'
+        s = '9' * size + SUBSTR
+        self.failUnless(s.isdigit())
+        s += 'z'
+        self.failIf(s.isdigit())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_islower(self, size):
+        chars = ''.join([ chr(c) for c in range(255) if not chr(c).isupper() ])
+        repeats = size // len(chars) + 2
+        s = chars * repeats
+        self.failUnless(s.islower())
+        s += 'A'
+        self.failIf(s.islower())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_isspace(self, size):
+        whitespace = ' \f\n\r\t\v'
+        repeats = size // len(whitespace) + 2
+        s = whitespace * repeats
+        self.failUnless(s.isspace())
+        s += 'j'
+        self.failIf(s.isspace())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_istitle(self, size):
+        SUBSTR = '123456'
+        s = ''.join(['A', 'a' * size, SUBSTR])
+        self.failUnless(s.istitle())
+        s += 'A'
+        self.failUnless(s.istitle())
+        s += 'aA'
+        self.failIf(s.istitle())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_isupper(self, size):
+        chars = ''.join([ chr(c) for c in range(255) if not chr(c).islower() ])
+        repeats = size // len(chars) + 2
+        s = chars * repeats
+        self.failUnless(s.isupper())
+        s += 'a'
+        self.failIf(s.isupper())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_join(self, size):
+        s = 'A' * size
+        x = s.join(['aaaaa', 'bbbbb'])
+        self.assertEquals(x.count('a'), 5)
+        self.assertEquals(x.count('b'), 5)
+        self.failUnless(x.startswith('aaaaaA'))
+        self.failUnless(x.endswith('Abbbbb'))
+
+    @bigmemtest(minsize=_2G + 10, memuse=1)
+    def test_ljust(self, size):
+        SUBSTR = ' abc def ghi'
+        s = SUBSTR.ljust(size)
+        self.failUnless(s.startswith(SUBSTR + '  '))
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.strip(), SUBSTR.strip())
+
+    @bigmemtest(minsize=_2G + 10, memuse=2)
+    def test_lower(self, size):
+        s = 'A' * size
+        s = s.lower()
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.count('a'), size)
+
+    @bigmemtest(minsize=_2G + 10, memuse=1)
+    def test_lstrip(self, size):
+        SUBSTR = 'abc def ghi'
+        s = SUBSTR.rjust(size)
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.lstrip(), SUBSTR.lstrip())
+        del s
+        s = SUBSTR.ljust(size)
+        self.assertEquals(len(s), size)
+        stripped = s.lstrip()
+        self.failUnless(stripped is s)
+
+    @bigmemtest(minsize=_2G + 10, memuse=2)
+    def test_replace(self, size):
+        replacement = 'a'
+        s = ' ' * size
+        s = s.replace(' ', replacement)
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.count(replacement), size)
+        s = s.replace(replacement, ' ', size - 4)
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.count(replacement), 4)
+        self.assertEquals(s[-10:], '      aaaa')
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_rfind(self, size):
+        SUBSTR = ' abc def ghi'
+        sublen = len(SUBSTR)
+        s = ''.join([SUBSTR, '-' * size, SUBSTR])
+        self.assertEquals(s.rfind(' '), sublen + size + SUBSTR.rfind(' '))
+        self.assertEquals(s.rfind(SUBSTR), sublen + size)
+        self.assertEquals(s.rfind(' ', 0, size), SUBSTR.rfind(' '))
+        self.assertEquals(s.rfind(SUBSTR, 0, sublen + size), 0)
+        self.assertEquals(s.rfind('i'), sublen + size + SUBSTR.rfind('i'))
+        self.assertEquals(s.rfind('i', 0, sublen), SUBSTR.rfind('i'))
+        self.assertEquals(s.rfind('i', 0, sublen + size),
+                          SUBSTR.rfind('i'))
+        self.assertEquals(s.rfind('j'), -1)
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_rindex(self, size):
+        SUBSTR = ' abc def ghi'
+        sublen = len(SUBSTR)
+        s = ''.join([SUBSTR, '-' * size, SUBSTR])
+        self.assertEquals(s.rindex(' '),
+                          sublen + size + SUBSTR.rindex(' '))
+        self.assertEquals(s.rindex(SUBSTR), sublen + size)
+        self.assertEquals(s.rindex(' ', 0, sublen + size - 1),
+                          SUBSTR.rindex(' '))
+        self.assertEquals(s.rindex(SUBSTR, 0, sublen + size), 0)
+        self.assertEquals(s.rindex('i'),
+                          sublen + size + SUBSTR.rindex('i'))
+        self.assertEquals(s.rindex('i', 0, sublen), SUBSTR.rindex('i'))
+        self.assertEquals(s.rindex('i', 0, sublen + size),
+                          SUBSTR.rindex('i'))
+        self.assertRaises(ValueError, s.rindex, 'j')
+
+    @bigmemtest(minsize=_2G + 10, memuse=1)
+    def test_rjust(self, size):
+        SUBSTR = ' abc def ghi'
+        s = SUBSTR.ljust(size)
+        self.failUnless(s.startswith(SUBSTR + '  '))
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.strip(), SUBSTR.strip())
+
+    @bigmemtest(minsize=_2G + 10, memuse=1)
+    def test_rstrip(self, size):
+        SUBSTR = ' abc def ghi'
+        s = SUBSTR.ljust(size)
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.rstrip(), SUBSTR.rstrip())
+        del s
+        s = SUBSTR.rjust(size)
+        self.assertEquals(len(s), size)
+        stripped = s.rstrip()
+        self.failUnless(stripped is s)
+
+    # The test takes about size bytes to build a string, and then about
+    # sqrt(size) substrings of sqrt(size) in size and a list to
+    # hold sqrt(size) items. It's close but just over 2x size.
+    @bigmemtest(minsize=_2G, memuse=2.1)
+    def test_split_small(self, size):
+        # Crudely calculate an estimate so that the result of s.split won't
+        # take up an inordinate amount of memory
+        chunksize = int(size ** 0.5 + 2)
+        SUBSTR = 'a' + ' ' * chunksize
+        s = SUBSTR * chunksize
+        l = s.split()
+        self.assertEquals(len(l), chunksize)
+        self.assertEquals(set(l), set(['a']))
+        del l
+        l = s.split('a')
+        self.assertEquals(len(l), chunksize + 1)
+        self.assertEquals(set(l), set(['', ' ' * chunksize]))
+
+    # Allocates a string of twice size (and briefly two) and a list of
+    # size.  Because of internal affairs, the s.split() call produces a
+    # list of size times the same one-character string, so we only
+    # suffer for the list size. (Otherwise, it'd cost another 48 times
+    # size in bytes!) Nevertheless, a list of size takes
+    # 8*size bytes.
+    @bigmemtest(minsize=_2G + 5, memuse=10)
+    def test_split_large(self, size):
+        s = ' a' * size + ' '
+        l = s.split()
+        self.assertEquals(len(l), size)
+        self.assertEquals(set(l), set(['a']))
+        del l
+        l = s.split('a')
+        self.assertEquals(len(l), size + 1)
+        self.assertEquals(set(l), set([' ']))
+
+    @bigmemtest(minsize=_2G, memuse=2.1)
+    def test_splitlines(self, size):
+        # Crudely calculate an estimate so that the result of s.split won't
+        # take up an inordinate amount of memory
+        chunksize = int(size ** 0.5 + 2) // 2
+        SUBSTR = ' ' * chunksize + '\n' + ' ' * chunksize + '\r\n'
+        s = SUBSTR * chunksize
+        l = s.splitlines()
+        self.assertEquals(len(l), chunksize * 2)
+        self.assertEquals(set(l), set([' ' * chunksize]))
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_startswith(self, size):
+        SUBSTR = ' abc def ghi'
+        s = '-' * size + SUBSTR
+        self.failUnless(s.startswith(s))
+        self.failUnless(s.startswith('-' * size))
+        self.failIf(s.startswith(SUBSTR))
+
+    @bigmemtest(minsize=_2G, memuse=1)
+    def test_strip(self, size):
+        SUBSTR = '   abc def ghi   '
+        s = SUBSTR.rjust(size)
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.strip(), SUBSTR.strip())
+        del s
+        s = SUBSTR.ljust(size)
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.strip(), SUBSTR.strip())
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_swapcase(self, size):
+        SUBSTR = "aBcDeFG12.'\xa9\x00"
+        sublen = len(SUBSTR)
+        repeats = size // sublen + 2
+        s = SUBSTR * repeats
+        s = s.swapcase()
+        self.assertEquals(len(s), sublen * repeats)
+        self.assertEquals(s[:sublen * 3], SUBSTR.swapcase() * 3)
+        self.assertEquals(s[-sublen * 3:], SUBSTR.swapcase() * 3)
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_title(self, size):
+        SUBSTR = 'SpaaHAaaAaham'
+        s = SUBSTR * (size // len(SUBSTR) + 2)
+        s = s.title()
+        self.failUnless(s.startswith((SUBSTR * 3).title()))
+        self.failUnless(s.endswith(SUBSTR.lower() * 3))
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_translate(self, size):
+        trans = string.maketrans('.aZ', '-!$')
+        SUBSTR = 'aZz.z.Aaz.'
+        sublen = len(SUBSTR)
+        repeats = size // sublen + 2
+        s = SUBSTR * repeats
+        s = s.translate(trans)
+        self.assertEquals(len(s), repeats * sublen)
+        self.assertEquals(s[:sublen], SUBSTR.translate(trans))
+        self.assertEquals(s[-sublen:], SUBSTR.translate(trans))
+        self.assertEquals(s.count('.'), 0)
+        self.assertEquals(s.count('!'), repeats * 2)
+        self.assertEquals(s.count('z'), repeats * 3)
+
+    @bigmemtest(minsize=_2G + 5, memuse=2)
+    def test_upper(self, size):
+        s = 'a' * size
+        s = s.upper()
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.count('A'), size)
+
+    @bigmemtest(minsize=_2G + 20, memuse=1)
+    def test_zfill(self, size):
+        SUBSTR = '-568324723598234'
+        s = SUBSTR.zfill(size)
+        self.failUnless(s.endswith('0' + SUBSTR[1:]))
+        self.failUnless(s.startswith('-0'))
+        self.assertEquals(len(s), size)
+        self.assertEquals(s.count('0'), size - len(SUBSTR))
+
+    @bigmemtest(minsize=_2G + 10, memuse=2)
+    def test_format(self, size):
+        s = '-' * size
+        sf = '%s' % (s,)
+        self.failUnless(s == sf)
+        del sf
+        sf = '..%s..' % (s,)
+        self.assertEquals(len(sf), len(s) + 4)
+        self.failUnless(sf.startswith('..-'))
+        self.failUnless(sf.endswith('-..'))
+        del s, sf
+
+        size = (size // 2)
+        edge = '-' * size
+        s = ''.join([edge, '%s', edge])
+        del edge
+        s = s % '...'
+        self.assertEquals(len(s), size * 2 + 3)
+        self.assertEquals(s.count('.'), 3)
+        self.assertEquals(s.count('-'), size * 2)
+
+    @bigmemtest(minsize=_2G + 10, memuse=2)
+    def test_repr_small(self, size):
+        s = '-' * size
+        s = repr(s)
+        self.assertEquals(len(s), size + 2)
+        self.assertEquals(s[0], "'")
+        self.assertEquals(s[-1], "'")
+        self.assertEquals(s.count('-'), size)
+        del s
+        # repr() will create a string four times as large as this 'binary
+        # string', but we don't want to allocate much more than twice
+        # size in total.  (We do extra testing in test_repr_large())
+        size = size // 5 * 2
+        s = '\x00' * size
+        s = repr(s)
+        self.assertEquals(len(s), size * 4 + 2)
+        self.assertEquals(s[0], "'")
+        self.assertEquals(s[-1], "'")
+        self.assertEquals(s.count('\\'), size)
+        self.assertEquals(s.count('0'), size * 2)
+
+    @bigmemtest(minsize=_2G + 10, memuse=5)
+    def test_repr_large(self, size):
+        s = '\x00' * size
+        s = repr(s)
+        self.assertEquals(len(s), size * 4 + 2)
+        self.assertEquals(s[0], "'")
+        self.assertEquals(s[-1], "'")
+        self.assertEquals(s.count('\\'), size)
+        self.assertEquals(s.count('0'), size * 2)
+
+    # This test is meaningful even with size < 2G, as long as the
+    # doubled string is > 2G (but it tests more if both are > 2G :)
+    @bigmemtest(minsize=_1G + 2, memuse=3)
+    def test_concat(self, size):
+        s = '.' * size
+        self.assertEquals(len(s), size)
+        s = s + s
+        self.assertEquals(len(s), size * 2)
+        self.assertEquals(s.count('.'), size * 2)
+
+    # This test is meaningful even with size < 2G, as long as the
+    # repeated string is > 2G (but it tests more if both are > 2G :)
+    @bigmemtest(minsize=_1G + 2, memuse=3)
+    def test_repeat(self, size):
+        s = '.' * size
+        self.assertEquals(len(s), size)
+        s = s * 2
+        self.assertEquals(len(s), size * 2)
+        self.assertEquals(s.count('.'), size * 2)
+
+    @bigmemtest(minsize=_2G + 20, memuse=1)
+    def test_slice_and_getitem(self, size):
+        SUBSTR = '0123456789'
+        sublen = len(SUBSTR)
+        s = SUBSTR * (size // sublen)
+        stepsize = len(s) // 100
+        stepsize = stepsize - (stepsize % sublen)
+        for i in range(0, len(s) - stepsize, stepsize):
+            self.assertEquals(s[i], SUBSTR[0])
+            self.assertEquals(s[i:i + sublen], SUBSTR)
+            self.assertEquals(s[i:i + sublen:2], SUBSTR[::2])
+            if i > 0:
+                self.assertEquals(s[i + sublen - 1:i - 1:-3],
+                                  SUBSTR[sublen::-3])
+        # Make sure we do some slicing and indexing near the end of the
+        # string, too.
+        self.assertEquals(s[len(s) - 1], SUBSTR[-1])
+        self.assertEquals(s[-1], SUBSTR[-1])
+        self.assertEquals(s[len(s) - 10], SUBSTR[0])
+        self.assertEquals(s[-sublen], SUBSTR[0])
+        self.assertEquals(s[len(s):], '')
+        self.assertEquals(s[len(s) - 1:], SUBSTR[-1])
+        self.assertEquals(s[-1:], SUBSTR[-1])
+        self.assertEquals(s[len(s) - sublen:], SUBSTR)
+        self.assertEquals(s[-sublen:], SUBSTR)
+        self.assertEquals(len(s[:]), len(s))
+        self.assertEquals(len(s[:len(s) - 5]), len(s) - 5)
+        self.assertEquals(len(s[5:-5]), len(s) - 10)
+
+        self.assertRaises(IndexError, operator.getitem, s, len(s))
+        self.assertRaises(IndexError, operator.getitem, s, len(s) + 1)
+        self.assertRaises(IndexError, operator.getitem, s, len(s) + 1<<31)
+
+    @bigmemtest(minsize=_2G, memuse=2)
+    def test_contains(self, size):
+        SUBSTR = '0123456789'
+        edge = '-' * (size // 2)
+        s = ''.join([edge, SUBSTR, edge])
+        del edge
+        self.failUnless(SUBSTR in s)
+        self.failIf(SUBSTR * 2 in s)
+        self.failUnless('-' in s)
+        self.failIf('a' in s)
+        s += 'a'
+        self.failUnless('a' in s)
+
+    @bigmemtest(minsize=_2G + 10, memuse=2)
+    def test_compare(self, size):
+        s1 = '-' * size
+        s2 = '-' * size
+        self.failUnless(s1 == s2)
+        del s2
+        s2 = s1 + 'a'
+        self.failIf(s1 == s2)
+        del s2
+        s2 = '.' * size
+        self.failIf(s1 == s2)
+
+    @bigmemtest(minsize=_2G + 10, memuse=1)
+    def test_hash(self, size):
+        # Not sure if we can do any meaningful tests here...  Even if we
+        # start relying on the exact algorithm used, the result will be
+        # different depending on the size of the C 'long int'.  Even this
+        # test is dodgy (there's no *guarantee* that the two things should
+        # have a different hash, even if they, in the current
+        # implementation, almost always do.)
+        s = '\x00' * size
+        h1 = hash(s)
+        del s
+        s = '\x00' * (size + 1)
+        self.failIf(h1 == hash(s))
+
+class TupleTest(unittest.TestCase):
+
+    # Tuples have a small, fixed-sized head and an array of pointers to
+    # data.  Since we're testing 64-bit addressing, we can assume that the
+    # pointers are 8 bytes, and that thus that the tuples take up 8 bytes
+    # per size.
+
+    # As a side-effect of testing long tuples, these tests happen to test
+    # having more than 2<<31 references to any given object. Hence the
+    # use of different types of objects as contents in different tests.
+
+    @bigmemtest(minsize=_2G + 2, memuse=16)
+    def test_compare(self, size):
+        t1 = (u'',) * size
+        t2 = (u'',) * size
+        self.failUnless(t1 == t2)
+        del t2
+        t2 = (u'',) * (size + 1)
+        self.failIf(t1 == t2)
+        del t2
+        t2 = (1,) * size
+        self.failIf(t1 == t2)
+
+    # Test concatenating into a single tuple of more than 2G in length,
+    # and concatenating a tuple of more than 2G in length separately, so
+    # the smaller test still gets run even if there isn't memory for the
+    # larger test (but we still let the tester know the larger test is
+    # skipped, in verbose mode.)
+    def basic_concat_test(self, size):
+        t = ((),) * size
+        self.assertEquals(len(t), size)
+        t = t + t
+        self.assertEquals(len(t), size * 2)
+
+    @bigmemtest(minsize=_2G // 2 + 2, memuse=24)
+    def test_concat_small(self, size):
+        return self.basic_concat_test(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=24)
+    def test_concat_large(self, size):
+        return self.basic_concat_test(size)
+
+    @bigmemtest(minsize=_2G // 5 + 10, memuse=8*5)
+    def test_contains(self, size):
+        t = (1, 2, 3, 4, 5) * size
+        self.assertEquals(len(t), size * 5)
+        self.failUnless(5 in t)
+        self.failIf((1, 2, 3, 4, 5) in t)
+        self.failIf(0 in t)
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_hash(self, size):
+        t1 = (0,) * size
+        h1 = hash(t1)
+        del t1
+        t2 = (0,) * (size + 1)
+        self.failIf(h1 == hash(t2))
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_index_and_slice(self, size):
+        t = (None,) * size
+        self.assertEquals(len(t), size)
+        self.assertEquals(t[-1], None)
+        self.assertEquals(t[5], None)
+        self.assertEquals(t[size - 1], None)
+        self.assertRaises(IndexError, operator.getitem, t, size)
+        self.assertEquals(t[:5], (None,) * 5)
+        self.assertEquals(t[-5:], (None,) * 5)
+        self.assertEquals(t[20:25], (None,) * 5)
+        self.assertEquals(t[-25:-20], (None,) * 5)
+        self.assertEquals(t[size - 5:], (None,) * 5)
+        self.assertEquals(t[size - 5:size], (None,) * 5)
+        self.assertEquals(t[size - 6:size - 2], (None,) * 4)
+        self.assertEquals(t[size:size], ())
+        self.assertEquals(t[size:size+5], ())
+
+    # Like test_concat, split in two.
+    def basic_test_repeat(self, size):
+        t = ('',) * size
+        self.assertEquals(len(t), size)
+        t = t * 2
+        self.assertEquals(len(t), size * 2)
+
+    @bigmemtest(minsize=_2G // 2 + 2, memuse=24)
+    def test_repeat_small(self, size):
+        return self.basic_test_repeat(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=24)
+    def test_repeat_large(self, size):
+        return self.basic_test_repeat(size)
+
+    # Like test_concat, split in two.
+    def basic_test_repr(self, size):
+        t = (0,) * size
+        s = repr(t)
+        # The repr of a tuple of 0's is exactly three times the tuple length.
+        self.assertEquals(len(s), size * 3)
+        self.assertEquals(s[:5], '(0, 0')
+        self.assertEquals(s[-5:], '0, 0)')
+        self.assertEquals(s.count('0'), size)
+
+    @bigmemtest(minsize=_2G // 3 + 2, memuse=8+3)
+    def test_repr_small(self, size):
+        return self.basic_test_repr(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=8+3)
+    def test_repr_large(self, size):
+        return self.basic_test_repr(size)
+
+class ListTest(unittest.TestCase):
+
+    # Like tuples, lists have a small, fixed-sized head and an array of
+    # pointers to data, so 8 bytes per size. Also like tuples, we make the
+    # lists hold references to various objects to test their refcount
+    # limits.
+
+    @bigmemtest(minsize=_2G + 2, memuse=16)
+    def test_compare(self, size):
+        l1 = [u''] * size
+        l2 = [u''] * size
+        self.failUnless(l1 == l2)
+        del l2
+        l2 = [u''] * (size + 1)
+        self.failIf(l1 == l2)
+        del l2
+        l2 = [2] * size
+        self.failIf(l1 == l2)
+
+    # Test concatenating into a single list of more than 2G in length,
+    # and concatenating a list of more than 2G in length separately, so
+    # the smaller test still gets run even if there isn't memory for the
+    # larger test (but we still let the tester know the larger test is
+    # skipped, in verbose mode.)
+    def basic_test_concat(self, size):
+        l = [[]] * size
+        self.assertEquals(len(l), size)
+        l = l + l
+        self.assertEquals(len(l), size * 2)
+
+    @bigmemtest(minsize=_2G // 2 + 2, memuse=24)
+    def test_concat_small(self, size):
+        return self.basic_test_concat(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=24)
+    def test_concat_large(self, size):
+        return self.basic_test_concat(size)
+
+    @bigmemtest(minsize=_2G // 5 + 10, memuse=8*5)
+    def test_contains(self, size):
+        l = [1, 2, 3, 4, 5] * size
+        self.assertEquals(len(l), size * 5)
+        self.failUnless(5 in l)
+        self.failIf([1, 2, 3, 4, 5] in l)
+        self.failIf(0 in l)
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_hash(self, size):
+        l = [0] * size
+        self.failUnlessRaises(TypeError, hash, l)
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_index_and_slice(self, size):
+        l = [None] * size
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-1], None)
+        self.assertEquals(l[5], None)
+        self.assertEquals(l[size - 1], None)
+        self.assertRaises(IndexError, operator.getitem, l, size)
+        self.assertEquals(l[:5], [None] * 5)
+        self.assertEquals(l[-5:], [None] * 5)
+        self.assertEquals(l[20:25], [None] * 5)
+        self.assertEquals(l[-25:-20], [None] * 5)
+        self.assertEquals(l[size - 5:], [None] * 5)
+        self.assertEquals(l[size - 5:size], [None] * 5)
+        self.assertEquals(l[size - 6:size - 2], [None] * 4)
+        self.assertEquals(l[size:size], [])
+        self.assertEquals(l[size:size+5], [])
+        
+        l[size - 2] = 5
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-3:], [None, 5, None])
+        self.assertEquals(l.count(5), 1)
+        self.assertRaises(IndexError, operator.setitem, l, size, 6)
+        self.assertEquals(len(l), size)
+        
+        l[size - 7:] = [1, 2, 3, 4, 5]
+        size -= 2
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-7:], [None, None, 1, 2, 3, 4, 5])
+        
+        l[:7] = [1, 2, 3, 4, 5]
+        size -= 2
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[:7], [1, 2, 3, 4, 5, None, None])
+        
+        del l[size - 1]
+        size -= 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-1], 4)
+        
+        del l[-2:]
+        size -= 2
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-1], 2)
+        
+        del l[0]
+        size -= 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[0], 2)
+        
+        del l[:2]
+        size -= 2
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[0], 4)
+
+    # Like test_concat, split in two.
+    def basic_test_repeat(self, size):
+        l = [] * size
+        self.failIf(l)
+        l = [''] * size
+        self.assertEquals(len(l), size)
+        l = l * 2
+        self.assertEquals(len(l), size * 2)
+
+    @bigmemtest(minsize=_2G // 2 + 2, memuse=24)
+    def test_repeat_small(self, size):
+        return self.basic_test_repeat(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=24)
+    def test_repeat_large(self, size):
+        return self.basic_test_repeat(size)
+
+    # Test repr-result of >2G
+    def basic_test_repr(self, size):
+        l = [0] * size
+        s = repr(l)
+        # The repr of a list of 0's is exactly three times the list length.
+        self.assertEquals(len(s), size * 3)
+        self.assertEquals(s[:5], '[0, 0')
+        self.assertEquals(s[-5:], '0, 0]')
+        self.assertEquals(s.count('0'), size)
+
+    @bigmemtest(minsize=_2G // 3 + 2, memuse=8 + 3)
+    def test_repr_small(self, size):
+        return self.basic_test_repr(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=8 + 3)
+    def test_repr_large(self, size):
+        return self.basic_test_repr(size)
+
+
+    @bigmemtest(minsize=_2G, memuse=8)
+    def test_append(self, size):
+        l = [object()] * size
+        l.append(object())
+        self.assertEquals(len(l), size+1)
+        self.failUnless(l[-3] is l[-2])
+        self.failIf(l[-2] is l[-1])
+
+    @bigmemtest(minsize=_2G // 5 + 2, memuse=8 * 5)
+    def test_count(self, size):
+        l = [1, 2, 3, 4, 5] * size
+        self.assertEquals(l.count(1), size)
+        self.assertEquals(l.count("1"), 0)
+
+    def basic_test_extend(self, size):
+        l = [file] * size
+        l.extend(l)
+        self.assertEquals(len(l), size * 2)
+        self.failUnless(l[0] is l[-1])
+        self.failUnless(l[size - 1] is l[size + 1])
+
+    @bigmemtest(minsize=_2G // 2 + 2, memuse=8)
+    def test_extend_small(self, size):
+        return self.basic_test_extend(size)
+
+    @bigmemtest(minsize=_2G + 2, memuse=8)
+    def test_extend_large(self, size):
+        return self.basic_test_extend(size)
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_index(self, size):
+        l = [1L, 2L, 3L, 4L, 5L] * (size // 5)
+        self.assertEquals(l.index(1), 0)
+        self.assertEquals(l.index(5, size - 5), size - 1)
+        self.assertEquals(l.index(5, size - 5, size), size - 1)
+        self.assertRaises(ValueError, l.index, 1, size - 4, size)
+        self.assertRaises(ValueError, l.index, 6L)
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_insert(self, size):
+        l = [1.0] * size
+        l.insert(size - 1, "A")
+        size += 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-3:], [1.0, "A", 1.0])
+
+        l.insert(size + 1, "B")
+        size += 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-3:], ["A", 1.0, "B"])
+
+        l.insert(1, "C")
+        size += 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[:3], [1.0, "C", 1.0])
+        self.assertEquals(l[size - 3:], ["A", 1.0, "B"])
+
+    @bigmemtest(minsize=_2G + 20, memuse=8)
+    def test_pop(self, size):
+        l = [u"a", u"b", u"c", u"d", u"e"] * (size // 5)
+        self.assertEquals(len(l), size)
+
+        item = l.pop()
+        size -= 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(item, u"e")
+        
+        item = l.pop(0)
+        size -= 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(item, u"a")
+        
+        item = l.pop(size - 2)
+        size -= 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(item, u"c")
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_remove(self, size):
+        l = [10] * size
+        self.assertEquals(len(l), size)
+
+        l.remove(10)
+        size -= 1
+        self.assertEquals(len(l), size)
+
+        l.append(5)
+        size += 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-2:], [10, 5])
+        l.remove(5)
+        size -= 1
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-2:], [10, 10])
+        
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_reverse(self, size):
+        l = [1, 2, 3, 4, 5] * (size // 5)
+        l.reverse()
+        self.assertEquals(len(l), size)
+        self.assertEquals(l[-5:], [5, 4, 3, 2, 1])
+        self.assertEquals(l[:5], [5, 4, 3, 2, 1])
+
+    @bigmemtest(minsize=_2G + 10, memuse=8)
+    def test_sort(self, size):
+        l = [1, 2, 3, 4, 5] * (size // 5)
+        l.sort()
+        self.assertEquals(len(l), size)
+        self.assertEquals(l.count(1), size // 5)
+        self.assertEquals(l[:10], [1] * 10)
+        self.assertEquals(l[-10:], [5] * 10)
+        
+def test_main():
+    test_support.run_unittest(StrTest, TupleTest, ListTest)
+
+if __name__ == '__main__':
+    if len(sys.argv) > 1:
+        test_support.set_memlimit(sys.argv[1])
+    test_main()

Lib/test/test_support.py

@@ -31,6 +31,8 @@ class ResourceDenied(TestSkipped):
 
 verbose = 1              # Flag set to 0 by regrtest.py
 use_resources = None     # Flag set to [] by regrtest.py
+max_memuse = 0           # Disable bigmem tests (they will still be run with
+                         # small sizes, to make sure they work.)
 
 # _original_stdout is meant to hold stdout at the time regrtest began.
 # This may be "the real" stdout, or IDLE's emulation of stdout, or whatever.
@@ -249,6 +251,72 @@ def open_urlresource(url):
     fn, _ = urllib.urlretrieve(url, filename)
     return open(fn)
 
+#=======================================================================
+# Big-memory-test support. Separate from 'resources' because memory use should be configurable.
+
+# Some handy shorthands. Note that these are used for byte-limits as well
+# as size-limits, in the various bigmem tests
+_1M = 1024*1024
+_1G = 1024 * _1M
+_2G = 2 * _1G
+
+def set_memlimit(limit):
+    import re
+    global max_memuse
+    sizes = {
+        'k': 1024,
+        'm': _1M,
+        'g': _1G,
+        't': 1024*_1G,
+    }
+    m = re.match(r'(\d+(\.\d+)?) (K|M|G|T)b?$', limit,
+                 re.IGNORECASE | re.VERBOSE)
+    if m is None:
+        raise ValueError('Invalid memory limit %r' % (limit,))
+    memlimit = int(float(m.group(1)) * sizes[m.group(3).lower()])
+    if memlimit < 2.5*_1G:
+        raise ValueError('Memory limit %r too low to be useful' % (limit,))
+    max_memuse = memlimit
+
+def bigmemtest(minsize, memuse, overhead=5*_1M):
+    """Decorator for bigmem tests.
+
+    'minsize' is the minimum useful size for the test (in arbitrary,
+    test-interpreted units.) 'memuse' is the number of 'bytes per size' for
+    the test, or a good estimate of it. 'overhead' specifies fixed overhead,
+    independant of the testsize, and defaults to 5Mb.
+    
+    The decorator tries to guess a good value for 'size' and passes it to
+    the decorated test function. If minsize * memuse is more than the
+    allowed memory use (as defined by max_memuse), the test is skipped.
+    Otherwise, minsize is adjusted upward to use up to max_memuse.
+    """
+    def decorator(f):
+        def wrapper(self):
+            if not max_memuse:
+                # If max_memuse is 0 (the default),
+                # we still want to run the tests with size set to a few kb,
+                # to make sure they work. We still want to avoid using
+                # too much memory, though, but we do that noisily.
+                maxsize = 1024*5
+                self.failIf(maxsize * memuse + overhead > 20 * _1M)
+            else:
+                maxsize = int((max_memuse - overhead) / memuse)
+                if maxsize < minsize:
+                    # Really ought to print 'test skipped' or something
+                    if verbose:
+                        sys.stderr.write("Skipping %s because of memory "
+                                         "constraint\n" % (f.__name__,))
+                    return
+                # Try to keep some breathing room in memory use
+                maxsize = max(maxsize - 50 * _1M, minsize)
+            return f(self, maxsize)
+        wrapper.minsize = minsize
+        wrapper.memuse = memuse
+        wrapper.overhead = overhead
+        return wrapper
+    return decorator
+
 #=======================================================================
 # Preliminary PyUNIT integration.
 

Misc/NEWS

@@ -171,6 +171,11 @@ Tests
   it separately and by hand.  It also wasn't cleaning up its changes to
   the current Decimal context.
 
+- regrtest.py now has a -M option to run tests that test the new limits of
+  containers, on 64-bit architectures. Running these tests is only sensible
+  on 64-bit machines with more than two gigabytes of memory. The argument
+  passed is the maximum amount of memory for the tests to use.
+
 Tools
 -----