Пример #1
0
 def new_buffer(self):
     buffer = os.urandom(
         self.fill_size) + bytes(self.settings.buffer_size - self.fill_size)
     self.last_data = TestData(buffer)
     self.test_function(self.last_data)
     self.last_data.freeze()
     self.update_fill_size()
Пример #2
0
 def incorporate_new_buffer(self, buffer: bytes) -> bool:
     if buffer[: self.last_data.index] == self.last_data.buffer[: self.last_data.index]:
         return False
     data = TestData(buffer)
     self.test_function(data)
     data.freeze()
     if self.consider_new_test_data(data):
         if self.last_data.status == Status.INTERESTING:
             self.shrinks += 1
         self.last_data = data
         self.changed += 1
         if self.shrinks >= self.settings.max_shrinks:
             raise StopShrinking()
         return True
     return False
Пример #3
0
def find(draw, check, settings=None):
    seen = False

    def test_function(data):
        nonlocal seen
        value = draw(data)
        try:
            if DEBUG:
                print(data.buffer[:data.index], "->",  value)
            if check(value):
                seen = True
                data.incur_cost(len(repr(data)))
                data.mark_interesting()
            elif not seen and DEBUG:
                print(data.buffer[:data.index], "->",  value)
        except UnsatisfiedAssumption:
            data.mark_invalid()
    buffer = find_interesting_buffer(
        test_function, settings=settings or Settings(
            generations=5000,
        ))
    if buffer is not None:
        result = draw(TestData(buffer))
        assert check(result)
        return result
    raise NoSuchExample()
Пример #4
0
 def incorporate_new_buffer(self, buffer: bytes) -> bool:
     if buffer == self.last_data.buffer:
         return False
     data = TestData(buffer)
     self.test_function(data)
     data.freeze()
     if self.consider_new_test_data(data):
         if self.last_data.status == Status.INTERESTING:
             self.shrinks += 1
         self.last_data = data
         self.update_fill_size()
         self.changed += 1
         if self.shrinks >= self.settings.max_shrinks:
             raise StopShrinking()
         return True
     return False
Пример #5
0
def find(draw, check, settings=None):
    def test_function(data):
        value = draw(data)
        if check(value):
            data.incur_cost(len(repr(data)))
            if DEBUG:
                print(data.buffer[:data.index], "->",  value)
            data.mark_interesting()
    buffer = find_interesting_buffer(
        test_function, settings=settings or Settings(
            mutations=50, generations=500,
        ))
    if buffer is not None:
        result = draw(TestData(buffer))
        assert check(result)
        return result
    raise NoSuchExample()
Пример #6
0
 def new_buffer(self):
     buffer = self.rand_bytes(self.settings.buffer_size)
     self.last_data = TestData(buffer)
     self.test_function(self.last_data)
     self.last_data.freeze()
Пример #7
0
class TestRunner(object):
    def __init__(
        self,
        test_function: typing.Callable[[TestData], type(None)],
        settings: typing.Optional[Settings]=None,
    ):
        self._test_function = test_function
        self.settings = settings or Settings()
        self.last_data = None
        self.changed = 0
        self.shrinks = 0
        self.random = Random()

    def new_buffer(self):
        buffer = self.rand_bytes(self.settings.buffer_size)
        self.last_data = TestData(buffer)
        self.test_function(self.last_data)
        self.last_data.freeze()

    def test_function(self, data):
        try:
            self._test_function(data)
        except StopTest:
            pass

    def consider_new_test_data(
        self, data: TestData
    ) -> bool:
        # Transition rules:
        #   1. Transition cannot decrease the status
        #   2. Any transition which increases the status is valid
        #   3. If the previous status was interesting, only shrinking
        #      transitions are allowed.
        if self.last_data.status < data.status:
            return True
        if self.last_data.status > data.status:
            return False
        if data.status == Status.INVALID:
            return data.index >= self.last_data.index
        if data.status == Status.OVERRUN:
            return data.index <= self.last_data.index
        if data.status == Status.INTERESTING:
            assert len(data.buffer) <= len(self.last_data.buffer)
            if len(data.buffer) == len(self.last_data.buffer):
                assert data.buffer < self.last_data.buffer
            return interest_key(data) < interest_key(self.last_data)
        return True

    def incorporate_new_buffer(
        self, buffer: bytes
    ) -> bool:
        if (
            buffer[:self.last_data.index] ==
            self.last_data.buffer[:self.last_data.index]
        ):
            return False
        data = TestData(buffer)
        self.test_function(data)
        data.freeze()
        if self.consider_new_test_data(data):
            if self.last_data.status == Status.INTERESTING:
                self.shrinks += 1
            self.last_data = data
            self.changed += 1
            if self.shrinks >= self.settings.max_shrinks:
                raise StopShrinking()
            return True
        return False

    def run(self):
        try:
            self._run()
        except StopShrinking:
            pass

    def _run(self):
        self.new_buffer()
        mutations = 0
        generation = 0
        while self.last_data.status != Status.INTERESTING:
            if mutations >= self.settings.mutations:
                generation += 1
                if generation >= self.settings.generations:
                    return
                mutations = 0
                self.new_buffer()
            else:
                self.incorporate_new_buffer(
                    self.mutate_data_to_new_buffer()
                )
            mutations += 1

        for c in range(256):
            if self.incorporate_new_buffer(bytes(
                min(c, b) for b in self.last_data.buffer
            )):
                break

        initial_changes = self.changed
        change_counter = -1
        while (
            initial_changes + self.settings.max_shrinks >=
            self.changed > change_counter
        ):
            assert self.last_data.status == Status.INTERESTING
            change_counter = self.changed
            interval_change_counter = -1
            while self.changed > interval_change_counter:
                interval_change_counter = self.changed
                i = 0
                while i < len(self.last_data.intervals):
                    u, v = self.last_data.intervals[i]
                    if not self.incorporate_new_buffer(
                        self.last_data.buffer[:u] +
                        self.last_data.buffer[v:]
                    ):
                        i += 1
            i = 0
            while i < len(self.last_data.intervals):
                u, v = self.last_data.intervals[i]
                self.incorporate_new_buffer(
                    self.last_data.buffer[:u] +
                    bytes(sorted(self.last_data.buffer[u:v])) +
                    self.last_data.buffer[v:]
                )
                i += 1

            local_changes = -1
            while local_changes < self.changed:
                local_changes = self.changed
                for c in range(1, 256):
                    buf = self.last_data.buffer
                    if buf.count(c) > 1:
                        if self.incorporate_new_buffer(bytes(
                            c - 1 if b == c else b
                            for b in buf
                        )):
                            buf = self.last_data.buffer
                            for d in range(c):
                                if self.incorporate_new_buffer(bytes(
                                    d if b == c - 1 else b
                                    for b in buf
                                )):
                                    break
            k = 8
            for i in range(len(self.last_data.buffer) - k):
                buf = self.last_data.buffer
                if i + k > len(buf):
                    break
                self.incorporate_new_buffer(
                    buf[:i] + bytes(k) + buf[i + k:]
                )
            i = 0
            while i < len(self.last_data.buffer):
                buf = self.last_data.buffer
                if not self.incorporate_new_buffer(
                    buf[:i] + buf[i+1:]
                ):
                    for c in range(buf[i]):
                        if self.incorporate_new_buffer(
                            buf[:i] + bytes([c]) + buf[i+1:]
                        ):
                            break
                        elif self.incorporate_new_buffer(
                            buf[:i] + bytes([c]) + self.rand_bytes((
                                len(buf) - i - 1))
                        ):
                            break
                i += 1
            i = 0
            while i + 1 < len(self.last_data.buffer):
                j = i + 1
                buf = self.last_data.buffer
                if buf[i] > buf[j]:
                    self.incorporate_new_buffer(
                        buf[:i] + bytes([buf[j], buf[i]]) + buf[j + 1:]
                    )
                i += 1
            if self.changed > change_counter:
                continue
            i = 0
            while i < len(self.last_data.buffer):
                buf = self.last_data.buffer
                if not self.incorporate_new_buffer(
                    buf[:i] + buf[i+1:]
                ):
                    if buf[i] == 0:
                        buf = bytearray(buf)
                        j = i
                        while j >= 0:
                            if buf[j] > 0:
                                buf[j] -= 1
                                self.incorporate_new_buffer(bytes(buf))
                                break
                            else:
                                buf[j] = 255
                            j -= 1
                i += 1
            if self.changed > change_counter:
                continue
            buckets = [[] for _ in range(256)]
            for i, c in enumerate(self.last_data.buffer):
                buckets[c].append(i)
            indices = []
            for bucket in buckets:
                if len(bucket) > 1:
                    indices.extend(
                        (j, k)
                        for j in bucket for k in bucket
                        if j < k
                    )
            for j, k in indices:
                buf = self.last_data.buffer
                if k >= len(buf):
                    continue
                if buf[j] == buf[k]:
                    c = buf[j]
                    if c == 0:
                        if j > 0 and buf[j - 1] > 0 and buf[k - 1] > 0:
                            self.incorporate_new_buffer(
                                buf[:j - 1] +
                                bytes([buf[j - 1] - 1, 255]) +
                                buf[j+1:k-1] +
                                bytes([buf[k - 1] - 1, 255]) +
                                buf[k+1:]
                            )
                    c = buf[j]
                    if c > 0:
                        bd = bytes([c - 1])
                        if self.incorporate_new_buffer(
                            buf[:j] + bd + buf[j+1:k] + bd +
                            buf[k+1:]
                        ):
                            for d in range(c - 1):
                                buf = self.last_data.buffer
                                bd = bytes([d])
                                if self.incorporate_new_buffer(
                                    buf[:j] + bd + buf[j+1:k] + bd +
                                    buf[k+1:]
                                ):
                                    break
            if self.changed > change_counter:
                continue
            buf = self.last_data.buffer
            for j in range(len(buf)):
                buf = self.last_data.buffer
                if j >= len(buf):
                    break
                if buf[j] == 0:
                    continue
                for k in range(j + 1, len(buf)):
                    buf = self.last_data.buffer
                    if k >= len(buf):
                        break
                    if buf[j] > buf[k]:
                        self.incorporate_new_buffer(
                            buf[:j] + bytes([buf[k]]) + buf[j+1:k] +
                            bytes([buf[j]]) + buf[k+1:]
                        )
                    buf = self.last_data.buffer
                    if k >= len(buf):
                        break
                    if buf[j] > 0 and buf[k] > 0 and buf[j] != buf[k]:
                        if self.incorporate_new_buffer(
                            buf[:j] + bytes([buf[j] - 1]) + buf[j+1:k] +
                            bytes([buf[k] - 1]) + buf[k+1:]
                        ):
                            break
                    if buf[j] == 0:
                        break
                    for t in range(256):
                        if self.incorporate_new_buffer(
                            buf[:j] + bytes([buf[j] - 1]) + buf[j+1:k] +
                            bytes([t]) + buf[k+1:]
                        ):
                            break

    def mutate_data_to_new_buffer(self):
        n = min(len(self.last_data.buffer), self.last_data.index)
        if not n:
            return b''
        if n == 1:
            return self.rand_bytes(1)

        if self.last_data.status == Status.OVERRUN:
            result = bytearray(self.last_data.buffer)
            for i, c in enumerate(self.last_data.buffer):
                t = self.random.randint(0, 2)
                if t == 0:
                    result[i] = 0
                elif t == 1:
                    result[i] = self.random.randint(0, c)
                else:
                    result[i] = c

        probe = self.random.randint(0, 255)
        if probe <= 200 or len(self.last_data.intervals) <= 1:
            c = self.random.randint(0, 2)
            i = self.random.randint(0, self.last_data.index - 1)
            result = bytearray(self.last_data.buffer)
            if c == 0:
                result[i] ^= (1 << self.random.randint(0, 7))
            elif c == 1:
                result[i] = 0
            else:
                result[i] = 255
            return bytes(result)
        else:
            int1 = None
            int2 = None
            while int1 == int2:
                i = self.random.randint(0, len(self.last_data.intervals) - 2)
                int1 = self.last_data.intervals[i]
                int2 = self.last_data.intervals[
                    self.random.randint(
                        i + 1, len(self.last_data.intervals) - 1)]
            return self.last_data.buffer[:int1[0]] + \
                self.last_data.buffer[int2[0]:int2[1]] + \
                self.last_data.buffer[int1[1]:]

    def rand_bytes(self, n):
        if n == 0:
            return b''
        return self.random.getrandbits(n * 8).to_bytes(n, 'big')
Пример #8
0
 def new_buffer(self):
     buffer = self.rand_bytes(self.settings.buffer_size)
     self.last_data = TestData(buffer)
     self.test_function(self.last_data)
     self.last_data.freeze()
Пример #9
0
class TestRunner(object):
    def __init__(
        self, test_function: typing.Callable[[TestData], type(None)], settings: typing.Optional[Settings] = None
    ):
        self._test_function = test_function
        self.settings = settings or Settings()
        self.last_data = None
        self.changed = 0
        self.shrinks = 0
        self.random = Random()

    def new_buffer(self):
        buffer = self.rand_bytes(self.settings.buffer_size)
        self.last_data = TestData(buffer)
        self.test_function(self.last_data)
        self.last_data.freeze()

    def test_function(self, data):
        try:
            self._test_function(data)
        except StopTest:
            pass

    def consider_new_test_data(self, data: TestData) -> bool:
        # Transition rules:
        #   1. Transition cannot decrease the status
        #   2. Any transition which increases the status is valid
        #   3. If the previous status was interesting, only shrinking
        #      transitions are allowed.
        if self.last_data.status < data.status:
            return True
        if self.last_data.status > data.status:
            return False
        if data.status == Status.INVALID:
            return data.index >= self.last_data.index
        if data.status == Status.OVERRUN:
            return data.index <= self.last_data.index
        if data.status == Status.INTERESTING:
            assert len(data.buffer) <= len(self.last_data.buffer)
            if len(data.buffer) == len(self.last_data.buffer):
                assert data.buffer < self.last_data.buffer
            return interest_key(data) < interest_key(self.last_data)
        return True

    def incorporate_new_buffer(self, buffer: bytes) -> bool:
        if buffer[: self.last_data.index] == self.last_data.buffer[: self.last_data.index]:
            return False
        data = TestData(buffer)
        self.test_function(data)
        data.freeze()
        if self.consider_new_test_data(data):
            if self.last_data.status == Status.INTERESTING:
                self.shrinks += 1
            self.last_data = data
            self.changed += 1
            if self.shrinks >= self.settings.max_shrinks:
                raise StopShrinking()
            return True
        return False

    def run(self):
        try:
            self._run()
        except StopShrinking:
            pass

    def _run(self):
        self.new_buffer()
        mutations = 0
        generation = 0
        while self.last_data.status != Status.INTERESTING:
            if mutations >= self.settings.mutations:
                generation += 1
                if generation >= self.settings.generations:
                    return
                mutations = 0
                self.new_buffer()
            else:
                self.incorporate_new_buffer(self.mutate_data_to_new_buffer())
            mutations += 1

        for c in range(256):
            if self.incorporate_new_buffer(bytes(min(c, b) for b in self.last_data.buffer)):
                break

        initial_changes = self.changed
        change_counter = -1
        while initial_changes + self.settings.max_shrinks >= self.changed > change_counter:
            assert self.last_data.status == Status.INTERESTING
            change_counter = self.changed
            interval_change_counter = -1
            while self.changed > interval_change_counter:
                interval_change_counter = self.changed
                i = 0
                while i < len(self.last_data.intervals):
                    u, v = self.last_data.intervals[i]
                    if not self.incorporate_new_buffer(self.last_data.buffer[:u] + self.last_data.buffer[v:]):
                        i += 1
            i = 0
            while i < len(self.last_data.intervals):
                u, v = self.last_data.intervals[i]
                self.incorporate_new_buffer(
                    self.last_data.buffer[:u] + bytes(sorted(self.last_data.buffer[u:v])) + self.last_data.buffer[v:]
                )
                i += 1

            local_changes = -1
            while local_changes < self.changed:
                local_changes = self.changed
                for c in range(1, 256):
                    buf = self.last_data.buffer
                    if buf.count(c) > 1:
                        if self.incorporate_new_buffer(bytes(c - 1 if b == c else b for b in buf)):
                            buf = self.last_data.buffer
                            for d in range(c):
                                if self.incorporate_new_buffer(bytes(d if b == c - 1 else b for b in buf)):
                                    break
            k = 8
            for i in range(len(self.last_data.buffer) - k):
                buf = self.last_data.buffer
                if i + k > len(buf):
                    break
                self.incorporate_new_buffer(buf[:i] + bytes(k) + buf[i + k :])
            i = 0
            while i < len(self.last_data.buffer):
                buf = self.last_data.buffer
                if not self.incorporate_new_buffer(buf[:i] + buf[i + 1 :]):
                    for c in range(buf[i]):
                        if self.incorporate_new_buffer(buf[:i] + bytes([c]) + buf[i + 1 :]):
                            break
                        elif self.incorporate_new_buffer(buf[:i] + bytes([c]) + self.rand_bytes((len(buf) - i - 1))):
                            break
                i += 1
            i = 0
            while i + 1 < len(self.last_data.buffer):
                j = i + 1
                buf = self.last_data.buffer
                if buf[i] > buf[j]:
                    self.incorporate_new_buffer(buf[:i] + bytes([buf[j], buf[i]]) + buf[j + 1 :])
                i += 1
            if self.changed > change_counter:
                continue
            i = 0
            while i < len(self.last_data.buffer):
                buf = self.last_data.buffer
                if not self.incorporate_new_buffer(buf[:i] + buf[i + 1 :]):
                    if buf[i] == 0:
                        buf = bytearray(buf)
                        j = i
                        while j >= 0:
                            if buf[j] > 0:
                                buf[j] -= 1
                                self.incorporate_new_buffer(bytes(buf))
                                break
                            else:
                                buf[j] = 255
                            j -= 1
                i += 1
            if self.changed > change_counter:
                continue
            buckets = [[] for _ in range(256)]
            for i, c in enumerate(self.last_data.buffer):
                buckets[c].append(i)
            indices = []
            for bucket in buckets:
                if len(bucket) > 1:
                    indices.extend((j, k) for j in bucket for k in bucket if j < k)
            for j, k in indices:
                buf = self.last_data.buffer
                if k >= len(buf):
                    continue
                if buf[j] == buf[k]:
                    c = buf[j]
                    if c == 0:
                        if j > 0 and buf[j - 1] > 0 and buf[k - 1] > 0:
                            self.incorporate_new_buffer(
                                buf[: j - 1]
                                + bytes([buf[j - 1] - 1, 255])
                                + buf[j + 1 : k - 1]
                                + bytes([buf[k - 1] - 1, 255])
                                + buf[k + 1 :]
                            )
                    c = buf[j]
                    if c > 0:
                        bd = bytes([c - 1])
                        if self.incorporate_new_buffer(buf[:j] + bd + buf[j + 1 : k] + bd + buf[k + 1 :]):
                            for d in range(c - 1):
                                buf = self.last_data.buffer
                                bd = bytes([d])
                                if self.incorporate_new_buffer(buf[:j] + bd + buf[j + 1 : k] + bd + buf[k + 1 :]):
                                    break
            if self.changed > change_counter:
                continue
            buf = self.last_data.buffer
            for j in range(len(buf)):
                buf = self.last_data.buffer
                if j >= len(buf):
                    break
                if buf[j] == 0:
                    continue
                for k in range(j + 1, len(buf)):
                    buf = self.last_data.buffer
                    if k >= len(buf):
                        break
                    if buf[j] > buf[k]:
                        self.incorporate_new_buffer(
                            buf[:j] + bytes([buf[k]]) + buf[j + 1 : k] + bytes([buf[j]]) + buf[k + 1 :]
                        )
                    buf = self.last_data.buffer
                    if k >= len(buf):
                        break
                    if buf[j] > 0 and buf[k] > 0 and buf[j] != buf[k]:
                        if self.incorporate_new_buffer(
                            buf[:j] + bytes([buf[j] - 1]) + buf[j + 1 : k] + bytes([buf[k] - 1]) + buf[k + 1 :]
                        ):
                            break
                    if buf[j] == 0:
                        break
                    for t in range(256):
                        if self.incorporate_new_buffer(
                            buf[:j] + bytes([buf[j] - 1]) + buf[j + 1 : k] + bytes([t]) + buf[k + 1 :]
                        ):
                            break

    def mutate_data_to_new_buffer(self):
        n = min(len(self.last_data.buffer), self.last_data.index)
        if not n:
            return b""
        if n == 1:
            return self.rand_bytes(1)

        if self.last_data.status == Status.OVERRUN:
            result = bytearray(self.last_data.buffer)
            for i, c in enumerate(self.last_data.buffer):
                t = self.random.randint(0, 2)
                if t == 0:
                    result[i] = 0
                elif t == 1:
                    result[i] = self.random.randint(0, c)
                else:
                    result[i] = c

        probe = self.random.randint(0, 255)
        if probe <= 200 or len(self.last_data.intervals) <= 1:
            c = self.random.randint(0, 2)
            i = self.random.randint(0, self.last_data.index - 1)
            result = bytearray(self.last_data.buffer)
            if c == 0:
                result[i] ^= 1 << self.random.randint(0, 7)
            elif c == 1:
                result[i] = 0
            else:
                result[i] = 255
            return bytes(result)
        else:
            int1 = None
            int2 = None
            while int1 == int2:
                i = self.random.randint(0, len(self.last_data.intervals) - 2)
                int1 = self.last_data.intervals[i]
                int2 = self.last_data.intervals[self.random.randint(i + 1, len(self.last_data.intervals) - 1)]
            return (
                self.last_data.buffer[: int1[0]]
                + self.last_data.buffer[int2[0] : int2[1]]
                + self.last_data.buffer[int1[1] :]
            )

    def rand_bytes(self, n):
        if n == 0:
            return b""
        return self.random.getrandbits(n * 8).to_bytes(n, "big")
Пример #10
0
class TestRunner(object):
    def __init__(
        self,
        test_function: typing.Callable[[TestData], type(None)],
        settings: typing.Optional[Settings] = None,
    ):
        self._test_function = test_function
        self.settings = settings or Settings()
        self.last_data = None
        self.changed = 0
        self.shrinks = 0
        self.fill_size = min(8, self.settings.buffer_size)

    def new_buffer(self):
        buffer = os.urandom(
            self.fill_size) + bytes(self.settings.buffer_size - self.fill_size)
        self.last_data = TestData(buffer)
        self.test_function(self.last_data)
        self.last_data.freeze()
        self.update_fill_size()

    def update_fill_size(self):
        self.fill_size = min(max(self.fill_size, self.last_data.index * 2),
                             self.settings.buffer_size)

    def test_function(self, data):
        try:
            self._test_function(data)
        except StopTest:
            pass

    def consider_new_test_data(self, data: TestData) -> bool:
        # Transition rules:
        #   1. Transition cannot decrease the status
        #   2. Any transition which increases the status is valid
        #   3. If the previous status was interesting, only shrinking
        #      transitions are allowed.
        if self.last_data.status < data.status:
            return True
        if self.last_data.status > data.status:
            return False
        if data.status == Status.INVALID:
            return data.index >= self.last_data.index
        if data.status == Status.OVERRUN:
            return data.index <= self.last_data.index
        if data.status == Status.INTERESTING:
            assert len(data.buffer) <= len(self.last_data.buffer)
            return interest_key(data) < interest_key(self.last_data)
        return True

    def incorporate_new_buffer(self, buffer: bytes) -> bool:
        if buffer == self.last_data.buffer:
            return False
        data = TestData(buffer)
        self.test_function(data)
        data.freeze()
        if self.consider_new_test_data(data):
            if self.last_data.status == Status.INTERESTING:
                self.shrinks += 1
            self.last_data = data
            self.update_fill_size()
            self.changed += 1
            if self.shrinks >= self.settings.max_shrinks:
                raise StopShrinking()
            return True
        return False

    def run(self):
        try:
            self._run()
        except StopShrinking:
            pass

    def _run(self):
        self.new_buffer()
        mutations = 0
        generation = 0
        while self.last_data.status != Status.INTERESTING:
            if mutations >= self.settings.mutations:
                generation += 1
                if generation >= self.settings.generations:
                    return
                mutations = 0
                self.incorporate_new_buffer(
                    mutate_data_to_new_buffer(self.last_data))
            else:
                self.new_buffer()
            mutations += 1

        initial_changes = self.changed
        change_counter = -1
        while (initial_changes + self.settings.max_shrinks >= self.changed >
               change_counter):
            assert self.last_data.status == Status.INTERESTING
            change_counter = self.changed
            interval_change_counter = -1
            while self.changed > interval_change_counter:
                interval_change_counter = self.changed
                i = 0
                while i < len(self.last_data.intervals):
                    u, v = self.last_data.intervals[i]
                    if not self.incorporate_new_buffer(
                            self.last_data.buffer[:u] +
                            self.last_data.buffer[v:]):
                        i += 1
            i = 0
            while i < len(self.last_data.intervals):
                u, v = self.last_data.intervals[i]
                self.incorporate_new_buffer(
                    self.last_data.buffer[:u] +
                    bytes(sorted(self.last_data.buffer[u:v])) +
                    self.last_data.buffer[v:])
                i += 1
            k = 8
            for i in range(len(self.last_data.buffer) - k):
                buf = self.last_data.buffer
                if i + k > len(buf):
                    break
                self.incorporate_new_buffer(buf[:i] + bytes(k) + buf[i + k:])
            i = 0
            while i < len(self.last_data.buffer):
                buf = self.last_data.buffer
                if not self.incorporate_new_buffer(buf[:i] + buf[i + 1:]):
                    for c in range(buf[i]):
                        if self.incorporate_new_buffer(buf[:i] + bytes([c]) +
                                                       buf[i + 1:]):
                            break
                        elif self.incorporate_new_buffer(
                                buf[:i] + bytes([c]) +
                                os.urandom(len(buf) - i - 1)):
                            break
                i += 1
            i = 0
            while i + 1 < len(self.last_data.buffer):
                j = i + 1
                buf = self.last_data.buffer
                if buf[i] > buf[j]:
                    self.incorporate_new_buffer(buf[:i] +
                                                bytes([buf[j], buf[i]]) +
                                                buf[j + 1:])
                i += 1
            if self.changed > change_counter:
                continue
            i = 0
            while i < len(self.last_data.buffer):
                buf = self.last_data.buffer
                if not self.incorporate_new_buffer(buf[:i] + buf[i + 1:]):
                    if buf[i] == 0:
                        j = i
                        while j > 0:
                            if buf[j] > 0:
                                self.incorporate_new_buffer(
                                    buf[:j] + bytes([buf[j] - 1]) +
                                    bytes([255]) * (i - j) + buf[i + 1:])
                                break
                            j -= 1
                i += 1
            if self.changed > change_counter:
                continue
            buckets = [[] for _ in range(256)]
            for i, c in enumerate(self.last_data.buffer):
                buckets[c].append(i)
            indices = []
            for bucket in buckets:
                if len(bucket) > 1:
                    indices.extend(
                        (j, k) for j in bucket for k in bucket if j < k)
            for j, k in indices:
                buf = self.last_data.buffer
                if k >= len(buf):
                    continue
                if buf[j] == buf[k]:
                    c = buf[j]
                    if c == 0:
                        if j > 0 and buf[j - 1] > 0 and buf[k - 1] > 0:
                            self.incorporate_new_buffer(
                                buf[:j - 1] + bytes([buf[j - 1] - 1, 255]) +
                                buf[j + 1:k - 1] +
                                bytes([buf[k - 1] - 1, 255]) + buf[k + 1:])
                    c = buf[j]
                    if c > 0:
                        bd = bytes([c - 1])
                        if self.incorporate_new_buffer(buf[:j] + bd +
                                                       buf[j + 1:k] + bd +
                                                       buf[k + 1:]):
                            for d in range(c - 1):
                                buf = self.last_data.buffer
                                bd = bytes([d])
                                self.incorporate_new_buffer(buf[:j] + bd +
                                                            buf[j + 1:k] + bd +
                                                            buf[k + 1:])
            if self.changed > change_counter:
                continue
            buf = self.last_data.buffer
            for j in range(len(buf)):
                buf = self.last_data.buffer
                if j >= len(buf):
                    break
                if buf[j] == 0:
                    continue
                for k in range(j + 1, len(buf)):
                    buf = self.last_data.buffer
                    if k >= len(buf):
                        break
                    if buf[j] > buf[k]:
                        self.incorporate_new_buffer(buf[:j] + bytes([buf[k]]) +
                                                    buf[j + 1:k] +
                                                    bytes([buf[j]]) +
                                                    buf[k + 1:])
                    buf = self.last_data.buffer
                    if k >= len(buf):
                        break
                    if buf[j] > 0 and buf[k] > 0:
                        self.incorporate_new_buffer(buf[:j] +
                                                    bytes([buf[j] - 1]) +
                                                    buf[j + 1:k] +
                                                    bytes([buf[k] - 1]) +
                                                    buf[k + 1:])