Limit delta copy length to 64K.

Version 2 packs are not supposed to contain delta copy opcodes that
encode lengths larger than 64K, so do not create them.  Version 3 packs
allow longer copy operations, but Dulwich and C Git always create
version 2 packs.

dulwich/pack.py

@@ -1593,9 +1593,10 @@ def _delta_encode_size(size):
     return ret
 
 
-# copy operations in git's delta format can be at most this long -
-# after this you have to decompose the copy into multiple operations.
-_MAX_COPY_LEN = 0xffffff
+# The length of delta compression copy operations in version 2 packs is limited
+# to 64K.  To copy more, we use several copy operations.  Version 3 packs allow
+# 24-bit lengths in copy operations, but we always make version 2 packs.
+_MAX_COPY_LEN = 0xffff
 
 def _encode_copy_operation(start, length):
     scratch = ''
@@ -1604,7 +1605,7 @@ def _encode_copy_operation(start, length):
         if start & 0xff << i*8:
             scratch += chr((start >> i*8) & 0xff)
             op |= 1 << i
-    for i in range(3):
+    for i in range(2):
         if length & 0xff << i*8:
             scratch += chr((length >> i*8) & 0xff)
             op |= 1 << (4+i)
@@ -1692,6 +1693,7 @@ def apply_delta(src_buf, delta):
                     index += 1
                     cp_off |= x << (i * 8)
             cp_size = 0
+            # Version 3 packs can contain copy sizes larger than 64K.
             for i in range(3):
                 if cmd & (1 << (4+i)):
                     x = ord(delta[index])

dulwich/tests/test_pack.py

@@ -168,18 +168,24 @@ class TestPackDeltas(TestCase):
     def test_nochange(self):
         self._test_roundtrip(self.test_string1, self.test_string1)
 
+    def test_nochange_huge(self):
+        self._test_roundtrip(self.test_string_huge, self.test_string_huge)
+
     def test_change(self):
         self._test_roundtrip(self.test_string1, self.test_string2)
 
     def test_rewrite(self):
         self._test_roundtrip(self.test_string1, self.test_string3)
 
-    def test_overflow(self):
+    def test_empty_to_big(self):
         self._test_roundtrip(self.test_string_empty, self.test_string_big)
 
-    def test_overflow_64k(self):
-        self.skipTest("big strings don't work yet")
-        self._test_roundtrip(self.test_string_huge, self.test_string_huge)
+    def test_empty_to_huge(self):
+        self._test_roundtrip(self.test_string_empty, self.test_string_huge)
+
+    def test_huge_copy(self):
+        self._test_roundtrip(self.test_string_huge + self.test_string1,
+                             self.test_string_huge + self.test_string2)
 
 
 class TestPackData(PackTests):