Patch: Add __str__ and __bytes__ for undecoded content.

Patch.patch assumes all content to be encoded in UTF-8 and forcefully
replaces any non-decodable sequences. This can lead to corruption for
content that either does not conform to any specific encoding altogether, or
uses an encoding that is incompatible with, or ambinuous to UTF-8.

This change adds __str__ and __bytes__ implementations to Patch that
return the unmodified, raw bytes.

Author: erikvanzijst

src/patch.c

@@ -182,7 +182,11 @@ Patch_create_from(PyObject *self, PyObject *args, PyObject *kwds)
 
 
 PyDoc_STRVAR(Patch_patch__doc__,
-    "Patch diff string. Can be None in some cases, such as empty commits.");
+    "Patch diff string. Can be None in some cases, such as empty commits. "
+    "Note that this decodes the content to unicode assuming UTF-8 encoding. "
+    "For non-UTF-8 content that can lead be a lossy, non-reversible process. "
+    "To access the raw, un-decoded patch, use `str(patch)` (Python 2), or "
+    "`bytes(patch)` (Python 3).");
 
 PyObject *
 Patch_patch__get__(Patch *self)
@@ -201,9 +205,55 @@ Patch_patch__get__(Patch *self)
     return py_patch;
 }
 
+PyObject *
+Patch__str__(PyObject *self)
+{
+    git_buf buf = {NULL};
+    int err;
+    PyObject *ret;
+
+    assert(self->patch);
+    err = git_patch_to_buf(&buf, ((Patch*)self)->patch);
+    if (err < 0)
+        return Error_set(err);
+
+#if PY_MAJOR_VERSION == 2
+    ret = Py_BuildValue("s#", buf.ptr, buf.size);
+#else
+    ret = to_unicode(buf.ptr, NULL, NULL);
+#endif
+    git_buf_free(&buf);
+    return ret;
+}
+
+PyDoc_STRVAR(Patch__bytes____doc__, "The raw bytes of the patch's contents.");
+
+PyObject *
+Patch__bytes__(PyObject *self)
+{
+#if PY_MAJOR_VERSION == 2
+    return Patch__str__(self);
+
+#else
+    git_buf buf = {NULL};
+    int err;
+
+    assert(self->patch);
+    err = git_patch_to_buf(&buf, ((Patch*)self)->patch);
+    if (err < 0)
+        return Error_set(err);
+
+    PyObject *bytes = PyBytes_FromStringAndSize(buf.ptr, buf.size);
+    git_buf_free(&buf);
+    return bytes;
+#endif
+}
+
 PyMethodDef Patch_methods[] = {
     {"create_from", (PyCFunction) Patch_create_from,
       METH_KEYWORDS | METH_VARARGS | METH_STATIC, Patch_create_from__doc__},
+    {"__bytes__", (PyCFunction) Patch__bytes__,
+            METH_NOARGS, Patch__bytes____doc__},
     {NULL}
 };
 
@@ -237,7 +287,7 @@ PyTypeObject PatchType = {
     0,                                         /* tp_as_mapping     */
     0,                                         /* tp_hash           */
     0,                                         /* tp_call           */
-    0,                                         /* tp_str            */
+    Patch__str__,                              /* tp_str            */
     0,                                         /* tp_getattro       */
     0,                                         /* tp_setattro       */
     0,                                         /* tp_as_buffer      */

test/data/encoding.tar

@@ -28,6 +28,8 @@
 from __future__ import absolute_import
 from __future__ import unicode_literals
 
+import six
+
 import pygit2
 from . import utils
 
@@ -84,6 +86,70 @@
 """
 
 
+class PatchEncodingTest(utils.AutoRepoTestCase):
+    repo_spec = 'tar', 'encoding'
+    expected_diff = b"""diff --git a/iso-8859-1.txt b/iso-8859-1.txt
+index e84e339..201e0c9 100644
+--- a/iso-8859-1.txt
++++ b/iso-8859-1.txt
+@@ -1 +1,2 @@
+ Kristian H\xf8gsberg
++foo
+"""
+
+    def test_patch_from_non_utf8(self):
+        # blobs encoded in ISO-8859-1
+        old_content = b'Kristian H\xf8gsberg\n'
+        new_content = old_content + b'foo\n'
+        patch = pygit2.Patch.create_from(
+            old_content,
+            new_content,
+            old_as_path='iso-8859-1.txt',
+            new_as_path='iso-8859-1.txt',
+        )
+
+        # `patch.patch` corrupted the ISO-8859-1 content as it forced UTF-8
+        # decoding, so assert that we cannot get the original content back:
+        self.assertNotEqual(patch.patch.encode('utf8'), self.expected_diff)
+
+        if six.PY2:
+            self.assertIsInstance(str(patch), str)
+            self.assertEqual(str(patch), self.expected_diff)
+
+            self.assertIsInstance(patch.__bytes__(), str)
+            self.assertEqual(patch.__bytes__(), self.expected_diff)
+
+        else:
+            self.assertIsInstance(str(patch), str)
+            self.assertEqual(bytes(patch), self.expected_diff)
+            self.assertEqual(str(patch),
+                             str(self.expected_diff, 'utf8', errors='replace'))
+
+    def test_patch_create_from_blobs(self):
+        patch = pygit2.Patch.create_from(
+            self.repo['e84e339ac7fcc823106efa65a6972d7a20016c85'],
+            self.repo['201e0c908e3d9f526659df3e556c3d06384ef0df'],
+            old_as_path='iso-8859-1.txt',
+            new_as_path='iso-8859-1.txt',
+        )
+        # `patch.patch` corrupted the ISO-8859-1 content as it forced UTF-8
+        # decoding, so assert that we cannot get the original content back:
+        self.assertNotEqual(patch.patch.encode('utf8'), self.expected_diff)
+
+        if six.PY2:
+            self.assertIsInstance(str(patch), str)
+            self.assertEqual(str(patch), self.expected_diff)
+
+            self.assertIsInstance(patch.__bytes__(), str)
+            self.assertEqual(patch.__bytes__(), self.expected_diff)
+
+        else:
+            self.assertIsInstance(str(patch), str)
+            self.assertEqual(bytes(patch), self.expected_diff)
+            self.assertEqual(str(patch),
+                             str(self.expected_diff, 'utf8', errors='replace'))
+
+
 class PatchTest(utils.RepoTestCase):
 
     def test_patch_create_from_buffers(self):