python hkdf

[dragonxue]

python hkdf which will be used as key derivation function

[xueyucoder]

add comment about where the test vectors come from

[zooko]

It looks like to me that Xue Yu attached the complete text as simple files in addition to the same text in the darcs patches. So, if you don't want to use darcs to apply them, then just inspect the darcs patches visually and compare them to the .py files, please, to make sure you're not missing anything.

[zooko]

The new Python hmac needs the hash function to have a DIGEST_SIZE attribute.

diff -rN -u old-darcsworld-hmac/pycryptopp/hash/sha256module.cpp new-darcsworld-hmac/pycryptopp/hash/sha256module.cpp
--- old-darcsworld-hmac/pycryptopp/hash/sha256module.cpp	2012-01-09 11:32:19.570933575 -0700
+++ new-darcsworld-hmac/pycryptopp/hash/sha256module.cpp	2012-01-09 11:32:19.612933575 -0700
@@ -28,6 +28,8 @@
 typedef struct {
     PyObject_HEAD
 
+	Py_ssize_t digest_size;
+
     /* internal */
     CryptoPP::SHA256* h;
     PyStringObject* digest;
@@ -59,7 +61,7 @@
 SHA256_digest(SHA256* self, PyObject* dummy) {
     if (!self->digest) {
         assert (self->h);
-        self->digest = reinterpret_cast<PyStringObject*>(PyString_FromStringAndSize(NULL, self->h->DigestSize()));
+        self->digest = reinterpret_cast<PyStringObject*>(PyString_FromStringAndSize(NULL, self->digest_size));
         if (!self->digest)
             return NULL;
         self->h->Final(reinterpret_cast<byte*>(PyString_AS_STRING(self->digest)));
@@ -100,6 +102,12 @@
     {NULL},
 };
 
+static PyMemberDef SHA256_members[] = {
+    {"number", T_PYSSIZET, offsetof(SHA256, digest_Size), READONLY,
+     "digest size"},
+    {NULL}  /* Sentinel */
+};
+
 static PyObject *
 SHA256_new(PyTypeObject* type, PyObject *args, PyObject *kwdict) {
     SHA256* self = reinterpret_cast<SHA256*>(type->tp_alloc(type, 0));
@@ -108,6 +116,7 @@
     self->h = new CryptoPP::SHA256();
     if (!self->h)
         return PyErr_NoMemory();
+	self->digest_size = self->h->DigestSize();
     self->digest = NULL;
     return reinterpret_cast<PyObject*>(self);
 }
@@ -162,7 +171,7 @@
     0,		               /* tp_iter */
     0,		               /* tp_iternext */
     SHA256_methods,      /* tp_methods */
-    0,                         /* tp_members */
+    SHA256_members,                         /* tp_members */
     0,                         /* tp_getset */
     0,                         /* tp_base */
     0,                         /* tp_dict */
diff -rN -u old-darcsworld-hmac/pycryptopp/test/test_sha256.py new-darcsworld-hmac/pycryptopp/test/test_sha256.py
--- old-darcsworld-hmac/pycryptopp/test/test_sha256.py	2012-01-09 11:32:19.574933575 -0700
+++ new-darcsworld-hmac/pycryptopp/test/test_sha256.py	2012-01-09 11:32:19.613933575 -0700
@@ -35,6 +35,11 @@
 h_5fd4 = a2b_hex("7c4fbf484498d21b487b9d61de8914b2eadaf2698712936d47c3ada2558f6788")
 
 class SHA256(unittest.TestCase):
+    def test_digest_size(self):
+        self.failUnless(hasattr(sha256.SHA256, 'digest_size'))
+        self.failUnless(isinstance(sha256.SHA256.digest_size, int))
+        self.failUnlessEqual(sha256.SHA256.digest_size, 32)
+
     def test_digest(self):
         empty_digest = sha256.SHA256().digest()
         self.failUnless(isinstance(empty_digest, str))

[warner]

I've got a cleaned up branch at ​https://github.com/warner/pycryptopp/tree/42-hkdf . It incorporates all these patches (including the .digest_size one), plus some necessary fixes. The result almost works: the (only?) remaining problem is that pycryptopp's SHA256 objects do not have a .copy() method (unlike hashlib.SHA256), and the stdlib HMAC code depends upon one. Not sure how to add this, it might need some C++ object-copying that's beyond my current comfort level.

[zooko]

I'll have a go at that. Thanks!

[zooko]

It looks like Crypto++'s sha256​ is an IteratedHashWithStaticTransform​, which is a Clonable​, and so the way to implement the .copy() method is probably to invoke .Clone() on the Crypto++ sha256 object.

[zooko]

We're going to go ahead and release pycryptopp v0.6.0 without this feature and then finish this feature for the next release.