protobuf: properly implement signed types (fixes #249)

tools/pb2py

@@ -70,8 +70,8 @@ def process_message(descriptor, protobuf_module, msg_id, indexfile, is_upy):
             field.TYPE_UINT64:  'p.UVarintType',
             field.TYPE_UINT32:  'p.UVarintType',
             field.TYPE_ENUM:    'p.UVarintType',
-            field.TYPE_SINT32:  'p.Sint32Type',
-            field.TYPE_SINT64:  'p.Sint64Type',
+            field.TYPE_SINT32:  'p.SVarintType',
+            field.TYPE_SINT64:  'p.SVarintType',
             field.TYPE_STRING:  'p.UnicodeType',
             field.TYPE_BOOL:    'p.BoolType',
             field.TYPE_BYTES:   'p.BytesType'

trezorlib/messages/NEMAggregateModification.py

@@ -6,5 +6,5 @@ from .NEMCosignatoryModification import NEMCosignatoryModification
 class NEMAggregateModification(p.MessageType):
     FIELDS = {
         1: ('modifications', NEMCosignatoryModification, p.FLAG_REPEATED),
-        2: ('relative_change', p.Sint32Type, 0),
+        2: ('relative_change', p.SVarintType, 0),
     }

trezorlib/protobuf.py

@@ -59,6 +59,8 @@ def load_uvarint(reader):
 
 
 def dump_uvarint(writer, n):
+    if n < 0:
+        raise ValueError("Cannot dump signed value, convert it to unsigned first.")
     buffer = _UVARINT_BUFFER
     shifted = True
     while shifted:
@@ -68,15 +70,47 @@ def dump_uvarint(writer, n):
         n = shifted
 
 
+# protobuf interleaved signed encoding:
+# https://developers.google.com/protocol-buffers/docs/encoding#structure
+# the idea is to save the sign in LSbit instead of twos-complement.
+# so counting up, you go: 0, -1, 1, -2, 2, ... (as the first bit changes, sign flips)
+#
+# To achieve this with a twos-complement number:
+# 1. shift left by 1, leaving LSbit free
+# 2. XOR with "all sign bits" - 0s for positive, 1s for negative
+#    This keeps positive number the same, and converts negative from twos-complement
+#    to the appropriate value, while setting the sign bit. Cute and efficient.
+#
+# The original algorithm makes use of the fact that arithmetic (signed) shift
+# keeps the sign bits, so for a n-bit number, (x >> n+1) gets us the "all sign bits".
+#
+# But this is harder in Python because we don't know the bit size of the number.
+# We could simply shift by 65, relying on the fact that the biggest type for other
+# languages is sint64. Or we could shift by 1000 to be extra sure.
+#
+# But instead, we'll do it less elegantly, with an if branch:
+# if the number is negative, do bitwise negation (which is the same as "xor all ones").
+
+def sint_to_uint(sint):
+    res = sint << 1
+    if sint < 0:
+        res = ~res
+    return res
+
+
+def uint_to_sint(uint):
+    sign = uint & 1
+    res = uint >> 1
+    if sign:
+        res = ~res
+    return res
+
+
 class UVarintType:
     WIRE_TYPE = 0
 
 
-class Sint32Type:
-    WIRE_TYPE = 0
-
-
-class Sint64Type:
+class SVarintType:
     WIRE_TYPE = 0
 
 
@@ -237,10 +271,8 @@ def load_message(reader, msg_type):
 
         if ftype is UVarintType:
             fvalue = ivalue
-        elif ftype is Sint32Type:
-            fvalue = (ivalue >> 1) ^ ((ivalue << 31) & 0xffffffff)
-        elif ftype is Sint64Type:
-            fvalue = (ivalue >> 1) ^ ((ivalue << 63) & 0xffffffffffffffff)
+        elif ftype is SVarintType:
+            fvalue = uint_to_sint(ivalue)
         elif ftype is BoolType:
             fvalue = bool(ivalue)
         elif ftype is BytesType:
@@ -289,11 +321,8 @@ def dump_message(writer, msg):
             if ftype is UVarintType:
                 dump_uvarint(writer, svalue)
 
-            elif ftype is Sint32Type:
-                dump_uvarint(writer, ((svalue << 1) & 0xffffffff) ^ (svalue >> 31))
-
-            elif ftype is Sint64Type:
-                dump_uvarint(writer, ((svalue << 1) & 0xffffffffffffffff) ^ (svalue >> 63))
+            elif ftype is SVarintType:
+                dump_uvarint(writer, sint_to_uint(svalue))
 
             elif ftype is BoolType:
                 dump_uvarint(writer, int(svalue))

trezorlib/tests/unit_tests/test_protobuf.py

@@ -0,0 +1,90 @@
+from io import BytesIO
+import pytest
+
+from trezorlib import protobuf
+
+
+class PrimitiveMessage(protobuf.MessageType):
+    FIELDS = {
+        0: ("uvarint", protobuf.UVarintType, 0),
+        1: ("svarint", protobuf.SVarintType, 0),
+        2: ("bool", protobuf.BoolType, 0),
+        3: ("bytes", protobuf.BytesType, 0),
+        4: ("unicode", protobuf.UnicodeType, 0),
+    }
+
+
+def load_uvarint(buffer):
+    reader = BytesIO(buffer)
+    return protobuf.load_uvarint(reader)
+
+
+def dump_uvarint(value):
+    writer = BytesIO()
+    protobuf.dump_uvarint(writer, value)
+    return writer.getvalue()
+
+
+def test_dump_uvarint():
+    assert dump_uvarint(0) == b'\x00'
+    assert dump_uvarint(1) == b'\x01'
+    assert dump_uvarint(0xff) == b'\xff\x01'
+    assert dump_uvarint(123456) == b'\xc0\xc4\x07'
+
+    with pytest.raises(ValueError):
+        dump_uvarint(-1)
+
+
+def test_load_uvarint():
+    assert load_uvarint(b'\x00') == 0
+    assert load_uvarint(b'\x01') == 1
+    assert load_uvarint(b'\xff\x01') == 0xff
+    assert load_uvarint(b'\xc0\xc4\x07') == 123456
+
+
+def test_sint_uint():
+    """
+    Protobuf interleaved signed encoding
+    https://developers.google.com/protocol-buffers/docs/encoding#structure
+    LSbit is sign, rest is shifted absolute value.
+    Or, by example, you count like so: 0, -1, 1, -2, 2, -3 ...
+    """
+    assert protobuf.sint_to_uint(0) == 0
+    assert protobuf.uint_to_sint(0) == 0
+
+    assert protobuf.sint_to_uint(-1) == 1
+    assert protobuf.sint_to_uint(1) == 2
+
+    assert protobuf.uint_to_sint(1) == -1
+    assert protobuf.uint_to_sint(2) == 1
+
+    # roundtrip:
+    assert protobuf.uint_to_sint(
+        protobuf.sint_to_uint(1234567891011)
+    ) == 1234567891011
+    assert protobuf.uint_to_sint(
+        protobuf.sint_to_uint(- 2 ** 32)
+    ) == - 2 ** 32
+
+
+def test_simple_message():
+    msg = PrimitiveMessage(
+        uvarint=12345678910,
+        svarint=-12345678910,
+        bool=True,
+        bytes=b'\xDE\xAD\xCA\xFE',
+        unicode="Příliš žluťoučký kůň úpěl ďábelské ódy 😊",
+    )
+
+    buf = BytesIO()
+
+    protobuf.dump_message(buf, msg)
+    buf.seek(0)
+    retr = protobuf.load_message(buf, PrimitiveMessage)
+
+    assert msg == retr
+    assert retr.uvarint == 12345678910
+    assert retr.svarint == -12345678910
+    assert retr.bool is True
+    assert retr.bytes == b'\xDE\xAD\xCA\xFE'
+    assert retr.unicode == "Příliš žluťoučký kůň úpěl ďábelské ódy 😊"