[BUG] StreamInput integer types are unsigned

[dbwiddis]

What is the bug?

The read_byte(), read_short(), read_int(), and read_long() methods in StreamInput return unsigned integers.

How can one reproduce the bug?

Attempt to deserialize any value with a negative sign bit, for example

input = StreamInput(b"\xff")
self.assertEqual(input.read_byte(), -1)

AssertionError: 255 != -1

input = StreamInput(b"\xff\xff\xff\xff\")
self.assertEqual(input.read_int(), -1)

AssertionError: 4294967295 != -1

What is the expected behavior?

Python int value matches the signed Java value from OpenSearch

Do you have any additional context?

[dbwiddis]

I see (at least) 3 possible approaches here, listed roughly in my preference order

• use int.from_bytes()
• use a conditional like return b - 256 if b >= 128 else b
• use numpy dependency

[dblock]

(1) works for me. Do you have an an XKCD for (2) and (3)?

[dbwiddis]

(1) works for me. Do you have an an XKCD for (2) and (3)?

2:

3:

[dbwiddis]

Bonus, the int.from_bytes() approach is so much better than this mess:

def read_int(self) -> int:
    return ((self.read_byte() & 0xFF) << 24) | ((self.read_byte() & 0xFF) << 16) | ((self.read_byte() & 0xFF) << 8) | (self.read_byte() & 0xFF)

def read_long(self) -> int:
    return self.read_int() << 32 | self.read_int() & 0xFFFFFFFF