varint

"varint", or long form variable-length integer, is a way to encode arbitrarily-large integer. It's used in a lot of places in many different variants, but the main idea is as follows:

• A varint is composed of multiple octets;
• The most-significant bit of every octet is the continuation bit; within an octet stream, if this bit is set to 1, then the next octet should still be considered as a part of the current varint; if this bit is set to 0, then this very octet should be considered to be the last octet of this varint.
• The actual value of the integer shall be composed by the other 7 bits of the octets, either in a little-endian or a big-endian manner.

For example, when seen as the most basic form of varint described above, the octet sequence 0xfb 0x67 is read as follows:

• The first octet, 0xfb, is 1111 1011 in binary, within which the continuation bit (i.e. most-significant bit) is 1. The other 7 bits is 111 1011.
• Since the continuation bit is 1, we need to read the next octet.
• The next octet, 0x67, is 0100 0011 in binary. Its continuation bit is 0, thus we will not consider the next octet to be a part of this varint. The other 7 bits of this octet is 100 0011.
• Now, depending on whether the specification calls for little-endian or big-endian, the result could be 111 1011 100 0011 which is 0x3dc3 (big-endian) or 100 0011 111 1011 which is 0x21fb (little-endian).

Big-endian varint can be seen in WBMP. Little-endian varint can be seen in Git, which curiously uses 3 different variants.

---

2025.4.15

Back