varints

Varints, or variable-length integers, are a family of binary encodings in which an integer is represented using a varying number of bytes depending on its magnitude. A common form is base-128 varints, often called LEB128-style varints, where each byte carries seven data bits and a continuation bit in the most significant bit. If the continuation bit is set, more bytes follow; if it is clear, that byte completes the value.

Decoding a varint involves reading bytes in sequence, collecting the seven data bits from each byte, and placing them at successive positions in the final value. The process ends when a byte with the continuation bit zero is read. This design makes small integers occupy only one byte, while larger values use more bytes. For a 64-bit integer, a typical maximum of ten bytes is required.

Varints are widely used in binary serialization formats because they save space when values are small and offer fast parsing. They appear in protocols and data formats such as Protocol Buffers, Apache Thrift, Cap'n Proto, Parquet, and several custom binary schemas. They are especially efficient when the data set contains many small values.

For signed integers, variants exist. Some systems store signed values as unsigned varints after applying a ZigZag encoding, which maps small negative and positive numbers to small unsigned values. Others use direct signed encodings such as SLEB128, which encodes signed values in a way similar to unsigned varints but with sign information preserved.

Limitations include potential variance in maximum length (e.g., up to ten bytes for 64-bit values) and the need for consistent decoding rules. Varints are not inherently endian-dependent, but correct handling of continuation bits is essential for interoperability.