rdiff

Rdiff refers to a family of diff formats and associated tools designed to store and apply incremental changes between two data sets, with a focus on backup and patching workflows. A rdiff-based workflow captures the differences between a base version and a target version in a compact form that can be used to reconstruct the target from the base.

The core idea behind rdiff is the use of rolling checksums to partition the target data into blocks and identify matching segments in the base data. When a block in the target matches a block in the base, a copy instruction is recorded; when there is no good match, the differing data is stored as an insert instruction. The resulting diff describes a sequence of operations—copy from the base and insert new data—that, when applied in order, yields the target data. This approach is efficient when the two versions share substantial similarity.

A rdiff file typically contains a header with version information followed by a stream of instructions that may include copy commands (referring to offsets and lengths in the base) and insert commands (carrying raw data). The patch is designed to be applied deterministically by a corresponding patch engine to reconstruct the updated data from the base.

Notable implementations and usage include the rdiff-backup project, which uses the rdiff format to store incremental snapshots of directory trees. Other libraries in the librdiff/librsync ecosystem provide tooling to generate and apply rdiff-style diffs in multiple programming languages. The format is widely used in open-source backup and archival tools for efficient storage of versioned data.

See also: librsync, rsync, incremental backups, patch formats.