Introduction

deltamesh answers one question: given two large structured documents, what is the smallest set of operations that turns the first into the second — and can we compute it without loading either into memory?

When you want this

• Replicating a dataset across a slow or metered link, where sending the full snapshot each cycle is wasteful.
• Building an audit trail of changes to configuration or catalogue data.
• Feeding a change-data-capture pipeline from systems that only expose full exports.
• Detecting drift between a declared desired state and the observed one.

When you do not

If your documents comfortably fit in memory and you already emit change events at the source, a plain JSON Patch library will be simpler and faster. deltamesh earns its complexity from about 50 MB per document upwards.

Model

Internally every input is projected onto the same tree: ordered sequences, unordered maps with comparable keys, and opaque leaves. The differ walks both trees in lockstep, emitting operations as soon as a subtree is known to differ — which is what keeps memory flat.

// the four operation kinds
type Op struct {
    Kind  OpKind  // Insert | Delete | Replace | Move
    Path  Path    // location in the target tree
    Value Value   // nil for Delete
}

Guarantees

Next

Head to the quickstart for a working example in about ten lines.