Dependency Graph

Xantham.Decoder.Types.Graph builds a dependency view over the decoded type/export maps. It is the analytical counterpart to the resolved object graph: while ArenaInterner gives you a navigable graph for rendering, Graph gives you a graph for analysis — topological ordering, cycle detection, and reachability.

What's in a Graph

The Graph record has two frozen lookups:

Field	Type	Meaning
Dependents	FrozenDictionary<TypeKey, FrozenSet<TypeKey>>	For each TypeKey, the set of keys it depends on.
Cycles	FrozenDictionary<TypeKey, TypeKey>	Pairs (typeKey, cycleParticipant) — keys that participate in a cycle, mapped to the partner that closes it.

A TypeKey not present in Cycles is acyclic with respect to its dependencies.

Building one

The most direct path is from a XanthamTree:

let graph = tree.GetDependencyGraph()

Internally this calls Graph.create excludeConditionalBranches decodedResult with excludeConditionalBranches = false — meaning the Check and Extends arms of TypeScript conditional types are not traced as dependency edges. Including them is occasionally useful for very thorough reachability analysis, but for the common case of "what would I have to declare before this type" they only add noise.

If you need the alternative, call Graph.create directly:

let graphAll = Graph.create true decodedResult

Typical uses

Topological emit order. Start from the leaves of Dependents and emit upward. Any key that appears as a value (a dependency) of an unemitted key must be emitted first.

Cycle reporting. Iterate Cycles to surface mutually-recursive type definitions to the user, or to decide where to break the cycle in generated output (e.g. by switching to interface-style declarations).

Reachability filtering. Walk Dependents from a chosen entry set to compute the transitive closure of types actually needed by a generation target — useful when the input contains a large standard library you don't want to emit wholesale.

Notes

• The graph is built over both TypeMap and ExportTypeMap keys; it does not distinguish between structural types and exported declarations.
• Edges are recorded against the pre-compression / post-sanitization key space — so behaviour depends on the Decoder.Settings used to load the tree.
• The graph is computed eagerly when GetDependencyGraph() is called; cache the result if you intend to query it more than once.