Java SDK: Draw a Java-authored Dag's edges with before and after

[jason810496]

• depends on: Java SDK: Honor TaskFlow arg bindings sent by the supervisor #71188
• Diff for early review: jason810496/airflow@feature/java-sdk-arg-bindings-runtime...feature/java-sdk-native-dag-edges

Merge order:

1. Java SDK: Honor TaskFlow arg bindings sent by the supervisor #71188 — Honor TaskFlow arg bindings sent by the supervisor
2. Java SDK: Draw a Java-authored Dag's edges with before and after #73594 — Draw a Java-authored Dag's edges with before and after (current one)
3. Java SDK: Generate Dag and task configuration from the serialization schema #73595 — Generate Dag and task configuration from the serialization schema
4. Java SDK: Resolve a task's arguments from the Dag's own wiring #73596 — Resolve a task's arguments from the Dag's own wiring
5. Java SDK: Declare a Dag's task graph with @Builder.Deps #71189 — Declare a Dag's task graph with @Builder.Deps
6. Java SDK: Serialize native Dags to DagSerialization v3 #71190 — Serialize native Dags to DagSerialization v3

Every PR targets main because GitHub cannot base a pull request on a branch that exists only on a fork, so these diffs are cumulative — the compare link above shows only this layer. #69757 and #71057, which the stack was originally built on, are merged.

Why

A Dag authored in Java had no way to say what runs after what. Without a Python Dag file to hold the graph every task was a root, so a Dag that Java owns end to end could not be expressed at all. This is the graph half of ADR-0002.

Example

var dag = new DagDef("java_etl");

var extract = dag.task("extract", Extract.class);
var transform = dag.task("transform", Transform.class);
var load = dag.task("load", Load.class);

transform.after(extract).before(load);

How

• dag.task(...) registers a task as it creates it and hands back the handle, so there is no second addTask to forget.
• before and after are Java's spelling of Python's >> and <<, named to match the TypeScript SDK and to leave then free for the conditional control flow that reads as a then-branch. They are the only way to draw an edge, so one graph has one spelling however it was built.
• Both are variadic, so a.before(b, c) fans out and d.after(b, c) fans in, and both return their own receiver: a fan-out has no single next task to hand back.
• Flow.of(a, b).before(c, d), from org.apache.airflow.sdk.Deps.Flow, draws every edge between two sets in one call.
• A Dag is checked as it is registered with a Bundle: an upstream that belongs to another Dag, and a cycle anywhere in the graph, both fail there rather than at the first task run. A Dag that fails the check is not added to the bundle.
• example/.../nativedag/InterfaceExample.java is a Dag with no Python counterpart.

---

Was generative AI tooling used to co-author this PR?

• Yes, with help of Claude Code Opus 5 following the guidelines

[uranusjr]

Since the dependent PR has been merged, I took the liberty to rebase this against main.

[uranusjr]

I added a couple of notes on implementation choices. This now looks good to me. (Pending CI and a final look from Jason.)

[jason810496]

Thanks for the update, TP, LTGM.