| title: Workflow Error Propagation | |
| emoji: 🌖 | |
| colorFrom: gray | |
| colorTo: purple | |
| sdk: gradio | |
| sdk_version: 6.20.0 | |
| python_version: '3.13' | |
| app_file: app.py | |
| pinned: false | |
| # Repro: downstream workflow nodes still run after an upstream node fails | |
| Minimal reproduction for a `gr.Workflow` bug: when an upstream node raises an exception, downstream nodes are not skipped. They execute with `None` for the input that the failed node was supposed to produce. | |
| ## Steps | |
| 1. Open the app and click Run on the canvas. | |
| 2. The `first` node fails as expected (it always raises `ValueError("boom")`). | |
| 3. The `second` node, whose only input comes from `first`, still runs: its node shows `got: None`, and the server log prints `second() called with None`. | |
| Expected: `second` should not run its body. It should be marked as failed with the existing "upstream node failed" message, like other node types. | |
| ## Why it happens | |
| The executor's fail-fast guard in `workflow-executor.ts` checks `port.required && inputs[port.id] === null`, but the fn-node ports generated from `bind=` never set `required`, so the guard is skipped and `null` flows into the function call. See the linked issue for details. | |