specs/012-step-outputs.md

Rama main · réplica de lectura del repositorio original.

# Spec: Step Outputs

## Status

Implemented.

## Scope

This spec defines the step-level top-level `outputs` field, `DAGU_OUTPUT_FILE`, and output publication.

The field shape is:

```yaml
steps:
  - id: build
    outputs:
      - name: image_tag
```

It does not define the existing singular `output` field, `stdout.outputs`, `outputs.write`, or the `outputs` schema in `dagu-action.yaml`.

Object-form `output`, `stdout.outputs`, `outputs.write`, and the `outputs` schema in `dagu-action.yaml` publish named outputs of their own, and Spec 007 defines how references read them. String-form `output: VAR` captures a variable rather than a named step output, so Spec 007 does not read it, and the rules below about failed attempts do not govern it: the variable is set from the attempt's stdout even when the step fails, which Spec 018 relies on for the aggregate of a `foreach` whose items partly failed.

Value-resolution references to published outputs are defined by [Spec 007: Value Resolution Steps](007-value-resolution-steps.md).

## Goal

Define a file-based way for a step to publish named values for later steps.

The producing step writes records to `DAGU_OUTPUT_FILE`. Dagu parses that file after the step succeeds. Later steps read the values through value resolution.

## Input

Input is a workflow YAML file accepted by the YAML schema spec.

Step output validation extends:

```sh
dagu validate <path/to/dag_file>
```

Workflow execution uses:

```sh
dagu start <path/to/dag_file>
```

Rules:

- Validation checks output declarations.
- Validation must not execute steps.

Example:

```yaml
steps:
  - id: build
    run: |
      printf 'image_tag=v1.2.3\n' >> "$DAGU_OUTPUT_FILE"
    outputs:
      - name: image_tag
```

## Output declarations

`outputs` is an optional step field.

Rules:

- When present, `outputs` must be a non-empty sequence.
- A step with `outputs` must define `id`.
- Output names must be unique inside one step.
- Unknown output item fields are invalid.
- JSON Schema is not part of this spec.

Output item fields:

| Field | Required | Rules |
| --- | --- | --- |
| `name` | Yes | Must match `^[A-Za-z][A-Za-z0-9_]*$`. |
| `type` | No | Must be `string` or `json` when present. Defaults to `string`. |

## Output reference inputs

Value-resolution references to step outputs use the published output names from this spec.

Rules:

- Output names are scoped to the producing step.
- Step outputs are scoped to one DAG document.
- Inline sub-DAG documents have independent output scopes.
- Reference syntax, escaping, dependency requirements, passive notices, and runtime lookup behavior are defined by [Spec 007: Value Resolution Steps](007-value-resolution-steps.md).
- Unsupported `${...}` text outside Dagu-owned reference forms is preserved silently by [Spec 003: Value Resolution and Field Evaluation](003-value-resolution.md).

## Output file lifecycle

For each step attempt, Dagu creates an empty output file before the executor starts.

Rules:

- Dagu exposes the output file path to the executor as `DAGU_OUTPUT_FILE`.
- Executor specs define how each executor receives `DAGU_OUTPUT_FILE`.
- Dagu must not read stdout, stderr, or logs as step outputs.
- Dagu parses the output file only after the executor exits successfully.
- If the executor fails, aborts, or times out, Dagu does not publish outputs from that attempt.
- If output parsing fails, the attempt fails.
- Normal retry behavior may retry the failed attempt.
- Failed attempts publish no outputs.
- Only the attempt that makes the step successful publishes outputs.
- Each retry attempt receives a new empty output file.

## Output file format

Dagu parses the output file as UTF-8 text. Invalid UTF-8 fails output parsing. Line endings are normalized to `\n` before parsing.

Record formats:

| Format | Syntax | Value |
| --- | --- | --- |
| Single-line | `name=value` | Text after the first `=`. The line ending is not part of the value. Empty values are valid. |
| Multi-line | `name<<DELIMITER`, value lines, then a closing `DELIMITER` line. | Text between the line ending after the opening record and the line ending before the delimiter line. |

Single-line rules:

- The output name is the text before the first `=`.
- The output value is the text after the first `=`.
- The line ending is not part of the value.
- Empty values are valid.

Multi-line rules:

- The delimiter must match `[A-Za-z_][A-Za-z0-9_]*`.
- The delimiter line must match the delimiter exactly.
- The delimiter line ending is not part of the value.
- An unclosed multi-line record fails output parsing.

Record validation rules:

- An empty line outside a multi-line value is invalid.
- Each emitted output name must be declared by the step.
- Each declared output must be emitted exactly once.
- An undeclared emitted output fails output parsing.
- A duplicate emitted output fails output parsing.

## Output values

Type behavior:

| Type | Behavior |
| --- | --- |
| `string` | Stored as emitted text after line ending normalization. |
| `json` | Must parse as JSON. Any JSON value is valid: object, array, string, number, boolean, or null. |

Rules:

- Invalid JSON fails output parsing.
- A JSON output reference inserts the emitted JSON text after line ending normalization.
- If `max_output_size` is specified, output file content larger than that many bytes fails output parsing.

## Outputs

After successful parsing, Dagu publishes declared step outputs for later value resolution.

Published reference form:

```text
${steps.step_id.outputs.output_name}
```

The published reference form must be unescaped to resolve.
Escaped `\${steps.step_id.outputs.output_name}` is literal string content under Spec 003 and Spec 007.

Step outputs are not stdout, stderr, logs, artifacts, or durable result storage. This spec does not define how the published values are stored on disk.

## Errors

Validation must fail when:

- `outputs` has an invalid shape.
- An output item omits `name`.
- An output `name` has invalid syntax.
- One step declares the same output name more than once.
- An output item contains an unknown field.
- An output item uses an invalid `type`.
- A step declares `outputs` but has no `id`.

Runtime output parsing must fail when:

- The output file is not valid UTF-8.
- The output file syntax is invalid.
- A declared output is missing.
- The output file emits an undeclared output.
- The output file emits the same output more than once.
- A `json` output emits invalid JSON.

## Examples

Valid string output:

```yaml
steps:
  - id: build
    run: |
      printf 'image_tag=v1.2.3\n' >> "$DAGU_OUTPUT_FILE"
    outputs:
      - name: image_tag

  - id: deploy
    depends: build
    env:
      IMAGE_TAG: ${steps.build.outputs.image_tag}
    run: ./deploy.sh "$IMAGE_TAG"
```

Valid JSON output:

```yaml
steps:
  - id: inspect
    run: |
      cat >> "$DAGU_OUTPUT_FILE" <<'EOF'
      metadata<<JSON
      {"image":"api","tag":"v1.2.3"}
      JSON
      EOF
    outputs:
      - name: metadata
        type: json

  - id: print
    depends: inspect
    run: printf '%s\n' '${steps.inspect.outputs.metadata}'
```

Runtime failure because stdout is not a step output:

```yaml
steps:
  - id: build
    run: echo image_tag=v1.2.3
    outputs:
      - name: image_tag
```

Warning-only undeclared output reference:

These examples show Dagu validation behavior.
Dagu preserves unresolved references as text.
Dagu does not shell-escape resolved values or preserved references.
Quote the reference when a shell-backed `run` field needs the literal text.

```yaml
steps:
  - id: build
    run: echo ok
  - id: deploy
    depends: build
    run: ./deploy.sh '${steps.build.outputs.image_tag}'
```

Inferred dependency without `depends`:

```yaml
steps:
  - id: build
    run: |
      printf 'image_tag=v1.2.3\n' >> "$DAGU_OUTPUT_FILE"
    outputs:
      - name: image_tag
  - id: deploy
    run: ./deploy.sh '${steps.build.outputs.image_tag}'
```