Pipelines

Renderflow has two related execution models.

1. Standard build pipeline

Implemented in src/commands/build.rs and src/pipeline/pipeline.rs, this is the default path for renderflow build.

flowchart LR
  A[Input file] --> B[Asset normalization]
  B --> C[Transform phase]
  C --> D[Per-output render strategies]
  D --> E[Files in output_dir]

Transform phase

The standard registry runs transforms in a fixed order:

1. EmojiTransform
2. VariableSubstitutionTransform
3. SyntaxHighlightTransform
4. optional YAML-defined command/AI transforms

Render phase

After transforms complete, Renderflow renders each configured output in parallel. The implementation uses Rayon to fan out per-output work once transformed content is ready.

Why this split matters

• transforms are pure string-to-string operations,
• render steps are output-specific and may call Pandoc, Tectonic, or FFmpeg,
• the split makes caching easier because transform outputs and final outputs can be cached independently.

2. Graph execution pipeline

renderflow build --target ... and renderflow build --all use src/commands/graph_build.rs.

flowchart LR
  A[Config + transforms.yaml] --> B[TransformGraph]
  B --> C[Path selection by optimization mode]
  C --> D[MultiTargetDag]
  D --> E[DagExecutor waves]
  E --> F[Produced formats]

Graph mode is best when conversion paths are dynamic or multi-hop, for example markdown -> html -> pdf.

Note

Graph execution currently builds its DagExecutor from YAML command transforms and AI transforms. The standalone plugin API exists, but graph construction does not currently wire plugin-backed edges in build_graph_and_executor_from_yaml.

Failure behavior

• Standard build uses fail-fast transform behavior.
• watch uses continue-on-error transform behavior.
• Render errors are aggregated per output so one failed output does not automatically hide sibling results.