/rap-pipeline — Build Automation with {targets}

/rap-pipeline — Build Automation with {targets}

User Input

$ARGUMENTS

Instructions

You are configuring build automation for a RAP R project using the {targets} package. Build automation is a Silver RAP requirement (S1: minimal manual intervention) — it replaces manually sourcing scripts in order with a dependency-aware pipeline that only re-runs what has changed.

1. Detect R project

Check the current directory for an R project (.R files, DESCRIPTION, or .Rproj).

If not an R project, display:

This directory does not appear to be an R project.
Run `/rap-init` to scaffold a new RAP-compliant project first.

Then stop.

2. Check for existing _targets.R

If _targets.R already exists in the project root:

1. Read it and display the current pipeline structure
2. Scan R/ for any functions not currently included as targets
3. Suggest additions or improvements:
   • New functions that could be added as targets
   • Missing dependency links between targets
   • Opportunities for branching (e.g. per-region analysis)
4. Ask the user whether to update the existing file or leave it unchanged
5. If the user agrees, make the suggested updates

Then skip to step 5 (Update DESCRIPTION).

3. Scan R/ for pipeline functions

Use Glob to find all .R files in R/. For each file:

1. Read the file and extract exported function names (lines matching ^[a-zA-Z_][a-zA-Z0-9_.]* <- function()
2. Identify function parameters to determine dependencies between stages
3. Build a dependency map:
   • Functions that take no data parameters → data ingestion (sources)
   • Functions that take raw data → processing stage
   • Functions that take processed data → analysis stage
   • Functions that produce files/output → output stage

4. Generate _targets.R

Read the template from plugins/rap/templates/pipeline/_targets.R.template.

Replace {{PROJECT_NAME}} with the project name.

If the project has custom functions beyond the standard four-stage pattern, customise the _targets.R to include them:

• Map each detected function to a tar_target() call
• Wire up dependencies based on the function parameter analysis
• Add appropriate comments explaining each stage

Write the result to _targets.R in the project root.

5. Update DESCRIPTION

If DESCRIPTION exists, check whether targets is listed in Imports: or Suggests:. If not, add it to Suggests::

Suggests:
    targets,
    tarchetypes

6. Display summary

## Pipeline Configured

**File**: `_targets.R`

### Pipeline stages

| Target | Function | Depends on |
|--------|----------|------------|
| raw_data | read_source_data() | — |
| clean_data | process_data() | raw_data |
| analysis_results | analyse_data() | clean_data |
| output | create_output() | analysis_results |

### Running the pipeline

```r
# Run the full pipeline (only re-runs changed stages)
targets::tar_make()

# Visualise the dependency graph
targets::tar_visnetwork()

# Read a cached result
targets::tar_read(analysis_results)

# Check which targets are outdated
targets::tar_outdated()

How {targets} helps RAP

• Reproducibility: The pipeline definition is code, not a manual process
• Efficiency: Only re-runs stages whose inputs have changed
• Transparency: tar_visnetwork() shows the full dependency graph
• Auditability: Results are cached with metadata for inspection

### 7. Edge Case Handling

Handle these situations gracefully:

- **No R/ directory**: If `R/` does not exist, display:

No R/ directory found. Run /rap-init to create the standard project structure, or create R/ and add your pipeline functions before running /rap-pipeline.

- **No functions found in R/**: If R files exist but contain no function definitions (only scripts), warn:

Your R files appear to be scripts rather than functions. {targets} works best with functions. Consider refactoring your scripts into functions: read_data <- function() { ... } Then run /rap-pipeline again.

- **targets not installed**: If the user runs `tar_make()` and targets is not available:

install.packages("targets")

- **Existing _targets.R**: Never overwrite without confirmation. Analyse the current pipeline and suggest improvements instead.

- **DESCRIPTION missing**: Skip the DESCRIPTION update and note that targets should be installed manually.

- **Circular dependencies detected**: If function analysis suggests circular dependencies, warn and suggest restructuring.

### 8. RAP context

End your response with:

> **Build automation** is Silver RAP requirement S1 (minimal manual intervention). The `{targets}` package by Will Landau is the recommended pipeline tool for R, endorsed by the [NHS RAP Community of Practice](https://nhsdigital.github.io/rap-community-of-practice/training_resources/R/R-functions-and-packages/). It replaces manually running numbered scripts with a dependency-aware, cached pipeline.
>
> **Learn more:**
> - [targets manual](https://books.ropensci.org/targets/)
> - [UK Government Analysis Function RAP page](https://analysisfunction.civilservice.gov.uk/support/reproducible-analytical-pipelines/)
> - [Building RAPs with R](https://raps-with-r.dev/)
> - [RAP Companion](https://ukgovdatascience.github.io/rap_companion/)