Skip to contents

Notes for developers

Source: DEVELOPERS.md

Requirements

  • R (>= 4.1)
  • just - runs the commands below (just list to see them all)
  • air - formats R code
  • gh - publishes releases to GitHub

Configuration

All in inst/config/:

  • tidy_data_sources.yml - current version and download url per dataset, read by get_*()
  • raw_<name>.yml - source urls per NHS Digital publication, read only when rebuilding data
  • shiny_app_datasets.yml - what shows up in the Shiny app’s dataset picker

Building the data

Each data-raw/<name>.R script downloads the source files from NHS Digital, cleans them, and writes a parquet file to data-raw/. Run one script directly, or just build-data to rebuild everything (this takes some time as it downloads the data).

Publishing a new version

  1. Add the new year’s source url to inst/config/raw_<name>.yml, then run data-raw/<name>.R (or just build-data) to rebuild the parquet in data-raw/
  2. Add a new entry at the top of the dataset’s list in tidy_data_sources.yml, with a placeholder url
  3. just release <dataset> <version> "<notes>" - uploads the parquet as a GitHub release
  4. Copy the real download url from the release page over the placeholder

Adding a new dataset

  1. Write data-raw/<name>.R and inst/config/raw_<name>.yml, following the existing pattern
  2. Add an entry to tidy_data_sources.yml. Set publication to the raw_<name>.yml base name and variant to usage or breakdowns. get_raw_source_periods() reads both.
  3. Export a get_<name>() function in R/get-data.R
  4. just release to publish it
  5. Add it to shiny_app_datasets.yml if it should appear in the app

Dataset names follow <source>_<coding system>. hesapc_icd10 means HES Admitted Patient Care, ICD-10. gp_snomed means GP, SNOMED CT. If a dataset has no single coding system, like QOF registers, use a short content word instead: qof_registers.

Running the app locally

run_app() downloads and caches data like any other user. Run cache_clear() first if you want to test against data you just published.

Deploying

just deploy copies the released parquet for every dataset into inst/app-data/ (this is gitignored), then deploys to Posit Connect Cloud. This means the hosted app reads data straight off disk and never calls GitHub at runtime, so a restart doesn’t need to download anything first. just deploy deploys the main app, just deploy beta deploys a separate beta app. The first time you deploy from a machine, you need to link your Posit Connect Cloud account by running rsconnect::connectCloudUser().