dataset_viewer() renders a SAS Studio-style interactive grid for a data frame, an artoo-conformed frame, or a file path read via artoo::read_dataset(). The data is sent to the browser as Parquet and queried in place with DuckDB-WASM, so filter, sort, and scroll stay fast on large datasets without sampling rows.
The shell provides a column-selection panel with char/num type icons and a collapse chevron (its list can be sorted by original order, name, or type and filtered by name to navigate wide datasets, without changing the grid column order), a property pane (Label, Name, Length, Type, Format), a names-versus-labels header toggle, and click-a-header to sort (the first click selects the column, then clicks cycle ascending, descending, unsorted) – Shift-click adds columns for a multi-column sort, each shown with its direction and priority. A right-click menu sorts per column (Sort Ascending/Descending add the column to the sort; Clear Sorting removes just that column) and offers per-column filters, copying a column or its header, and size-to-content, alongside a free-text row filter and CSV export of the current view.
Missing values display as a muted NA in every column (numeric, character, and date), keeping them distinct from a genuine empty string and from the NaN floating-point value, which renders as NaN. They can be filtered with the COL is na / COL is not na free-text predicate or the per-column “(Missing)” option in the Add Filter dialog; the generated dplyr code uses is.na().
The “Show code” toolbar button emits the runnable, air-formatted dplyr pipeline (filter, arrange, then select, with SQL-to-R translation) that reproduces the current view.
The DuckDB-WASM engine loads from a CDN by default and is fetched into the package at install time when reachable, so a Shiny app can serve it to browsers with no internet at runtime (offline / corporate deployment). The install-time fetch honours DATASETVIEWER_DUCKDB_DIR, DATASETVIEWER_DUCKDB_URL, DATASETVIEWER_DUCKDB_EXT_URL, and DATASETVIEWER_DUCKDB_OFFLINE; the install never fails if the engine cannot be fetched (it falls back to the CDN at runtime).
options(datasetviewer.use_local_engine = FALSE) forces the CDN even when the engine is present locally, keeping self-contained HTML documents small.