Materialise a lazy object locally and stay lazy.

The checkpoint verb (dbplyr's compute() for rasters): execute the current graph, write the results to local raw-BSQ cubes (.vrt + .bin, a format any GDAL tool reads and garry re-reads much faster than tiled GeoTIFF), and return the SAME KIND of lazy object rebuilt over the local files. Everything downstream continues unchanged; nothing upstream (network reads, warps, masking, model inference) runs again.

Usage

materialise(
  x,
  dir = NULL,
  name = "garry",
  nodata = NULL,
  overwrite = FALSE,
  distributed = garry_daemons_set()
)

Arguments

x

A LazyDataset or LazyRaster.

dir

Directory for the cubes (created if missing); default: a unique session-temporary directory.

name

File-name stem (default "garry").

nodata

Optional sentinel for the written files, as in write_tif().

overwrite

Replace existing files at the target paths?

distributed

As in collect().

Value

A lazy object of the same class as x, reading the local cubes.

Details

A LazyDataset writes one multiband cube per time slice through a single multi-sink plan (all slices' reads drain together), carrying band names, slice dates, and the mask_asset into the rebuilt dataset; ragged bands (a band missing some slices) survive. A LazyRaster writes one cube and reopens it. A computed raster cannot be warped directly, so materialise-then-rewarp is the supported route: align(materialise(x, dir), grid).

Files land at dir/name-<slice>.vrt (dataset) or dir/name.vrt (raster). Existing files are refused unless overwrite = TRUE: the graph may have changed since they were written, and silently reusing stale pixels is the failure mode a checkpoint must not have.

dir defaults to a fresh unique directory under the session's tempdir(), announced by a message: convenient, but session-scoped (the files vanish when R exits), and every call writes a NEW copy, so repeated interactive re-runs accumulate until the session ends. For large cubes, or to keep or reuse a checkpoint, give a real directory (note some systems mount /tmp in RAM).

See also

collect() to execute and return the result in the R session; write_tif() to execute and stream to a GeoTIFF.