NEWS

altarr 0.1.0.9000

New vignette, vignette("fetch", package = "altarr"): writing a fetch function, for maintainers of Zarr, netCDF, GDAL and object-store readers.
pkgdown site at https://hypertidy.github.io/altarr/, built by GitHub Actions.

First release-shaped version: the package is usable on its own, not only as a record of the base R proposal (tag v0.0.1).

Bounded chunk cache: least recently used chunks are evicted to keep each array within getOption("altarr.cache_bytes") (default 256 MiB). A copy shares its original's cache (#5, #10).
Planned region reads (ALTREP Get_region): mean(), prod(), anyNA() and reductions on R's wrapper class read whole chunk layers in batched fetch calls instead of one chunk per round trip (#6, #11).
Integer and logical arrays, via altarr(type = ). altarr_zarr_v2() picks the type from the Zarr dtype; CF scale and offset give doubles (#7, #12).
?altarr_contract documents the fetch contract, pinned sources, saving recipes safely, and a measured table of which base R operations are planned, element by element, or materializing (#8, #13).
New vignette, vignette("walkthrough", package = "altarr"): ten steps showing what base R asks a lazy array for.
Examples for every exported function; altarr_zarr_v2() is labelled a demonstration reader.
useDynLib is declared in roxygen, so document() keeps it (#14).
GitHub Actions R CMD check on Linux, macOS and Windows.

The state of the package for the base R ArraySubset hook proposal and the post "What R does today with an ALTREP that has a dim and no class".