zr

Zarr V2/V3 arrays for R, via the zarrs Rust crate.

Features

• Read Zarr V2 and V3 multidimensional arrays
• Filesystem, HTTP, and S3 stores
• Local write for testing and prototyping (remote write out of scope)
• Chunk-level read/write access

We've done some limited testing on creating icechunk stores .

Installation

Requires Rust toolchain (rustc >= 1.85.0, cargo):

# install.packages("pak")
pak::pak("hypertidy/zr")

Quick start

Read from store

library(zr)

## Local filesystem
s <- zr_store("/path/to/store.zarr")
arr <- zr_array(s, "/temperature")
m <- zr_read(arr)

## Remote HTTP (any server, no WebDAV needed)
s <- zr_http_store("https://example.com/data.zarr")
arr <- zr_array(s, "/temperature")
m <- zr_read(arr, offset = c(0, 0, 0), count = c(1, 100, 200))

## S3 (credentials from environment)
s <- zr_s3_store("s3://bucket/store.zarr")
arr <- zr_array(s, "/temperature")
m <- zr_read(arr)

Write to local stores

d <- tempfile(fileext = ".zarr")
dir.create(d)
s <- zr_store(d)
zr_create_group(s, "/")

arr <- zr_create_array(s, "/data",
  shape = c(100, 200), chunks = c(50, 50),
  dtype = "float64", fill_value = NaN,
  dimension_names = c("y", "x"))

zr_write(s, arr, rnorm(100 * 200), offset = c(0, 0), count = c(100, 200))

m <- zr_read(arr)
# m is a 100x200 matrix, oriented for ximage()/rasterImage()

Indexing

All indexing is 0-based (matching zarr/Python convention).

Design

zr is a thin wrapper — no R6/S4 classes, plain functions with zr_ prefix, extendr external pointers for Rust objects. Remote stores (HTTP, S3) are read-only by design. Write-to-local is for testing and prototyping.