Changelog • Rarr

Rarr 1.9

New functions write_zarr_attributes() and read_zarr_attributes() provide support for reading, writing and updating Zarr v2 attributes.
This package now has a pkgdown website, available at https://huber-group-embl.github.io/Rarr/.
Zarr v3 arrays are now supported for reading metadata via zarr_overview().

zarr_overview(as_data_frame = TRUE) now returns information more in line with the output of zarr_overview(as_data_frame = FALSE). In particular:
- a new endianness column has been added to indicate the byte order of the array data.
- the nchunks column is now a list column specifying the number of chunks in each dimension, rather than a single integer giving the total number of chunks.

An explicit error message is now given when attempting to read a Zarr array version 3. This version will be supported in a future release of Rarr.

.url_parse_other() now accounts for port numbers in host name and colons in S3 buckets.
writeZarrArray() now allows writing character arrays, and no longer errors complaining about null ‘nchar’ argument value. Default of ‘nchar’ is now NULL.
writeZarrArray() no longer silently and incorrectly fills the last rows/columns when dim is not divisible by chunk_dim.
The object name is no longer repeated (e.g., name.zarrname.zarr) when writing a Zarr array to a file in the current working directory.
Invalid URLs for examples with S3 storage in read_zarr_array() and zarr_overview() have been updated.
read_zarr_array() no longer errors on arrays with numeric values other than float, int, uint and complex.
zarr_overview() now returns an explicit error message when the .zarray file is absent

Coding style throughout the package has been harmonized using the air tool. Contributors using RStudio, Positron or VS Code should have their code styled automatically on save.
Continuous integration checks have been made stricter by setting biocCheck() error level to “error” rather than “never”, and R CMD check error level to “warning” rather than “error”.
Static analysis via the lintr package is now performed on each push and PR. It should mostly be invisible to users but might result in slightly increased performance in some cases.
The superseded httr dependency has been replaced with the lighter curl package, thus reducing the total number of dependencies for the package from 42 to 40.
The unused stringr dependency has been removed, reducing the total number of dependencies for the package from 40 to 38.
A minor PROTECT()/UNPROTECT() imbalance in the C code, exposed by rchk, has been fixed. It is not likely to cause problems in real-world situations but it could theoretically lead to crashes in some cases.
Argument path in internal function read_array_metadata() has been renamed to zarr_path for consistency with other internal functions
Some internal functions have been renamed with a leading dot, in line with the officially recommended style for Bioconductor packages.

Added path() method for ZarrArray class that returns the location of the zarr array root.
Removed used of non-API call SETLENGTH in C code.
Small changes to compilation of internal blosc libraries to cope with the C23 compiler becoming the default in R-4.5.0

Fixed bug when creating an empty array with a floating datatype. The fill value would be interpreted as an integer by read_metadata() and create and array of the wrong type.
Fixed bug in update_zarr_array() when NULL was provided to one or more dimensions in the index argument. This was parsed incorrectly and the underlying zarr was not modified.
Fixed bug in reading 64-bit integer arrays compressed with ZLIB or LZ4. The calculated decompression buffer size was too small and reading would fail. (Thanks to Dan Auerbach for the report: https://github.com/grimbough/Rarr/issues/10)
Added support for the ZarrArray S4 class and the DelayedArray framework.
Improvements to read and write performance.

Fixed bug when reading an array if the fill value in .zarray was null.
Addressed bug in makevars where Rarr.so could be compiled before libblosc.a was ready. Also backported to Rarr 1.0.2. (Thanks to Michael Sumner for reporting this issue: https://github.com/grimbough/Rarr/issues/5)
Corrected issue where fixed length string datatypes would be written with null terminators, resulting in strings that were one byte longer than the dtype value written in the .zarray metadata. Also backported to Rarr 1.0.3.
Added support for reading and writing the fixed length Unicode datatype, and for reading variable length UTF-8 datatype.

Patch compression libraries to remove R CMD check warnings about C functions that might crash R or write to something other than the R console. Working in Linux only.

Fixed bug when specifying nested chunks, where the chunk couldn’t be written unless the directory already existed.

When writing chunks that overlap the array edge, even the undefined overhang region should be written to disk.