Reading structured data types

[Artur-man]

Here is a case, from zarr-python which we do not read in Rarr at the moment.

>>> import zarr
>>> zarr.__version__
'2.18.7'
>>> z = zarr.open("example2.zarr")
>>> za = z["uns/rank_genes_groups/logfoldchanges"]
>>> za.attrs.asdict()
{'encoding-type': 'rec-array', 'encoding-version': '0.2.0'}
>>> za.info
Type               : Array
Zarr format        : 2
Data type          : [('0', '<f4'), ('1', '<f4'), ('2', '<f4'), ('3', '<f4'), ('4', '<f4'), ('5', '<f4')]
Shape              : (100,)
Chunk shape        : (100,)
Order              : C
Read-only          : False
Store type         : LocalStore
Filters            : ()
Compressors        : (Blosc(cname='lz4', clevel=5, shuffle=SHUFFLE, blocksize=0),)
No. bytes          : 2400 (2.3K)
>>> za[:]
array([( 1.2091126 ,  1.3671894 ,  1.03600216e+00,  1.25505447e+00,  1.21497381e+00,  0.8379856 ),
       ( 1.0016283 ,  1.520212  ,  8.02816689e-01,  8.07398200e-01,  1.11363661e+00,  1.2618493 ),
       ( 1.2513808 ,  0.9569721 ,  1.06660414e+00,  7.27090359e-01,  7.70881355e-01,  0.6068869 ),
       ( 0.8667367 ,  0.8978347 ,  7.36395359e-01,  6.78247988e-01,  7.51340449e-01,  1.3400772 ),
       ( 0.8735813 ,  0.76945794,  9.31681454e-01,  7.42217124e-01,  1.22592008e+00,  0.8190094 ),
       ( 0.8352304 ,  0.57019335,  6.17068052e-01,  8.51450741e-01,  1.17840135e+00,  0.58131063),

in Rarr

> library(Rarr)
> za <- read_zarr_array("example2.zarr/uns/rank_genes_groups/logfoldchanges")
Error in strsplit(typestr, "", fixed = TRUE) : non-character argument
> read_zarr_attributes("example2.zarr/uns/rank_genes_groups/logfoldchanges/")
$`encoding-type`
[1] "rec-array"

$`encoding-version`
[1] "0.2.0"

https://github.com/keller-mark/anndataR/blob/keller-mark/zarr/inst/extdata/example2.zarr.zip

[Bisaloo]

Notes for myself:

Related portion of the spec: https://zarr-specs.readthedocs.io/en/latest/v2/v2.0.html#data-type-encoding

See "structured data types".

[Bisaloo]

Just a quick note that this issue is likely to take a bit longer than some other feature requests as it requires a change in the backend since base arrays only support elements of length 1:

x <- array(NA_character_, dim = c(2, 2))
x[1, 1] <- letters
#> Error in x[1, 1] <- letters: number of items to replace is not a multiple of replacement length

^{Created on 2025-12-08 with reprex v2.1.1}

[Artur-man]

We have dealt with most issues I believe (regarding anndataR-zarr compatibility at least). This is just a small request, we have time, Luke and others still need to review my changes scverse/anndataR#190 (comment)

[Bisaloo]

Actually, this works:

x <- list(
  c("a", "b", "c"),
  c("d", "e", "f"),
  c("g", "h", "i"),
  c("j", "k", "l")
)

dim(x) <- c(2, 2)

x
#>      [,1]        [,2]       
#> [1,] character,3 character,3
#> [2,] character,3 character,3

x[1, 1]
#> [[1]]
#> [1] "a" "b" "c"

is.array(x)
#> [1] TRUE

^{Created on 2025-12-10 with reprex v2.1.1}

And should behave as an array for most applications. I believe it's a bit of a hack because arrays are intended to have atomic single-length elements but that's probably a good option for this edge case.