Dense `[<-` write fails for 3D arrays ("Subarray should have num_dims * 2 values")

[kokitsuyuzaki]

Hi,

When writing to a dense 3D TileDB array using bracket assignment ([<-), I get the following error:

Error: Subarray should have num_dims * 2 values: (low, high) for each dimension.

This was originally encountered through the TileDBArray Bioconductor package, but the error comes from the tiledb package's [<- method. I've also filed LTLA/TileDBArray#30.

Minimal reproducible example

library(tiledb)

# Create a dense 3D array
tmp <- tempfile()
dom <- tiledb_domain(dims = list(
    tiledb_dim("d1", c(1L, 10L), 10L, type="INT32"),
    tiledb_dim("d2", c(1L, 10L), 10L, type="INT32"),
    tiledb_dim("d3", c(1L, 10L), 10L, type="INT32")
))
schema <- tiledb_array_schema(dom,
    attrs=list(tiledb_attr("val", type="FLOAT64")),
    sparse=FALSE)
tiledb_array_create(tmp, schema)

# Write using [<- : FAIL
arr <- tiledb_array(tmp, query_type="WRITE")
arr[1:10, 1:10, 1:10] <- array(runif(1000), dim=c(10,10,10))
#> Error: Subarray should have num_dims * 2 values: (low, high) for each
dimension.
tiledb_array_close(arr)

# Note: selected_ranges works as a workaround (per #259)
arr2 <- tiledb_array(tmp, query_type="WRITE")
selected_ranges(arr2) <- list(cbind(1L,10L), cbind(1L,10L), cbind(1L,10L))
# ... but not sure how to combine with value assignment
tiledb_array_close(arr2)

Context

I'm developing DelayedTensor, a Bioconductor package for out-of-core tensor arithmetic. We'd like to use TileDB as a storage backend (via TileDBArray) for 3D+ arrays. The [<- method works for 2D, and the sparse write path (via data.frame) works for 3D, so this seems specific to the dense subarray write path.

This appears to be the same issue as #259, which was closed as completed in 2021. However, the error still occurs with tiledb 0.34.0.

Session info

R version 4.5.3 (2026-03-11)
Platform: x86_64-conda-linux-gnu
tiledb 0.34.0

[kokitsuyuzaki]

Hi, I'm hitting multiple issues with [<- on dense arrays with a single attribute, as investigated by @LTLA in LTLA/TileDBArray#31. The behavior depends on ncol:

ncol	Behavior
1	No error, but data is silently not written
2	Works correctly
3	Error: misinterpreted as sparse triplets
4+	Works correctly

Minimal example:

library(tiledb)

tmp <- tempfile()
dom <- tiledb_domain(dims = list(
    tiledb_dim("d1", c(1L, 4L), 4L, type = "INT32"),
    tiledb_dim("d2", c(1L, 3L), 3L, type = "INT32")
))
schema <- tiledb_array_schema(
    dom,
    attrs = list(tiledb_attr("x", type = "FLOAT64")),
    sparse = FALSE
)
tiledb_array_create(tmp, schema)

# ncol=1: silent failure
arr <- tiledb_array(tmp, query_type = "WRITE")
selected_ranges(arr) <- list(cbind(1, 4), cbind(1, 1))
arr[] <- matrix(runif(4), 4, 1)
tiledb_array_close(arr)
arr <- tiledb_array(tmp, query_type = "READ")
arr[1:4, 1:3]$x  # all NaN
tiledb_array_close(arr)

# ncol=3: error
arr2 <- tiledb_array(tmp, query_type = "WRITE")
arr2[1:4, 1:3] <- matrix(runif(12), 4, 3)
# Error: Assigned data.frame does not contain all required
#   attribute and dimension columns.
tiledb_array_close(arr2)

Both selected_ranges() and direct [i,j] assignment show the same behavior, so this appears to be in the core [<- logic.

This is blocking the use of TileDBArray as a DelayedArray backend for 2D writes in my package (DelayedTensor).

Session Information

• tiledb 0.34.0
• R 4.5.3

[kokitsuyuzaki]

To add to this, as @LTLA noted in LTLA/TileDBArray#31 (comment: LTLA/TileDBArray#31 (comment)), this is not just an edge case — it's a major problem for block processing. Any matrix where the block width or its remainder is 1 or 3 columns will trigger these bugs. This makes [<- effectively unusable as a general-purpose dense write interface for DelayedArray backends.

[cgiachalis]

As you figured out, [<- method for dense arrays does not recognize all cases. See this part of [<- code that sets the subarray:

TileDB-R/R/TileDBArray.R

Lines 1533 to 1538 in 3760c68

	## case of dense array with subarray writes needs to set the subarray
	if (!sparse && !is.null(i) && !is.null(j) && length(allnames) == 1) {
	if (!is.vector(i) || !is.vector(j)) message("'i' and 'j' should be simple vectors.")
	subarr <- as.integer(c(range(i), range(j)))
	qryptr <- libtiledb_query_set_subarray(qryptr, subarr)
	}

I had similar issue when working with 1D dense time-series; eventually I opted to implement a custom [<- method that sets the subarray with type.