[BenjaminDEMAILLE]

Fixes satijalab/seurat#9138.

The bug

Converting a v5 assay to a v3 assay fails when no feature-level metadata has been computed yet:

obj <- CreateSeuratObject(counts = data)
obj <- NormalizeData(obj)
obj[["RNA"]] <- methods::as(obj[["RNA"]], Class = "Assay")
#> Error in .subset2(x, i, exact = exact) : subscript out of bounds

Running FindVariableFeatures() first makes it work, which is what made the report read as a variable-features problem. It is not. FindVariableFeatures() simply happens to be what first populates feature metadata: before it, assay[[]] is a data frame of n features and zero columns; after it, eight.

Root cause

The coercion copies feature metadata with to[[]] <- from[[]], and [[<- unpacks the incoming frame column by column:

value <- lapply(
  X = 1:ncol(x = value),
  FUN = function(index) { ... value[[index]] ... }
)

For zero columns 1:0 is c(1, 0), so the first iteration asks for value[[1]] of a frame that has no columns. Hence subscript out of bounds, from a loop that should not have run at all.

The change

seq_len(ncol(value)).

I audited the remaining 1:ncol / 1:nrow / 1:length loops in the package rather than assume this was the only one: the others are either inside a length(i) > 1 guard or not reachable with an empty input, so this is the only site changed.

Verification

	main	this PR
as(assay5, "Assay") with no feature metadata	subscript out of bounds	succeeds, counts and data preserved
same, after FindVariableFeatures	succeeds	succeeds, 200 variable features kept

Tests

New tests/testthat/test_assay_meta_features.R: assigning an empty feature metadata frame is a no-op rather than an error, a v5 assay with no feature metadata coerces to v3 with its matrices intact, metadata that is present is still carried across, and assigning a non-empty frame still replaces the metadata.

Against main: 2 failures and 3 errors. With this change: 15 passes. Full suite 191 passing, R CMD check Status OK.

🤖 Generated with Claude Code

[BenjaminDEMAILLE]

Superseded by #318, which groups the correctness PRs into one so that a single review covers the subject. The commit is carried over unchanged; closing here to keep the queue down rather than because anything was wrong with it.