File-backed Daf store.

A Daf store backed by a directory of small self-describing files, bidirectionally compatible with Julia's DataAxesFormats.FilesDaf. Writes are non-atomic; only one writer may touch a store at a time.

Usage

files_daf(path, mode = c("r", "r+", "w", "w+"), name = NULL, packed = FALSE)

Arguments

path

Directory path.

mode

One of "r" (read-only, store must exist), "r+" (read-write, store must exist), "w" (create; fails if store already exists), "w+" (create or open an existing store; if the store already exists, its scalars, axes, vectors, matrices subdirectories and daf.json are truncated, but unrelated files in the directory are preserved).

name

Human-readable identifier. Defaults to basename(path).

packed

When TRUE (writeable modes only), dense vectors, dense matrices, and large sparse components that exceed the packing threshold are written as compressed dual-format .zip shards (DataAxesFormats.jl 0.3.0 "indexed+zipped" packed format) instead of flat binary payloads. Small components and string properties always stay flat. Tuning is read from the dafr.packed_compression, dafr.packed_compression_level, and dafr.packed_target_chunk_kb options. Defaults to FALSE (flat layout).

Value

A FilesDaf instance (DafWriter under "r+"/"w"/"w+", FilesDafReadOnly/DafReadOnly under "r").

Concurrent access

files_daf does not lock the store. Two writers opening the same path in mode "r+" or "w+" will race on metadata.zip rebuilds and per-entry JSON writes, with no guarantee of last-writer-wins consistency. The supported pattern is single-writer plus arbitrary read-only readers; cross-process concurrency must be coordinated externally (e.g., a job scheduler).

Examples

path <- tempfile("dafr-example-")
d <- files_daf(path, mode = "w")
add_axis(d, "cell", c("c1", "c2"))
set_scalar(d, "organism", "human")
rm(d)
unlink(path, recursive = TRUE)