buffer: Shared Memory Buffers

In shard: Deterministic, Zero-Copy Parallel Execution for R

Shared Memory Buffers

Description

Create typed writable output buffers backed by shared memory for cross-process writes during parallel execution.

Creates a typed output buffer backed by shared memory that can be written to by parallel workers using slice assignment.

Usage

buffer(
  type = c("double", "integer", "logical", "raw"),
  dim,
  init = NULL,
  backing = c("auto", "mmap", "shm")
)

Arguments

type	Character. Data type: "double" (default), "integer", "logical", or "raw".
dim	Integer vector. Dimensions of the buffer. For a vector, specify the length. For a matrix, specify c(nrow, ncol). For arrays, specify all dimensions.
init	Initial value to fill the buffer. Default is type-appropriate zero (0, 0L, FALSE, or raw(0)).
backing	Backing type for shared memory: "auto" (default), "mmap", or "shm".

Details

Buffers provide an explicit output mechanism for shard_map. Instead of returning results from workers (which requires serialization and memory copying), workers write directly to shared buffers.

Supported types:

• "double": 8-byte floating point (default)

• "integer": 4-byte signed integer

• "logical": 4-byte logical (stored as integer)

• "raw": 1-byte raw data

Buffers support slice assignment using standard R indexing: buf[1:100] <- values

Value

An S3 object of class "shard_buffer" that supports:

• Slice assignment: buf[idx] <- values

• Slice reading: buf[idx]

• Full extraction: buf[]

• Conversion to R vector: as.vector(buf), as.double(buf), etc.

See Also

segment_create for low-level segment operations, share for read-only shared inputs

Examples

out <- buffer("double", dim = 100)
out[1:10] <- rnorm(10)
result <- out[]

shard documentation built on April 3, 2026, 9:08 a.m.