shard_share_hook: Deep Sharing Hook for Custom Classes

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

Deep Sharing Hook for Custom Classes

Description

S3 generic that allows classes to customize deep sharing behavior. Override this for your class to control which slots/elements are traversed, force sharing of small objects, or transform objects before traversal.

Usage

shard_share_hook(x, ctx)

## Default S3 method:
shard_share_hook(x, ctx)

Arguments

x	The object being traversed during deep sharing.
ctx	A context list containing: path Current node path string (e.g., "<root>$data@cache") class class(x) - the object's class vector mode 'strict' or 'balanced' - sharing mode min_bytes Minimum size threshold for sharing types Character vector of enabled types for sharing deep Logical, always TRUE when hook is called

Value

A list with optional fields:

skip_slots

Character vector of S4 slot names to not traverse

skip_paths

Character vector of paths to not traverse

force_share_paths

Character vector of paths to force share (ignore min_bytes)

rewrite

Function(x) -> x to transform object before traversal

Return an empty list for default behavior (no customization).

Examples

shard_share_hook.MyModelClass <- function(x, ctx) {
    list(
        skip_slots = "cache",
        force_share_paths = paste0(ctx$path, "@coefficients")
    )
}

shard_share_hook.LazyData <- function(x, ctx) {
    list(
        rewrite = function(obj) {
            obj$data <- as.matrix(obj$data)
            obj
        }
    )
}

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