rho.compute

rho.compute adapts mirai worker evaluation to rho.async. Expressions and function calls are typed specifications; the agent sees only a RhoTask and does not acquire hidden worker policy.

Worker evaluation

library(rho.async)
library(rho.compute)

task <- rho_mirai_eval(value * 2L, args = list(value = 21L))
rho_is_task(task)
#> [1] TRUE
rho_await(task, timeout = 5000)
#> [1] 42

failure <- rho_mirai_eval(stop("worker failed", call. = FALSE)) |>
  rho_await(timeout = 5000)
list(
  typed_error = S7::S7_inherits(failure, RhoComputeErrorValue),
  message_preserved = grepl("worker failed", failure@message, fixed = TRUE)
)
#> $typed_error
#> [1] TRUE
#>
#> $message_preserved
#> [1] TRUE

Worker failures resolve as typed values, so callers can dispatch on their semantics. A tool chooses whether it uses this backend and separately declares whether multiple calls may overlap.

rho.graphics uses the same backend to render away from the interactive device. See the rho.compute reference.