The 'future' package is now pre-loaded on the parallel workers when cluster workers are created

Author: HenrikBengtsson

DESCRIPTION

@@ -1,5 +1,5 @@
 Package: future
-Version: 1.58.0-9004
+Version: 1.58.0-9005
 Title: Unified Parallel and Distributed Processing in R for Everyone
 Depends:
     R (>= 3.2.0)

NEWS.md

@@ -6,6 +6,11 @@
    launched was unnecessarily expensive due to a thinko, e.g. it would
    take ~0.1-0.2 seconds for a multisession future, whereas after the
    fix it is effectly 0.0 seconds.
+   
+ * The **future** package is now pre-loaded on the parallel workers as
+   soon as a new cluster or multisession future backend is set
+   up. Previously, it was loaded with the first future being evaluated
+   on a worker.
 
  * Calling `resolved()` on a lazy `ClusterFuture` would collect the
    result for the first _resolved_ future in order to free up one

R/backend_api-11.ClusterFutureBackend-class.R

@@ -209,6 +209,8 @@ launchFuture.ClusterFutureBackend <- function(backend, future, ...) {
   ## Get the index of a free cluster node, which has been validated to
   ## be functional. If the existing worker was found to be non-functional,
   ## it was re-launched by requestNode()
+  ## FIXME: Is this wrong? Does it get the future-registry slot index
+  ## rather than worker index? /HB 2025-07-04
   node_idx <- requestNode(await = function() {
     FutureRegistry(reg, action = "collect-first", earlySignal = TRUE, debug = debug)
   }, backend = backend, timeout = timeout, delta = delta, alpha = alpha)
@@ -473,6 +475,10 @@ nbrOfFreeWorkers.ClusterFutureBackend <- function(evaluator, ...) {
 
   cl <- do.call(makeClusterPSOCK, args = args, quote = TRUE)
   cl <- addCovrLibPath(cl)
+
+  ## Pre-attach 'future' package
+  void <- clusterCall(cl = cl, fun = requireNamespace, "future", quietly = TRUE)
+
   cl
 } ## .makeCluster()