Summary

Relands Remove-Reconfigure after #5997

Commits

1. Revert of #5997 (8e54770cf)

Restores the post-#5944 state — Remove-Reconfigure + cascade fix together — by reverting the previous PR. The git revert produced merge conflicts in four files due to drift on main during the bake window (primarily from #6041, RSDK-13417, "skip weak/optional dep reconfigure when unchanged"). Resolved by preserving #6041's snapshot-tracking + skip-when-unchanged optimization adapted to Remove-Reconfigure semantics.

Main conflicts:

robot/impl/resource_manager.go — 3 conflict regions:

• Remote-resource discovery (around line 305): nodeAlreadyExists branching. HEAD had an inverted if !nodeAlreadyExists condition with an unconditional SwapResource later in the function; parent had if nodeAlreadyExists { SwapResource } else { ... }. Took HEAD's structure — its end-of-block SwapResource advances the clock for both new and existing nodes, which snapshot tracking depends on. Parent only advanced for existing. Side fix: parent's graph_node.go adds an incrementClock bool parameter to SwapResource, so the post-merge call site needed , true appended.

• processResource dependency fetch (around line 1125). HEAD used getDependenciesWithWeakOptionalSnapshot; parent used the simpler getDependencies. Initially merged hybrid (snapshot variant, discard deps), later reverted to parent verbatim (_, err := lr.getDependencies(...)) — see the fix in commit 3.

• processResource rebuild path (lines 1140-1167): in-place reconfigure vs. always-rebuild. HEAD had a big gNode.ResourceModel() == conf.Model block calling moduleManager.ReconfigureResource or currentRes.Reconfigure with MustRebuildError fallback, returning (resource.Resource, bool, error). Parent had no in-place path and always rebuilt, returning (resource.Resource, error). Took parent entirely — Remove-Reconfigure removes the public Reconfigure method and moduleManager.ReconfigureResource, so the in-place branch can't exist. Function signature is back to 2-value; caller's markChildrenForUpdate becomes unconditional.

robot/impl/local_robot.go — 1 conflict region:

• updateWeakAndOptionalDependents action (around line 1180). HEAD had skip-when-unchanged guard (weakOptionalDepClocksEqual) plus an in-place Reconfigure. Parent had no skip guard and a BuiltInReconfigure for internal resources and inline rebuild for any other resource. Hybrid: kept HEAD's skip guard + "handling weak/optional update" log and took parent's BuiltInReconfigure / inline-rebuild branch.

Test-file conflicts in robot/impl/optional_dependencies_test.go and robot/impl/robot_reconfigure_weak_dependencies_test.go are not enumerated here (many); they were resolved by resetting both test files to parent and selectively restoring the new tests/helpers #6041 added.

2. Cleanup (d73416078)

I noticed some rebuild/cascade logic duplication within rebuildResource, cascadeRebuildDependentsOf, and ReconfigureResource so I took the chance to clean it up. I merged rebuildResource and rebuildResourceWithVisited into a single rebuildResource that takes a visited parameter (nil from outer callers, threaded from recursive ones), and replace the inline cascade walk inside the rebuild path with a call to cascadeRebuildDependentsOf (now also taking a visited parameter). Both the addResource post-add hook and the rebuildResource post-rebuild step now go through the same cascade implementation.

3. Post-merge audit (257182945)

A combing pass over the merge product to fix problems my initial conflict resolution introduced:

• Removed redundant snapshot capture in processResource. My initial merge for conflict 2 + conflict 3 captured weakOptionalSnapshot in processResource and called gNode.SetLastWeakOptionalDepsClocks(weakOptionalSnapshot) after a successful rebuild. But newResource already records the snapshot with its own freshly-captured value taken at construction time. Reverted both conflicts to parent verbatim. newResource is the single authoritative place that records the snapshot.

• Restored missing tests dropped during the conflict reset.

• Comment audit across both test files. Many comments retained pre-[APP-16062] Reapply Remove Reconfigure + Rebuild stale module-internal dependents when a dependency is re-added #5944 narrative referring to a resource-level Reconfigure or pre-RSDK-13417: Skip weak/optional dependency reconfigure when resolved set is unchanged #6041 narrative claiming updateWeakAndOptionalDependents always fires a rebuild. I rewrote these to reflect the latest set of changes.