feat(LF-12): Pipeline DAG with StepId derivation + OrchestrationBridge adapter#300
Merged
Conversation
Add `StepId` type alias and `id`/`depends_on` fields to `UnifiedStep` in the contract crate, enabling DAG-based step dependency tracking. Implement topological executor in `lance-graph-planner::pipeline` using Kahn's algorithm with cycle detection, missing-dep validation, and execute_with callback. 10 tests cover empty DAG, single step, linear chain, diamond, cycle detection (2-node and 3-node), missing dependency, duplicate StepId, step failure halting, and wide fan-out. Unblocks: PR-F4, PR-G2, PR-G6 https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj
…dmine Closes the three loose ends called out by the S1 reviewer on the PipelineDag PR: 1. (Tier 1) Remove the `pub id: StepId` field from `UnifiedStep`. Every caller was hard-coding `id: 0`, collapsing the DAG to a single colliding node. Replace with a derived `id()` method that FNV-1a-hashes `step_id`. Two distinct step_id strings now produce distinct numeric ids deterministically. All 4 callers (cypher_bridge.rs, serve.rs, lance_membrane.rs, orchestration_impl.rs) drop the `id: 0` line; `pipeline.rs` calls `step.id()` instead of `step.id`. The type alias `pub type StepId = u64` stays — pipeline keys its DAG and dependency edges off it. 2. (Tier 2) Add `PipelineDag::execute_via_bridge<B: OrchestrationBridge>` adapter so PR-F4 / PR-G2 cross-domain routing does not fork into a parallel closure-based executor. Tests cover end-to-end routing through a counting bridge. 3. (Tier 2) Add `cycle_detected_self_loop` test for the 1 -> 1 case; `CycleDetected.involved` now provably contains the self-looping step. Critical: the regression test `test_distinct_step_ids_when_step_id_strings_differ` was written FIRST and verified to fail under the old design (compile error: `id` is a field, not a method) before the fix landed. Tests: contract 307 pass (was 306); planner 182 pass (was 170, +12 in pipeline module covering the three loose ends); callcenter 23 pass; cognitive-shader-driver 46 pass (standalone build). https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5e568c34e4
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+353
to
+357
| pub fn id(&self) -> StepId { | ||
| // FNV-1a over step_id bytes — deterministic, collision-resistant. | ||
| let mut h: u64 = 0xcbf29ce484222325; | ||
| for b in self.step_id.as_bytes() { | ||
| h ^= *b as u64; |
There was a problem hiding this comment.
UnifiedStep::id() is recomputed from the mutable step_id string, but PipelineDag::build caches IDs once in index and later APIs (get, error reporting, execution_order) assume that identity is stable. Because execute_with passes &mut UnifiedStep to user/bridge callbacks, any callback that normalizes or rewrites step_id will desynchronize the cached index from the step’s current ID, causing lookups by current ID to fail and failures to be attributed to a different ID than the one validated in the DAG.
Useful? React with 👍 / 👎.
AdaWorldAPI
pushed a commit
that referenced
this pull request
Apr 30, 2026
- Fix 4× `#[deprecated(since = "next")]` invalid semver in context_chain.rs — drop `since` field (G3 refactor artifact) - Fix `actor.role <= u8::MAX` tautological comparison in lance_membrane.rs:768 — replace with meaningful `< 32` guard - Document Wave-1 LOC audit in EPIPHANIES.md: recovery (#275-#283) = +8,728; Wave 1 (#300-#306) = +3,156; combined = +11,807; zero LOC lost from G1 rebase cargo fmt --check: clean cargo clippy (4 crates): warnings only, 0 errors https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj
6 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Vec<UnifiedStep>withdepends_onDAG edgesUnifiedStep::id()computes FNV-1a overstep_idbytes — eliminates theid: 0landmine across all 4 callers (cypher_bridge, serve, lance_membrane, orchestration_impl)execute_via_bridge<B>(&self, bridge: &B)wires PipelineDag into the canonical contract pattern — F4/G2 won't fork the executorPipelineError::{MissingDependency, CycleDetected, StepFailed, DuplicateStepId}Review notes
type StepId = u64alias kept in contract;depends_on: Vec<StepId>added to UnifiedStepTest plan
test_distinct_step_ids_when_step_id_strings_differ— regression test (failing-first proven on oldid: 0code)https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj
Generated by Claude Code