feat(pipeline-core): add fgumi-pipeline-core, the typed-step pipeline framework

[nh13]

What this is

PR 3 of the feat-runall landing series. It adds fgumi-pipeline-core, the typed-step pipeline framework: the Step / Step2 traits, the bounded queue layer, the reorder stage, and the work-stealing runtime that schedules and drives them. Like the crates in #672 and #674 it is additive — nothing on main-runall changes behaviour, and the crate has no in-tree consumer yet. The chains / steps layers that drive these primitives arrive in a later PR.

The crate is deliberately free of anything that reads or writes sequencing data: it depends on ahash, crossbeam-queue, parking_lot, log, and exactly one noodles::sam type (the shared header handle). It compiles in ~8s standalone.

Ported as-is, with these changes

The source is feat-runall's crates/fgumi-pipeline-core verbatim, plus:

Dependencies centralized (H7). The branch declared ahash = "0.8", log = "0", rstest = "0" etc. inline; all nine now come from [workspace.dependencies]. trybuild gained a second consumer with this crate (fgumi-cli-macros was the first), so per the root Cargo.toml's own rule — "third-party crates shared by two or more members declared once here" — it moves to the workspace table and fgumi-cli-macros switches to { workspace = true }.

WS-LINT (H9). Five collapsible_if errors, fixed as let-chains rather than suppressed, per CLAUDE.md's "adopt idioms the newer compiler unlocks" rule. This matches the tracker's prediction of 5 for this crate exactly.

publish.yml (H11). fgumi-pipeline-core added to the CRATES array after fgumi-cli-common. It is a leaf, so any position before its future dependents is topologically valid. Verified locally with the workflow's own completeness, stale-entry and topological-order checks.

Documentation that described a state this repo is not in. The crate root's module doc opened by explaining that the crate had been "extracted from the fgumi crate's pipeline::core module" and that "the fgumi crate re-exports this as pipeline::core" — neither is true here, and both would still be wrong after the series lands. Rewritten to describe what the crate is. Two doc comments referenced design docs that exist in no branch (docs/design/unified-pipeline-step-dsl.md, docs/design/unified-pipeline-typed-step-migration.md); those pointers are dropped. Three others used the pre-extraction core/queues.rs module paths, now intra-doc links.

Twenty rustdoc link defects. cargo ci-doc runs with -D warnings; as a private module inside fgumi these never surfaced. Nine unresolved links (StepCtx2, OutputQueueSet, RawOccupancy, TypedStep/TypedStep2/ErasedStep, Step/Step2) now carry crate paths; eight public-doc links to private items (retry_held_impl, Self::to_result, Self::push_metrics, DetachedPlaceholder) are demoted to code spans; three redundant explicit link targets are shortened.

Review tooling follows the code out of src/lib/

Two things silently stopped covering this code the moment it became a crate, and both are fixed here rather than left for whoever notices:

.coderabbit.yaml. The pipeline path_instructions entry — the one naming this code's actual failure modes (deadlock, lost output, unbounded memory, the must-accept next_serial exemption, StepDrainCounter output-close accounting, and "any new unsafe in the typed-step dispatch hot path not matching the documented #[allow(unsafe_code)] sites") — matched only src/lib/{unified_pipeline,pipeline}/**/*.rs. Nothing in this PR matched it. Of 42 changed files, 4 were instructed; 38 were not — including the file carrying the unsafe that clause was written for. The glob now covers the crate too, bringing it to 33 of 42 (the 9 remaining are yaml/toml/md/lock/.stderr, correctly outside a *.rs rule). The config's own MAINTENANCE comment asks for exactly this on a rename.

miri.yml. The workflow's stated inclusion criterion is "contains the raw-pointer … walks and has no FFI", which this crate meets — and Stacked Borrows is precisely what checks the lifetime-extension transmutes below. Added a step for fgumi-pipeline-core::erased, where all four sites live: 25 tests, clean, ~6s. The rest of the crate is excluded deliberately, and the workflow comment says why: the builder/runtime tests spawn threads and run full pipelines, which takes Miri well past ten minutes even with -Zmiri-disable-isolation, and two wedge tests (detached_two_sided_no_deadlock, driver_round_robins_all_live_before_parking) guard against a hang with a wall-clock watchdog that process::abort()s — under Miri's slowdown that fires on a healthy run. Widening the scope means giving those watchdogs a #[cfg(miri)] budget first.

The one thing worth a close look: four unsafe sites

erased.rs carries four #[allow(unsafe_code)] transmutes. This is hazard H4 arriving at P1 rather than P3 where the tracker expected it.

The runtime hands each step its input/output handles as &dyn Any, so the adapter must downcast_ref them back per dispatch. That sits on the per-item hot path — the crate docs record ≈1.6% of samples on TypeId compares in a 4-thread CODEC 8M run — so the adapter resolves once and caches. Caching a reference whose real lifetime the struct cannot name is what the unsafe buys: &'a Handle is widened to &'static for storage and narrowed back to &'a on read. No pointer is ever dereferenced through the 'static form.

Soundness rests on three invariants documented on TypedStep: the handle boxes are owned by ChainContexts (alive for the whole Pipeline::run), every step instance is dropped before those contexts are, and every dispatch passes the same box for a given step_idx. The third is the one the compiler cannot check, so this PR adds a debug_assert! on the cached-hit path that re-resolves the handle and compares pointers — a violation now panics in tests instead of becoming a use-after-free in release. Release builds are unchanged.

CLAUDE.md's unsafe-code section requires every unsafe site to be listed with a justification, so it gains an entry for these four. Worth knowing: feat-runall has its own entry for this ("Approved typed-step DSL hot path"), but it documents two sites where the code has four — it predates the TypedStep2 pair and was never updated. The entry added here covers all four, so it supersedes rather than duplicates the branch's version.

If you would rather not take the unsafe at all, the alternative is dropping the cache and paying the downcast_ref per dispatch. That is a one-function change and I am happy to make it — the ≈1.6% figure comes from the source branch and has not been re-measured here.

Tests

284 tests arrive with the crate (283 unit + 1 trybuild suite of 3 compile-fail goldens, which pass unmodified under rustc 1.93). Workspace total goes 7,169 → 7,456, all passing, 27 skipped.

I added 23 more to clear the 90% patch-coverage gate, all against public API that arrived with zero coverage rather than to chase the number:

• Fan-out output shapes. arity() was the only thing tested for (A,B,C), (A,B,C,D), OrderedBytesTuple2/3 and () — nothing built their queues. Each now round-trips build_queues → per-branch typed input → mark_all_drained, which catches a build_tupleN_queues that wires a branch twice or drops one. This is what arity() alone cannot see. All four assert branches are open before and closed after, so they cannot pass against an inert mark_all_drained.
• Chain::into_multi for 3 and 4 branches. Only the 2-branch case was covered. Each branch must get its own BranchIdx; the tests assert both that wiring all branches builds and that dropping one is reported by index.
• The type-erased append_source / append_step / append_step2 API — the assembly path the parent crate's ChainBuilder will drive, previously untested end to end. Includes that append_step2 consumes both upstream branches.
• HeapSize for String / Vec<T> / Option<T>. These decide how much budget a ByteBoundedQueue charges; the nested-Vec case (spine + per-element heap) is the one whose absence would let a queue blow past its byte budget.
• BuildError rendering, the PipelineConfig builder setters, Step2's trait defaults, and the two framework-bug panics in erased.rs that guard input-arity dispatch. The setter test is a 6-case #[rstest] table: each case asserts the field its setter targets took the value and that a field it does not target is still at its default — and asserts that second probe against PipelineConfig::default() first, so a case cannot silently encode a wrong baseline. with_scheduler's case asserts the trait's observable name(), not allocation identity, since both schedulers are zero-sized and a pointer comparison would pass even if the setter stored the wrong one.

Fidelity vs feat-runall

Audited rather than assumed. The 37-file set is identical to origin/feat-runall, and every one of the 67 removed lines is a dependency version string, a doc comment rewritten above, or a re-indented let-chain — no functional code was dropped. The unsafe surface diffs byte-identical: four sites on both sides, none added, none removed. No #[ignore], todo!, unimplemented!, or feature gate is hiding anything.

Gates

Gate	Result
ci-fmt / ci-lint / ci-test / ci-doctest / ci-doc	all exit 0
--no-default-features / --all-features checks	exit 0
G3 test loss	zero — no existing test file is touched
G8 Cargo.lock	regenerated
G10 patch coverage	see below

Every non-new file in the diff is a registration: Cargo.toml, Cargo.lock, publish.yml, crates/fgumi-cli-macros/Cargo.toml, CLAUDE.md.

Note on the base

main-runall is 2 commits behind origin/main (#676, #681). Rebasing the shared branch is a separate step; this PR targets main-runall as it stands.

Summary by CodeRabbit

• New Features

  • Added a pipeline framework supporting typed construction, branching, joining, ordering, cancellation, diagnostics, statistics, and configurable scheduling.
  • Added bounded and unbounded queues with memory budgeting, backpressure, detached processing, and optimized single-thread execution.
  • Added shared header resolution and pipeline finalization hooks.
  • Added compile-time validation for incompatible connections and output requirements.

• Documentation

  • Expanded guidance for supported pipeline locations, runtime behavior, and safety invariants.

• Tests

  • Added comprehensive execution, queue, ordering, cancellation, diagnostics, and compile-time validation coverage.

• Chores

  • Integrated the new package into workspace builds, publishing, and Miri validation.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

Walkthrough

Added fgumi-pipeline-core with typed construction, erased dispatch, bounded and ordered queues, worker and detached execution, cancellation, monitoring, statistics, and compile-time validation. Workspace, publishing, Miri, and review configuration now include the crate.

Changes

Pipeline core

Layer / File(s)	Summary
Pipeline contracts and transport crates/fgumi-pipeline-core/src/{step,item,outputs,queues,reorder,handles,topology,header,held,finalize}.rs	Adds typed steps, graph wiring, output shapes, queue limits, ordering, retries, header resolution, and finalization hooks.
Assembly and runtime adapters crates/fgumi-pipeline-core/src/{builder,erased}.rs, crates/fgumi-pipeline-core/src/runtime/contexts.rs	Builds typed and two-input chains, validates graphs, adapts steps, and constructs queue contexts.
Execution and observability crates/fgumi-pipeline-core/src/runtime/*, crates/fgumi-pipeline-core/src/{builder,signal}.rs	Adds worker, fused, and detached execution with scheduling, drain coordination, cancellation, monitoring, statistics, and queue-memory management.
Integration and repository wiring crates/fgumi-pipeline-core/src/tests.rs, crates/fgumi-pipeline-core/tests/*, Cargo.toml, .github/workflows/*, CLAUDE.md, .coderabbit.yaml	Adds integration and compile-fail tests, registers and publishes the crate, adds focused Miri checks, and documents pipeline paths and typed-handle cache invariants.

Estimated code review effort: 5 (Critical) | ~120 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Client
  participant PipelineBuilder
  participant WorkerPool
  participant Queue
  participant PipelineSignal
  Client->>PipelineBuilder: build and configure pipeline
  PipelineBuilder->>WorkerPool: create contexts and workers
  WorkerPool->>Queue: push and pop items
  WorkerPool->>PipelineSignal: report completion, cancellation, or error
  PipelineSignal-->>Client: return pipeline result

Loading

Possibly related PRs

• fulcrumgenomics/fgumi#461: Refines APIs and runtime modules introduced by this pipeline-core implementation.
• fulcrumgenomics/fgumi#542: Hardens duplicate-ordinal handling in the reorder implementation.
• fulcrumgenomics/fgumi#581: Refines Miri coverage for pipeline code.

Suggested labels: fgumi runall

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the new fgumi-pipeline-core crate and its typed-step pipeline framework, which are the main changes.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch nh/runall-03-pipeline-core

---

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

❌ Patch coverage is 93.81093% with 419 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main-runall@7576f13). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
crates/fgumi-pipeline-core/src/handles.rs	93.03%	70 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/stats.rs	90.76%	68 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/detached.rs	86.05%	52 Missing ⚠️
crates/fgumi-pipeline-core/src/erased.rs	94.15%	37 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/fused.rs	91.86%	34 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/contexts.rs	90.47%	26 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/driver.rs	96.83%	21 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/sampler.rs	93.23%	18 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/storage.rs	90.60%	17 Missing ⚠️
crates/fgumi-pipeline-core/src/runtime/pool.rs	90.06%	15 Missing ⚠️
... and 10 more

Additional details and impacted files

@@              Coverage Diff               @@
##             main-runall     #697   +/-   ##
==============================================
  Coverage               ?   93.86%           
==============================================
  Files                  ?      206           
  Lines                  ?   117683           
  Branches               ?        0           
==============================================
  Hits                   ?   110468           
  Misses                 ?     7215           
  Partials               ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[nh13]

@coderabbitai pause

[coderabbitai]

✅ Action performed

Reviews paused.

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 24

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-pipeline-core/src/builder.rs`:
- Around line 1320-1334: Update the monitor limitation documentation and both
referenced test comments to name ensure_monitor_visible_transports instead of
assert_monitor_visible_transports, and describe that it returns
PipelineError::MonitorBlindTransport in every build rather than only enforcing
the invariant through debug assertions. Preserve the existing explanation of
monitor visibility and production transport coverage.
- Around line 2636-2699: Update ParallelCountingSink to retain each popped input
value in shared storage instead of discarding it, then have
pipeline_run_with_threads_1_drains_chain,
pipeline_run_with_threads_4_drains_chain, and
detached_step_runs_and_drains_multithreaded sort the collected values and
compare them with the expected 1..=N sequence. Keep the existing count checks if
useful, but ensure the tests verify every emitted ordinal arrives exactly once.
- Around line 216-256: Register source steps with explicit input arity 0 in both
chain and append_source, instead of relying on register_step’s default arity of
1. Update the register_step calls in chain and append_source while preserving
their existing output arity and return behavior, so wiring an input into a
source is rejected at wire time.
- Around line 2848-2906: Strengthen both panic tests,
pipeline_run_reraises_worker_panic_single_threaded and
pipeline_run_reraises_worker_panic_with_monitor_enabled, by asserting that
catch_unwind returns the expected worker-panic payload rather than merely
checking result.is_err(). Preserve the existing setup while distinguishing the
intended PanickingSink panic from framework, spawn, or monitor-visibility
panics.

In `@crates/fgumi-pipeline-core/src/erased.rs`:
- Around line 851-903: The existing tests do not exercise cached handle
resolution or the Detached step kind. Add a test near the dispatch tests that
builds an AddOne consumer, invokes try_run_erased twice with the same
ErasedStepCtx, and verifies both dispatches succeed and produce the expected
outputs, thereby executing the cached resolve_input and resolve_outputs paths.
Extend both cached_kind_and_sticky_match_profile and
cached_kind_and_sticky_match_profile_step2 with Detached cases for false and
true.

In `@crates/fgumi-pipeline-core/src/handles.rs`:
- Around line 326-341: Split the documentation immediately before edge_metrics:
keep the build_branch description and its # Panics section attached to
build_branch, then start a separate doc comment for edge_metrics containing only
its per-edge metrics behavior. Ensure build_branch is documented and
edge_metrics no longer inherits unrelated panic documentation.
- Around line 629-642: Update the ordered transport path around ReorderStage and
its transport queue so memory is bounded by bytes, using ByteBoundedQueue or
equivalent byte accounting and a configured cap rather than an unbounded or
count-only queue. Keep the existing per-edge budget wiring through
BranchBudgetHandles and reorder_cap, and preserve the next_serial exemption.

In `@crates/fgumi-pipeline-core/src/header.rs`:
- Around line 23-48: Refactor HeaderHandle into producer-side HeaderSetter and
consumer-side HeaderReader halves, preserving one-shot header resolution.
Implement Drop for HeaderSetter so an unresolved slot is poisoned with a
descriptive io::Error, while normal set/poison completion suppresses the
fallback. Update HeaderHandle construction and accessors so consumers receive
HeaderReader and observe Some(Err(..)) instead of polling forever.

In `@crates/fgumi-pipeline-core/src/reorder.rs`:
- Around line 81-97: Ensure single-input Serial, Exclusive, and Detached steps
preserve BranchOrdering::ByItemOrdinal when receiving items from a None edge; do
not collapse the output ordering to None. Keep the reorder stage for these
paths, or explicitly enforce and document an ordered-input precondition where
that behavior is intended.

In `@crates/fgumi-pipeline-core/src/runtime/contexts.rs`:
- Around line 498-519: Strengthen assert_linear_chain_invariants by verifying
downstream handle identity, not only its BranchInputHandle<u32> type: create
distinct per-producer values, push each through ctx.outputs[i], and assert the
matching value is received from ctx.inputs[i + 1]. Import OutputHandles and
Single in the test module as needed, while retaining the existing source and
type assertions.

In `@crates/fgumi-pipeline-core/src/runtime/detached.rs`:
- Around line 78-97: Update DetachedPlaceholder::profile() so it does not
silently return incomplete metadata: either make the method fail loudly when
called or preserve the original detached step’s output_queues and
branch_ordering metadata in the placeholder. Ensure extract_detached_steps
callers cannot receive misleading empty metadata for detached steps that declare
outputs.

In `@crates/fgumi-pipeline-core/src/runtime/drain.rs`:
- Around line 81-93: Update concurrent_decrements_have_exactly_one_winner so the
iterator first spawns and collects all thread handles, then joins those handles
in a separate step. Ensure every worker can execute observe_drain concurrently
before results are collected, preserving the exactly-one-winner assertion.

In `@crates/fgumi-pipeline-core/src/runtime/driver.rs`:
- Around line 1004-1053: Create a separate StepDrainCounter for the
pool-dispatch portion of driver_dispatch_records_detached_busy_by_own_step, and
pass it to the is_driver=false dispatch_one_step call so that both dispatches
exercise the output-close path independently.

In `@crates/fgumi-pipeline-core/src/runtime/pool.rs`:
- Around line 82-94: Update assign_sticky_owners to resolve each sticky step’s
worker through the shared Affinity::target_worker helper instead of matching
Affinity variants locally. Preserve skipping affinities without a target and
only assign sticky[target] when the resolved target is within n_workers and
currently unset, keeping the existing StepIdx assignment behavior.

In `@crates/fgumi-pipeline-core/src/runtime/storage.rs`:
- Around line 136-146: Update the StepKind::Exclusive arm in
build_worker_storage to assert that owner is less than the worker count before
indexing entries[owner], matching the always-on range validation used by the
Serial arm. Keep the existing owner-assignment expectation and worker-entry
replacement behavior unchanged, but provide a clear diagnostic for out-of-range
owners.

In `@crates/fgumi-pipeline-core/src/tests.rs`:
- Around line 540-544: Update the HeapSize implementations for OrderedU32 and
OrderedU64 so each item reports a realistic nonzero heap payload instead of 0,
then adjust the ByteBounded limits in both output queue specifications so the
test’s item count exceeds those limits. Preserve the test setup while ensuring
pushes reach the byte cap and force the step to retry.
- Around line 309-312: The direct Pipeline::run calls in tests.rs need a shared
timeout harness to prevent deadlocks from hanging the test suite. Extract the
spawn-and-recv_timeout logic from run_reports_finished_pipeline into a small
closure-taking helper, then use that helper for the PairSummer run at
crates/fgumi-pipeline-core/src/tests.rs lines 309-312, the OrderedPairSummer run
at lines 686-687, and the existing run_reports_finished_pipeline test; preserve
the "DEADLOCKED or stalled at --threads 4" failure message for the PairSummer
case.
- Around line 314-327: Update both Step2 end-to-end assertions in
crates/fgumi-pipeline-core/src/tests.rs at lines 314-327 and 691-697: remove
sort_unstable, assert the collected values in emit order as [55, 44, 33, 22, 11]
and [10, 8, 6, 4, 2] respectively, and revise the comments to state that Serial
dispatch preserves FIFO pairing and output order. If either sequence is unstable
with threads: 4, leave the test unsorted and report the ordering issue rather
than masking it.
- Around line 142-157: Update
step2_build_two_input_handles_rejects_duplicate_producer to describe the actual
invariant: the same producer index is allowed only when the two branch indices
differ. Add a positive test using a producer output with two branches, call
build_two_input_handles with the same producer index and distinct branches, and
assert it succeeds; retain the existing panic case for identical producer and
branch indices.
- Around line 53-61: Update SumPairStep::try_run to avoid eagerly popping both
branches and discarding a value when only one is available. Reuse the
pending-item buffering approach documented by PairSummer, retaining any
unmatched item until the other branch provides a value while preserving Progress
and NoProgress outcomes.
- Around line 576-582: Update the push handling in OrderedSource and
OrderedPairSummer to assert or otherwise propagate failure from
ctx.outputs.push(...) instead of discarding its result. Preserve the existing
decrement and ordinal behavior only when the item is successfully enqueued,
matching WideQueueSource’s rejected-push handling so queue rejection cannot
produce Progress with a missing item.
- Around line 445-459: Remove the `new_worker_copy` override from the `Serial`
step implementation, and update the stale `DrainReproSource` reference
associated with this logic to reflect that serial steps use one shared instance.
Preserve the existing completion and buffering behavior in the surrounding step
logic.

In `@crates/fgumi-pipeline-core/src/topology.rs`:
- Around line 267-295: Add a test alongside
branch_offsets_route_multi_branch_consumers_correctly that creates two producers
and a two-input consumer, wires them with wire_to_slot into slots 0 and 1, and
asserts consumer_input_slot returns the matching slot for each edge and
input_arity reports 2.

In `@crates/fgumi-pipeline-core/tests/compile_fail.rs`:
- Around line 8-12: Update pipeline_core_compile_fail to verify that
tests/compile-fail/*.rs matches at least one fixture before invoking trybuild,
failing explicitly when the set is empty; also replace the “Phase 0 design (line
530)” reference with the specific document path.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 3fb309e2-8a42-4191-8363-e1909981d0bc

📥 Commits

Reviewing files that changed from the base of the PR and between 7576f13 and 27d5d46.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (43)

• .coderabbit.yaml
• .github/workflows/miri.yml
• .github/workflows/publish.yml
• CLAUDE.md
• Cargo.toml
• crates/fgumi-cli-macros/Cargo.toml
• crates/fgumi-pipeline-core/Cargo.toml
• crates/fgumi-pipeline-core/src/builder.rs
• crates/fgumi-pipeline-core/src/erased.rs
• crates/fgumi-pipeline-core/src/finalize.rs
• crates/fgumi-pipeline-core/src/handles.rs
• crates/fgumi-pipeline-core/src/header.rs
• crates/fgumi-pipeline-core/src/held.rs
• crates/fgumi-pipeline-core/src/item.rs
• crates/fgumi-pipeline-core/src/lib.rs
• crates/fgumi-pipeline-core/src/outputs.rs
• crates/fgumi-pipeline-core/src/queues.rs
• crates/fgumi-pipeline-core/src/reorder.rs
• crates/fgumi-pipeline-core/src/runtime/contexts.rs
• crates/fgumi-pipeline-core/src/runtime/detached.rs
• crates/fgumi-pipeline-core/src/runtime/drain.rs
• crates/fgumi-pipeline-core/src/runtime/driver.rs
• crates/fgumi-pipeline-core/src/runtime/fused.rs
• crates/fgumi-pipeline-core/src/runtime/live.rs
• crates/fgumi-pipeline-core/src/runtime/metrics.rs
• crates/fgumi-pipeline-core/src/runtime/mod.rs
• crates/fgumi-pipeline-core/src/runtime/pool.rs
• crates/fgumi-pipeline-core/src/runtime/sampler.rs
• crates/fgumi-pipeline-core/src/runtime/scheduler.rs
• crates/fgumi-pipeline-core/src/runtime/stats.rs
• crates/fgumi-pipeline-core/src/runtime/storage.rs
• crates/fgumi-pipeline-core/src/runtime/worker_core.rs
• crates/fgumi-pipeline-core/src/signal.rs
• crates/fgumi-pipeline-core/src/step.rs
• crates/fgumi-pipeline-core/src/tests.rs
• crates/fgumi-pipeline-core/src/topology.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
• crates/fgumi-pipeline-core/tests/compile_fail.rs

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 13

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/miri.yml:
- Around line 72-84: Update the Miri workflow step around the command for the
erased test filter to first run cargo with “-- --list”, count entries ending in
“: test”, and fail the step when the count is zero; only invoke the existing
Miri test command when at least one matching test is discovered.

In `@CLAUDE.md`:
- Around line 225-234: Update the cached-hit handle identity checks described in
the TypedStep cache path so mismatches are rejected unconditionally in release
builds, rather than relying solely on the four debug_assert! checks. Preserve
each debug_assert! as an additional diagnostic, and ensure a different
ChainContexts or handle identity cannot use the stale transmuted reference.

In `@crates/fgumi-pipeline-core/src/handles.rs`:
- Around line 199-207: Update the Ordered branch handling in the Unpushed
conversion logic so the ordinal=None invariant violation always fails loudly in
release builds; remove the self.push(item) fallback that allocates a new
ordinal, while preserving the existing debug assertion context and Direct-branch
behavior.

In `@crates/fgumi-pipeline-core/src/runtime/contexts.rs`:
- Around line 313-328: Update find_producer and/or build_consumer_map to detect
multiple producer branches targeting a consumer whose input_arity is 1, and
reject the graph instead of retaining only the first edge. Preserve the existing
producer lookup for valid graphs while ensuring duplicate incoming edges cannot
leave an unpopped branch during context construction.

In `@crates/fgumi-pipeline-core/src/runtime/driver.rs`:
- Around line 105-149: The sticky loop around dispatch_one_step must not retry
indefinitely when Progress leaves the output held. Bound sticky retries and
ensure a held-output Progress yields to downstream/round-robin scheduling
without immediately restarting the sticky owner; preserve normal sticky
continuation when progress can proceed. Add a one-worker regression test using a
full bounded output and a draining downstream step to verify forward progress.

In `@crates/fgumi-pipeline-core/src/runtime/fused.rs`:
- Around line 129-152: Update build_fused_output_set so fused transport queues
retain each profile’s configured count or byte bound instead of replacing it
with QueueSpec::Unbounded. Preserve the existing no-reorder optimization and
ensure run_fused_single_thread continues using the bounded queue specifications.

In `@crates/fgumi-pipeline-core/src/runtime/stats.rs`:
- Around line 24-33: The module documentation in the “What’s not recorded yet”
section is stale: remove or revise the queue-depth bullet to reflect that
runtime::sampler::sample_once records depth through EdgeMetrics::record_depth
and snapshot_with_edges exposes edge occupancy, without claiming
InputHandle::depth() is still required.
- Around line 641-659: Update StatsSnapshot::detached and
PipelineStats::detached_snapshot to carry the detached step index as the first
tuple field, and adjust write_detached plus its documentation for the new shape.
In the SPIN detection loop, match detached entries by StepIdx rather than step
name so duplicate names remain independent. Update
verdict_skips_spin_for_detached_steps and any affected tuple destructuring to
assert the index-based behavior.

In `@crates/fgumi-pipeline-core/src/tests.rs`:
- Around line 725-727: Update the doc comment for the Step2 merger test near the
ordered-input description to state that the test preserves and verifies the
emitted order. Remove the claims that cross-branch pairing is interleave-based
and that the assertion checks only a multiset, while leaving the exact ordered
assertion unchanged.
- Around line 548-559: Update run_with_deadlock_timeout to distinguish channel
disconnection from an actual timeout: when recv_timeout reports disconnection,
join the worker handle immediately so a panic in build_and_run is propagated;
only report the deadlock/stalled assertion for a true timeout, while preserving
successful completion handling.
- Around line 331-342: Update the completion guards in both the shown step
implementation and OrderedPairSummer so StepOutcome::Finished is returned only
when ctx.a and ctx.b are drained and pending_a and pending_b are both None.
Preserve the existing pair-processing and NoProgress behavior, ensuring any
buffered item is handled before completion.

In
`@crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr`:
- Around line 12-21: Remove the rustc trait-implementation help blocks from both
compile-fail fixtures: in
crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
at lines 12-21 and 39-48, remove both HeapSize inventory copies; in
crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
at lines 12-16 and 34-38, remove both Ordered-for-Sequenced copies. Preserve the
E0277 diagnostics and the required by a bound in OrderedBytesSingle notes.
- Around line 12-21: Remove the inventory-sensitive “other types implement trait
HeapSize” lists from both compile-fail diagnostic expectations, while preserving
the E0277 errors and the “required by a bound in OrderedBytesSingle” notes. Keep
the normalized “and $N others” placeholder only if it remains part of the
retained diagnostic output.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 461ff2ba-e3c8-490e-aa6d-b73a22d60930

📥 Commits

Reviewing files that changed from the base of the PR and between 27d5d46 and 986059e.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (43)

• .coderabbit.yaml
• .github/workflows/miri.yml
• .github/workflows/publish.yml
• CLAUDE.md
• Cargo.toml
• crates/fgumi-cli-macros/Cargo.toml
• crates/fgumi-pipeline-core/Cargo.toml
• crates/fgumi-pipeline-core/src/builder.rs
• crates/fgumi-pipeline-core/src/erased.rs
• crates/fgumi-pipeline-core/src/finalize.rs
• crates/fgumi-pipeline-core/src/handles.rs
• crates/fgumi-pipeline-core/src/header.rs
• crates/fgumi-pipeline-core/src/held.rs
• crates/fgumi-pipeline-core/src/item.rs
• crates/fgumi-pipeline-core/src/lib.rs
• crates/fgumi-pipeline-core/src/outputs.rs
• crates/fgumi-pipeline-core/src/queues.rs
• crates/fgumi-pipeline-core/src/reorder.rs
• crates/fgumi-pipeline-core/src/runtime/contexts.rs
• crates/fgumi-pipeline-core/src/runtime/detached.rs
• crates/fgumi-pipeline-core/src/runtime/drain.rs
• crates/fgumi-pipeline-core/src/runtime/driver.rs
• crates/fgumi-pipeline-core/src/runtime/fused.rs
• crates/fgumi-pipeline-core/src/runtime/live.rs
• crates/fgumi-pipeline-core/src/runtime/metrics.rs
• crates/fgumi-pipeline-core/src/runtime/mod.rs
• crates/fgumi-pipeline-core/src/runtime/pool.rs
• crates/fgumi-pipeline-core/src/runtime/sampler.rs
• crates/fgumi-pipeline-core/src/runtime/scheduler.rs
• crates/fgumi-pipeline-core/src/runtime/stats.rs
• crates/fgumi-pipeline-core/src/runtime/storage.rs
• crates/fgumi-pipeline-core/src/runtime/worker_core.rs
• crates/fgumi-pipeline-core/src/signal.rs
• crates/fgumi-pipeline-core/src/step.rs
• crates/fgumi-pipeline-core/src/tests.rs
• crates/fgumi-pipeline-core/src/topology.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
• crates/fgumi-pipeline-core/tests/compile_fail.rs

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 10

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/miri.yml:
- Around line 86-112: Add a pre-run location assertion in the Miri workflow step
that verifies every approved unsafe site remains within the erased module,
rather than only checking that the erased test filter matches tests. Validate
the source locations of the four targeted #[allow(unsafe_code)] sites and fail
with a clear error if any moves outside erased; keep the existing test-count
guard and Miri command unchanged.

In `@crates/fgumi-pipeline-core/src/erased.rs`:
- Around line 835-839: Update the chain-building code around
StepOutputs::build_queues to bind self.inner.profile() once, then reuse that
profile for both output_queues and branch_ordering instead of calling profile()
separately. Match the existing single-input adapter’s one-time profile binding
pattern.

In `@crates/fgumi-pipeline-core/src/handles.rs`:
- Around line 1383-1384: Replace every plain tuple builder’s branch construction
with build_branch_byte_aware: update the two calls at
crates/fgumi-pipeline-core/src/handles.rs lines 1383-1384, the three calls at
lines 1451-1453, and the four calls at lines 1533-1536. This must cover all
branches while preserving build_branch behavior for non-byte queue
specifications.

In `@crates/fgumi-pipeline-core/src/queues.rs`:
- Around line 530-548: Update the byte_bounded_slot_cap_reject_is_observable
test to fill all BYTE_BOUNDED_QUEUE_SLOT_CAPACITY slots with nonzero-size Heavy
values, using a sufficiently large byte limit. Capture current_bytes() before
the extra push, assert that the slot-cap rejection occurs, and verify
current_bytes() remains unchanged afterward to cover byte rollback.

In `@crates/fgumi-pipeline-core/src/runtime/sampler.rs`:
- Around line 40-47: The sampler loop currently reads ordered-edge depth
separately through sample_once and write_row, causing two state-lock
acquisitions and potentially inconsistent values. Update the sampler flow around
sample_once and write_row to compute each edge’s buffer depth once per tick,
pass the captured depths to both consumers, and revise the module documentation
to reflect the resulting lock/read behavior; preserve existing sampling and
trace output semantics.

In `@crates/fgumi-pipeline-core/src/runtime/scheduler.rs`:
- Around line 20-22: Correct the module documentation near run_worker_loop to
state that sources and sinks remain on the sticky fast-path while the sticky
owner is still included in the round_robin_dispatch walk; only its priority
restart is suppressed. Remove the claim that sticky steps are unaffected or
skipped by the round-robin scheduling, including under Reverse ordering.

In `@crates/fgumi-pipeline-core/src/runtime/storage.rs`:
- Around line 178-188: Update affinity_in_range to validate the resolved worker
returned by Affinity::target_worker against n_workers, rather than matching
affinity variants directly. Preserve the predicate’s existing boolean behavior
and use target_worker as the sole affinity-to-worker mapping.
- Around line 379-387: Add a test alongside
serial_out_of_range_worker_panics_in_storage that constructs an Exclusive step
with an out-of-range Worker affinity, invokes build_worker_storage with the same
invalid owner index, and uses should_panic to assert the “Exclusive owner-range”
failure message, ensuring the always-on assert remains covered.

In `@crates/fgumi-pipeline-core/src/step.rs`:
- Around line 59-90: Update the `Affinity` enum documentation to state that
affinity hints are ignored for `Parallel`, `Exclusive`, and `Detached` step
kinds, matching the behavior in `build_worker_storage` where affinity is handled
only for `StepKind::Serial`.

In `@crates/fgumi-pipeline-core/src/tests.rs`:
- Around line 724-751: Update OrderedPairSummer::try_run so rejected
ctx.outputs.push operations return StepOutcome::NoProgress instead of
StepOutcome::Contention in both the held-item retry branch and the newly
generated item branch, while preserving the existing held-item retry behavior
and item retention.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 08cd39cc-a3ee-415c-9058-a1d50057f778

📥 Commits

Reviewing files that changed from the base of the PR and between 986059e and 4e849b1.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (43)

• .coderabbit.yaml
• .github/workflows/miri.yml
• .github/workflows/publish.yml
• CLAUDE.md
• Cargo.toml
• crates/fgumi-cli-macros/Cargo.toml
• crates/fgumi-pipeline-core/Cargo.toml
• crates/fgumi-pipeline-core/src/builder.rs
• crates/fgumi-pipeline-core/src/erased.rs
• crates/fgumi-pipeline-core/src/finalize.rs
• crates/fgumi-pipeline-core/src/handles.rs
• crates/fgumi-pipeline-core/src/header.rs
• crates/fgumi-pipeline-core/src/held.rs
• crates/fgumi-pipeline-core/src/item.rs
• crates/fgumi-pipeline-core/src/lib.rs
• crates/fgumi-pipeline-core/src/outputs.rs
• crates/fgumi-pipeline-core/src/queues.rs
• crates/fgumi-pipeline-core/src/reorder.rs
• crates/fgumi-pipeline-core/src/runtime/contexts.rs
• crates/fgumi-pipeline-core/src/runtime/detached.rs
• crates/fgumi-pipeline-core/src/runtime/drain.rs
• crates/fgumi-pipeline-core/src/runtime/driver.rs
• crates/fgumi-pipeline-core/src/runtime/fused.rs
• crates/fgumi-pipeline-core/src/runtime/live.rs
• crates/fgumi-pipeline-core/src/runtime/metrics.rs
• crates/fgumi-pipeline-core/src/runtime/mod.rs
• crates/fgumi-pipeline-core/src/runtime/pool.rs
• crates/fgumi-pipeline-core/src/runtime/sampler.rs
• crates/fgumi-pipeline-core/src/runtime/scheduler.rs
• crates/fgumi-pipeline-core/src/runtime/stats.rs
• crates/fgumi-pipeline-core/src/runtime/storage.rs
• crates/fgumi-pipeline-core/src/runtime/worker_core.rs
• crates/fgumi-pipeline-core/src/signal.rs
• crates/fgumi-pipeline-core/src/step.rs
• crates/fgumi-pipeline-core/src/tests.rs
• crates/fgumi-pipeline-core/src/topology.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
• crates/fgumi-pipeline-core/tests/compile_fail.rs

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 6

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-pipeline-core/src/handles.rs`:
- Around line 282-290: Update BranchInputHandle::pop so record_pop receives
heap_size() only for byte-bounded edges; pass zero for CountBoundedQueue and
UnboundedQueue cases. Preserve the existing record_empty behavior when no item
is popped and reordering is not blocked.

In `@crates/fgumi-pipeline-core/src/held.rs`:
- Around line 78-89: Update the double_put_panics test for HeldSlot::put to
assert the specific expected panic message rather than only checking that
catch_unwind returned an error. Match the established always-on guard pattern,
using the exact duplicate-put message emitted by put while preserving coverage
in both debug and release builds.

In `@crates/fgumi-pipeline-core/src/outputs.rs`:
- Around line 340-354: Add build_queues/mark_all_drained round-trip tests for
the missing Single<T> and OrderedBytesSingle<T> output shapes, alongside the
existing shape tests. In each test, verify one branch is created, its typed
input starts open, and StepOutputs::mark_all_drained closes it; keep the comment
accurate by ensuring all declared shapes are covered.

In `@crates/fgumi-pipeline-core/src/queues.rs`:
- Around line 325-332: Update ByteBoundedQueue::set_limit_bytes to clamp
new_limit to at least 1 before storing it, using the existing atomic limit_bytes
field and preserving the current relaxed ordering.

In `@crates/fgumi-pipeline-core/src/runtime/detached.rs`:
- Around line 578-608: Update detached_two_sided_no_deadlock to retain each
value popped by the consumer instead of only incrementing received, then assert
the sorted collected values match the sorted input multiset. Preserve the
existing completion and deadlock checks, and use the producer’s input values to
validate record identity despite retry-induced reordering.
- Around line 110-114: Update detached_group() to preserve and return the
original DetachedGroup label instead of unconditionally returning PerStep.
Extend the detached placeholder state and the extract_detached_steps swap site
so the real group, including Shared labels, is carried into each placeholder
while retaining the existing behavior for other metadata.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 6bcba3f9-2104-46ee-a0af-db9e2aee6edf

📥 Commits

Reviewing files that changed from the base of the PR and between 4e849b1 and 72c4efa.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (43)

• .coderabbit.yaml
• .github/workflows/miri.yml
• .github/workflows/publish.yml
• CLAUDE.md
• Cargo.toml
• crates/fgumi-cli-macros/Cargo.toml
• crates/fgumi-pipeline-core/Cargo.toml
• crates/fgumi-pipeline-core/src/builder.rs
• crates/fgumi-pipeline-core/src/erased.rs
• crates/fgumi-pipeline-core/src/finalize.rs
• crates/fgumi-pipeline-core/src/handles.rs
• crates/fgumi-pipeline-core/src/header.rs
• crates/fgumi-pipeline-core/src/held.rs
• crates/fgumi-pipeline-core/src/item.rs
• crates/fgumi-pipeline-core/src/lib.rs
• crates/fgumi-pipeline-core/src/outputs.rs
• crates/fgumi-pipeline-core/src/queues.rs
• crates/fgumi-pipeline-core/src/reorder.rs
• crates/fgumi-pipeline-core/src/runtime/contexts.rs
• crates/fgumi-pipeline-core/src/runtime/detached.rs
• crates/fgumi-pipeline-core/src/runtime/drain.rs
• crates/fgumi-pipeline-core/src/runtime/driver.rs
• crates/fgumi-pipeline-core/src/runtime/fused.rs
• crates/fgumi-pipeline-core/src/runtime/live.rs
• crates/fgumi-pipeline-core/src/runtime/metrics.rs
• crates/fgumi-pipeline-core/src/runtime/mod.rs
• crates/fgumi-pipeline-core/src/runtime/pool.rs
• crates/fgumi-pipeline-core/src/runtime/sampler.rs
• crates/fgumi-pipeline-core/src/runtime/scheduler.rs
• crates/fgumi-pipeline-core/src/runtime/stats.rs
• crates/fgumi-pipeline-core/src/runtime/storage.rs
• crates/fgumi-pipeline-core/src/runtime/worker_core.rs
• crates/fgumi-pipeline-core/src/signal.rs
• crates/fgumi-pipeline-core/src/step.rs
• crates/fgumi-pipeline-core/src/tests.rs
• crates/fgumi-pipeline-core/src/topology.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
• crates/fgumi-pipeline-core/tests/compile_fail.rs

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-pipeline-core/src/lib.rs`:
- Around line 7-10: Update the dependency summary in the crate-level
documentation to include anyhow alongside the existing documented dependencies,
reflecting the anyhow::Result import used by finalize.rs.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: c3b16a72-f6a7-4158-9f9f-91bd660c4ec6

📥 Commits

Reviewing files that changed from the base of the PR and between 7576f13 and e019868.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (43)

• .coderabbit.yaml
• .github/workflows/miri.yml
• .github/workflows/publish.yml
• CLAUDE.md
• Cargo.toml
• crates/fgumi-cli-macros/Cargo.toml
• crates/fgumi-pipeline-core/Cargo.toml
• crates/fgumi-pipeline-core/src/builder.rs
• crates/fgumi-pipeline-core/src/erased.rs
• crates/fgumi-pipeline-core/src/finalize.rs
• crates/fgumi-pipeline-core/src/handles.rs
• crates/fgumi-pipeline-core/src/header.rs
• crates/fgumi-pipeline-core/src/held.rs
• crates/fgumi-pipeline-core/src/item.rs
• crates/fgumi-pipeline-core/src/lib.rs
• crates/fgumi-pipeline-core/src/outputs.rs
• crates/fgumi-pipeline-core/src/queues.rs
• crates/fgumi-pipeline-core/src/reorder.rs
• crates/fgumi-pipeline-core/src/runtime/contexts.rs
• crates/fgumi-pipeline-core/src/runtime/detached.rs
• crates/fgumi-pipeline-core/src/runtime/drain.rs
• crates/fgumi-pipeline-core/src/runtime/driver.rs
• crates/fgumi-pipeline-core/src/runtime/fused.rs
• crates/fgumi-pipeline-core/src/runtime/live.rs
• crates/fgumi-pipeline-core/src/runtime/metrics.rs
• crates/fgumi-pipeline-core/src/runtime/mod.rs
• crates/fgumi-pipeline-core/src/runtime/pool.rs
• crates/fgumi-pipeline-core/src/runtime/sampler.rs
• crates/fgumi-pipeline-core/src/runtime/scheduler.rs
• crates/fgumi-pipeline-core/src/runtime/stats.rs
• crates/fgumi-pipeline-core/src/runtime/storage.rs
• crates/fgumi-pipeline-core/src/runtime/worker_core.rs
• crates/fgumi-pipeline-core/src/signal.rs
• crates/fgumi-pipeline-core/src/step.rs
• crates/fgumi-pipeline-core/src/tests.rs
• crates/fgumi-pipeline-core/src/topology.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
• crates/fgumi-pipeline-core/tests/compile_fail.rs

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-pipeline-core/src/tests.rs`:
- Around line 1166-1187: Update the manifest header parsing in the dependency
collection loop so `[dependencies.<name>]` is recognized as a runtime dependency
and the `<name>` segment is added to `dependencies`. Preserve the existing
handling for `[dependencies]` and target-specific `*.dependencies` sections,
while continuing to exclude dev- and build-dependency sections.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 08c48695-6c5d-4407-99b2-26ecd4297797

📥 Commits

Reviewing files that changed from the base of the PR and between 7576f13 and a581ae9.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (43)

• .coderabbit.yaml
• .github/workflows/miri.yml
• .github/workflows/publish.yml
• CLAUDE.md
• Cargo.toml
• crates/fgumi-cli-macros/Cargo.toml
• crates/fgumi-pipeline-core/Cargo.toml
• crates/fgumi-pipeline-core/src/builder.rs
• crates/fgumi-pipeline-core/src/erased.rs
• crates/fgumi-pipeline-core/src/finalize.rs
• crates/fgumi-pipeline-core/src/handles.rs
• crates/fgumi-pipeline-core/src/header.rs
• crates/fgumi-pipeline-core/src/held.rs
• crates/fgumi-pipeline-core/src/item.rs
• crates/fgumi-pipeline-core/src/lib.rs
• crates/fgumi-pipeline-core/src/outputs.rs
• crates/fgumi-pipeline-core/src/queues.rs
• crates/fgumi-pipeline-core/src/reorder.rs
• crates/fgumi-pipeline-core/src/runtime/contexts.rs
• crates/fgumi-pipeline-core/src/runtime/detached.rs
• crates/fgumi-pipeline-core/src/runtime/drain.rs
• crates/fgumi-pipeline-core/src/runtime/driver.rs
• crates/fgumi-pipeline-core/src/runtime/fused.rs
• crates/fgumi-pipeline-core/src/runtime/live.rs
• crates/fgumi-pipeline-core/src/runtime/metrics.rs
• crates/fgumi-pipeline-core/src/runtime/mod.rs
• crates/fgumi-pipeline-core/src/runtime/pool.rs
• crates/fgumi-pipeline-core/src/runtime/sampler.rs
• crates/fgumi-pipeline-core/src/runtime/scheduler.rs
• crates/fgumi-pipeline-core/src/runtime/stats.rs
• crates/fgumi-pipeline-core/src/runtime/storage.rs
• crates/fgumi-pipeline-core/src/runtime/worker_core.rs
• crates/fgumi-pipeline-core/src/signal.rs
• crates/fgumi-pipeline-core/src/step.rs
• crates/fgumi-pipeline-core/src/tests.rs
• crates/fgumi-pipeline-core/src/topology.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.rs
• crates/fgumi-pipeline-core/tests/compile-fail/chain_input_type_mismatch.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_heapsize.stderr
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.rs
• crates/fgumi-pipeline-core/tests/compile-fail/ordered_bytes_single_requires_ordered.stderr
• crates/fgumi-pipeline-core/tests/compile_fail.rs