execution: Add anti-cycle validation

[rattus128]

I fat fingered a cycle into a large flow without noticing tonight and it took ages to figure out what was wrong.

Currently if the graph contains a cycle, the just inifitiate recursions, hits a catch all then throws a generic error against the output node that seeded the validation. Instead, fail the offending cycling node chain and handling it as an error in its own right.

Example test case:

A workflow with a graph cycle.

Before:

After:

Regression Tests:

Firered template (I have none of the models downloaded): Errors look right:

Linux 5090, stable-cascade -> Flux2 ✅
Linux 5090, LTX2.3 FLF2V ✅

[coderabbitai]

📝 Walkthrough

Walkthrough

The change adds cycle detection to the validate_inputs function by introducing a visiting parameter that tracks the current recursion path. When a circular dependency is detected (a node appears twice in the same traversal), the function records validation failures for all nodes in the cycle and returns immediately. The implementation uses push/pop operations on the visiting list during recursive traversal, protected by try/finally blocks to maintain stack integrity. The validation result logic is updated to merge locally detected errors with pre-recorded cycle-detection results.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'execution: Add anti-cycle validation' clearly and concisely describes the main change—adding cycle detection to prevent infinite recursion in node graphs.
Description check	✅ Passed	The description clearly relates to the changeset by explaining the problem (cycles causing generic errors), the solution (detecting cycles and reporting them explicitly), and providing test evidence.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@execution.py`:
- Around line 793-810: The current order checks validated before visiting, which
causes cached cycle results (validated[unique_id]) to hide later back-edges and
fail to mark all nodes in overlapping cycles; change the logic in the function
that uses unique_id, visiting, and validated so that the visiting membership
test (if unique_id in visiting) runs before the cache lookup (if unique_id in
validated), compute and populate validated entries for all nodes in the detected
cycle as currently done, and only then return validated[unique_id]; ensure you
update references to visiting and validated in that function consistently so
overlapping cycles like B->C->B and B->D->B are both reported.

🪄 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: CHILL

Plan: Pro

Run ID: 8b319c51-8869-40b2-990b-a0b4bd0c385f

📥 Commits

Reviewing files that changed from the base of the PR and between 2a1f402 and 61e29f2.

📒 Files selected for processing (1)

• execution.py

[christian-byrne]

LGTM