Skip to content

common-issues.md

Latest commit

 

History

History
109 lines (80 loc) · 3.56 KB

common-issues.md

File metadata and controls

109 lines (80 loc) · 3.56 KB

Common Issues

For response sequencing and closure criteria, start with incident-playbook.md and use this file for pattern-specific fixes.

1. make check fails with type/test errors

Symptoms:

  • Nonzero exit
  • Compiler or assertion failures

Actions:

  1. Re-run make check.
  2. Run failing test target directly for focused logs.
  3. Fix root cause.
  4. Re-run make check and governance gates.

2. ./scripts/dp ... returns ok: false

Symptoms:

  • JSON output contains failing step(s)

Actions:

  1. Identify first failed step in JSON.
  2. Run that command directly.
  3. Resolve failure.
  4. Re-run full dp chain.

3. Python binding build failures

Symptoms:

  • maturin develop errors

Actions:

  1. Verify Python and uv versions.
  2. Re-run uv sync.
  3. Ensure Rust build succeeds first.
  4. Re-run binding install.

4. Go binding test failures

Symptoms:

  • go test ./... fails in rlm-core/go/rlmcore

Actions:

  1. Confirm static Rust library exists in expected path.
  2. Rebuild Rust library.
  3. Re-run Go tests.

5. Drift detection output appears unstable

Symptoms:

  • Different drift ordering/results on repeated runs

Actions:

  1. Re-run targeted sync/drift tests.
  2. Verify input ordering assumptions.
  3. Check symbol interning / temporary allocation code paths.
  4. Capture deterministic repro fixture before patching.

6. Spec-agent output includes placeholders unexpectedly

Symptoms:

  • Generated artifacts include draft: annotations when strict baseline output was expected

Actions:

  1. Verify CompletenessMode configuration.
  2. Use Baseline for placeholder-free stubs.
  3. Re-run formalization and tests.

7. "It works locally" but gates fail elsewhere

Symptoms:

  • Local ad hoc run passes, policy gates fail

Actions:

  1. Trust gate outputs over ad hoc confidence.
  2. Capture exact environment details.
  3. Align local commands with policy commands.
  4. Re-run from clean state.

8. Ignored Lean/REPL integration tests hang or leave subprocesses behind

Symptoms:

  • cargo test ... -- --ignored appears stuck
  • Later runs show odd REPL startup errors
  • Background rlm_repl or Lean repl processes linger

Actions:

  1. Run ignored subprocess tests serially: cd /Users/rand/src/loop/rlm-core && cargo test --no-default-features --features gemini test_repl_spawn -- --ignored --test-threads=1 && cargo test --no-default-features --features gemini test_lean_repl_spawn -- --ignored --test-threads=1
  2. Treat runtime > 120s as a failure signal and capture logs to evidence.
  3. Scan for leftovers: ps -axo pid=,command=,rss= -ww | rg -n "rlm_repl|lake env repl|\\brepl\\b" -S
  4. If environment prerequisites are missing (Python package path, Lean toolchain), record the actionable stderr and classify as environment blocker, not runtime success.
  5. Re-run after fixing prerequisites; unattended runs should complete without manual process cleanup.

9. Proof automation reports "Missing proof state"

Symptoms:

  • Tactic execution fails immediately with a deterministic missing proof-state error
  • AI tactic validation returns failure before trying any tactic

Actions:

  1. Ensure you are targeting a sorry location with a real proof_state_id (protocol session target selection prefers these).
  2. Run the protocol path (execute_tactic_with_feedback / execute_tactic_with_repl) instead of ad hoc tactic calls without state.
  3. Confirm operation IDs are stateful (sorry:<proof_state>:<idx>) rather than sorry:missing:<idx>.
  4. If Lean response omitted proof state, treat as runtime blocker and capture the raw response in evidence.

Because gravity is also optional until it is not.