CI fix: failing checks on PR #82 (branch technocore/issue-81-fix) #83

Activity

1. BnJam

   added

   agent-taskApproved for agent processingApproved for agent processing

   technocore/generated

   and removed

   technocore/generated

   on Aug 1, 2026
2. BnJam commented on Aug 1, 2026

   

   BnJam

   on Aug 1, 2026

   OwnerAuthor

   More actions

   Starting work on this issue. I'll investigate the failing tox (Python 3.12) checks on PR #82, branch technocore/issue-81-fix, then open a fix-up PR targeting that branch.
3. BnJam commented on Aug 1, 2026

   

   BnJam

   on Aug 1, 2026

   OwnerAuthor

   More actions

   Opened PR: #84
4. BnJam commented on Aug 1, 2026

   

   BnJam

   on Aug 1, 2026

   OwnerAuthor

   More actions

   technocore completed pipeline issue-handler.

   Found it: segfault (exit code -11) in trend_analysis at python/eo_processor/__init__.py:378 during test_trend_analysis_detects_break. Let me examine the code.
   The segfault is a stack overflow from infinite recursion in src/trends.rs:73-75. When max_residual_index == 0, y.split_at(0) returns left=[] and right=y (the whole array), so recursive_trend_analysis recurses forever on the same slice. Let me reproduce it.
   No Python/Rust toolchain available locally, so I'll simulate the exact Rust algorithm in JS to find the recursion failure.
   Confirmed — infinite recursion / stack overflow on the exact test input. Let me see the recursion path.
   Root cause confirmed: for the "∧"-shaped test input, the full-array regression (a flat line at 5) has max-abs residual 5.0 at indices 0, 49, and 99; Rust's max_by keeps the first, so max_residual_index = 0. Then y.split_at(0) gives left=[] and right=y — no progress → infinite recursion → stack overflow (segfault, exit -11).

   Let me verify a fix in simulation: split at max(1, max_residual_index) so both sides always strictly shrink.
   The fix terminates but produces only 1 segment [47,99], which would fail the test's len(segments) >= 2 assertion. Let me check what PR #82 actually changed and whether the Rust function is new.
   The Rust trend_analysis (added in 8841a10) has a latent infinite-recursion bug exposed by PR #82's new test. Let me examine the exact residuals to understand the right fix.
   In JS (also IEEE-754 f64), index 0's residual is -5.000000000000003 — a hair larger than the peak's +5, so maxIdx=0. The data has two peaks (indices 49 and 50 both equal 10). Let me trace the peeling sequence with the fix.
   The clamp fix peels 53 points from the start (all dropped) and yields 1 segment — still failing the test's >= 2 requirement. The real issue: for a symmetric "∧", the flat-line fit's max-abs residual is at an endpoint, not the interior peak. The correct fix is to only split at interior points (1..n-2), which both guarantees progress (no degenerate split_at(0)) and finds the true break at the peak. Let me verify.
   The interior-only approach produces exactly the 2 correct segments [0,48] and [49,99]. Let me re-simulate with the exact proposed Rust structure (base case len<2 returns, max<=threshold pushes, else interior split).
   Fix verified: 2 segments [0,48],[49,99] for the "∧" test, 1 segment for the no-break test, and no stack overflow. Let me check for a CHANGELOG and docs references before implementing.
   No CHANGELOG. Now implementing the fix in src/trends.rs.
   The fix is in place. Let me check if I can validate the Rust code compiles — try installing a Rust toolchain.
   Network is available. Let me install a minimal Rust toolchain to validate the actual Rust algorithm (as a standalone harness, since pyo3 needs Python headers) and run cargo fmt --check.
   No C compiler, so a full build isn't possible. But cargo fmt --check only needs rustfmt — let me run it.
   The fmt diffs are pre-existing in src/indices.rs (unrelated to my change). Let me check if src/trends.rs specifically has any formatting issues.
   No formatting issues in src/trends.rs. Let me confirm the src/indices.rs fmt diffs are pre-existing on main (not from PR #82).
   PR #82 only touches Python/docs/test files; the src/indices.rs fmt diff is pre-existing on main. My change to src/trends.rs is the only fix needed. Let me commit.
   Done. Summary of the fix for issue #83:

   Root cause: test_trend_analysis_detects_break (tests/test_trends.py:34) segfaulted (exit -11) due to infinite recursion in recursive_trend_analysis at src/trends.rs:37. For the symmetric "∧" input, the whole-array regression is a flat line whose max-abs residual lands on an endpoint (index 0, from floating-point rounding: -5.000000000000003 vs peak +5). y.split_at(0) yields left=[] and right=y — an unchanged slice — so the recursion never shrank and overflowed the stack.

   Fix (src/trends.rs:66-74): restrict the split candidate to interior indices (1..len-2) via .filter, guaranteeing both sides strictly shrink (left ≥ 1, right ≥ 2 elements) so recursion always terminates, and the true break at the interior peak (index 49) is found. Verified via an exact IEEE-754 simulation: "∧" → 2 segments [0,48]/[49,99]; no-break → 1 segment.

   PR: #84 (base technocore/issue-81-fix, as required — branched from the failing PR's branch, not main).
5. BnJam

   removed

   agent-taskApproved for agent processingApproved for agent processing

   on Aug 1, 2026