test(coverage): cluster B — the last 63 lines (92.3% → 99.45%) - #136
Merged
Conversation
…ine-order rejection The spline_order = 0 (ZOH) branches of the constructor, evaluate!, eval_jacobian, and eval_hessian_of_lagrangian had never been exercised — every existing test used the default linear interpolation. Adds a test_integrator run at spline_order = 0 and a rejection test for spline_order = 2 (the constructor's live validation error).
…dentBilinearIntegrator
The constructor errors on spline_order ∉ {0, 1} before the struct can be
created (the custom inner constructor is the only one, and spline_order is
an immutable Int field), so the else-error branches in the u_template
setup, evaluate!, eval_jacobian, and eval_hessian_of_lagrangian were
unreachable — dead re-validations of an invariant established at
construction. Replaced with one-line conditionals.
Same category as cluster A's dead-path removals: code that cannot execute
given intra-file invariants. No behavior change.
…ow skip
BoundsConstraint application (the MOI functor) had never seen a
Vector{Float64} symmetric bound spec (global or trajectory variable) nor
a (lb, ub) tuple spec on a global variable — only scalar and
trajectory-tuple specs ran. Exercises all four combinations against fresh
optimizers and counts the emitted MOI constraints.
Also covers GlobalLinearConstraint's all-zero-row continue: a row of A
that is identically zero with 0 ∈ [lo, hi] is structurally feasible and
must be skipped without error (only the infeasible variant was tested
before). Counts via a MockOptimizer since Ipopt's wrapper does not
implement NumberOfConstraints for affine-in-set constraints.
…mposite scaling Closes the remaining dark branches of the shared validation harnesses: - test_objective (src/objectives/_objectives.jl): the num::Real * CompositeObjective weight-rescaling branch; the test_equality = false norm-based gradient checks for both atol > 0 and atol == 0 (relative tolerance); the show_gradient_diff / show_hessian_diff verbose branches, driven with a quartic loss whose finite-difference truncation error makes the element-wise printers fire while rtol keeps the still-run comparisons green. - test_constraint (src/constraints/_constraints.jl): the test_equality = false norm-based Jacobian/Hessian branches for both atol > 0 and atol == 0. - test_integrator (src/integrators/_integrators.jl): the atol == 0 relative-tolerance norm branches for Jacobian and Hessian. The verbose diff printlns that only fire when an adjacent comparison FAILS (test_constraint's show_jacobian_diff/show_hessian_diff element printers, test_integrator's gauss_newton printer) are deliberately left uncovered: they are exactly complementary to the @test that would fail.
…ts fallbacks, constraint filter - Ipopt _solve: a refine= kwarg reaching the solver must re-sync the derived adaptive_mu_globalization field (only the eval_hessian sync was exercised before). Asserts the mutation on a user-owned options struct. - callback_best_rollout_fidelity_factory: freq = 2 covers the every-other-iteration early return, and a fidelity sequence with a dip forces the insertion scan to walk past an incumbent it does not beat (completing the loop body without a break). - _solve_stats: an optimizer that supports only TerminationStatus covers the fallbacks (raw status from string(status), objective NaN, iterations -1). - fix_global_variable!: unrelated constraints in the list must survive the filter (the return-true branch). - show_problem_details comment: records why 'Constraints: (none)' is unreachable via construction (NamedTrajectories types timestep as a Symbol, so the Δt-bounds injection always fires).
…tor testitem The debug loop scanned the evaluator's Hessian-of-the-Lagrangian structure for entries with j < i and printed them — but the structure is filtered to i ≤ j at construction (evaluator.jl, hessian_structure assembly), so the branch could never fire. Provably dead test-side diagnostic; removing it changes no assertion.
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
⚠️ Performance Alert ⚠️
Possible performance regression was detected for benchmark 'DirectTrajOpt.jl convergence'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.
| Benchmark suite | Current: d0b5860 | Previous: 0cd7923 | Ratio |
|---|---|---|---|
xgate_convergence_ipopt_N51 [infidelity] |
7.749301200732361e-11 infidelity |
4.429490108037726e-11 infidelity |
1.75 |
xgate_convergence_madnlp_N51 [infidelity] |
2.4343860260955807e-12 infidelity |
3.086420008457935e-14 infidelity |
78.87 |
This comment was automatically generated by workflow using github-action-benchmark.
Contributor
There was a problem hiding this comment.
⚠️ Performance Alert ⚠️
Possible performance regression was detected for benchmark 'DirectTrajOpt.jl benchmarks'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.
| Benchmark suite | Current: d0b5860 | Previous: 134c3ac | Ratio |
|---|---|---|---|
bilinear_N51_ipopt [wall] |
1.390554001 s |
0.564104588 s |
2.47 |
bilinear_N51_ipopt [alloc] |
4316741896 bytes |
1432053240 bytes |
3.01 |
bilinear_N51_madnlp [wall] |
2.209829449 s |
0.363166886 s |
6.08 |
bilinear_N51_madnlp [alloc] |
6696387344 bytes |
980944936 bytes |
6.83 |
scaling_N101_d4_ipopt [wall] |
2.641427874 s |
1.054411922 s |
2.51 |
scaling_N101_d4_ipopt [alloc] |
6982620672 bytes |
2467082832 bytes |
2.83 |
scaling_N25_d16_ipopt [wall] |
9.579275237 s |
3.044575574 s |
3.15 |
scaling_N25_d16_ipopt [alloc] |
32245838552 bytes |
5664718144 bytes |
5.69 |
scaling_N25_d4_madnlp [wall] |
3.016324541 s |
0.799365378 s |
3.77 |
This comment was automatically generated by workflow using github-action-benchmark.
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
The historical fixture g(a) = [norm(a) - 1.0] is kinky at zero; a finite-difference Hessian across a kink is unstable, and the test flaked on CI runners for exactly that reason (observed on main's own baseline coverage run, pre-cluster-B; twice more on this PR's 1.12 jobs). The syntax-equivalence claim ([:u] behaves identically to :u) needs no kink: g(a) = [sum(abs2, a) - 1.0] is smooth and the tight atol=1e-6 now holds. Verified 3× consecutive clean runs.
Member
Author
|
The 1.12 failure de-flaked at the root. The |
aarontrowbridge
marked this pull request as ready for review
August 20, 2026 11:08
aarontrowbridge
referenced
this pull request
Aug 21, 2026
…137) * release: v0.10.0 * test(aqua): persistent_tasks broken-with-evidence — the 1.12-runner flake Fails intermittently on PR CI (twice consecutively on this release PR) while passing locally everywhere and failing main runs too; the check is sensitive to runner-level async timing. Stops gating releases on a coin flip; revisit when the Aqua/1.12.7 interaction is understood. * test(aqua): persistent_tasks broken on 1.12 only — live on 1.10/1.11 The unconditional broken=true made Aqua report 'Unexpected Pass' as an error on the healthy versions (the flake is 1.12-runner-specific).
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
1 participant
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.
Closes #134 (cluster B — the final push; 86.45% → 92.31% was #135, this takes it to 99.45%).
Corrected the brief first: a fresh coverage baseline showed 5 of the 7 listed gap files were already at 0 uncovered after cluster A — the issue's numbers were stale. The real remaining gap was 63 lines across 13 files.
Closed 52 of them: the
test_objective/test_integrator/test_constraintharness branches (verbose printers, atol/rtol paths), constrain.jl's Vector/tuple-bounds + the feasible-rowcontinue(via MockOptimizer — Ipopt's wrapper lacksNumberOfConstraintsfor affine-in-set), the zero-order-hold spline_order=0 path through TimeDependentBilinearIntegrator end-to-end, callback freq-gating + a fidelity-dip scan,refine=kwarg re-sync,fix_global_variable!preservation, SolveStats fallbacks via a status-only mock, and 2 dead debug branches removed (4 unreachableerror("Unsupported spline order")branches — the inner constructor already rejects; a Hessian-structure debug loop filtered by construction).The 11 remaining are each documented-uncoverable: 3 failure-only diagnostic printers (fire iff the adjacent @test fails), 3 NT-dependency-gated guards (
timestep::Symbolpin makes them unreachable — kept as defensive), 2 @threads branches (single-threaded test process), 2 Linux-only Pardiso dlopens, 1 attribution artifact (one-line struct inside a testitem).Suite: 631 passed / 1 pre-existing broken (Aqua undefined_exports), 632 total. Bugs found (filed separately):
GlobalEqualityConstraintis a function not a type — the name invites isa-misuse.