📐algorithm-binary-search-tests

The groupant beside algorithms/src/binary_search/tests.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.

Facts

Components

Obernants

The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.

Model text

descriptant

Test extraction epoch 2026-10-29

The existing two inline tests were moved into adjacent tests.rs without changing their assertions. Two new tests extend the admitted fixture domain; production behavior is unchanged. Expected cut is the first linear >= position and first match is linear equality position; no interval narrowing is reused.

The mathematical combiner reads Rust/partial syntax, finite acceptance relation, typed models and actual pass/fail observations. This epoch preserves earlier leaf evidence and adds successful repeated runs and a rejected source mutant. Failure fibers remain ambiguous; no automated proof or universal bug-localizer is claimed.

Replacing midpoint < with <= is rejected by wider_duplicate_runs_linear_oracle (compiled external copy, exit 101).

The shared parent must wire this test source as a registered sibling and refresh source bindings. Local models do not establish that integration. Macro bodies remain opaque in the partial syntax projection.

model

binary_search test groupant

Forward

For the explicit fixture set F, run existing and new assertions against super's implementation. Accept iff all observed outputs equal oracle expectations and metamorphic equalities hold. Execution assumes ordinary successful allocations and lawful traits.

Finite domain: Old exhaustive sorted ternary sequences and boundaries; new six lengths {8,31,64,127,257,1025} with 17-value runs and targets -310..380; descending Reverse values -70..70 repeated three times and targets -72..72; translations +1000 for targets -102..202.

Invariant

Production code is accessed through super::*; existing assertions are retained. Expected values do not call the algorithm under test except on the explicitly declared metamorphic sides. All generated values use deterministic arithmetic with explicit wrapping where intended.

Cost

4,146 wider duplicate cut/first cases, 145 Reverse targets and 305 translation targets, in addition to preserved bounded tests. Linear expected search costs O(n) per query; all lengths and target ranges fixed.

nonbehavior

Type: behavior_complement. Let F be the explicit finite fixture set below, O(f) the oracle expectations and U the candidate implementations paired with pass/fail observations over F. R(P,pass) iff every assertion succeeds under the fixed execution assumptions; R(P,fail) iff the terminating execution encounters an assertion mismatch or a caught panic. These two branches define R on the declared terminating, compiled-execution universe. U minus R consists of false pass observations for any mismatch and false failure observations when every assertion holds. Conditional exact harness model; panics count as failures, harness crashes/timeouts are outside this two-outcome universe.

failure_fibers

Type: reverse_reconstruction. Given a failing fixture and observed mismatch, retain all candidate implementations whose observations violate that assertion. Wrong tie policy, wrong bounds, wrong arithmetic and fixture-oracle defects may share the same failure. Passing admits every implementation agreeing on this finite F and leaves all off-fixture behavior unconstrained; no unique fault reconstruction.

oracle_independence

Type: oracle_correspondence. Expected cut is the first linear >= position and first match is linear equality position; no interval narrowing is reused. Agreement is finite evidence, with oracle bugs still possible; the designated mutation is a falsifiable disagreement witness.

transformations

Type: metamorphic_relation. Reverse wrapper changes the lawful order; translation adds 1000 to both sorted values and targets and must preserve cut indices. This relation holds only within admitted domains; it does not identify a unique input from output.

finite_schedule

Type: resource_observation. 4,146 wider duplicate cut/first cases, 145 Reverse targets and 305 translation targets, in addition to preserved bounded tests. Linear expected search costs O(n) per query; all lengths and target ranges fixed. The harness fixes fixture order, seeds and assertions; wall-clock duration and allocation behavior are not deterministic outputs.

evidence

Evidence 2026-10-29

Four focused tests pass twice with identical normalized test statuses: two preserved tests and two new tests. Full crate check at this isolated base passes 36 tests plus one doctest. Owned-source rustfmt check and strict all-target crate Clippy pass. Replacing midpoint < with <= is rejected by wider_duplicate_runs_linear_oracle (compiled external copy, exit 101).

Old exhaustive sorted ternary sequences and boundaries; new six lengths {8,31,64,127,257,1025} with 17-value runs and targets -310..380; descending Reverse values -70..70 repeated three times and targets -72..72; translations +1000 for targets -102..202.

Expected cut is the first linear >= position and first match is linear equality position; no interval narrowing is reused.

4,146 wider duplicate cut/first cases, 145 Reverse targets and 305 translation targets, in addition to preserved bounded tests. Linear expected search costs O(n) per query; all lengths and target ranges fixed.

Commands: cargo test -p mova-algorithms --offline binary_search::tests:: -- --test-threads=1 (twice); cargo clippy -p mova-algorithms --offline --all-targets -- -D warnings. External mutation uses rustc --edition 2024 --test on a copied mod.rs with adjacent tests.rs, filtering only the named new test. Source production prefix is preserved; only test module placement changes. Parent registry and source-binding integration remain separate.

This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/algorithms/src/binary_search/tests.rs.groupant/