Split orbit-core command/search.rs (2,963 lines) into a search/ module

[danieljhkim]

Task

ORB-00354 — Split orbit-core command/search.rs (2,963 lines) into a search/ module

Description

Problem

crates/orbit-core/src/command/search.rs is 2,963 lines: one ~860-line impl OrbitRuntime block (lines ~172-1030) plus a long tail of pure free helper functions that do not depend on OrbitRuntime and extract cleanly. A single file this size is hard to navigate and review.

Why It Matters

Low-risk, high-readability win. The helpers are free functions, so moving them is mechanical (move + adjust use / pub(crate) visibility). Rust allows multiple impl OrbitRuntime blocks across files, so the orchestration impl can stay in mod.rs and later be peeled per-domain (task/doc/learning/adr search) without touching the type.

Proposed Change

Convert search.rs into search/mod.rs and extract the free-function tail into submodules. Group by symbol; line ranges below are approximate (* = approximate, confirm at edit time):

New file	Moves (approx current lines)	Owns
search/types.rs	37-170	GlobalSearchKind / GlobalSearchMode / GlobalSearchParams / GlobalSearchResponse / GlobalSearchHit + their impls
search/path_match.rs	~1496-1557	selector_matches_path, normalize_path_for_match, paths_overlap, is_within, learning_scope_contains_path
search/hybrid.rs	~1095-1352*	*HybridCandidate structs, blend_*, normalized_doc_scores, compare_global_hits_by_score, warn_*_fallback
search/convert.rs	~1031-1062, ~1357-1492*	*_hit_to_global, *_result_to_global, adr_to_global_hit*, lexical_*_hit
search/filters.rs	~1393-1429, ~1559-1714*	*_has_all_tags, SearchStatusFilters, resolve_*_statuses, task_open_statuses, push/extend_unique
search/mod.rs	172-1030	the impl OrbitRuntime orchestration (stays)

Start with path_match.rs and hybrid.rs — most self-contained and independently testable, so they give the biggest readability win for the least churn.

Constraints / Notes

• Pure refactor: no behavior change, no public API change. The orchestration impl OrbitRuntime body stays byte-identical aside from use adjustments.
• crates/orbit-core/src/command/mod.rs:32 declares pub mod search;, which resolves to either search.rs or search/mod.rs — likely no change needed; verify.
• Default new items to pub(crate); keep pub only on symbols already part of the crate's public surface. Anything previously reachable via command::search::... must stay re-exported at the same path from search/mod.rs.
• Follow the sibling test layout (docs/design-patterns/test_layout.md): there is no sibling search test file today; any test extracted alongside a moved symbol goes under search/tests/<file>.rs.
• Do not modify call sites in other crates.

Acceptance Criteria

• crates/orbit-core/src/command/search.rs is removed and replaced by crates/orbit-core/src/command/search/mod.rs plus the extracted submodules types.rs, path_match.rs, hybrid.rs, convert.rs, and filters.rs.
• No resulting file under search/ exceeds ~1,100 lines (the impl OrbitRuntime orchestration in mod.rs is the largest).
• cargo build -p orbit-core succeeds with no new warnings.
• cargo test -p orbit-core passes with no fewer tests than before the split.
• cargo clippy -p orbit-core --all-targets is clean under the workspace lints.
• make ci-fast passes.
• No public symbol path changes: rg 'command::search::' crates resolves to the same symbols (re-exported from search/mod.rs where they were previously reachable), and no call sites in other crates are edited.
• git diff over the moved code shows move/extraction only (import adjustments), with no logic changes to the impl OrbitRuntime body.

Execution Summary

Click to expand

Outcome: success

Changes:

• Split crates/orbit-core/src/command/search.rs into search/mod.rs and focused helper modules: types, path_match, hybrid, convert, and filters.
• Moved the existing inline search tests into sibling search/tests/ modules while preserving all 36 search tests.
• Preserved public command::search::* re-exports and verified the impl OrbitRuntime body is byte-identical to the pre-split source.

Assessment: Pure module extraction with no intended behavior changes; all new files are under the requested size limit and no other crate call sites were edited.

Validation:

• cargo build -p orbit-core: passed.
• cargo test -p orbit-core command::search::tests: passed (36 tests).
• cargo clippy -p orbit-core --all-targets: passed.
• make ci-fast: passed.
• cargo test -p orbit-core: attempted; 442 passed, 3 failed, 1 ignored. The failures are unrelated job owner/process identity tests and match known environment issue L-0025: ps -p $$ -o lstart= returns operation not permitted in this runner.

Design weaknesses / risks:

• Full local cargo test -p orbit-core could not be green in this execution environment because process identity tests require ps; CI or a runner with ps access should re-run the full suite.

Validation

• Not reported

Branch Freshness

• Base ref: origin/agent-main
• Head ref: orbit/ORB-00354-6a1be149
• Behind base: 0
• Ahead of base: 1