Skip to content

ADR-0016 — Dependency-First Implementation

Status

Accepted

Context

OntoIndex/OntoCode spans indexing, diagnostics, authoring, reasoning, diff, and IDE integration. Reimplementing parsers, reasoners, triple stores, or format libraries would duplicate mature Rust ecosystem work and drift from sibling project OntoLogos.

Early milestones already delegate:

Remaining specs (ontoindex-owl, ontoindex-diagnostics, ontoindex-diff, OBO, SHACL) lacked a single policy for choosing external crates vs in-house code.

Decision

Adopt a dependency-first strategy. Canonical inventory: DEPENDENCY_MATRIX.md.

Rules

  1. Do not reimplement parsers, reasoners, triple stores, OBO parsers, or SHACL engines when a maintained Rust crate covers the profile.
  2. ontoindex-* crates are thin facades — catalog bridge, LSP JSON, CLI, caching, limits, VS Code integration. They do not embed ontology calculus.
  3. OntoLogos is the sole reasoner orchestration layer (ADR-0015). No parallel EL/RL/DL stacks (whelk-rs, reasonable, owl-dl-core direct).
  4. External CLIs (ROBOT) are subprocess adapters only — never reimplement merge/report logic (OBO_ROBOT_SPEC.md).
  5. When no crate exists, document logic as intentionally in-house with explicit scope (quality lints, semantic diff UX, LSP custom protocol).
  6. Upstream gaps — track issues/PRs upstream; do not silently fork (aligned with OntoLogos dependency-first).

Adding a dependency

  1. Update DEPENDENCY_MATRIX.md.
  2. Note license in LICENSES.md if not MIT/Apache-2.0.
  3. Prefer workspace [workspace.dependencies] pin in root Cargo.toml.

Consequences

Positive:

  • Smaller OntoIndex surface; faster milestones.
  • Shared quality gates with OntoLogos conformance tests for reasoning.
  • Clear contributor guidance before inventing new crates.

Negative:

  • Multiple ontology representations in memory (Oxigraph triples, Horned-OWL axioms, ontologos_core::Ontology) until bridges mature.
  • LGPL-3.0 horned-owl requires license documentation (LICENSES.md).
  • MSRV rises to 1.88 when Horned-OWL / OntoLogos integrate.

Appendix A — ontoindex-owl parsing: horned-owl vs ontologos-parser

Question: Should ontoindex-owl depend on horned-owl directly or on ontologos-parser?

Criterion horned-owl direct ontologos-parser
Authoring / patch write-back Native Ontology axiom API Extra bridge from ontologos_core
Manchester / functional syntax horned-functional integration Indirect
Reasoner input alignment Requires horned-owl → ontologos bridge at classify time Same model as ontoindex-reasoner
License LGPL-3.0 MIT OR Apache-2.0 wrapper
Coupling OWL library only Couples authoring crate to OntoLogos release cycle

Decision:

  • ontoindex-owl → horned-owl + horned-functional direct for axiom model, Manchester, patch write-back, and semantic diff (ADR-0013).
  • ontoindex-reasoner → ontologos-parser for loading workspace files into OntoLogos (ADR-0015).
  • ontoindex-owl (or shared bridge module) converts Horned-OWL ontology → OntoLogos input when invoking the reasoner; avoid a second in-tree parser.

Rationale: authoring needs fine-grained Horned-OWL axiom objects; reasoning already standardizes on OntoLogos. Splitting parse paths matches the dual-stack rule (Oxigraph for triples, Horned-OWL for edit/diff, OntoLogos for classify).

Appendix B — v1.0 SQL joins: sqlparser vs DataFusion

Question: How to deliver joins and aggregations at v1.0 (v1.0_BACKLOG.md)?

Approach Pros Cons
Extend sqlparser virtual tables Consistent with v0.2; lightweight; no Arrow dep Limited optimizer; manual join implementation
Revive DataFusion (ADR-0004) Mature execution, aggregations, export Heavy dependency; virtual table integration cost

Decision:

  1. v1.0 default: extend hand-rolled virtual-table joins/aggregations in ontoindex-query using existing sqlparser AST.
  2. Trigger DataFusion revisit if any of: (a) join queries exceed maintainability, (b) Parquet/Arrow export becomes P0, (c) aggregations need optimizer behavior beyond hand-rolled plans.
  3. Document evaluation outcome in DEPENDENCY_MATRIX.md when decided.