Skip to main content

Roadmap

The v1 ship landed bundles + prefixes + cas-cache + cli + tests + CI

  • fuzz/mutation infra. Several smaller items surfaced during implementation got tagged "v1.1" or "deferred." This page summarises those; the full priority + sequencing + effort breakdown lives in PLAN-followups.

P0 — blocking​

  • CI green-up. Several recent pushes went red on real GitHub Actions despite local act-based verification (cargo fmt drift, tokio-on-wasm, sibling-repo path-deps for compression-multiplexer
    • tvm-wasm). All sequentially fixed; tracking the latest run.

P1 — high value, parallel-able​

  • Prefix v1.1 migration sweep. All 217 in-tree extensions currently run on the synthetic-expansion fallback. Before v1.1 makes the manifest fields a hard error, every extension needs real (preferred-prefix, prefix-expansion) assigned. ~1.5 days mechanical sweep.
  • Live .prefix prefer. Pin row writes today but bare-name re-registration only fires next session. Live application would walk the existing register_X cache + re-register. ~0.5 day.
  • Cosmetic warnings. Two pre-existing 1-line fixes: non_snake_case in sqlink-host bin + unused import at cli/src/lib.rs:771. ~15 minutes.

P2 — medium​

  • Single-slot hook audit recording. Authorizer / update / commit / wal hooks aren't recorded in __sqlink_prefix_function (intentional v1 call). v1.1 can add diagnostic-only recording.
  • install_loaded_extension refactor. The recovery fork's Host::ensure_prefix_for_extension + record_function_for_extension helpers landed but install_loaded_extension still has its own inline logic. DRY-up.
  • Bundles browser dispatch driver. Browser composed-cli has no dispatch_dot_command driver in v1; composed-bundle.spec.js uses test.skip(). ~1-2 days.
  • Bundle multi-name aliasing. v1's __cas_bundle.name UNIQUE constraint blocks true aliasing. v1.1 adds __cas_bundle_alias.
  • .load auto-cache by content-hash. So --bundle-load after a .load doesn't hit the cache-miss path.
  • HTTPS resolver mocking. cargo-mutants flagged 3 surviving mutants in resolver.rs::HttpsResolver::resolve; tests need mockito/wiremock to close.
  • Hostile-fs + migration-arm fixtures. Two more architectural- mutant gaps the v1 mutation round flagged.
  • cargo-mutants re-run. Confirm the estimated 92-94% kill rate after the new test fixtures land.

P3 — nice-to-have​

  • More fuzz targets (spawn-build path validation, bundle-cli argv, sqlink-cli-argv).
  • Longer fuzz runs (hours per target) — sharply diminishing returns past the 20-min mark observed in the extended round.

Genuinely deferred to v2​

These are explicitly out of any near-term roadmap and listed only for visibility:

  • Cross-target builds (wasi-sdk / zig-cc / cross).
  • Bundle registry / publishing.
  • Per-query prefix overrides.
  • Prefix-scoped permissions (gate system__* to operator-only).
  • Prefix registry hosting (org-wide expansion uniqueness).
  • Cross-database prefix sync.
  • sqlink-host as rlib for production install.
  • OSS-Fuzz integration.

See the full plan for effort + dependencies

  • fan-out strategy.