Drop the reasonable rdf12-passthrough fork patch — upstream #50 merged, 0.4.3/0.4.4 released (E-011 cleanup)

[styk-tv]

Summary

pgRDF still consumes reasonable via a [patch.crates-io] fork pin
(styk-tv/reasonable@rdf12-passthrough, commit 8974c110). The reason that fork
exists has been resolved upstream for ~5 weeks, so the patch can now be dropped
in favour of the released crate. Tracked internally as ERRATA.v0.4 E-011
(see Cargo.toml).

Current state (v0.6.17, latest)

Cargo.toml:

• reasonable = { version = "0.4", features = ["rdf-12"] }
• [patch.crates-io] reasonable = { git = "https://github.com/styk-tv/reasonable", branch = "rdf12-passthrough" }

Cargo.lock:

• reasonable 0.4.1 from git+…rdf12-passthrough#8974c110…

The fork commit (8974c110, 2026-05-15) adds a feature-gated TermRef::Triple(_)
arm so reasonable compiles when shacl 0.3.x (via rudof_rdf) hard-enables
oxrdf/rdf-12 workspace-wide.

Why it can be dropped now

• Upstream feat(rdf-12): add TermRef::Triple(_) arm for downstream coexistence gtfierro/reasonable#50 — "feat(rdf-12): add TermRef::Triple(_) arm
  for downstream coexistence" — merged 2026-05-22, touching lib/Cargo.toml +
  lib/src/common.rs (the same change the fork carries).
• crates.io has shipped reasonable 0.4.3 (2026-05-27) and 0.4.4 (2026-05-28) —
  both post-merge.

So the fork is now functionally identical to a released crate, and the git
dependency is pure supply-chain debt (hurts reproducibility / SLSA cleanliness for
no functional gain).

Proposed change

1. Confirm the released crate exposes the feature pgRDF opts into (PR Aggregates over expressions + BIND-produced variables (SUM(?a*?b), SUM over Extend output) #50 added it
   to lib/Cargo.toml; verify the published feature name matches features = ["rdf-12"]).
2. Remove the [patch.crates-io] reasonable = … block.
3. Pin reasonable = { version = "0.4.4", features = ["rdf-12"] } (or current).
4. Drop the E-011 comment block.

Verification

• cargo build / cargo check clean with no [patch.crates-io] for reasonable.
• cargo tree -i reasonable shows the crates.io source (not git+…).
• Cargo.lock reasonable source is registry+…, not git+….

Notes

• Functionally a no-op (the fork is the merged PR), so no behaviour change is
  expected — this is build-hygiene / supply-chain.
• Filed as part of moving cross-repo NOTIFIES → GitHub issues.

[styk-tv]

Reopened 2026-07-03 — deferred here is now wanted (target v0.6.20), and a second, related symptom surfaced: CI recompile cost.

1. The fork drop (unchanged, re-verified)

crates.io reasonable 0.4.4 (2026-05-28, not yanked) exposes "rdf-12": ["oxrdf/rdf-12"] — the exact feature pgRDF opts into. So the drop is clean:

• remove the [patch.crates-io] reasonable = { git = "…rdf12-passthrough" } block in Cargo.toml
• pin reasonable = { version = "0.4.4", features = ["rdf-12"] }
• cargo update -p reasonable → Cargo.lock resolves to the registry crate
• full reasoning-suite re-verification is the gate (0.4.1-fork → 0.4.4 is a reasoner version bump, not just a source swap): LUBM, materialize, RDFS/OWL-RL profile fixtures (48/49), type-closure (134/135) must stay green.

2. The CI recompile symptom (related)

The regression CI job's .so build (compose/builder.Containerfile → cargo pgrx package) uses a BuildKit --mount=type=cache that is build-scoped and not exported to the type=gha cache — so every run re-downloads + recompiles all deps, including reasonable (~140s). Two independent improvements:

• The git dependency is the most cache-hostile kind; dropping the fork (§1) alone removes the git-fetch and lets the registry cache it cleanly.
• The docker build could export its cargo cache to gha (buildkit-cache-dance or an actions/cache on the BuildKit state). The "clean-room" comment justifying the current no-cache applies to release.yml (the attested build, correctly separate + uncached), not to the CI test .so — so caching it does not touch attestation integrity.

Scope for v0.6.20: §1 is the headline (supply-chain hygiene); §2 is the build-infra follow-on it enables. Keep them in one increment since both center on the reasonable dependency. Does NOT block v0.6.19 (that release ships with the fork; it is correctable debt, not a defect — the fork commit is pinned so today's build is reproducible).