Backport Receiving ledger compatibility to 5.x

[colll78]

Summary

Support protected addresses and per-output Receiving through the existing consensus adapters. Purposes use each output’s original body-local index; identical outputs retain separate redeemers and budgets, including parent/child capacity accounting.

wire transaction → applyTx → recipient authorization + ledger-table state
parent + child indexed redeemers → combined execution-unit capacity

Receiving tracing uses raw indexes. Missing-redeemer diagnostics preserve every purpose sharing a script hash: singleton JSON keeps its shape, while multiple purposes form a source-order array. Tracing also covers UnsupportedOutputAddresses; regressions include singleton, forward/reversed ordering and updated goldens. Tools fixtures use the required 369-entry V4 model and correct genesis hash, with independent temporary databases for synthesis, analysis and truncation.

Follow the upstream six-field NewEpochState: Praos and both ticked/unticked Peras views read the current set-snapshot pool distribution through nesStakePoolDistrG. Initial test-network snapshot seeding delegates to the ledger’s injectIntoTestState, replacing consensus duplication. The normal project pins ledger 4b06e253 and includes eras/shelley/test-suite, selecting the migrated 1.10.0.1 rather than the stale published test library.

Genuine owning encoders refresh the 16 disk goldens for NES 7→6 and Mark 3→4. Published release successor bb55e647 changes only those fixtures on parent 8268ddc7; the same refreshed bytes are validated on main.

This backport targets ouroboros-consensus-5.x-backports, retaining version 5.1.0.0 under the dependency-bound policy. API/node consumers retain their 5.1 bounds; main proposal #2 carries the corresponding changes. The compatibility source and its fixture-only successor bb55e647 are published.

Evidence

GitHub Actions runs in this fork. Earlier upstream CI links are historical evidence.

Local compilation/runtime uses GHC 9.6.7 and the coordinated native graph (crypto -external-libsodium-vrf, blockio +serialblockio). Default-flag public-source solver/Nix evidence is separate.

The preceding release checkpoint passed the complete 240-test tracing suite and two Dijkstra encoding goldens, plus the original admission/capacity and parallel-tools regressions. Its ordinary compiler/native checks passed, while the exact-seed CSJ issue below remained unresolved. These full-suite results retain their pre-snapshot-refactor scope.

• The current release cardano library and complete owning Cardano/tools test targets compile strictly. All 11 admission/capacity tests pass: actual wire decoding, missing-recipient rejection, successful applyTx, ledger-table equality, and distinct output budgets through parent/child capacity accounting.
• The genuine fixture-only migration passes all 18 ordinary disk comparisons (two Shelley and 16 Cardano), 16 independently checked codec changes and four unchanged Byron byte controls. Production/test assertions and all other files are unchanged.
• All 20 tools tests pass with two workers. All three Dijkstra Header/GenTx/GenTxId CDDL checks pass using genuine Cuddle 1.8.1.0 and cddlc 0.4.2 with current data directories. Actual binary/source/plan/tool hashes and logs are retained.
• Changed Haskell files pass Fourmolu 0.20 and git diff --check. Distinct main 5.0 compilation and selected 18/11/20/3 runtimes also pass, using its default flags with scoped private native VRF directories and recorded runtime library paths.
• Both normal native projects pass default/all and tests-enabled resolution: 622 components, including all 46 owning components. Both isolated default Nix devShell derivation evaluations pass with 621/46 components. Public ledger source/archive identity and selection of the migrated Shelley test suite were verified. This establishes resolution/evaluation, without a completed default Nix build or runtime claim.
• The ledger Foreign-interface group passes 19 tests, including actual genesis set/go seeding with the live epoch-zero committee and pool cache. This supports the delegated seeding change; it does not establish broad forecasting or Leios protocol behavior.
• Fresh release-backport consumers pass the retries-disabled real Dijkstra Receiving/same-database relay lifecycle: missing-recipient rejection, inclusion of four protected outputs with two distinct V4 redeemers, complete LSQ/fresh API replay identity through restart, and spending through the restored relay. This is current-format restoration, without snapshot-specific or old-format restoration claims.

Depends on ledger #1, Plutus #1, keys #2 and config #2. The recorded proposal SRPs keep this fork reproducible. Upstream integration would separately require accepted commits or released packages under upstream SRP policy.

Merge Danger

Door: two-way

Blast Radius: Consensus

Old persisted NES/Mark states require replay after the upstream state-format transition. Current-format same-database restart does not establish old-format or snapshot restoration. The fixture correction is public at bb55e647. At the historical upstream 15:52 UTC checkpoint, three checks pass, four GHC jobs are running and required Linux Hydra is queued, with no failed check at that head. Upstream dependency acceptance, scheduled activation/rollback and Leios endorser-block lifecycle coverage remain external gates.

Parent release 8268ddc7 has a separate GHC 9.10 storage failure without an emitted assertion/counterexample/signal or available test-log artifact; its cause is unclassified. The same exact release storage suites pass 112 tests on GHC 9.6 and 9.14. Those peer passes do not classify the failure as a flake. It is documented here, with no storage repair included.

The release backport retains a deterministically reproduced CSJ counterexample: the normalized exact-seed trace matches CI and identifies accepted-prefix loss after the documented new-dynamo reset. Whether the cumulative test bound exceeds the protocol guarantee remains unresolved; a later random pass does not discharge it. No peer-selection policy or assertion is changed here.