[carbolymer]

This PR is stacked on the GHC 9.12 PR and needs it to merge first.

What changes for users

• Kupo now needs cardano-node 11.1 or newer. ouroboros-network 2.0 only speaks node-to-client V23, so nodeToClientVMin is raised to V23 to match.
• Kupo now indexes Dijkstra blocks. Before this change, it stopped with an error on the first one.
• Kupo does not index a Dijkstra transaction that carries sub-transactions yet. It stops with a typed UnsupportedSubTransactions error that names the transaction, rather than silently dropping the inputs and outputs of its sub-transactions. The metadata endpoint fails that request with the same error. How to number sub-transactions in the index is still undecided.

Dependencies

• The Hackage index-state moves to 2026-10-23, and the CHaP index-state to 2026-10-30. The CHaP input in flake.lock is bumped to match. The solver picks cardano-ledger-conway 1.24, cardano-ledger-dijkstra 0.4, ouroboros-consensus 5.3.0.0, ouroboros-network 2.0.0.0 and cardano-diffusion 1.3.0.1.
• The Hackage bump also forces the move off the old cardano stack. On GHC 9.12, vector-map 1.2.0.0 and the newer nothunks both define NoThunks (VP.Vector e), so that stack no longer builds.
• contra-tracers now comes from IntersectMBO/ogmios master at 2f9b5d14d, which includes the PR below. Only its unpublished contra-tracers 1.0.1 builds against contra-tracer 0.2, which ouroboros-consensus 4.0 and later requires.
• fast-bech32 now comes from Hackage, pinned at 1.0.1.
• The unit tests use QuickCheck's withNumTests instead of the deprecated withMaxSuccess, which the newer QuickCheck requires.

CI

• The integration test now runs against cardano-node 11.1.3 instead of 11.0.1.

Related

• Replace the Alonzo-based eons with ShelleyBasedEra ogmios#19
• Bump GHC to 9.12.4 #10

[Jimbo4350]

LGTM a couple of comments

[Jimbo4350]

In a follow up PR can you propose how to store and access sub transactions in the database? Have a look at other indexers to get an idea of how the community may want to retrieve these from said indexers.

[carbolymer]

I've opened a spike for this. It collects the requirements for storing and querying sub-transactions, and includes a look at how other indexers expose them.

• Spike: index Dijkstra sub-transactions #12

[Jimbo4350]

Could we remove the default implementation and require each IsBlock instance to implement this explicitly? Right () reports success without inspecting the block, so an instance that omits the method silently bypasses the check. PartialBlock currently inherits this default, and the Ogmios decoder neither rejects nor preserves sub-transactions. We should also validate those during decoding, before that information is discarded; removing the default alone would not close that gap.

[carbolymer]

The default is gone, and every IsBlock instance now defines checkSubTransactions. There is nothing to validate in the Ogmios JSON yet, because Ogmios does not serialise Dijkstra transactions. It sends only their id and spends, so sub-transactions never reach Kupo.

decodeBlock now reads the block's era. For a Dijkstra block, the PartialBlock check fails with the new UnverifiableSubTransactions error, so Kupo stops with a message pointing to a direct cardano-node connection instead of indexing empty transactions. Hydra decodes Conway transactions only, so it cannot carry sub-transactions.

Once Ogmios encodes Dijkstra transactions in full, the decoder can check the sub-transactions themselves.

• https://github.com/IntersectMBO/ogmios/blob/2f9b5d14daeccfece3a7d33f8a10e2136a91fbef/server/src/Ogmios/Data/Json/Dijkstra.hs#L107-L122

[Jimbo4350]

Can we not have a default definition and explicitly define instances where appropriate?

[carbolymer]

Each instance now defines it explicitly. Details are in the other thread on this line.