Repository navigation
fix: deploy on Dash Core 24 as well as Core 23, with a Core version matrix in CI - #30
Conversation
CI pinned one Core image (dashpay/dashd:23), so a Core release that changed an RPC, a default or an error code was first met on a live deployment: Core 24 makes descriptor wallets the default and the spork-key step failed there. Add a Core compatibility workflow with a matrix over dashpay/dashd:23 and dashpay/dashd:24. Per image it runs: - test_core_compat.py: the real worker against a lone regtest dashd. Every RPC the workers name still exists; the error codes they branch on; the wallet and spork identity; a Core restart with the spork key; funding; an EvoNode registration and its replay; spork activation signed with the recorded key; and two resume cases (a deployment holding a key exported from a legacy wallet, and one interrupted before the key was recorded). - the existing dashmate-rendered devnet contract (core_integration.py and join_contract.py), moved here unchanged from ci.yml's core-contract job. The workflow can be dispatched by hand with any other dashpay/dashd image, so a release candidate is checked before it is deployed. On this commit the Core 24 leg fails at the spork key (dumpprivkey is refused on a descriptor wallet). The next commit fixes it. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The wallet step created an address in the dashnet wallet and exported its
key with dumpprivkey to use as the spork key. Core 24 creates descriptor
wallets by default, and dumpprivkey is refused on them ("Only legacy wallets
are supported by this command"), so a deployment on Core 24 stopped there.
The spork key never needed to be the wallet's: Core signs sporks with the
key in its configuration. The worker now makes the key itself and asks Core
for its address through the wallet-less descriptor RPCs (getdescriptorinfo,
deriveaddresses), so Core still validates the key and derives the address.
The same code runs on Core 23 and Core 24, on legacy and descriptor wallets.
Deployments made before this keep the key they exported: its address is
derived the same way and is the address they always had. A deployment that
stopped at the failed export resumes with a new key; nothing had recorded
the old address.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Activation required six sporks to be listed by Core and refused the whole profile otherwise. Core 24 removed SPORK_3_INSTANTSEND_BLOCK_FILTERING and SPORK_9_SUPERBLOCKS_ENABLED (dashpay/dash#7278): what they gated is always enabled there and they no longer appear in the spork RPC, so activation failed with unsupported-spork-profile once the spork key step passed. Those two are now activated where Core still has them (Core 23) and skipped where it does not. Any other missing spork is still refused. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 24 minutes. View limit details
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
…does The devnet contract activated the sporks before funding and registering. A deployment does the opposite (registrations, fast-forward, then activate), and on Core 24 the test's order cannot work on a lone node. With InstantSend on, Core keeps a transaction that has no InstantSend lock out of blocks for ten minutes. Core 23 skipped that rule while the node's masternode sync was incomplete (RejectConflictingBlocks, which also needed SPORK_3); Core 24 removed that exception together with SPORK_3. A lone node has no quorum to lock anything, so a registration sent after activation was never mined and the readback failed with rpc-protx:-8. Move both activation calls after the second registration. Every step still runs, on both Core versions. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Why
Dash Core 24 broke a deployment (Palmatum) at the spork-key step: the wallet step exports a key with
dumpprivkey, and Core 24 creates descriptor wallets by default, where that RPC is refused ("Only legacy wallets are supported by this command"). CI pinned a single Core image (dashpay/dashd:23), so nothing checked a new Core before it was deployed.Running the code against real Core 23.1.10 and 24.0.1 nodes turned up a second break behind the first: Core 24 removed
SPORK_3_INSTANTSEND_BLOCK_FILTERINGandSPORK_9_SUPERBLOCKS_ENABLED(dashpay/dash#7278), and the activation step refuses a Core that does not list all six sporks it wants (unsupported-spork-profile).Core 23 keeps working: every change here runs against both versions.
What changed
test(core): a Core version matrix. New workflow.github/workflows/core-compat.yml, matrix overdashpay/dashd:23anddashpay/dashd:24. Per image:internal/node/testdata/test_core_compat.py(new): the realWorkeragainst a lone regtestdashd, nothing scripted. Every RPC the workers name still exists; the error codes they branch on; wallet and spork identity; a Core restart withsporkaddr/sporkkey; funding; an EvoNode registration and its replay; spork activation signed with the recorded key; and two resume cases (a deployment holding a key exported from a legacy wallet, and one interrupted before the key was recorded, which is the state Core 24 left behind). When Core refuses a call, the failure prints Core's own message, which the worker's client drops.core_integration.py,join_contract.py), moved fromci.yml'score-contractjob so it runs once per Core.mainhas no branch protection, so no required check name changes.image=dashpay/dashd:<tag>to check a release candidate or patch release before deploying it.fix(node): the spork key no longer comes out of the wallet. The worker makes the key itself and asks Core for its address with the wallet-less descriptor RPCs (getdescriptorinfo "pkh(<key>)", thenderiveaddresses).dumpprivkeyis gone; thedashnetwallet stays whatever type Core creates.fix(node): activation on Core 24. Sporks 3 and 9 are activated where Core has them (23) and skipped where it does not (24). Any other missing spork is still refused.test(core): the devnet contract activates sporks after the registrations, aslifecycle/runner.godoes (registrations, fast-forward, activate). The contract did it the other way round, which Core 24 does not allow on a lone node: with InstantSend on, Core keeps a transaction without an InstantSend lock out of blocks for ten minutes. Core 23 skipped that rule while masternode sync was incomplete (RejectConflictingBlocks,src/node/miner.cpp); Core 24 removed that exception together with SPORK_3. No step was dropped.A search for every key export/import or legacy-wallet dependence (
dumpprivkey,importprivkey,dumpwallet,createwallet,getnewaddressfollowed by an export) found only the onedumpprivkeyininternal/node/worker.py.createwalletis called with the name only, which both versions accept. The managed, join, upgrade and seed-TLS workers use no wallet RPC. The infraclaw clone and its worktrees were checked read-only for unpushed work on this path: there is none.Which approach, and why
Both suggested approaches were tried against real nodes first (probe run, jobs
wallet-keys (dashpay/dashd:23)and(dashpay/dashd:24)):createwalletdumpprivkeyon the default walletdescriptors=false)load_on_startup=true;dumpprivkeyworks on itlistdescriptors true+ BIP32 derivation of the address's keygetdescriptorinfo "pkh(<own key>)"+deriveaddressessporkaddr/sporkkeyacceptssporkupdateprotx register_fund_evofrom the default walletmigratewalletand makes legacy creation opt-in). It also does nothing for adashnetwallet that Core 24 already created as a descriptor wallet, which is exactly the interrupted deployment.dumpprivkey: two code paths, each tested on one version. It also needs BIP32 and secp256k1 arithmetic in the worker and parsing of descriptor strings, whose format already differs between the two versions (44'/1'/0'vs44h/1h/0h). Andlistdescriptors truereturns the wallet's mnemonic, far more than the one key needed.secrets, wallet import format) and Core validates it and derives the address. One code path, identical on both versions and both wallet types, no dependence on legacy wallets, no elliptic-curve code here. Core refuses a damaged key or one for another chain (code -5), and did not start when given asporkkeythat does not matchsporkaddr.Effect on existing deployments:
secrets.jsonalready holds the exported key. Its address is now derived from it rather than read from the wallet label, and is the same address (test_7proves this on a real legacy wallet on both versions). No migration, and upgrading such a network's Core to 24 needs no key export.test_8). Nothing had recorded the olddashnet:sporkaddress: the controller records it only after this step returns. The unused labelled address stays in the wallet.What was run and seen
mainagainst Core 24 and passes on Core 23. Run against the worker exactly as onmain: run 43497807750.dashpay/dashd:23: 8 of 8 ok.dashpay/dashd:24:test_3andtest_8fail withrpc-dumpprivkey:-4 on dashpay/dashd:24: Core answered {"code": -4, "message": "Only legacy wallets are supported by this command"}. (That run used the test a few lines before its final form: premine 1000 instead of 600, the spork list before Core 24's was known, and a legacy-wallet setup call that Core 24 rejected, sotest_7was skipped there.)main's existing devnet contract fails the same way on Core 24: jobmain-core-contractin the probe run,worker.RPCFailure: rpc-dumpprivkey:-4right aftercore-start passed.4332ac5.dashpay/dashd:23(Core 23.1.10, legacy default wallet) anddashpay/dashd:24(Core 24.0.1, descriptor default wallet) each: 8 of 8 regtest tests ok, then the devnet contract throughwallet,core-finalize,fund,register(twice, with replay), miner start/pause, stop/resume andactivate, then the join contract.334d7ce) was green on Core 23 and on Core 24's regtest step, and red on Core 24's devnet step:registerafteractivateended inrpc-protx:-8. That is item 4 above; the cause was read from Core's source, not observed directly.checks/checkis red, and not because of this change.internal/seedtls/testdata/test_tls.py::test_real_served_tls_validates_both_identitiesfails withgateway-certificate-not-served. It fails 3 times out of 3 onmain's own code on today's runner image (Ubuntu 26.04, Python 3.14.4, OpenSSL 3.5.5): probe run 43501625847.checklast passed onmainon 2026-10-30. Not fixed here. In that same job the Go tests, the DynamoDB proof and the node and managed worker tests (including the new fixture tests) passed before it;test_console_operation.pyand the "Examples and command boundary" step come after the failing test and did not run. Every otherchecksjob passed at4332ac5(run 43501673707):platform-contract,managed-contract,bootstrap-recipeand the fourbinaries.Not covered
getbestchainlockandmasternode statusare checked only as "exist and fail with a code the worker expects". No multi-node network was deployed.servicefield andgetblockchaininfo.softforksbehind-deprecatedrpc=service/-deprecatedrpc=softforks; Platform below 5.0.0 needs both, and dashmate 5.0.0+ adds the first itself. This tool renders Core's configuration with the release's own dashmate, so a release pairing Core 24 with Platform or dashmate below 5 would start Core without them. Nothing here adds or checks those flags.platform-contractstill runs Core 23 with Platform 4.2.0-beta.3; it passes unmodified withdashd:24too (probe jobmain-platform-contract), but only reaches "Drive waits for Core readiness", so it says nothing either way.dashmate-helper:4.2.0-beta.3for both Core versions, the only helpermainpins. Core 24 boots from that render, but it is not the dashmate that ships with Core 24.activatewaits ten minutes for a block, and the worker reports that asrpc-protx:-8. A deployment registers before it activates, so this is not on its path; the worker was not changed for it.imageinput can first be tried after merge. A dispatched tag that needs another dashmate may fail the devnet step for dashmate reasons; the regtest step does not use dashmate and is reported separately.fcntl); the fixture tests were run locally only through a shim, so the Linux CI run is the reference.🤖 Generated with Claude Code