core, eth, tests: v1.17.3 upstream merge, develop gap & remaining CI/UT fixes#4
Draft
zlacfzy wants to merge 76 commits into
Draft
core, eth, tests: v1.17.3 upstream merge, develop gap & remaining CI/UT fixes#4zlacfzy wants to merge 76 commits into
zlacfzy wants to merge 76 commits into
Conversation
…test after v1.17.3 merge core/state/database_ubt.go: NoTries() now returns false instead of panicking. It is called on the StateDB commit hot path (updateStateObject/IntermediateRoot/commit); a unified binary trie always maintains tries, so panicking there broke TestBinaryGenesisCommit and any UBT genesis commit. core/genesis_test.go: - resolve leftover v1.17.3 merge conflict markers (keep BSC Chapel case, drop upstream Sepolia/Hoodi); - rename TestGenesis_Commit -> TestGenesisCommit to match upstream; - keep the total-difficulty assertion: unlike upstream (which removed TD entirely), BSC still writes TD at genesis (genesis.go) and reads it in production (headerchain.go), so the assertion still guards live behavior; - remove TestConfigOrDefault: the inline config-defaulting it covered was reworked into params.GetBuiltInChainConfig (upstream #30907), leaving the test a tautological switch check. core/state_prefetcher_test.go: remove TestPrefetchLeaking. The channel-send goroutine leak it guarded (added in bnb-chain#1079) is structurally impossible in the current Prefetch, which was rewritten to errgroup.Group + Wait(); the test does not cover PrefetchMining (the method still using the manual goroutine/channel pattern). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ructions tests Both BenchmarkInterpreter (interpreter_test.go) and TestOpTstore (instructions_test.go) still had `<<<<<<<`/`>>>>>>>` conflict markers, so the entire core/vm test package failed to compile (`[setup failed]`). Resolve contract construction to the current API: keep BSC's pooled `GetContract` (used throughout evm.go) and adopt upstream's gas type change (`GasBudget` via `NewGasBudget(...)`). The HEAD side still passed a bare `uint64`, which no longer matches the `gas GasBudget` parameter. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…merge conflicts Resolve the remaining `<<<<<<<`/`>>>>>>>` conflict markers in three test packages left by the v1.17.3 merge: * core/rawdb/accessors_chain_test.go: drop TestCanonicalHashIteration — ReadAllCanonicalHashes was removed upstream and no longer exists. Also restore the two ReadAllHashes edge-case assertions in TestHashesInRange that the resolution had dropped; upstream keeps them and ReadAllHashes still exists, so they still guard live behavior. * core/state/statedb_test.go: TestDeleteStorage — combine both sides: keep BSC's 6-arg snapshot.New(..., cap, withoutTrie) and adopt upstream's NewMPTDatabase(tdb, nil).WithSnapshot(snaps) construction. * triedb/pathdb/layertree_test.go: TestLookupZeroBaseRootFallback — both sides added this regression test for the zero-base-root sentinel fix; keep the more thorough version (fall-through account/storage, happy path, and stale-state cases) and drop the duplicate. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
core/state, core: fix UBTDatabase.NoTries panic, clean up genesis_test
zlacfzy
changed the title
core, triedb: resolve leftover v1.17.3 unit-test merge conflicts
Process() called postExecution() — which parses EIP-6110 deposit requests from the block logs — while allLogs was still empty; allLogs was only populated from the receipts *after* postExecution and Finalize. So during block verification the deposit request was always dropped, yielding a wrong requests hash (TestPragueRequests failed with local != remote, remote carrying deposit+withdrawal+consolidation, local only withdrawal+consolidation). Populate allLogs from the user-transaction receipts before postExecution (matching upstream ordering), then append the system-transaction logs added by Finalize so ProcessResult.Logs still carries the full set. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* ethclient: resolve some ut conflicts * worker: add is-not-in-bsc check to make ut succeed
… fork context BSC's ValidateBlobTx is invoked with nil head/opts from the concurrent MEV blob-validation paths (miner_mev, bid_simulator, bid_block) and from TestValidateBlobTx_InvalidProof. The v1.17.3 merge grafted an `opts.Config.IsOsaka(head.Number, head.Time)` sidecar-version check onto the top of the function, which nil-dereferences for those callers and panics. Gate that fork-version enforcement behind an `opts != nil && head != nil` check. Without fork context the sidecar is still validated against its own declared version by the fork-specific checks further down (restoring pre-merge behavior); the real txpool path (validation.go:151, non-nil head/opts) keeps the check. Additionally pin the expected sidecar version to V0 on BSC chains via IsNotInBSC(): Osaka is already live on BSC mainnet/Chapel, and without the pin the pool would demand version-1 (cell proof) sidecars — which BSC does not support — rejecting every valid V0 blob transaction at admission. TestValidateBlobTx_BSCOnlyV0PostOsaka pins this policy for both networks. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…r v1.17.3 merge
The v1.17.3 merge replaced the TypeMux downloader events with the downloader's
SyncEvent feed and rewired miner.update/VoteManager.loop through an inline
`eth.(interface{ Downloader() *downloader.Downloader })` assertion. That broke
the miner and vote test suites in three ways:
1. The test backends never implemented the asserted method, so every test
panicked with an interface-conversion error. Narrow the dependency instead:
miner.Backend and vote.Backend now declare
`SubscribeSyncEvents(chan<- downloader.SyncEvent) event.Subscription`,
implemented by *Ethereum (forwarding to the downloader) and by the test
backends (event.Feed), which also makes sync events injectable from tests
again (they previously posted to a TypeMux nothing listens to anymore).
2. worker.commit hard-cast `w.engine.(*parlia.Parlia).FinalizeAndAssemble`,
panicking for the clique/ethash-based worker tests. Use core.AssembleBlock
like generateWork already does: engines implementing FinalizeAndAssemble
(parlia) keep the BSC path, others fall back to Finalize + NewBlock.
3. The merge dropped the one-shot semantics of miner.update: after the first
successful sync the miner must stop reacting to downloader events (the
documented DOS protection). Unsubscribe from the sync feed on
SyncCompleted, restoring the pre-merge `events.Unsubscribe()` behavior.
With this, the miner package (excluding the two blob-validation cases fixed by
the ValidateBlobTx PR) and the core/vote package pass again.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
rlp/ethapi: fix failed/conflict rlp/ethapi UTs
core: parse EIP-6110 deposit requests before Finalize in state processor
… context Instead of skipping the sidecar-version check when head/opts are nil, default the expectation to version 0 and only raise it to version 1 on non-BSC chains past Osaka. The MEV blob-validation paths (miner_mev, bid_simulator, bid_block) therefore now reject cell-proof (v1) sidecars outright, matching the simulator's explicit "cell proof is not supported yet" stance and closing the blind-signing gap where validateBidBlockBlobTxs would otherwise accept a builder-supplied v1 sidecar into a block that the network's V0-only data availability check then rejects (costing the proposer its slot). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
core/txpool: fix ValidateBlobTx nil-deref when called without fork context
testVotePool still posted downloader.SyncCompleted to a local TypeMux that nothing subscribes to anymore, silently dropping coverage of the sync-event path through VoteManager.loop (the test only kept passing because startVote defaults to true). Give testBackend a retained sync feed and fire the event through it, and drop the now-dead local mux and testBackend.EventMux. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
They restate the switch cases verbatim and still use the retired TypeMux event terminology. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…PoSA engines (#10) BSC overrides CurrentFinalBlock() to derive finality from the Parlia fast-finality attestations, returning nil for any other engine. That made the SetFinalized marker write-only outside PoSA — the engine API (catalyst forkchoiceUpdated) and the syncer do call SetFinalized, but nothing could read the result — and it broke the upstream regression test TestSetHeadBeyondRootFinalizedBug (ethereum/go-ethereum#33486), whose setup nil-panicked under its ethash engine. Fall back to the marker (upstream semantics) when the engine is not PoSA. Parlia chains are unaffected: the PoSA branch is taken unchanged. Non-PoSA configurations regain upstream behavior, and the upstream test now passes verbatim, keeping coverage for the setHeadBeyondRoot finalized-cleanup fix. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
will-2012
marked this pull request as draft
July 6, 2026 02:13
* tests: fix txfetcher_fuzzer conflict * tests: fix state_test conflicts
eth/fetcher: adapt tx fetcher tests to upstream apis
fix: fix p2p api in test case
miner, core/vote, eth: fix sync-event wiring and worker assembly after v1.17.3 merge
opDifficulty now returns PREVRANDAO when Context.Random is set (post-merge) and DIFFICULTY otherwise, mirroring bsc-erigon's unified opDifficulty. Parlia chains leave Context.Random nil (their block difficulty is non-zero, see NewEVMBlockContext), so BSC keeps DIFFICULTY semantics byte-for-byte, while the standard state tests set Random and observe PREVRANDAO. This lets the Shanghai instruction set keep its London base on BSC without a test patch swapping it to the Merge base. Combined with the earlier in-tree precompile split (rules.IsInBSC), run-evm-tests.sh no longer applies ANY patch: 0001 is deleted (0002 was removed earlier). Verified with zero patches applied: bcRandomBlockhashTest, stSStoreTest, stRandom and the core/vm suite all pass. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Trim the opDifficulty comment and drop the now-stale "why no patch" notes in run-evm-tests.sh (the test patches are gone). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Re-enable ci.go's downloadSpecTestFixtures (fixtures pinned to v5.1.0 in build/checksums.txt, sha256-verified) and run the spec tests through `ci.go test ./tests/` directly, matching upstream go-ethereum's CI. The manual-wget-plus-source-patch wrapper is no longer needed now that BSC precompiles are split in-tree (IsInBSC) and fixtures are back on v5.1.0. - build/ci.go: uncomment the cachedir flag and the downloadSpecTestFixtures call that BSC had disabled while it relied on the wrapper. - .github/workflows/evm-tests.yml: non-recursive submodule checkout so TestLegacyState self-skips (its Constantinople collision fixtures don't apply to BSC); run `ci.go test -timeout 1h ./tests/` (non-short, needed so fixtures download; 10m default timeout is too short for statetests). - Remove tests/run-evm-tests.sh. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The four …ForBSC precompile maps kept upstream's pre-rename wording ("the
default set … used in the X release"), which no longer explained the ForBSC
suffix. State the convention once on the Istanbul map and reference it from
the others. Also tighten the two evm-tests.yml comments.
No behavior change (comments only).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Three independent unit-test failures surfaced by the go-ethereum v1.17.3 merge, none of them production bugs: - cmd/devp2p ethtest: statusExchange replied only after breaking out of the read loop, so BSC's two-phase handshake68 (Status then UpgradeStatus) stalled on the node side and every TestEthSuite subtest hit EOF. Reply with our Status and only break after answering UpgradeStatusMsg. - rlp/rlpgen pkgclash testdata referenced GetReceiptsPacket69, which does not exist (BSC ships only eth/68 and eth/70); point both the input and the expected output at GetReceiptsPacket68. - cmd/geth db command registered dbInspectTrieCmd twice, so urfave/cli rejected the duplicate subcommand at startup and geth failed to launch (TestAttach). Drop the stray upstream-block entry, keep the BSC one. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This reverts commit 7dcba88.
tests, core/vm: stabilize spec-tests after the v1.17.3 merge
This reverts commit 5cc0049.
This reverts commit cbf7151.
…SC split The precompile split repurposed PrecompiledContractsOsaka (and siblings) to mean the standard, non-BSC set, moving the BSC set to PrecompiledContractsOsakaForBSC. bnb-chain#3726's tests still assumed the old meaning where those names were the BSC set: - precompile_pasteur_test.go asserted 0x64/0x65/0x67 live in PrecompiledContractsOsaka; point it at PrecompiledContractsOsakaForBSC. - contracts_lightclient_test.go built params.Rules{IsOsaka:true} and expected the BSC 0x67; add IsInBSC:true so activePrecompiledContracts selects the BSC set (otherwise 0x67 is absent -> nil-deref panic). Production is unaffected: chainConfig.Rules() sets IsInBSC, so live precompile selection was always correct; only these hand-built test rules needed the flag. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…plit The precompile split makes non-BSC chains (Parlia == nil — e.g. this test's Ethash Hoodi config) expose the standard precompile set rather than the BSC set. eth_config (EIP-7910) reports the active precompiles, so the golden files had been listing the BSC-specific 0x64-0x69 and the early 0x100 p256 on a non-BSC chain — which was exactly the pre-split bug. Regenerate the golden to the correct standard set (matches upstream Ethereum: no tendermint precompiles, p256 only from Osaka). WRITE_TEST_FILES=1 go test ./internal/ethapi/ -run TestEIP7910Config Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
tests, miner, cmd: fix failed ci actions
…mpile_tests core/vm, internal/ethapi: fix tests broken by the ForBSC precompile split
The v1.17.3 merge branches no longer auto-run CI; workflows trigger only on master/develop again. Removes the temporary '- merge-v1.17.3*' entry from commit-lint, lint, integration-test, evm-tests, unit-test and build-test workflows. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
zlacfzy
changed the title
allformless
reviewed
Jul 13, 2026
| defer tester.terminate() | ||
|
|
||
| chain := testChainBase.shorten(blockCacheMaxItems - 15) | ||
| peer := tester.newPeer("corrupt", eth.ETH69, chain.blocks[1:]) |
|
enable ETH70 |
The field lacked a toml omitempty tag, so dumpconfig/init-network always wrote "MaxBlobsPerBlock = 0" even when unset (0 = protocol default), and geth's strict TOML decoding makes binaries predating the field fatal on such configs. Tag it omitempty like the other optional fields. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
zlacfzy
force-pushed
the
merge-v1.17.3_ci_fix
branch
from
July 13, 2026 09:09
590f9ef to
b48d947
Compare
eth/tracers, core: derive system tx traced gas
eth, cmd/devp2p: enable eth/70 with TD status and single-phase handshake
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Tracking PR for the go-ethereum v1.17.3 upstream merge.
422b270876unblocked the binary build (make all); this PR tracks the remaining test/CI breakage and the sub-PRs resolving it.Status
On
merge-v1.17.3_ci_fixthe code is healthy: the unit-test, evm-test, golang-lint and truffle-test jobs are green (verified locally + in sub-PR CI). The two remaining red checks on this tracking PR are non-code:build— the fork has noOSSINDEX_USERNAME/OSSINDEX_TOKENsecrets, so the Nancy Security Scan step fails (affects every PR on this fork, not the merge).commitlint— a few historical commit headers exceed the 80-char limit.Fixed (sub-PRs)
UBTDatabase.NoTriespanic +genesis_testcleanupcore/vm,core/rawdb,core/state,triedb/pathdbtest conflicts…ForBSC/IsInBSC), run viaci.goeth_configtests to the ForBSC splitdb inspect-trieduplicate, miner gas reservation)Note
The temporary CI
merge-v1.17.3*branch triggers have been removed from the workflows (766e2e4bd9); these branches no longer auto-run CI.