fix-nethtest

Fix failing Ethereum Foundation tests

Diagnose and fix tests run with Nethermind.Test.Runner (nethtest). The test runner executes Ethereum Foundation state tests and blockchain tests against the Nethermind EVM.

Arguments: $ARGUMENTS — one or more test file paths.

---

Phase 1 — Run the test

Detect test type

Infer the test type from the JSON structure — do not require the user to specify it:

• Blockchain test — the test object contains "blocks", "genesisBlockHeader", or "network" keys
• State test — the test object contains "transaction" and "post" keys

A quick check: grep -l '"blocks"' <file> or grep -l '"transaction"' <file>.

Build and run

# Build (only if needed)
dotnet build src/Nethermind/Nethermind.Test.Runner/Nethermind.Test.Runner.csproj -c release --verbosity quiet

# Run state test
dotnet run --project src/Nethermind/Nethermind.Test.Runner/Nethermind.Test.Runner.csproj -c release -- -t -i "<test-file>"

# Run blockchain test (add -b flag)
dotnet run --project src/Nethermind/Nethermind.Test.Runner/Nethermind.Test.Runner.csproj -c release -- -b -t -i "<test-file>"

Output structure:

• Trace lines (one per EVM opcode): {"pc":N,"op":N,"gas":"0x...","gasCost":"0x...","stack":[...],"depth":N,"opName":"...","error":"..."}
• Execution result: {"output":"0x...","gasUsed":"0x...","time":N,"error":"..."}
• State root: {"stateRoot":"0x..."}
• Test result JSON: [{"name":"...","pass":bool,"fork":"...","stateRoot":"0x..."}]

If pass is true, report success and stop. Otherwise proceed to Phase 2.

---

Phase 2 — Classify the failure

Read the test JSON file to understand:

1. Target fork — from the post section key (e.g., "Osaka")
2. Transaction type — legacy, EIP-1559, blob, SetCode (EIP-7702 if authorizationList present)
3. Expected state root — from post.<fork>[0].hash (all-zeros = placeholder, not a real expected value)
4. Environment — which optional fields are present/absent in env (e.g., currentExcessBlobGas, currentBeaconRoot)

Then classify the failure from the trace:

---

Phase 3 — Root cause analysis

For BadInstruction failures

This is the most common failure pattern for new forks. Follow these steps:

1. Identify the opcode from the trace entry where error is "BadInstruction":

   • op field = opcode byte value
   • opName field = human-readable name

2. Check opcode registration in src/Nethermind/Nethermind.Evm/Instructions/EvmInstructions.cs:

   • Find the line lookup[(int)Instruction.<OPNAME>] = ...
   • Note the spec flag gate: if (spec.<FlagName>) wrapping it

3. Check if the spec flag is true for the target fork:

   • Find the flag in src/Nethermind/Nethermind.Core/Specs/IReleaseSpecExtensions.cs or IReleaseSpec.cs
   • Trace it through the fork hierarchy in src/Nethermind/Nethermind.Specs/Forks/
   • Each fork class calls Apply(ReleaseSpec spec) to set its flags; the chain is replayed from root

4. If the opcode IS registered but still fails — the instruction implementation has a runtime guard. Check the instruction handler in src/Nethermind/Nethermind.Evm/Instructions/EvmInstructions.*.cs:

   • Look for patterns like if (!context.Header.SomeField.HasValue) goto BadInstruction;
   • These guards fail when the test harness doesn't set the header field

5. Check the test harness header construction:

   • State tests: src/Nethermind/Ethereum.Test.Base/GeneralStateTestBase.cs — look at the BlockHeader initializer
   • Blockchain tests: src/Nethermind/Ethereum.Test.Base/BlockchainTestBase.cs
   • Look for conditional defaults that use type checks (is Cancun, is Prague) instead of spec flag checks (IsEip4844Enabled, IsEip7702Enabled) — type checks only match the exact class, not subclasses or later forks
   • Look for missing defaults for new fork fields (e.g., ExcessBlobGas, RequestsHash, BlockAccessListHash)

For state root mismatches (no EVM error)

1. Check if expected hash is all-zeros — this is a placeholder test. The test was generated but the expected root hasn't been filled in. Report this to the user; it's not a code bug.

2. Check conditional defaults in header construction — a missing header field (e.g., null ExcessBlobGas) can change how the EVM executes even without causing BadInstruction.

3. Analyze the trace for unexpected behavior:

   • Storage writes at unexpected values
   • Calls that revert unexpectedly
   • Gas costs that differ from expected (compare with EIP specs)

4. Check fork spec configuration:

   • Missing EIP flags in the fork's Apply() method
   • Wrong gas parameters (e.g., blob schedule not applied from test config.blobSchedule)

5. Check blob schedule override — test JSON may have a config.blobSchedule section with per-fork overrides for target, max, and baseFeeUpdateFraction. Verify these are applied via OverridableReleaseSpec in JsonToEthereumTest.LoadSpec().

For block validation failures (blockchain tests)

1. Check if required header fields are set for the target fork
2. Look for new consensus rules that affect block validity (e.g., EIP-7928 BlockAccessListHash)
3. Check if genesis block processing uses the correct spec (see genesisUsesTargetFork logic in BlockchainTestBase.cs)

For load/parse failures

1. Check if the test JSON uses fields not yet supported in the deserialization models
2. Look at Ethereum.Test.Base/JsonToEthereumTest.cs and the JSON model classes
3. Check SpecNameParser.cs for unsupported fork names

---

Phase 4 — Fix and verify

1. Apply the minimal fix — prefer spec flag checks over type checks, add missing defaults, or fix gas calculations
2. Re-run the failing test to verify the fix resolves the issue
3. If test still fails:
   • Re-read the trace output
   • Return to Phase 2 with the new trace
   • Repeat until pass or root cause is confirmed as upstream
4. Add a regression test when the root cause was a Nethermind bug (AGENTS.md requires one for every bug fix) — extend an existing test file with a [TestCase] where possible; create a new test file only if no suitable one exists
5. Report the result — include:
   • Root cause (one sentence)
   • What was fixed (file:line)
   • Verification result (pass/fail + new state root)
   • Whether the test has a real expected hash or a placeholder (all-zeros)

---

Key files reference

Common bug patterns

1. RLP deserialization — incorrect RLP decoding can either accept malformed data (missing validation, wrong type expectations) or reject valid data (overly strict length checks, wrong sequence handling). For blockchain tests, blocks arrive as RLP — check Nethermind.Serialization.Rlp/ decoders for the relevant types. Common issues: not handling optional fields added by new EIPs, wrong list vs single-item decoding, incorrect length prefix validation.

2. Gas accounting — wrong gas costs for opcodes, missing gas charges for memory expansion, incorrect intrinsic gas calculation for new transaction types, or wrong refund logic. Compare against the EIP specification. Check GasCostOf.cs for static costs, instruction handlers for dynamic costs, and IntrinsicGasCalculator for transaction-level gas.

3. State root mismatch — execution completes without EVM errors but produces the wrong state root. Causes include: incorrect storage writes (wrong slot or value), wrong balance updates (fee burning, coinbase rewards, value transfer edge cases), missing or extra account touches that affect empty-account cleanup (EIP-158), or wrong nonce increments.

4. Missing header field default — covered in Phase 3, "Check the test harness header construction".

5. Instruction runtime guard — covered in Phase 3, "For BadInstruction failures" step 4.

6. Validation errors — missing validation that should reject invalid data (e.g., not checking transaction type constraints, accepting out-of-range values) or overly strict validation that rejects valid inputs (e.g., rejecting new transaction fields not yet accounted for, wrong bounds on EIP-introduced parameters). For blockchain tests, check block and transaction validators. For state tests, check TransactionProcessor validation and BlockValidator.

7. Missing blob schedule override — covered in Phase 3, "For state root mismatches" step 5.