Make builders non-validating staked actors

[StefanBratanov]

PR Description

As per ethereum/consensus-specs#4788

Also enabling back al the Gloas consensus-specs reference tests from: https://github.com/ethereum/consensus-specs/releases/tag/v1.7.0-alpha.1

The major change TLDR is that we track builders in the state separate from validators and process their deposits/withdrawals separate

Fixed Issue(s)

N/A

Documentation

• I thought about documentation and added the doc-change-required label to this PR if updates are required.

Changelog

• I thought about adding a changelog entry, and added one if I deemed necessary.

---

Note

Implements builders as non-validating staked actors and aligns with consensus-specs v1.7.0-alpha.1.

• Add Builder SSZ container and registry in BeaconState with caches (buildersPubKeys, builderIndexCache); add next_withdrawal_builder_index and payload_expected_withdrawals
• Replace validator-based builder fields with BuilderIndex (incl. BUILDER_INDEX_SELF_BUILD), update ExecutionPayloadBid/Envelope and signature checks to use builder pubkeys; special handling for self-builds
• Rework withdrawals: queue/process builder pending withdrawals and a builders sweep; apply withdrawals to builder balances; advance next_withdrawal_* indices; extend ExpectedWithdrawals
• Handle deposits for builders immediately and add builder exits (initiateBuilderExit); map builder/validator indices via new helpers
• Update predicates/accessors/mutators for builders (balance checks, active status, index/pubkey lookups); adjust SSZ schemas (Builder.version uint8; remove withdrawable_epoch from BuilderPendingWithdrawal)
• Enable Gloas ssz_*, fork-choice tests; update test fixtures/utilities to use BUILDER_INDEX_SELF_BUILD; bump specrefs to v1.7.0-alpha.1

^{Written by Cursor Bugbot for commit cbd4670. This will update automatically on new commits. Configure here.}

[StefanBratanov]

Thanks a lot @jtraglia actually all reference tests should be passing now after the last 3 commits (apart from few minimal failing tests)

[zilm13]

LGTM, just few nits

[cursor]

Builder index corrupts next withdrawal validator index calculation

High Severity

WithdrawalsHelpersGloas does not override updateNextWithdrawalValidatorIndex, inheriting the Capella implementation that assumes all withdrawals have real validator indices. In Gloas, builder withdrawals use convertBuilderIndexToValidatorIndex which sets BUILDER_INDEX_FLAG (2^40). When the withdrawal list is full and the last withdrawal is from a builder, withdrawals.getLast().getValidatorIndex() returns a large value (e.g., 2^40 + builderIndex). Computing .mod(state.getValidators().size()) on this produces an arbitrary position, causing validators to be skipped or processed out of order in subsequent withdrawal sweeps.

Additional Locations (1)

• ethereum/spec/src/main/java/tech/pegasys/teku/spec/logic/versions/capella/withdrawals/WithdrawalsHelpersCapella.java#L272-L291

 

[jtraglia]

Fixed here, will be included in the next release:

• Reserve space for one validator sweep withdrawal ethereum/consensus-specs#4832

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

[cursor]

Missing override for validator sweep index with builder withdrawals

High Severity

WithdrawalsHelpersGloas inherits updateNextWithdrawalValidatorIndex from Capella without override. When the withdrawal limit is reached during builder withdrawals (before validator sweep), the last withdrawal has a validatorIndex with BUILDER_INDEX_FLAG (2^40) set. The inherited method computes withdrawals.getLast().getValidatorIndex().plus(1).mod(validators.size()), producing an incorrect next validator sweep index since the builder index is astronomically larger than the validator count.

Additional Locations (1)

• ethereum/spec/src/main/java/tech/pegasys/teku/spec/logic/versions/capella/withdrawals/WithdrawalsHelpersCapella.java#L272-L290

 

[mehdi-aouadi]

LGTM
Just few nits

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

[cursor]

Signed byte to int conversion fails for values >= 128

Low Severity

In getBuilderFromDeposit, withdrawalCredentials.get(0) returns a Java signed byte which is automatically promoted to a signed int when passed to the Builder constructor. For byte values >= 0x80 (128), this produces a negative int (e.g., 0x80 → -128), which will fail the checkArgument(value >= 0 && value <= 255) validation in SszByte.asUInt8. The getter Builder.getVersion() already uses ByteUtil.toUnsignedInt for correct unsigned conversion, but the setter path doesn't.

 

[cursor]

Missing override causes wrong next validator index with builder withdrawals

Medium Severity

WithdrawalsHelpersGloas doesn't override updateNextWithdrawalValidatorIndex. The inherited Capella implementation assumes when withdrawals.size() == maxWithdrawals, the last withdrawal is from validator sweep and uses withdrawals.getLast().getValidatorIndex() to set the next index. In Gloas, if builder withdrawals fill the list first, the last withdrawal has BUILDER_INDEX_FLAG set (a value around 2^40), causing getValidatorIndex().plus(1).mod(validatorCount) to produce an essentially arbitrary validator index for the next sweep.

Additional Locations (1)

• ethereum/spec/src/main/java/tech/pegasys/teku/spec/logic/versions/capella/withdrawals/WithdrawalsHelpersCapella.java#L272-L290