fix(staking): read the joined validator set a page at a time

[kirilaa]

What broke

The consensus resolution-kernel train withdraws the unpaged validator reads this CLI is built on. Three separate footholds went at once:

Surface	Status on the train	CLI caller
activeValidators()	removed	staking active-validators, staking validators
validatorsRoot()	removed (survives only in a test-helper mock)	StakingAction.getAllValidatorsFromTree()
validatorView(address)	tuple shrank 12 → 9 fields; left / right / parent dropped	the same tree walk

None of these degrade gracefully. The unpaged read reverts rather than truncating, so the command exits non-zero rather than returning a short list.

Evidence — composed e2e run against the train, shard retry job 97932864617:

genlayer staking active-validators --staking-address 0xdE14C2233118651c4140F959aC4d27E1AA925cFf --rpc http://127.0.0.1:9168
exited 1

What this changes

The joined validator set is read from the append-only registry, one page at a time — validatorsJoinedCount() bounds the walk, getValidatorsJoined(start, pageSize) returns each page.

• The count is read first, so a set that grows underneath the walk cannot spin the loop; a short page means it shrank instead, so the walk stops and lets the next read see the settled set.
• Page size is 64 — the convention the contract's own paged reads are written around. Committee capacity is 1,543 seats, which is why an address[] that long had to go: it overruns the return-size limit.
• getAllValidatorsFromTree() becomes getJoinedValidators(). The registry is the same superset the tree walk was after (all joined validators, primed or not), so callers keep their semantics.
• staking validators loses its one-call answer for "in the current draw". The active marker is derived from what is still readable: joined, and neither banned nor quarantined.

This mirrors the two reference adaptations, both of which also map "active validators" onto the paged joined registry: genlayerlabs/genlayer-node#1800 (pagedValidatorsJoined, same page size of 64) and genlayerlabs/genlayer-dev-env#138 (VALIDATORS_JOINED_PAGE_SIZE).

What this does NOT fix

The reads above are ones the CLI issues itself, against its own ABI fragment, so they are fixable here. The rest of the CLI's train exposure is inside genlayer-js (pinned github:genlayerlabs/genlayer-js#v2-dev) and has to be fixed there — this PR alone does not turn the CLI e2e green:

• getTransactionData(bytes32,uint256) was removed from ConsensusData (successors: getStoredTransactionData(bytes32), getStoredTransactionStatus(bytes32), getTransactionAllData(bytes32)). transactionActions.getTransaction still calls it, so waitForTransactionReceipt / getTransactionReceipt fail — this is the same Function "getTransactionData" not found on ABI seen in the explorer and wallet lanes. Blocks 024_cli_counter_single_inc.
• activeValidatorsCount() was removed, and getEpochInfo() reads it inside a Promise.all, so the whole call reverts. Blocks staking epoch-info (090_cli_network_and_balances) and every client.getEpochInfo() caller.
• validatorView output drift (above) also breaks the SDK's getValidatorInfo(), so staking validator-info and the detail rows of staking validators stay blocked.
• BalancesAction.getKnownValidatorSet() calls client.getActiveValidators() via the vesting client. Left alone deliberately: once the SDK pages internally (as node#1800 and dev-env#138 both do while keeping the method name), that call site is correct as written.

Also worth a decision: .github/e2e-track in this repo pins v0.5, while the train work runs on the v0.6 track. The CLI's own e2e will not exercise the train until that moves.

Validation

• npm run build — clean
• npx vitest run — 796 passed, 72 files
• npx tsc --noEmit — 37 errors, all pre-existing (38 before this change; one implicit any went away with the deleted tree walk)

Not yet run against a live train deployment — that is what the acceptance below is for.

Acceptance

genlayer staking active-validators and staking validators return the validator set against a deployment carrying the train, and the CLI e2e lane passes once the genlayer-js items above land.

Depends-On: genlayerlabs/genlayer-consensus#1307

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[kirilaa]

The four SDK-side items under What this does NOT fix are now covered by
genlayerlabs/genlayer-js#212 — leaving a note rather than editing the body,
since that PR is yours.

Item listed here	#212
getTransactionData successors	detected against the chain's own ABI, so a testnet keeps the old call
activeValidatorsCount() inside getEpochInfo()	reads validatorsJoinedCount() instead
validatorView drift breaking getValidatorInfo()	ABI resynced to the train's nine fields
getKnownValidatorSet() → client.getActiveValidators()	paged walk over getValidatorsJoined

One correction worth carrying across: the paged reads are not train-only.
Pre-train StakingReadFacet already exposes validatorsJoinedCount and
getValidatorsJoined (and activeValidatorsCount), so moving to them is safe
on both surfaces — activeValidators, the one the train removed, lived on
StakingConfig. That narrows the shared compat risk to validatorView 12→9,
which is the one read with no chain ABI to detect against.