test(fast-inbox): drive l1 publisher inbox consumption through the streaming bucket selector (A-1387)

spalladino · spalladino · commit 92c4a70e6af2 · 2026-07-21T19:00:22.000-03:00
The A-1387 rewrite that removed inHash left this test's inbox handling half-adapted: it hardcoded
previousInboxRollingHash to Fr.ZERO and proposed with bucketHint 0n while consuming real L1-&gt;L2 messages, so
message-consuming checkpoints reverted on L1 with Rollup__InvalidInboxRollingHash / UnconsumedInboxMessages.
INBOX_LAG_SECONDS is an L1-time censorship cutoff (not whole checkpoints), so no checkpoint-depth shift
register can reproduce the aged bucket set. Mirror the real Inbox buckets into the test's
MockL1ToL2MessageSource, then reuse the production selectInboxBucketForBlock (which mirrors
ProposeLib.validateInboxConsumption) to pick exactly the buckets each checkpoint must consume, deriving the
consumed bundle, the propose bucket hint, and the header rolling hash from that one selection so header,
world state, and L1 agree. Matches the A-1388 rewrite of this file so the segment rebase dissolves to zero.
diff --git a/yarn-project/sequencer-client/src/publisher/l1_publisher.integration.test.ts b/yarn-project/sequencer-client/src/publisher/l1_publisher.integration.test.ts
@@ -1,4 +1,5 @@
 import type { ArchiverDataSource } from '@aztec/archiver';
+import { MockL1ToL2MessageSource } from '@aztec/archiver/test';
 import { AztecAddress } from '@aztec/aztec.js/addresses';
 import { Fr } from '@aztec/aztec.js/fields';
 import { createLogger } from '@aztec/aztec.js/log';
@@ -13,6 +14,8 @@ import {
 } from '@aztec/blob-lib';
 import {
   GENESIS_ARCHIVE_ROOT,
+  INBOX_LAG_SECONDS,
+  MAX_L1_TO_L2_MSGS_PER_BLOCK,
   MAX_L1_TO_L2_MSGS_PER_CHECKPOINT,
   MAX_NULLIFIERS_PER_TX,
   MAX_PROCESSABLE_L2_GAS,
@@ -47,7 +50,7 @@ import { retryUntil } from '@aztec/foundation/retry';
 import { sleep } from '@aztec/foundation/sleep';
 import { hexToBuffer } from '@aztec/foundation/string';
 import { TestDateProvider } from '@aztec/foundation/timer';
-import { RollupAbi } from '@aztec/l1-artifacts';
+import { InboxAbi, RollupAbi } from '@aztec/l1-artifacts';
 import { getVKTreeRoot } from '@aztec/noir-protocol-circuits-types/vk-tree';
 import { ProtocolContractsList, protocolContractsHash } from '@aztec/protocol-contracts';
 import { LightweightCheckpointBuilder } from '@aztec/prover-client/light';
@@ -86,11 +89,12 @@ import { NativeWorldStateService, ServerWorldStateSynchronizer, type WorldStateC
 
 import { beforeEach, describe, expect, it, jest } from '@jest/globals';
 import { type MockProxy, mock } from 'jest-mock-extended';
-import { type Address, encodeFunctionData, getAbiItem, getAddress, multicall3Abi } from 'viem';
+import { type Address, encodeFunctionData, getAbiItem, getAddress, getContract, multicall3Abi } from 'viem';
 import { type PrivateKeyAccount, privateKeyToAccount } from 'viem/accounts';
 import { foundry } from 'viem/chains';
 
 import { type SequencerClientConfig, getConfigEnvVars } from '../config.js';
+import { selectInboxBucketForBlock } from '../sequencer/inbox_bucket_selector.js';
 import { sendL1ToL2Message } from './l1_to_l2_messaging.js';
 import { SequencerPublisherMetrics } from './sequencer-publisher-metrics.js';
 import { SequencerPublisher } from './sequencer-publisher.js';
@@ -111,7 +115,8 @@ const logger = createLogger('integration_l1_publisher');
 // depending on @aztec/aztec-node, which would create a sequencer-client <-> aztec-node cycle.
 const config: SequencerClientConfig & L1ContractsConfig = { ...getL1ContractsConfigEnvVars(), ...getConfigEnvVars() };
 
-// Must exceed the inbox lag (network default 2) so at least one checkpoint consumes a real L1->L2 message.
+// Several consecutive checkpoints, each consuming the L1->L2 messages sent while it was being built, so real
+// messages are genuinely consumed and validated on L1 (AZIP-22 Fast Inbox).
 const numberOfConsecutiveBlocks = 3;
 
 jest.setTimeout(1000000);
@@ -138,6 +143,11 @@ describe('L1Publisher integration', () => {
 
   let builderDb: NativeWorldStateService;
 
+  // Backs the blockSource mock's streaming L1->L2 message queries. The world-state synchronizer reconstructs each
+  // block's consumed message bundle from Inbox buckets (AZIP-22 Fast Inbox) when it syncs a block back, so the test
+  // registers one bucket per published block here (see buildAndPublishBlock).
+  let messageSource: MockL1ToL2MessageSource;
+
   // The header of the last block
   let prevHeader: BlockHeader;
 
@@ -274,6 +284,21 @@ describe('L1Publisher integration', () => {
       checkpointNumber: CheckpointNumber.ZERO,
       indexWithinCheckpoint: IndexWithinCheckpoint(0),
     };
+    // Seed the genesis sentinel bucket (seq 0, no messages) so the world-state synchronizer can resolve a
+    // totalMsgCount of 0 to a bucket when reconstructing the first block's message bundle.
+    messageSource = new MockL1ToL2MessageSource(0);
+    messageSource.setInboxBucket(
+      {
+        seq: 0n,
+        inboxRollingHash: Fr.ZERO,
+        totalMsgCount: 0n,
+        timestamp: 0n,
+        msgCount: 0,
+        lastMessageIndex: 0n,
+        isOpen: false,
+      },
+      [],
+    );
     blockSource = mock<ArchiverDataSource>({
       getBlocks(query: BlocksQuery) {
         if (!('from' in query)) {
@@ -349,6 +374,14 @@ describe('L1Publisher integration', () => {
       getBlockNumber(): Promise<BlockNumber> {
         return Promise.resolve(BlockNumber(blocks.at(-1)?.number ?? BlockNumber.ZERO));
       },
+      // Streaming L1->L2 message reconstruction (AZIP-22 Fast Inbox): the world-state synchronizer resolves each
+      // block's consumed message bundle from the Inbox buckets registered per published block in buildAndPublishBlock.
+      getInboxBucketByTotalMsgCount(totalMsgCount: bigint) {
+        return messageSource.getInboxBucketByTotalMsgCount(totalMsgCount);
+      },
+      getL1ToL2MessagesBetweenBuckets(fromExclusive: bigint, toInclusive: bigint) {
+        return messageSource.getL1ToL2MessagesBetweenBuckets(fromExclusive, toInclusive);
+      },
     });
 
     const worldStateConfig: WorldStateConfig = {
@@ -444,13 +477,16 @@ describe('L1Publisher integration', () => {
 
   /**
    * Build a checkpoint with a single block using the LightweightCheckpointBuilder.
-   * This properly computes all checkpoint header fields (blobsHash, blockHeadersHash, inHash, epochOutHash, etc.)
+   * This properly computes all checkpoint header fields (blobsHash, blockHeadersHash, inboxRollingHash,
+   * epochOutHash, etc.). `previousInboxRollingHash` is the previous checkpoint's rolling hash (zero at genesis), so
+   * the header's `inboxRollingHash` continues the on-chain Inbox chain over `l1ToL2Messages`.
    */
   const buildCheckpoint = async (
     globalVariables: GlobalVariables,
     txs: ProcessedTx[],
     l1ToL2Messages: Fr[],
     previousCheckpointOutHashes: Fr[] = [],
+    previousInboxRollingHash: Fr = Fr.ZERO,
   ): Promise<Checkpoint> => {
     await worldStateSynchronizer.syncImmediate();
     const tempFork = await worldStateSynchronizer.fork(BlockNumber(globalVariables.blockNumber - 1));
@@ -472,7 +508,7 @@ describe('L1Publisher integration', () => {
       checkpointConstants,
       l1ToL2Messages,
       previousCheckpointOutHashes,
-      Fr.ZERO,
+      previousInboxRollingHash,
       tempFork,
     );
 
@@ -486,7 +522,8 @@ describe('L1Publisher integration', () => {
   const buildSingleCheckpoint = async (
     opts: { l1ToL2Messages?: Fr[]; blockNumber?: BlockNumber; slot?: SlotNumber } = {},
   ) => {
-    const l1ToL2Messages = opts.l1ToL2Messages ?? new Array(MAX_L1_TO_L2_MSGS_PER_CHECKPOINT).fill(Fr.ZERO);
+    // By default a single checkpoint consumes no Inbox messages (bucketHint 0 against the genesis bucket).
+    const l1ToL2Messages = opts.l1ToL2Messages ?? [];
 
     const txs = await Promise.all([makeProcessedTx(0x1000), makeProcessedTx(0x2000)]);
     const ts = (await l1Client.getBlock()).timestamp;
@@ -503,7 +540,6 @@ describe('L1Publisher integration', () => {
       new GasFees(0, await rollup.getManaMinFeeAt(timestamp, true)),
     );
     const checkpoint = await buildCheckpoint(globalVariables, txs, l1ToL2Messages);
-    blockSource.getL1ToL2Messages.mockResolvedValueOnce(l1ToL2Messages);
     return { checkpoint, l1ToL2Messages };
   };
 
@@ -517,10 +553,8 @@ describe('L1Publisher integration', () => {
 
   describe('block building', () => {
     beforeEach(async () => {
-      // This suite proposes consecutive checkpoints and models the inbox lag by hand (a checkpoint
-      // consumes the L1->L2 messages sent inboxLag checkpoints earlier -- see the shift register in
-      // buildAndPublishBlock). It inherits the network default lag and proposes enough checkpoints
-      // that a real message is actually consumed and validated on L1.
+      // This suite proposes consecutive checkpoints, each consuming the streaming-Inbox messages sent while it was
+      // being built (AZIP-22 Fast Inbox), so real messages are genuinely consumed and validated on L1.
       await setup();
     });
 
@@ -535,11 +569,24 @@ describe('L1Publisher integration', () => {
         '0x1647b194c649f5dd01d7c832f89b0f496043c9150797923ea89e93d5ac619a93',
       );
 
-      // The deployed rollup consumes L1->L2 messages with a lag of `inboxLag` checkpoints: a message
-      // inserted while building checkpoint N is only consumable at checkpoint N + inboxLag. Model that
-      // with a shift register so a real message is genuinely consumed and validated on L1.
-      const inboxLag = getL1ContractsConfigEnvVars().inboxLag;
-      const messagesInFlight: Fr[][] = Array.from({ length: inboxLag }, () => []);
+      // Streaming Inbox consumption (AZIP-22 Fast Inbox): the L1 Rollup only lets a checkpoint consume Inbox buckets
+      // that have aged past the censorship cutoff (`toTimestamp(slot - 1) - INBOX_LAG_SECONDS`), measured in L1 time,
+      // not whole checkpoints. Each checkpoint mirrors the real Inbox buckets into messageSource, then reuses the
+      // production `selectInboxBucketForBlock` (which mirrors `ProposeLib.validateInboxConsumption`) to pick exactly
+      // the buckets it must consume, deriving the consumed bundle, the propose bucket hint, and the header rolling
+      // hash from that one selection so header, world state, and L1 agree by construction.
+      const inbox = getContract({
+        address: getAddress(l1ContractAddresses.inboxAddress.toString()),
+        abi: InboxAbi,
+        client: l1Client,
+      });
+      // Every message sent to the Inbox, in insertion order, so each bucket's leaves can be mirrored into messageSource.
+      const allSentMessages: Fr[] = [];
+      let mirroredThroughSeq = 0n;
+      let mirroredThroughTotal = 0n;
+      // The last Inbox bucket this checkpoint chain has consumed through; genesis sentinel to start.
+      let parent = { seq: 0n, totalMsgCount: 0n };
+      let previousInboxRollingHash = Fr.ZERO;
       const blobFieldsPerCheckpoint: Fr[][] = [];
       // The below batched blob is used for testing different epochs with 1..numberOfConsecutiveBlocks blocks on L1.
       // For real usage, always collect ALL epoch blobs first then call .batch().
@@ -554,10 +601,29 @@ describe('L1Publisher integration', () => {
         for (let j = 0; j < l1ToL2Content.length; j++) {
           sentThisCheckpoint.push(await sendToL2(l1ToL2Content[j], recipientAddress));
         }
-
-        // Consume the messages sent inboxLag checkpoints ago, then enqueue this checkpoint's messages.
-        const currentL1ToL2Messages = messagesInFlight.shift()!;
-        messagesInFlight.push(sentThisCheckpoint);
+        allSentMessages.push(...sentThisCheckpoint);
+
+        // Mirror the Inbox's new buckets (seq, timestamp, rolling hash, totals) and their leaves into messageSource,
+        // so the selector, the world-state synchronizer, and L1 all read the same bucket state.
+        const currentBucketSeq = await inbox.read.getCurrentBucketSeq();
+        for (let seq = mirroredThroughSeq + 1n; seq <= currentBucketSeq; seq++) {
+          const bucket = await inbox.read.getBucket([seq]);
+          const bucketMessages = allSentMessages.slice(Number(mirroredThroughTotal), Number(bucket.totalMsgCount));
+          messageSource.setInboxBucket(
+            {
+              seq,
+              inboxRollingHash: Fr.fromString(bucket.rollingHash),
+              totalMsgCount: bucket.totalMsgCount,
+              timestamp: bucket.timestamp,
+              msgCount: Number(bucket.msgCount),
+              lastMessageIndex: bucket.totalMsgCount - 1n,
+              isOpen: seq === currentBucketSeq,
+            },
+            bucketMessages,
+          );
+          mirroredThroughTotal = bucket.totalMsgCount;
+        }
+        mirroredThroughSeq = currentBucketSeq;
 
         // Ensure that each transaction has unique (non-intersecting nullifier values)
         const totalNullifiersPerBlock = 4 * MAX_NULLIFIERS_PER_TX;
@@ -580,14 +646,42 @@ describe('L1Publisher integration', () => {
           new GasFees(0, await rollup.getManaMinFeeAt(timestamp, true)),
         );
 
-        const checkpoint = await buildCheckpoint(globalVariables, txs, currentL1ToL2Messages);
+        // Reuse the production streaming selector to pick the buckets this single-block (hence last-block) checkpoint
+        // must consume, then derive the consumed bundle, the propose bucket hint, and the rolling-hash cursor from
+        // that one selection so the header, world state, and L1 all agree.
+        const buildFrameStart = await rollup.getTimestampForSlot(SlotNumber(slot - 1));
+        const cutoffTimestamp = buildFrameStart - BigInt(INBOX_LAG_SECONDS);
+        const selection = await selectInboxBucketForBlock({
+          messageSource,
+          now: buildFrameStart,
+          lagSeconds: BigInt(INBOX_LAG_SECONDS),
+          parent,
+          checkpointStartTotalMsgCount: parent.totalMsgCount,
+          perBlockCap: MAX_L1_TO_L2_MSGS_PER_BLOCK,
+          perCheckpointCap: MAX_L1_TO_L2_MSGS_PER_CHECKPOINT,
+          isLastBlock: true,
+          cutoffTimestamp,
+        });
+        const currentL1ToL2Messages = selection.consume ? selection.bundle : [];
+        const bucketHint = selection.consume ? selection.bucket.seq : parent.seq;
+
+        const checkpoint = await buildCheckpoint(
+          globalVariables,
+          txs,
+          currentL1ToL2Messages,
+          [],
+          previousInboxRollingHash,
+        );
+        previousInboxRollingHash = checkpoint.header.inboxRollingHash;
         const block = checkpoint.blocks[0];
+        if (selection.consume) {
+          parent = { seq: selection.bucket.seq, totalMsgCount: selection.bucket.totalMsgCount };
+        }
 
         const totalManaUsed = txs.reduce((acc, tx) => acc.add(new Fr(tx.gasUsed.billedGas.l2Gas)), Fr.ZERO);
         expect(totalManaUsed.toBigInt()).toEqual(block.header.totalManaUsed.toBigInt());
 
         prevHeader = block.header;
-        blockSource.getL1ToL2Messages.mockResolvedValueOnce(currentL1ToL2Messages);
 
         const checkpointBlobFields = checkpoint.toBlobFields();
         const blockBlobs = await getBlobsPerL1Block(checkpointBlobFields);
@@ -615,7 +709,7 @@ describe('L1Publisher integration', () => {
           checkpoint,
           CommitteeAttestationsAndSigners.empty(getSignatureContext()),
           Signature.empty(),
-          0n,
+          bucketHint,
         );
         // Align chain time so the bundle simulate and the L1 send both run at the header's slot.
         await progressToSlot(BigInt(checkpoint.header.slotNumber));
@@ -661,7 +755,7 @@ describe('L1Publisher integration', () => {
               oracleInput: {
                 feeAssetPriceModifier: 0n,
               },
-              bucketHint: 0n,
+              bucketHint,
             },
             CommitteeAttestationsAndSigners.packAttestations([]),
             [],
