fix: Fix eth_getLogs pending state handling, and more e2e pending state coverage

[sieniven]

Description

Two related changes to the flashblocks RPC layer.

Fix eth_getLogs pending-state handling (crates/rpc/src/filter.rs)

toBlock=Pending/Latest delegated the whole filter to the canonical filter whenever get_rpc_block returned None — which happens in the brief window after a pending sequence is promoted (pending_cache = None) or after canonical catch-up evicts the confirm-cache entry. The canonical filter then resolves Pending/Latest against canonical and drops logs from confirm-cache blocks ahead of canonical.

Now resolves Pending/Latest directly to pending_height/confirm_height from the flashblocks state cache; the existing range scan handles cross-boundary fallback via get_canonical_logs. Earliest/Safe/Finalized delegate the whole filter to canonical (they are not served by the FB cache), distinguished from "no bound supplied" so the bound is not silently collapsed into the FB cache range.

More e2e pending-state coverage (crates/tests/flashblocks-tests/main.rs)

• fb_pending_native_balance_delta_test — asserts sender/recipient balance deltas and sender nonce increment on Pending immediately after a native transfer.
• fb_pending_erc20_balance_delta_test — asserts sender/recipient ERC20 balance deltas on Pending via eth_call(balanceOf) immediately after an ERC20 transfer.
• sleep_one_blocktime() (~1 block) at the start of pending-state tests that send transactions to avoid block-boundary races.
• Reworked fb_pending_get_logs_test and fb_get_logs_cross_boundary_range_test to query via Pending / Number(tx_block) ahead-of-canonical ranges instead of waiting for canonical catch-up — exercising the actual flashblocks overlay.

Type of Change

• Bug fix
• Refactoring (no functional changes) — simplified FlashblocksEthFilterExt::logs flow

Testing

• cargo check -p xlayer-rpc, cargo clippy -p xlayer-rpc --all-targets -- -D warnings, cargo fmt -p xlayer-rpc -- --check
• cargo check/clippy/fmt -p xlayer-e2e-test --test flashblocks_tests
• E2E: fb_pending_get_logs_test reliably passes against a running devnet, including in the pending_cache = None window that previously caused intermittent failures.

Additional Notes

None — no public API changes outside FlashblocksEthFilterExt::new dropping the unused eth_api parameter (bin/node/src/main.rs updated in the same commit).

[sieniven]

@claude review for me

[XLayer-Bot]

The core filter.rs fix is correct and well-designed. Resolving Pending/Latest directly to the cached heights eliminates the delegation race during the pending_cache = None window, and resolve_fb_bound cleanly separates FB-served tags from canonical-only ones (Earliest/Safe/Finalized). Removing the unused eth_api field is a good clean-up. Two concerns in the new test code below.

workflow run

[sieniven]

@claude review again

[XLayer-Bot]

Both previously flagged concerns have been fully addressed:

1. L1 data fee in sender-balance assertion — accepted design: XLayer devnet uses zero L1 fee oracle scalars, so is effectively zero and the simplified gas cost computation is correct.

2. **Stale guard causing empty-result race in ** — fixed in commit 63fd06e: the guard is removed, now resolves to via , and is a valid range served directly from the FB cache. Only now returns .

The rest of the changes look clean:

• Removing the now-unused field from is a good simplification.
• covers all variants correctly (// delegate to canonical).
• The additions reduce block-boundary flakiness across the pending-state tests.
• New and are well-structured and correctly account for gas cost.
• The ABI encoding (selector + 12-byte padding + 20-byte address = 36 bytes) is correct.

workflow run

[XLayer-Bot]

🔧 CI running — workflow run

[XLayer-Bot]

✅ XLayer Reth CI passed for commit a968f0b2bb962e0a1a02294dbc345e3b1b4a54be

Step	Result
format-check	✅ success
compile-check	✅ success
clippy	✅ success
tests	✅ success

View run