test(sae): Address legacypool race

[alarso16]

Why this should be merged

There is a bug in the legacypool that's not worth fixing. Specifically, at the end of block execution, an event is sent to the TxPool to indicate a reorg. This is expected to update nonces so that transactions already in the mempool are either evicted or promoted using this new data. However, the internal nonce tracking happens after the promotions, so if there is already a pending tx for some account in the pool that is still executable, a dependent queued transaction will be incorrectly gapped. If the pending tx is ever repriced or executed (and maybe some other cases), the queued transaction can later be moved to pending.

This is a rather complicated problem, and still exists upstream in go-ethereum. There is no simple way to inject a couple lines in libevm to address this, so we should probably just deal with it here. Most importantly, this doesn't significantly affect any production processes, since the transaction will just be included in a later block. It only affects testing, because we want to rely on adding transactions to a block.

How this works

This race requires a few conditions:

1. A block is currently executing (i.e. the NewHead event could come anytime)
2. A transaction for an address is marked as pending
3. A transaction for the same address is marked as queued, but a reorg is scheduled

This is pretty specific, so there's several ways that we can in general fix this. I only know of two flakes of this manner, fixed in this PR. The first flake, TestWorstCase, relies on execution in the background, so we must just mark it as flaky. The second flake is easily avoided by not issuing transactions while a block is executing.

I manually reviewed all tests in sae/saevm, and if there were any blocks that expected multiple transactions from a single account, I ensured that they are either listed above, or are the first block created (aka race avoided)

How this was tested

Lots and lots of debugging... if the flakes come back, I'm happy to further investigate

Need to be documented in RELEASES.md?

No

[Copilot]

Pull request overview

This PR reduces flakes in SAE VM tests caused by a known race in libevm’s legacy txpool nonce tracking during concurrent block execution/reorg signaling.

Changes:

• Mark TestWorstCase as explicitly flaky (opt-in via env var) and move it into a Bazel flaky-only test target.
• Ensure receipt-related tests wait for block execution before asserting/using execution-dependent state.
• Document the “pending tx” race caveat on helper methods that wait for mempool pending state.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
vms/saevm/sae/worstcase_test.go	Makes TestWorstCase opt-in flaky and removes CLI-flag plumbing in favor of fixed parameters.
vms/saevm/sae/vm_test.go	Removes flag usage from TestMain; adds warnings about pending-state wait helpers under concurrent execution.
vms/saevm/sae/rpc_test.go	Waits for blocks to finish executing before continuing receipt-based assertions.
vms/saevm/sae/accept_block_test.go	Unifies flaky gating under SAEVM_TEST_FLAKY.
vms/saevm/sae/BUILD.bazel	Moves flaky tests into a dedicated flaky_test Bazel target and excludes them from the main test target.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[JonathanOppenheimer]

This look pretty good to me. You clearly did your research.

[alarso16]

Symbolic approval of Stephen's edits

[JonathanOppenheimer]

Restoring the fuzz testing is good IMO.