[Nic-dorman]

Fixes #144.

Before this PR, nothing on main exercised #172's resumable external-signer finalize against a live network: no e2e test called finalize_upload_resumable, finalize_upload_merkle_multi_resumable or finalize_resume (the Tier 1 test #144 cites belonged to #143, which #172 replaced). Only the pure assemble_* unit tests covered the retain/hand-off contract.

Approach: a reversible outage, not node restarts

#144 listed three ways to get "fails after payment, then recovers": restart nodes in MiniTestnet, add per-node stop/start to LocalDevnet (ant-node 0.21's Devnet has neither), or a reversible network block. It called the last one least flaky, but Linux-only via iptables.

The harness already had the mechanism: PointerSilence makes a node read and drop pointer requests while staying in the network. This generalizes it to Silence { pointer_gets, pointer_puts, chunk_puts } and adds MiniTestnet::silence_chunk_puts(bool).

• During the outage nodes are up but never answer a chunk PUT.
• Lifting it restores them with the same PeerIds and addresses. Every paid wave chunk is bound to the peers that quoted it, so that's the requirement.
• No port reuse, no DHT re-convergence, and it's cross-platform.

e2e_pointer.rs changes only for the rename (pointer_silence.gets → silence.pointer_gets, etc.).

Tests

Both run in the existing CI jobs (e2e_file in E2E Tests, e2e_merkle in Merkle E2E). Each one:

1. Prepares a public upload and pays as an external signer with the testnet wallet.
2. Silences chunk PUTs on every node. Nothing answers a silenced PUT, so every attempt waits out its timeout:
   • Wave uses the fixed 10s STORE_RESPONSE_TIMEOUT, which adds up to about 90s.
   • Merkle uses merkle_store_timeout_secs. The test lowers it from 270s to 2s for the outage only. At 270s this phase alone took 92 minutes.
3. Finalizes, and asserts FinalizeOutcome::Partial with 0 stored, all failed, and the data_map_address carried through.
4. Lifts the silence, restores the merkle timeout, and runs finalize_resume in a loop bounded to 3 attempts, as the API documents. Asserts Complete with everything stored and the same data_map_address.
5. Asserts the wallet token balance is unchanged across the failed finalize and the resume (no second payment).
6. Runs data_map_fetch(public address) → file_download and checks the bytes equal the original.

• e2e_file::test_finalize_resume_after_post_payment_outage_wave_batch: wave-batch, 4 KB file, the default 14-node testnet.
• e2e_merkle::test_external_merkle_finalize_resume_after_post_payment_outage: merkle multi-batch (cap 3, so ≥2 sub-batches), 500 KB, 35 nodes.

Linear issue

Closes V2-1519

Risk tier

• T0 — docs / tooling / CI / pure UX-output. Repo CI only.
• T1 — client-only, no network-facing behavior change. CI + prod compat smoke.
• T2 — node/client logic with behavioral surface, no protocol/format/economics change. Dev testnet + ADR.
• T3 — protocol / storage format / payments / routing. T2 evidence + adversarial testing.

Test-only: harness and e2e tests, no library code.

Compatibility

• Wire: none
• Storage: none
• API: none (test harness only)

Semver impact

• breaking
• feature
• fix

Test evidence

Rebased 2026-10-23 onto main at 50bc653 (0.3.12 promoted). tests/support/mod.rs conflicted with #218's ChunkGetCounts and sentinel-barrier changes, which also thread through spawn_node. The resolution keeps both sides: main's chunk_gets plumbing, plus this PR's PointerSilence → Silence generalisation with chunk_puts and silence_chunk_puts(). This PR's added and removed lines are otherwise identical to before the rebase. Re-run on Linux (dev2, stable):

• cargo fmt --all -- --check and cargo clippy --all-targets --all-features -- -D warnings: pass.
• test_finalize_resume_after_post_payment_outage_wave_batch: passed (144 s).
• test_external_merkle_finalize_resume_after_post_payment_outage: passed (593 s).
• e2e_pointer, which exercises the renamed silence switches: 17 passed.

Original evidence follows.
Local runs on Windows 11, against the in-process MiniTestnet and Anvil:

• e2e_file::test_finalize_resume_after_post_payment_outage_wave_batch: passed in 159s. The finalize during the outage returned Partial after 91s, and the resume completed on its first attempt.
• e2e_merkle::test_external_merkle_finalize_resume_after_post_payment_outage: passed in 596s. The outage finalize returned after 140s (5505s before the test lowered merkle_store_timeout_secs), and the resume completed on its first attempt.
• e2e_pointer, all 14 tests: passed after the Silence rename. CI doesn't run this file.
• cargo fmt --all -- --check: clean.
• cargo clippy --all-targets --all-features -- -D warnings: didn't finish locally because the machine ran short of memory. CI's Clippy job is the lint check here.

CI runs both new tests on ubuntu-latest and macos-latest (E2E Tests, Merkle E2E). Merkle E2E has a 90-minute limit and has recently taken 48–54 minutes; this test adds about 10.

New dependency

none

ADR

n/a

Mitigation / rollback

Revert the commit. It only touches ant-core/tests/.

🤖 Generated with Claude Code

[MaidSafe-Hermes]

APPROVE

Reviewed head 56cc7c4eb04ccad205bf8e16cd930428cde62ec3 against base ccd4106dd8fd60545bbd31c40d46cc041f85fb8d. No actionable defects found. The request filter drops chunk PUTs before handling; the tests verify partial finalize, resume against the original payment, unchanged token balance, and byte-exact download.

2/2 seats reviewed (sol-1 gpt-6.1-sol xhigh; glm-1 z-ai/glm-5.3); both seats agree (two model families). All 18 CI checks passed for this head. git diff --check passed.

Non-blocking clarity note: the wave test says “a fresh retriever” but reuses its client. The address-based fetch still exercises the network chunk-get path, but not a separately created client.