test: pin the first two-phase cargo-mutants misses

[rearden-grok]

Summary

• Pins the 46 misses from the first successful two-phase nightly mutants run (Actions 37895507758): fee-filter equality, getblocktxn depth 10, txid notfound, the send-budget read pause, and the comparison sites in BIP68, median time, prevout lookup, script jobs, and IBD batch meters.
• Deletes the dead BIP34 negative-height encoder and the redundant coinbase clause beside OutPoint::is_null. The no-precompute value sum uses the money-range check.

Test plan

• cargo test -p rbitcoin-consensus --lib -- --quiet
• Targeted rbitcoin-net tests for the fee-filter journey, getblocktxn depth, txid notfound, and the over-budget read pause
• Local CI except coverage (fmt, clippy -D warnings, cargo deny, ast-grep, cargo test --workspace, ci-os-smoke)
• Required GitHub checks: qc, test, windows, macos, coverage

Made with Cursor

[average-gary]

Reviewed for correctness and performance. Nothing blocking: production always passes Some(&meta.pres) (confirm_run/phases.rs:174), so the changed no-pres branch only runs in unit tests or on a corrupt short pres. The production path performs the same; tx_to_apply drops a redundant 32-byte compare per input.

Inline: error-type mismatch plus stale comments in assemble_tx_value_out, a dead branch left in bip34_height_script, assert messages that read backwards, and a coinbase test that can't fail.

Optional, existing code outside this diff: phases.rs:190 checks meta.pres.len() == block.txdata.len() only after assemble_block_prevouts. A short pres uses the no-pres path for the extra txs before the corrupt-store error fires. Moving that check above the call means the fallback never runs in production.

Checked and fine: the over-budget ping test. V2SessionReader is cancel-safe (state lives on self.state), and over_budget is recomputed every 50ms heartbeat, so the 200ms timeout can't tear a frame. The getblocktxn depth 10/11 boundary matches peer.rs:3344.

🤖 Generated with Claude Code

[average-gary]

The structure gate rejects the same condition as BadBlock("bad-txns-vout-toolarge") (line 245). This branch returns BadTx, so the two paths give different reject types for the same input. Suggest BadBlock to match.

This also makes two comments stale: line 241 ("Only money-range gate") and the exceeds_max_money doc ("That cast is not a second range check"). Both need updating now that this branch range-checks too.

[average-gary]

is_err() would also pass on the old "negative output" / overflow errors. Suggest matches!(.., Err(ConsensusError::BadBlock("bad-txns-vout-toolarge"))) (or whatever variant line 1469 settles on), so the reject reason is pinned too.

[average-gary]

With the negative path gone this else is dead: n >= 17 here, so num is never empty. That's a fresh unkillable mutant in a PR that's killing them. abs / as i64 aren't needed anymore either. Possible form: loop on the u32, then if num.last().unwrap() & 0x80 != 0 { num.push(0) }.

[average-gary]

The message reads backwards. The peer's filter is rate + 1, so it's the filter that sits above the fee ("a filter above the fee is not announced"). The same goes for lines 376 and 389.

[average-gary]

The removed clause (txid == 0 && vout == MAX) was exactly OutPoint::is_null(), so the deletion changes no behaviour and this case passes on both versions. The deletion is right (the original mutant was equivalent), but this test doesn't pin anything. Optional: drop it.