Optimize Firo Spark mint tx generation and fix fee < vSize error

[reubenyap]

Summary

Two changes to _createSparkMintTransactions in spark_interface.dart:

1. Performance: Pre-compute signing keys, addresses, and wallet-owned address set once before the main loop, instead of per-UTXO per-fee-iteration.
2. Bug fix: Scale the fee-estimation buffer with input count to avoid the fee is less than vSize error on many-input mints (e.g., anonymize-all).

Performance optimization

Hot-loop operations moved out of the nested loops:

Operation	Before	After
getRootHDNode()	per UTXO × per fee iteration	once
mainDB.getAddress(walletId, ...)	per UTXO × per fee iteration	once per unique address
mainDB...walletIdEqualTo(...).valueEqualTo(...).findFirst() for walletOwns	per output × per fee iteration	once, as a Set<String> lookup
getCurrentReceivingSparkAddress() / getCurrentChangeAddress()	per inner iteration	once

For an N-input, M-fee-iteration mint this eliminates ~N×M expensive async calls.

Fee bug fix

Root cause. The dummy tx built for fee estimation is signed with the real keys over different data than the final tx. bitcoindart's ECDSA path (bip32-dart/ecurve.sign) uses RFC 6979 and enforces low-S but not low-R, so r's high bit is random — DER signatures vary by up to ~4 bytes per input. For Firo P2PKH (bip44) inputs this counts at full weight; with ~10+ inputs the dummy-vs-real vSize delta exceeds the original 10-byte buffer and trips nFeeRet < data.vSize.

Fix. Scale the buffer with input count:

final nBytesBuffer = 10 + 4 * setCoins.length;

Extra cost: ~4 sats per input at 1 sat/byte. Matches what Firo's own C++ wallet does — see DummySignatureCreator in src/wallet/wallet.h:1436 ("max-sized low-S signatures (eg 72 bytes)" → fee estimation is always an upper bound).

Subsidiary fixes

• mintedValue <= BigInt.zero (was == BigInt.zero): catches negative mintedValue when a UTXO group's total is less than the computed fee and subtractFeeFromAmount=false. Matches the C++ reference's !MoneyRange(mintedValue) || mintedValue == 0.
• Clarified the fee < vSize error message and added a comment: the check is a disguised min-relay-fee check (feeRate < 1 sat/byte), not a fee/size accounting error.

Why not retry?

An earlier iteration used a retry loop that restored UTXOs, outputs, and fee state when the check fired. Investigation confirmed the size variance has a deterministic upper bound (ECDSA DER max per input), so a scaled buffer is cleaner — no state to restore, no wasted mint-proof generation, no risk of loop bugs.

Test plan

• Single-recipient Spark mint (1–3 inputs): verify no regression.
• Anonymize-all with 10+ small UTXOs: verify the tx builds without fee is less than vSize.
• Sanity check: signing-key cache miss (corrupted DB) throws a clear rescan message.

🤖 Generated with Claude Code