Treat errors as first-class: don’t blanket-suppress them with `try?`, `try!`, `_ =` result discards, or `2>/dev/null` unless you can prove the failure is expected and harmless. For unexpected failures, either propagate (`throw`) or explicitly log and recover in a deterministic way. Apply this rule: - **Narrow suppression only**: when falling back for backward-compatibility, suppress only specific, expected cases (e.g. `keyNotFound`), and **throw** for malformed payloads. - **Check system/OS calls**: capture return codes (and log warnings if you intentionally continue). Never ignore calls that can fail silently (e.g. `setsockopt`, iptables/systemctl). - **Preserve failure signals in tests/cleanup**: don’t use `try?` where failures would corrupt subsequent state; either throw or record the failure. - **Validate preconditions early**: ensure required inputs exist (e.g. “exactly one DNS question”, container running) and return an appropriate error/response variant. - **Avoid crashing on bad inputs**: replace `fatalError` with typed thrown errors. Example (narrow fallback instead of swallowing decode errors): ```swift do { ipv4Address = try container.decode(CIDRv4.self, forKey: .ipv4Address) } catch let DecodingError.keyNotFound(key, _) where key.stringValue == CodingKeys.address.stringValue { ipv4Address = try container.decode(CIDRv4.self, forKey: .address) // expected legacy fallback } // any other decoding error should be rethrown ``` Example (check OS call result / log warnings): ```swift var sendSize = Self.socketSendBufferSize let rc = setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sendSize, socklen_t(MemoryLayout.size)) if rc != 0 { log.warning("setsockopt(SO_SNDBUF) failed with errno \(errno)") // either throw or continue deliberately after logging } ``` This standard reduces silent misconfiguration, hidden data corruption, and hard-to-debug runtime/test failures while still allowing intentional, documented recovery paths.