Parse Email Addresses from a Log String

Single-pass character scanner over a long log string with `;`-separated email records. Each record may have an optional display name and comments anywhere; deduplicate and return the list of emails. Requirements Input is a single (potentially very large) string. Records are separated by ;. Each record represents an email address but may contain: A leading display name ("John Doe" <john@example.com>). Parenthesized comments anywhere (john(work)@example.com or (internal) john@example.com). Quoted regions containing literal semicolons that must not be treated as record separators. Return the list of unique email addresses parsed out, in order of first appearance. Notes The problem reduces to a single-character state machine. The state set candidates use: NORMAL — outside quotes and parens, accumulating an email candidate. IN_QUOTES — inside "...", semicolons are literal, \" escapes a quote. IN_PAREN — inside (...), semicolons are literal, parens nest. IN_ANGLE — inside <...>, the angle contents are the email. At each character, look at the current state, advance / change state, and decide whether to append to the current email accumulator or discard. On ; while in NORMAL, finalize the accumulator (extract the @-containing token, strip surrounding whitespace), insert into an ordered set, reset. Regex is a trap — the quoted / parenthesized / angle-bracketed nesting is not a regular language, and any regex you write will mis-handle at least one of the interviewer's adversarial inputs. For deduplication, an OrderedDict (Python) or LinkedHashSet (Java) preserves insertion order while keeping O(1) membership. Common reported failure mode: candidates start with a regex pass, get it 80% right in 20 minutes, then spend the rest of the round patching nested-paren edge cases. The state machine is faster to write correctly even though it looks longer. Preparation Write a four-state character scanner on paper; type out the (state, char) → next_state transition table before opening the editor. Pre-write the "extract @-containing token from a <...>-or-bareword string" helper. Drill on adversarial inputs: "John;Doe" <a@b>; (c@d); e@f, nested parens (a (b) c), escaped quotes inside quoted display names.