import * as nip19 from "nostr-tools/nip19"; export const NEWLINE = "newline"; export const TEXT = "text"; export const TOPIC = "topic"; export const LINK = "link"; export const INVOICE = "invoice"; export const NOSTR_NOTE = "nostr:note"; export const NOSTR_NEVENT = "nostr:nevent"; export const NOSTR_NPUB = "nostr:npub"; export const NOSTR_NPROFILE = "nostr:nprofile"; export const NOSTR_NADDR = "nostr:naddr"; const first = (list: any) => (list ? list[0] : undefined); export const fromNostrURI = (s: string) => s.replace(/^[\w+]+:\/?\/?/, ""); export const urlIsMedia = (url: string) => !url.match(/\.(apk|docx|xlsx|csv|dmg)/) && url.split("://").pop()?.includes("/"); export type ParsedPart = { type: string; value: any; }; export const isImage = (url: string) => url?.match(/^.*\.(jpg|jpeg|png|webp|gif|avif|svg)/gi); export const isVideo = (url: string) => url?.match(/^.*\.(mov|mkv|mp4|avi|m4v|webm)/gi); export const isAudio = (url: string) => url?.match(/^.*\.(ogg|mp3|wav)/gi); type ContentArgs = { content: string; tags?: Array<[string, string, string]>; }; export const parseContent = ({ content, tags = [] }: ContentArgs): ParsedPart[] => { const result: ParsedPart[] = []; let text = content.trim(); let buffer = ""; const parseNewline = () => { const newline = first(text.match(/^\n+/)); if (newline) { return [NEWLINE, newline, newline]; } }; const parseMention = () => { const mentionMatch = text.match(/^#\[(\d+)\]/i); if (mentionMatch) { const i = parseInt(mentionMatch[1]); if (tags[i]) { const [tag, value, url] = tags[i]; const relays = url ? [url] : []; let type, data, entity; try { if (tag === "p") { type = "nprofile"; data = { pubkey: value, relays }; entity = nip19.nprofileEncode(data); } else { type = "nevent"; data = { id: value, relays, pubkey: null }; entity = nip19.neventEncode(data); } } catch { /**/ } return [`nostr:${type}`, mentionMatch[0], { ...data, entity }]; } } }; const parseTopic = () => { const topic = first(text.match(/^#\w+/i)); if (topic && !topic.match(/^#\d+$/)) { return [TOPIC, topic, topic.slice(1)]; } }; const parseBech32 = () => { const bech32 = first( text.match(/^(web\+)?(nostr:)?\/?\/?n(event|ote|profile|pub|addr)1[\d\w]+/i), ); if (bech32) { try { const entity = fromNostrURI(bech32); const { type, data } = nip19.decode(entity) as { type: string; data: object; }; let value = data; if (type === "note") { value = { id: data }; } else if (type === "npub") { value = { pubkey: data }; } return [`nostr:${type}`, bech32, { ...value, entity }]; } catch (e) { console.log(e); } } }; const parseUrl = () => { const raw = first(text.match(/^([a-z+:]{2,30}:\/\/)?[^\s]+\.[a-z]{2,6}[^\s]*[^.!?,:\s]/gi)); if (raw) { const prev = result[result.length - 1]; if (prev?.type === "text" && prev.value.endsWith("/")) { return; } let url = raw; if (url.match(/\.\./)) { return; } if (!url.match("://")) { url = "https://" + url; } return [LINK, raw, { url, isMedia: urlIsMedia(url) }]; } }; while (text) { const part = parseNewline() || parseMention() || parseTopic() || parseBech32() || parseUrl(); if (part) { if (buffer) { result.push({ type: "text", value: buffer }); buffer = ""; } const [type, raw, value] = part; result.push({ type, value }); text = text.slice(raw.length); } else { const match = first(text.match(/^[\w\d]+ ?/i)) || text[0]; buffer += match; text = text.slice(match.length); } } if (buffer) { result.push({ type: TEXT, value: buffer }); } return result; };