| 1 | // Media, Links & Docs — the pure extraction + in-memory per-thread index |
| 2 | // behind the thread's media overlay (mediaview.js). Same posture as the |
| 3 | // search index (search.js): rows decrypt lazily into MEMORY, never to disk — |
| 4 | // the store deliberately holds only ciphertext at rest. decryptRow is |
| 5 | // injected so this module stays free of keys and Node-testable. |
| 6 | // Relative import: Node tests load this file directly. |
| 7 | import { tokenizeText, blockTokens } from "./format.js"; |
| 8 | |
| 9 | // One decrypted payload -> its media/links/docs entries, or null when the |
| 10 | // message contributes nothing. Typed app events and reactions are invisible |
| 11 | // here (same rule as search); tombstones never reach us (index skips them). |
| 12 | export function extractEntries(data, meta) { |
| 13 | if (!data || data.type) return null; |
| 14 | const base = { sid: meta.id, ts: data.ts || meta.ts || 0, uid: data.uid || "", name: data.name || "" }; |
| 15 | const media = []; |
| 16 | const docs = []; |
| 17 | const links = []; |
| 18 | for (const a of data.attachments || []) { |
| 19 | if (!a || !a.id) continue; |
| 20 | const t = a.type || ""; |
| 21 | if (t.startsWith("image/") || t.startsWith("video/")) media.push({ ...base, att: a }); |
| 22 | else docs.push({ ...base, att: a }); |
| 23 | } |
| 24 | // Every link in the prose — the feed previews only the first (firstLink), |
| 25 | // but the Links tab is an inventory. Fenced code is skipped for the same |
| 26 | // reason firstLink skips it: a URL pasted as code isn't a shared link. |
| 27 | if (data.text) { |
| 28 | for (const b of blockTokens(data.text)) { |
| 29 | if (b.type === "code") continue; |
| 30 | for (const tok of tokenizeText(b.text)) if (tok.href) links.push({ ...base, url: tok.href }); |
| 31 | } |
| 32 | } |
| 33 | if (!media.length && !docs.length && !links.length) return null; |
| 34 | return { media, links, docs }; |
| 35 | } |
| 36 | |
| 37 | export function monthLabel(ts) { |
| 38 | return new Date(ts).toLocaleDateString([], { month: "long", year: "numeric" }); |
| 39 | } |
| 40 | |
| 41 | // Group an already-sorted (newest-first) entry list into month sections. |
| 42 | // Consecutive grouping is enough BECAUSE the input is sorted. |
| 43 | export function bucketByMonth(entries) { |
| 44 | const out = []; |
| 45 | for (const e of entries) { |
| 46 | const d = new Date(e.ts || 0); |
| 47 | const key = d.getFullYear() + "-" + String(d.getMonth() + 1).padStart(2, "0"); |
| 48 | if (!out.length || out[out.length - 1].key !== key) out.push({ key, label: monthLabel(e.ts || 0), items: [] }); |
| 49 | out[out.length - 1].items.push(e); |
| 50 | } |
| 51 | return out; |
| 52 | } |
| 53 | |
| 54 | export class ThreadMediaIndex { |
| 55 | constructor(decryptRow) { |
| 56 | this.decryptRow = decryptRow; // async (row, tid) -> payload | null |
| 57 | this.threads = new Map(); // tid -> { seen:Set<sid>, media:[], links:[], docs:[] } |
| 58 | } |
| 59 | |
| 60 | bucket(tid) { |
| 61 | let b = this.threads.get(tid); |
| 62 | if (!b) { |
| 63 | b = { seen: new Set(), media: [], links: [], docs: [] }; |
| 64 | this.threads.set(tid, b); |
| 65 | } |
| 66 | return b; |
| 67 | } |
| 68 | |
| 69 | // Fold rows not yet seen. Idempotent per sid; a failed decrypt still marks |
| 70 | // seen (same as search — a row that won't decrypt now won't decrypt later). |
| 71 | async ensureThread(tid, rows) { |
| 72 | const b = this.bucket(tid); |
| 73 | for (const row of rows || []) { |
| 74 | if (!row || !row.id || b.seen.has(row.id) || row.tombstone_of) continue; |
| 75 | b.seen.add(row.id); |
| 76 | let data = null; |
| 77 | try { |
| 78 | data = await this.decryptRow(row, tid); |
| 79 | } catch {} |
| 80 | if (!data) continue; |
| 81 | const ex = extractEntries(data, { id: row.id, ts: (row.created_at || 0) * 1000 }); |
| 82 | if (!ex) continue; |
| 83 | b.media.push(...ex.media); |
| 84 | b.links.push(...ex.links); |
| 85 | b.docs.push(...ex.docs); |
| 86 | } |
| 87 | for (const k of ["media", "links", "docs"]) b[k].sort((x, y) => y.sid - x.sid); |
| 88 | } |
| 89 | |
| 90 | // Moderation filters at QUERY time so un-muting needs no re-scan. |
| 91 | tab(tid, name, { hiddenUids }) { |
| 92 | return this.bucket(tid)[name].filter((e) => !hiddenUids.has(e.uid)); |
| 93 | } |
| 94 | |
| 95 | counts(tid, opts) { |
| 96 | return { |
| 97 | media: this.tab(tid, "media", opts).length, |
| 98 | links: this.tab(tid, "links", opts).length, |
| 99 | docs: this.tab(tid, "docs", opts).length, |
| 100 | }; |
| 101 | } |
| 102 | |
| 103 | // An edit re-extracts (links can change); a delete just never re-folds. |
| 104 | invalidate(tid, sid) { |
| 105 | const b = this.threads.get(tid); |
| 106 | if (!b) return; |
| 107 | b.seen.delete(sid); |
| 108 | for (const k of ["media", "links", "docs"]) b[k] = b[k].filter((e) => e.sid !== sid); |
| 109 | } |
| 110 | |
| 111 | dropThread(tid) { |
| 112 | this.threads.delete(tid); |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | // Search within a tab: a case-insensitive substring over what each entry |
| 117 | // actually shows — the sender's name (every tab), the URL and its unfurl |
| 118 | // title (links; `titleFor(url)` returns the title if it's been fetched, else |
| 119 | // ""), and the filename (docs, and media that carry one). Empty/blank query = |
| 120 | // everything. Pure, so the three UIs share one definition of "matches". |
| 121 | export function filterEntries(entries, query, titleFor = () => "") { |
| 122 | const q = (query || "").trim().toLowerCase(); |
| 123 | if (!q) return entries; |
| 124 | return entries.filter((e) => { |
| 125 | if ((e.name || "").toLowerCase().includes(q)) return true; |
| 126 | if (e.url) { |
| 127 | if (e.url.toLowerCase().includes(q)) return true; |
| 128 | if ((titleFor(e.url) || "").toLowerCase().includes(q)) return true; |
| 129 | } |
| 130 | if (e.att && (e.att.name || "").toLowerCase().includes(q)) return true; |
| 131 | return false; |
| 132 | }); |
| 133 | } |
| 134 | |
amadan