@gryt/core 0.1.0

package/README.md

@@ -0,0 +1,63 @@
+# @gryt/core
+
+Gryt's shared application logic. One implementation of the things the desktop
+app and the phone app both do.
+
+## What goes in here
+
+Two questions, and a module needs both answers to be yes:
+
+- Would it compile with no DOM and no React Native?
+- Would both apps otherwise need a copy?
+
+`tsconfig.json` enforces the first by leaving `DOM` out of `lib`, so `document`,
+`window` and `localStorage` fail to typecheck rather than working in one app and
+breaking in the other. The second is a judgement call, and the surface guard is
+what stops it drifting: `scripts/check-public-surface.mjs` fails when an export
+is missing *and* when one appears that nobody listed.
+
+## What does not
+
+Anything that needs a platform. Fetching is the clearest case — the desktop and
+the phone reach a server differently, and neither way belongs here — so this
+package decides what a link preview *means* and each app fetches it.
+
+Artwork is the other one. A brand logo is a React component on the desktop and
+an SVG path or a favicon on the phone. This package owns the hostnames, the
+colours and the path rules; each app maps a provider `id` to its own icon.
+
+`@gryt/voice` splits web and native behind two entry points because a media
+engine genuinely cannot avoid the platform. This package has no such entry on
+purpose. If one becomes necessary, that is a decision to make out loud.
+
+## What is in it now
+
+**Reports** — the shape `Gryt-chat/reports` takes, and how an app fills it in.
+The two apps had a copy each with the same seven exports and different insides.
+Most of the difference was real: a renderer knows its Chrome build, a phone
+knows whether it is a simulator. Some of it was not — `MESSAGE_MAX` was 8000 on
+the desktop and 4000 on the phone, and nothing recorded a reason.
+
+**Links** — which site a URL belongs to, what colour to draw its card, what line
+to read out of its path, and which of four shapes a preview earns. 73 sites,
+seven of which also read a detail from the path.
+
+## Working on it
+
+```bash
+npm install
+npm test          # node --test, no runner to install
+npm run typecheck
+npm run build     # tsc to dist/, then rewrite specifiers for ESM
+npm run check-surface
+```
+
+`prepublishOnly` runs the build, the tests and the surface guard, so a release
+that would have shipped a missing export fails before it leaves.
+
+## Versioning
+
+Both apps pin this package. They should pin the *same* version: the point is one
+implementation, and two apps on different versions is two implementations with
+extra steps. `@gryt/voice` is currently `0.4.4` in the client and `^0.3.2` in
+mobile, which is the failure this is meant to avoid.

package/dist/index.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Gryt's shared application logic.
+ *
+ * One implementation of the things the desktop and the phone both do. What
+ * lives here is decided by a single test: could this file compile with no DOM
+ * and no React Native, and would the two apps otherwise both need a copy? If
+ * either answer is no, it belongs in the app.
+ *
+ * The barrel is flat and named rather than a set of subpath exports, because
+ * the two apps import from it by name and `check-public-surface.mjs` asserts
+ * that everything they import is actually here. Subpaths would need that guard
+ * per path for no gain at this size.
+ */
+export { buildReport, describeAttached, MESSAGE_MAX, TITLE_MAX, type Diagnostics, type Report, type ReportType, } from "./reports/report.js";
+export { getAccentColor, getLinkProvider, getProviderDetail, hostnameOf, LINK_PROVIDERS, type LinkProvider, } from "./links/providers.js";
+export { describePreviewFailure, extractUrls, getCardSubtitle, getLinkCardLayout, type LinkCardLayout, type LinkPreviewData, } from "./links/preview.js";

package/dist/index.js

@@ -0,0 +1,16 @@
+/**
+ * Gryt's shared application logic.
+ *
+ * One implementation of the things the desktop and the phone both do. What
+ * lives here is decided by a single test: could this file compile with no DOM
+ * and no React Native, and would the two apps otherwise both need a copy? If
+ * either answer is no, it belongs in the app.
+ *
+ * The barrel is flat and named rather than a set of subpath exports, because
+ * the two apps import from it by name and `check-public-surface.mjs` asserts
+ * that everything they import is actually here. Subpaths would need that guard
+ * per path for no gain at this size.
+ */
+export { buildReport, describeAttached, MESSAGE_MAX, TITLE_MAX, } from "./reports/report.js";
+export { getAccentColor, getLinkProvider, getProviderDetail, hostnameOf, LINK_PROVIDERS, } from "./links/providers.js";
+export { describePreviewFailure, extractUrls, getCardSubtitle, getLinkCardLayout, } from "./links/preview.js";

package/dist/links/preview.d.ts

@@ -0,0 +1,62 @@
+/**
+ * What the server found out about a link, and what a card should do with it.
+ *
+ * The data comes from `GET /api/link-preview`; the shape is the server's, and
+ * the decisions below are the ones both apps were making separately. Fetching
+ * is not here — the desktop and the phone reach a server differently and neither
+ * way belongs in a package with no platform.
+ */
+export interface LinkPreviewData {
+    url: string;
+    title: string | null;
+    description: string | null;
+    image: string | null;
+    imageWidth: number | null;
+    imageHeight: number | null;
+    siteName: string | null;
+    favicon: string | null;
+    imageAlt?: string | null;
+    /** The colour the page declares for itself, used when we know no brand. */
+    themeColor?: string | null;
+    /** `og:type`: "article", "video.other", "music.song". */
+    type?: string | null;
+    author?: string | null;
+    publishedAt?: string | null;
+    /** Present when the page advertises a real player. */
+    oembedUrl?: string | null;
+    /** What the page answered with, so a 404 can say so rather than guess. */
+    status?: number | null;
+}
+/**
+ * The links in a message, minus the ones that are already something else.
+ *
+ * Code spans and fences are stripped first: a URL inside backticks is being
+ * quoted rather than shared, and a card under it would be noise. Markdown
+ * images are stripped because the picture is already drawn.
+ */
+export declare function extractUrls(text: string | null): string[];
+/**
+ * How much of a card a preview can fill.
+ *
+ * A wide image wants to sit under the text at full width. A small or square one
+ * wants to be a thumbnail beside it. No image wants no space set aside for a
+ * picture at all. Drawing all three the same way is what produced a hostname
+ * next to an empty grey rectangle.
+ */
+export type LinkCardLayout = "large" | "thumbnail" | "text" | "bare";
+export declare function getLinkCardLayout(data: LinkPreviewData): LinkCardLayout;
+/**
+ * Why a page gave us nothing, in words worth showing somebody.
+ *
+ * Only for statuses that mean something to a reader. A 500 is the site's
+ * problem and saying so helps nobody, so it returns null and the card falls
+ * back to showing the link on its own.
+ */
+export declare function describePreviewFailure(status: number | null | undefined): string | null;
+/**
+ * The line under the title, where the path says something the title does not.
+ *
+ * Wikipedia titles its WebRTC page "WebRTC" and the detail read out of
+ * `/wiki/WebRTC` is "WebRTC", so showing both printed the word twice.
+ */
+export declare function getCardSubtitle(title: string | null, detail: string | null): string | null;

package/dist/links/preview.js

@@ -0,0 +1,81 @@
+/**
+ * What the server found out about a link, and what a card should do with it.
+ *
+ * The data comes from `GET /api/link-preview`; the shape is the server's, and
+ * the decisions below are the ones both apps were making separately. Fetching
+ * is not here — the desktop and the phone reach a server differently and neither
+ * way belongs in a package with no platform.
+ */
+const URL_REGEX = /https?:\/\/[^\s<>[\](){}'"`,]+[^\s<>[\](){}'"`,.:;!?)]/gi;
+/**
+ * The links in a message, minus the ones that are already something else.
+ *
+ * Code spans and fences are stripped first: a URL inside backticks is being
+ * quoted rather than shared, and a card under it would be noise. Markdown
+ * images are stripped because the picture is already drawn.
+ */
+export function extractUrls(text) {
+    if (!text)
+        return [];
+    let cleaned = text.replace(/```[\s\S]*?```/g, "");
+    cleaned = cleaned.replace(/`[^`]+`/g, "");
+    cleaned = cleaned.replace(/!\[[^\]]*\]\([^)]+\)/g, "");
+    const matches = cleaned.match(URL_REGEX);
+    if (!matches)
+        return [];
+    return [...new Set(matches)];
+}
+/** Wide enough, and landscape enough, to be a header rather than a thumbnail. */
+const LARGE_IMAGE_MIN_WIDTH = 400;
+const LARGE_IMAGE_MIN_ASPECT = 1.2;
+export function getLinkCardLayout(data) {
+    const hasText = Boolean(data.title || data.description);
+    if (!data.image)
+        return hasText ? "text" : "bare";
+    const w = data.imageWidth;
+    const h = data.imageHeight;
+    /* Unknown dimensions count as large. A site that sets og:image and says
+       nothing about its size has almost always set a share card, and the ones
+       that have not lose less by being drawn big than a real share card loses by
+       being shrunk into a corner. */
+    if (!w || !h)
+        return "large";
+    if (w >= LARGE_IMAGE_MIN_WIDTH && w / h >= LARGE_IMAGE_MIN_ASPECT)
+        return "large";
+    return hasText ? "thumbnail" : "large";
+}
+/**
+ * Why a page gave us nothing, in words worth showing somebody.
+ *
+ * Only for statuses that mean something to a reader. A 500 is the site's
+ * problem and saying so helps nobody, so it returns null and the card falls
+ * back to showing the link on its own.
+ */
+export function describePreviewFailure(status) {
+    if (status == null)
+        return null;
+    if (status === 401)
+        return "Sign-in only";
+    /* Not "private": a 403 is as often a site refusing our fetcher as it is a
+       page somebody is not allowed to see. Stack Overflow answers 403 to the
+       preview fetch and 200 to a browser. A private GitHub repository answers
+       404 and is covered below. */
+    if (status === 403)
+        return "The site would not let us look";
+    if (status === 404 || status === 410)
+        return "Page not found";
+    if (status === 429)
+        return "The site is rate limiting us";
+    return null;
+}
+/**
+ * The line under the title, where the path says something the title does not.
+ *
+ * Wikipedia titles its WebRTC page "WebRTC" and the detail read out of
+ * `/wiki/WebRTC` is "WebRTC", so showing both printed the word twice.
+ */
+export function getCardSubtitle(title, detail) {
+    if (!detail || !title)
+        return null;
+    return title.toLowerCase().includes(detail.toLowerCase()) ? null : detail;
+}

package/dist/links/providers.d.ts

@@ -0,0 +1,43 @@
+/**
+ * Which site a link belongs to, worked out from the URL alone.
+ *
+ * Knowing means a card can carry a site's identity before its metadata
+ * arrives, and can still carry it when the metadata never does — a private
+ * repository, a page behind Cloudflare, a site that blocks whatever fetches
+ * previews. Those otherwise draw a bare hostname.
+ *
+ * This is deliberately not the list of sites that "work". Any site with
+ * OpenGraph tags gets a full card without appearing here, and one that declares
+ * a `theme-color` gets a coloured one. Being on this list buys a real logo
+ * instead of a favicon, a brand colour that does not depend on the site
+ * declaring one, and for a few of them a line read straight out of the path.
+ *
+ * The logos are not here. A brand icon is a React component on the desktop and
+ * an SVG path or a favicon on the phone, and neither belongs in a package that
+ * must compile without a DOM. Each app maps `id` to its own artwork; this file
+ * owns the hostnames, the colours and the path rules, which is the part that
+ * was drifting.
+ */
+export interface LinkProvider {
+    id: string;
+    label: string;
+    brand: string;
+    /** For brands that are essentially black and vanish on a dark card. */
+    brandDark?: string;
+    hosts: string[];
+    hostSuffixes?: string[];
+    detail?: (url: URL) => string | null;
+}
+/** Seventy-three sites, and the seven that also read a line out of the path. */
+export declare const LINK_PROVIDERS: LinkProvider[];
+/** The hostname without `www.`, or the input when that was not a URL. */
+export declare function hostnameOf(url: string): string;
+export declare function getLinkProvider(url: string): LinkProvider | null;
+/** The line a provider can read out of the path, if it knows how. */
+export declare function getProviderDetail(url: string): string | null;
+/**
+ * The accent a card is drawn with: the brand where we know it, the colour the
+ * page declared where we do not, and null for the app's own accent when
+ * neither is on offer.
+ */
+export declare function getAccentColor(url: string, themeColor: string | null | undefined, appearance: "light" | "dark"): string | null;

package/dist/links/providers.js

@@ -0,0 +1,212 @@
+/**
+ * Which site a link belongs to, worked out from the URL alone.
+ *
+ * Knowing means a card can carry a site's identity before its metadata
+ * arrives, and can still carry it when the metadata never does — a private
+ * repository, a page behind Cloudflare, a site that blocks whatever fetches
+ * previews. Those otherwise draw a bare hostname.
+ *
+ * This is deliberately not the list of sites that "work". Any site with
+ * OpenGraph tags gets a full card without appearing here, and one that declares
+ * a `theme-color` gets a coloured one. Being on this list buys a real logo
+ * instead of a favicon, a brand colour that does not depend on the site
+ * declaring one, and for a few of them a line read straight out of the path.
+ *
+ * The logos are not here. A brand icon is a React component on the desktop and
+ * an SVG path or a favicon on the phone, and neither belongs in a package that
+ * must compile without a DOM. Each app maps `id` to its own artwork; this file
+ * owns the hostnames, the colours and the path rules, which is the part that
+ * was drifting.
+ */
+const seg = (url) => url.pathname.split("/").filter(Boolean);
+function githubDetail(url) {
+    const p = seg(url);
+    if (p.length === 0)
+        return null;
+    if (p.length === 1)
+        return `@${p[0]}`;
+    const repo = `${p[0]}/${p[1]}`;
+    const kind = p[2];
+    const number = p[3];
+    if (kind === "pull" && number)
+        return `${repo} · pull request #${number}`;
+    if (kind === "issues" && number)
+        return `${repo} · issue #${number}`;
+    if (kind === "releases")
+        return `${repo} · releases`;
+    if (kind === "commit" && number)
+        return `${repo} · commit ${number.slice(0, 7)}`;
+    return repo;
+}
+function modrinthDetail(url) {
+    const p = seg(url);
+    const kinds = {
+        mod: "Mod", plugin: "Plugin", datapack: "Data pack",
+        shader: "Shader", resourcepack: "Resource pack", modpack: "Modpack",
+    };
+    if (p.length >= 2 && kinds[p[0]])
+        return `${kinds[p[0]]} · ${p[1]}`;
+    if (p[0] === "user" && p[1])
+        return `@${p[1]}`;
+    return null;
+}
+function redditDetail(url) {
+    const p = seg(url);
+    if (p[0] === "r" && p[1])
+        return p[2] === "comments" ? `r/${p[1]} · post` : `r/${p[1]}`;
+    if ((p[0] === "u" || p[0] === "user") && p[1])
+        return `u/${p[1]}`;
+    return null;
+}
+function npmDetail(url) {
+    const p = seg(url);
+    if (p[0] !== "package" || !p[1])
+        return null;
+    return p[1].startsWith("@") && p[2] ? `${p[1]}/${p[2]}` : p[1];
+}
+function wikipediaDetail(url) {
+    const p = seg(url);
+    const i = p.indexOf("wiki");
+    const article = i !== -1 ? p[i + 1] : undefined;
+    if (!article)
+        return null;
+    try {
+        return decodeURIComponent(article).replace(/_/g, " ");
+    }
+    catch {
+        return article.replace(/_/g, " ");
+    }
+}
+function steamDetail(url) {
+    const p = seg(url);
+    if (p[0] === "app" && p[2])
+        return decodeURIComponent(p[2]).replace(/_/g, " ");
+    if (p[0] === "app" && p[1])
+        return `App ${p[1]}`;
+    return null;
+}
+/** Seventy-three sites, and the seven that also read a line out of the path. */
+export const LINK_PROVIDERS = [
+    { id: "github", label: "GitHub", brand: "#181717", brandDark: "#8B949E", hosts: ["github.com", "gist.github.com"], detail: githubDetail },
+    { id: "gitlab", label: "GitLab", brand: "#FC6D26", hosts: ["gitlab.com"] },
+    { id: "npm", label: "npm", brand: "#CB3837", hosts: ["npmjs.com"], detail: npmDetail },
+    { id: "pypi", label: "PyPI", brand: "#3775A9", hosts: ["pypi.org"] },
+    { id: "crates", label: "crates.io", brand: "#B7410E", hosts: ["crates.io", "docs.rs"] },
+    { id: "docker", label: "Docker Hub", brand: "#2496ED", hosts: ["hub.docker.com"] },
+    { id: "stackoverflow", label: "Stack Overflow", brand: "#F58025", hosts: ["stackoverflow.com", "superuser.com", "serverfault.com"], hostSuffixes: [".stackexchange.com"] },
+    { id: "codepen", label: "CodePen", brand: "#111111", brandDark: "#C9CDD3", hosts: ["codepen.io"] },
+    { id: "codesandbox", label: "CodeSandbox", brand: "#151515", brandDark: "#B9BEC5", hosts: ["codesandbox.io"] },
+    { id: "stackblitz", label: "StackBlitz", brand: "#1269D3", hosts: ["stackblitz.com"] },
+    { id: "replit", label: "Replit", brand: "#F26207", hosts: ["replit.com"] },
+    { id: "vercel", label: "Vercel", brand: "#000000", brandDark: "#B4B4B4", hosts: ["vercel.com"] },
+    { id: "netlify", label: "Netlify", brand: "#00C7B7", hosts: ["netlify.com", "netlify.app"] },
+    { id: "cloudflare", label: "Cloudflare", brand: "#F38020", hosts: ["cloudflare.com"] },
+    { id: "modrinth", label: "Modrinth", brand: "#00AF5C", hosts: ["modrinth.com"], detail: modrinthDetail },
+    { id: "curseforge", label: "CurseForge", brand: "#F16436", hosts: ["curseforge.com"] },
+    { id: "nexusmods", label: "Nexus Mods", brand: "#D98F40", hosts: ["nexusmods.com"] },
+    { id: "steam", label: "Steam", brand: "#1B2838", brandDark: "#8BA5C4", hosts: ["store.steampowered.com", "steamcommunity.com"], detail: steamDetail },
+    { id: "itch", label: "itch.io", brand: "#FA5C5C", hosts: ["itch.io"], hostSuffixes: [".itch.io"] },
+    { id: "gog", label: "GOG", brand: "#86328A", hosts: ["gog.com"] },
+    { id: "epic", label: "Epic Games", brand: "#2A2A2A", brandDark: "#B0B0B0", hosts: ["store.epicgames.com", "epicgames.com"] },
+    { id: "playstation", label: "PlayStation", brand: "#0070D1", hosts: ["playstation.com", "store.playstation.com"] },
+    { id: "nintendo", label: "Nintendo", brand: "#E60012", hosts: ["nintendo.com"] },
+    { id: "youtube", label: "YouTube", brand: "#FF0000", hosts: ["youtube.com", "youtu.be", "m.youtube.com", "music.youtube.com"] },
+    { id: "twitch", label: "Twitch", brand: "#9146FF", hosts: ["twitch.tv", "clips.twitch.tv"] },
+    { id: "vimeo", label: "Vimeo", brand: "#1AB7EA", hosts: ["vimeo.com", "player.vimeo.com"] },
+    { id: "kick", label: "Kick", brand: "#0FA33C", hosts: ["kick.com"] },
+    { id: "rumble", label: "Rumble", brand: "#85C742", hosts: ["rumble.com"] },
+    { id: "spotify", label: "Spotify", brand: "#1DB954", hosts: ["open.spotify.com"] },
+    { id: "applemusic", label: "Apple Music", brand: "#FA243C", hosts: ["music.apple.com"] },
+    { id: "soundcloud", label: "SoundCloud", brand: "#FF5500", hosts: ["soundcloud.com", "on.soundcloud.com"] },
+    { id: "bandcamp", label: "Bandcamp", brand: "#408294", hosts: ["bandcamp.com"], hostSuffixes: [".bandcamp.com"] },
+    { id: "tidal", label: "Tidal", brand: "#1A1A1A", brandDark: "#B0B0B0", hosts: ["tidal.com"] },
+    { id: "mixcloud", label: "Mixcloud", brand: "#5000FF", hosts: ["mixcloud.com"] },
+    { id: "lastfm", label: "Last.fm", brand: "#D51007", hosts: ["last.fm"] },
+    { id: "x", label: "X", brand: "#0F1419", brandDark: "#C4CDD5", hosts: ["x.com", "twitter.com"] },
+    { id: "bluesky", label: "Bluesky", brand: "#0285FF", hosts: ["bsky.app"] },
+    { id: "mastodon", label: "Mastodon", brand: "#6364FF", hosts: ["mastodon.social", "mastodon.online", "fosstodon.org"] },
+    { id: "reddit", label: "Reddit", brand: "#FF4500", hosts: ["reddit.com", "old.reddit.com", "redd.it"], detail: redditDetail },
+    { id: "facebook", label: "Facebook", brand: "#0866FF", hosts: ["facebook.com", "fb.com", "fb.watch", "m.facebook.com"] },
+    { id: "instagram", label: "Instagram", brand: "#E4405F", hosts: ["instagram.com"] },
+    { id: "threads", label: "Threads", brand: "#101010", brandDark: "#BFBFBF", hosts: ["threads.net", "threads.com"] },
+    { id: "tiktok", label: "TikTok", brand: "#EE1D52", hosts: ["tiktok.com", "vm.tiktok.com", "vt.tiktok.com"] },
+    { id: "linkedin", label: "LinkedIn", brand: "#0A66C2", hosts: ["linkedin.com"] },
+    { id: "pinterest", label: "Pinterest", brand: "#BD081C", hosts: ["pinterest.com"], hostSuffixes: [".pinterest.com"] },
+    { id: "snapchat", label: "Snapchat", brand: "#B8B400", hosts: ["snapchat.com"] },
+    { id: "telegram", label: "Telegram", brand: "#26A5E4", hosts: ["t.me", "telegram.me"] },
+    { id: "discord", label: "Discord", brand: "#5865F2", hosts: ["discord.com", "discord.gg", "discordapp.com"] },
+    { id: "wikipedia", label: "Wikipedia", brand: "#3366CC", hosts: ["wikipedia.org"], hostSuffixes: [".wikipedia.org"], detail: wikipediaDetail },
+    { id: "mdn", label: "MDN Web Docs", brand: "#1B76C4", hosts: ["developer.mozilla.org"] },
+    { id: "hackernews", label: "Hacker News", brand: "#FF6600", hosts: ["news.ycombinator.com"] },
+    { id: "medium", label: "Medium", brand: "#1A1A1A", brandDark: "#C0C0C0", hosts: ["medium.com"], hostSuffixes: [".medium.com"] },
+    { id: "devto", label: "DEV", brand: "#3B49DF", hosts: ["dev.to"] },
+    { id: "arxiv", label: "arXiv", brand: "#B31B1B", hosts: ["arxiv.org"] },
+    { id: "archive", label: "Internet Archive", brand: "#4A4A4A", brandDark: "#B5B5B5", hosts: ["archive.org", "web.archive.org"] },
+    { id: "wolfram", label: "Wolfram Alpha", brand: "#DD1100", hosts: ["wolframalpha.com"] },
+    { id: "goodreads", label: "Goodreads", brand: "#75633F", hosts: ["goodreads.com"] },
+    { id: "imdb", label: "IMDb", brand: "#C9A227", hosts: ["imdb.com"] },
+    { id: "letterboxd", label: "Letterboxd", brand: "#00B020", hosts: ["letterboxd.com"] },
+    { id: "imgur", label: "Imgur", brand: "#1BB76E", hosts: ["imgur.com", "i.imgur.com"] },
+    { id: "giphy", label: "Giphy", brand: "#FF6666", hosts: ["giphy.com"] },
+    { id: "flickr", label: "Flickr", brand: "#0063DC", hosts: ["flickr.com"] },
+    { id: "figma", label: "Figma", brand: "#F24E1E", hosts: ["figma.com"] },
+    { id: "dribbble", label: "Dribbble", brand: "#EA4C89", hosts: ["dribbble.com"] },
+    { id: "behance", label: "Behance", brand: "#1769FF", hosts: ["behance.net"] },
+    { id: "notion", label: "Notion", brand: "#2F2F2F", brandDark: "#BFBFBF", hosts: ["notion.so", "notion.site"], hostSuffixes: [".notion.site"] },
+    { id: "linear", label: "Linear", brand: "#5E6AD2", hosts: ["linear.app"] },
+    { id: "trello", label: "Trello", brand: "#0052CC", hosts: ["trello.com"] },
+    { id: "jira", label: "Jira", brand: "#0052CC", hosts: ["atlassian.net"], hostSuffixes: [".atlassian.net"] },
+    { id: "obsidian", label: "Obsidian", brand: "#7C3AED", hosts: ["obsidian.md"] },
+    { id: "googledrive", label: "Google Drive", brand: "#1A73E8", hosts: ["drive.google.com", "docs.google.com"] },
+    { id: "googlemaps", label: "Google Maps", brand: "#34A853", hosts: ["maps.google.com", "maps.app.goo.gl"] },
+    { id: "openstreetmap", label: "OpenStreetMap", brand: "#5A8B3E", hosts: ["openstreetmap.org"] },
+];
+const BY_HOST = new Map();
+for (const provider of LINK_PROVIDERS) {
+    for (const host of provider.hosts)
+        BY_HOST.set(host, provider);
+}
+/** The hostname without `www.`, or the input when that was not a URL. */
+export function hostnameOf(url) {
+    try {
+        return new URL(url).hostname.replace(/^www\./, "");
+    }
+    catch {
+        return url;
+    }
+}
+export function getLinkProvider(url) {
+    const host = hostnameOf(url).toLowerCase();
+    const exact = BY_HOST.get(host);
+    if (exact)
+        return exact;
+    for (const provider of LINK_PROVIDERS) {
+        if (provider.hostSuffixes?.some((suffix) => host.endsWith(suffix)))
+            return provider;
+    }
+    return null;
+}
+/** The line a provider can read out of the path, if it knows how. */
+export function getProviderDetail(url) {
+    const provider = getLinkProvider(url);
+    if (!provider?.detail)
+        return null;
+    try {
+        return provider.detail(new URL(url));
+    }
+    catch {
+        return null;
+    }
+}
+/**
+ * The accent a card is drawn with: the brand where we know it, the colour the
+ * page declared where we do not, and null for the app's own accent when
+ * neither is on offer.
+ */
+export function getAccentColor(url, themeColor, appearance) {
+    const provider = getLinkProvider(url);
+    if (provider) {
+        return appearance === "dark" ? (provider.brandDark ?? provider.brand) : provider.brand;
+    }
+    return themeColor ?? null;
+}

package/dist/reports/report.d.ts

@@ -0,0 +1,153 @@
+/**
+ * The shape `Gryt-chat/reports` takes, and how an app fills it in.
+ *
+ * `POST /v1/reports`. Only `type` and `message` are required — everything else
+ * is diagnostics, and the service is explicit that a field an app gets wrong is
+ * truncated or dropped rather than a reason to reject: "a report lost to a
+ * validation error is a bug nobody hears about."
+ *
+ * That cuts both ways, and it is why nothing here throws and nothing here is
+ * required to succeed. A diagnostic an app cannot work out is left off rather
+ * than sent as a guess, because a wrong Electron version in a bug report is
+ * worse than no Electron version.
+ *
+ * ## Why this is one file and not two
+ *
+ * The desktop and the phone had a copy each, with the same seven exports and
+ * different insides. Most of the difference was real: a renderer knows its
+ * Chrome build, a phone knows whether it is an emulator, and neither should
+ * send the other's fields empty. Some of it was not. `MESSAGE_MAX` was 8000 on
+ * the desktop and 4000 on the phone, and nothing recorded a reason — the two
+ * files even carried the same sentence about not cutting people off at a tweet.
+ *
+ * So `Diagnostics` below is the union of both, every field optional, and each
+ * app fills what it can actually find out. A field neither app fills is dead
+ * and can be seen to be dead, which is the other half of the point.
+ */
+export type ReportType = "bug" | "feedback";
+/** The fields an app can fill. The service accepts more. */
+export interface Report {
+    type: ReportType;
+    message: string;
+    title?: string;
+    app?: {
+        version?: string;
+        /** The build number, where the platform has one distinct from the version. */
+        build?: string;
+        channel?: string;
+        /** Random per install, not per person. What rate limits are counted against. */
+        installId?: string;
+        locale?: string;
+    };
+    device?: {
+        platform?: string;
+        osVersion?: string;
+        /** A crash only simulators produce is worth telling apart from a real one. */
+        isEmulator?: boolean;
+        screen?: {
+            width: number;
+            height: number;
+            scale: number;
+        };
+        timezone?: string;
+    };
+    runtime?: {
+        engine?: string;
+        chromeVersion?: string;
+        electronVersion?: string;
+        reactNativeVersion?: string;
+        expoVersion?: string;
+        userAgent?: string;
+    };
+    context?: {
+        route?: string;
+        serverVersion?: string;
+        connected?: boolean;
+        voiceActive?: boolean;
+        online?: boolean;
+        networkType?: string;
+        /** "It broke twenty minutes in" and "it broke on launch" are different bugs. */
+        sessionUptimeSec?: number;
+    };
+    /** The tail of the app's own log, where it keeps one. */
+    logs?: string[];
+    extra?: Record<string, unknown>;
+}
+/**
+ * What an app knows about itself when somebody opens the form.
+ *
+ * Passed in rather than read here, so the assembly below stays readable and
+ * testable: every one of these comes from somewhere that needs a running app —
+ * the Electron bridge, `expo-constants`, `Platform`, the socket layer.
+ */
+export interface Diagnostics {
+    version?: string | null;
+    build?: string | null;
+    channel?: string | null;
+    installId?: string | null;
+    locale?: string | null;
+    platform?: string | null;
+    /** A number on some platforms: Android's API level arrives as one. */
+    osVersion?: string | number | null;
+    isEmulator?: boolean | null;
+    screen?: {
+        width: number;
+        height: number;
+        scale: number;
+    } | null;
+    timezone?: string | null;
+    engine?: string | null;
+    chromeVersion?: string | null;
+    electronVersion?: string | null;
+    reactNativeVersion?: string | null;
+    expoVersion?: string | null;
+    userAgent?: string | null;
+    route?: string | null;
+    serverVersion?: string | null;
+    connected?: boolean | null;
+    voiceActive?: boolean | null;
+    online?: boolean | null;
+    networkType?: string | null;
+    sessionUptimeSec?: number | null;
+    /** Whether this app is running the server it is connected to. */
+    embeddedServer?: boolean | null;
+    /** The version of that embedded server, which is not the app's. */
+    embeddedServerVersion?: string | null;
+    logs?: string[];
+}
+/**
+ * The service truncates, but an app should not send a novel either.
+ *
+ * Generous rather than tight: somebody describing a bug properly is the good
+ * case, and cutting them off at a tweet is how you get "it broke" instead. The
+ * desktop's 8000 wins over the phone's 4000 on exactly that argument, which
+ * both files made and only one of them followed.
+ */
+export declare const MESSAGE_MAX = 8000;
+export declare const TITLE_MAX = 120;
+/**
+ * Assemble what gets sent.
+ *
+ * `type` and `message` always; every diagnostic only if it is actually known.
+ * An empty `device` object says "this app does not collect device
+ * information", which is a different and wronger claim than leaving it off.
+ */
+export declare function buildReport(type: ReportType, input: {
+    message: string;
+    title?: string;
+}, diagnostics?: Diagnostics): Report;
+/**
+ * The same diagnostics, as lines to show somebody before they send.
+ *
+ * A form that quietly ships a route, a server version and a log tail is worse
+ * than one that says so — and this is the list that makes "what is attached"
+ * answerable without reading the source. It is built from the report rather
+ * than from the inputs, so it cannot drift from what actually goes.
+ *
+ * Every row is conditional, so an app that fills half the fields shows half the
+ * rows rather than a column of blanks.
+ */
+export declare function describeAttached(report: Report): {
+    label: string;
+    value: string;
+}[];