DoM-Que-Choisir/sw.js

"use strict";

const OFFLINE_DATA_FILE = "offline.js";
const CACHE_NAME_PREFIX = "c2offline";
const BROADCASTCHANNEL_NAME = "offline";
const CONSOLE_PREFIX = "[SW] ";

// Create a BroadcastChannel if supported.
const broadcastChannel = (typeof BroadcastChannel === "undefined" ? null : new BroadcastChannel(BROADCASTCHANNEL_NAME));

//////////////////////////////////////
// Utility methods
function PostBroadcastMessage(o)
{
	if (!broadcastChannel)
		return;		// not supported
	
	// Impose artificial (and arbitrary!) delay of 3 seconds to make sure client is listening by the time the message is sent.
	// Note we could remove the delay on some messages, but then we create a race condition where sometimes messages can arrive
	// in the wrong order (e.g. "update ready" arrives before "started downloading update"). So to keep the consistent ordering,
	// delay all messages by the same amount.
	setTimeout(() => broadcastChannel.postMessage(o), 3000);
};

function Broadcast(type)
{
	PostBroadcastMessage({
		"type": type
	});
};

function BroadcastDownloadingUpdate(version)
{
	PostBroadcastMessage({
		"type": "downloading-update",
		"version": version
	});
}

function BroadcastUpdateReady(version)
{
	PostBroadcastMessage({
		"type": "update-ready",
		"version": version
	});
}

function GetCacheBaseName()
{
	// Include the scope to avoid name collisions with any other SWs on the same origin.
	// e.g. "c2offline-https://example.com/foo/" (won't collide with anything under bar/)
	return CACHE_NAME_PREFIX + "-" + self.registration.scope;
};

function GetCacheVersionName(version)
{
	// Append the version number to the cache name.
	// e.g. "c2offline-https://example.com/foo/-v2"
	return GetCacheBaseName() + "-v" + version;
};

// Return caches.keys() filtered down to just caches we're interested in (with the right base name).
// This filters out caches from unrelated scopes.
function GetAvailableCacheNames()
{
	return caches.keys()
	.then(cacheNames =>
	{
		const cacheBaseName = GetCacheBaseName();
		return cacheNames.filter(n => n.startsWith(cacheBaseName));
	});
};

// Identify if an update is pending, which is the case when we have 2 or more available caches.
// One must be an update that is waiting, since the next navigate that does an upgrade will
// delete all the old caches leaving just one currently-in-use cache.
function IsUpdatePending()
{
	return GetAvailableCacheNames()
	.then(availableCacheNames => availableCacheNames.length >= 2);
};

// Automatically deduce the main page URL (e.g. index.html or main.aspx) from the available browser windows.
// This prevents having to hard-code an index page in the file list, implicitly caching it like AppCache did.
function GetMainPageUrl()
{
	return clients.matchAll({
		includeUncontrolled: true,
		type: "window"
	})
	.then(clients =>
	{
		for (let c of clients)
		{
			// Parse off the scope from the full client URL, e.g. https://example.com/index.html -> index.html
			let url = c.url;
			if (url.startsWith(self.registration.scope))
				url = url.substring(self.registration.scope.length);
			
			if (url && url !== "/")		// ./ is also implicitly cached so don't bother returning that
			{
				// If the URL is solely a search string, prefix it with / to ensure it caches correctly.
				// e.g. https://example.com/?foo=bar needs to cache as /?foo=bar, not just ?foo=bar.
				if (url.startsWith("?"))
					url = "/" + url;
				
				return url;
			}
		}
		
		return "";		// no main page URL could be identified
	});
};

// Hack to fetch optionally bypassing HTTP cache until fetch cache options are supported in Chrome (crbug.com/453190)
function fetchWithBypass(request, bypassCache)
{
	if (typeof request === "string")
		request = new Request(request);
	
	if (bypassCache)
	{
		// bypass enabled: add a random search parameter to avoid getting a stale HTTP cache result
		const url = new URL(request.url);
		url.search += Math.floor(Math.random() * 1000000);

		return fetch(url, {
			headers: request.headers,
			mode: request.mode,
			credentials: request.credentials,
			redirect: request.redirect,
			cache: "no-store"
		});
	}
	else
	{
		// bypass disabled: perform normal fetch which is allowed to return from HTTP cache
		return fetch(request);
	}
};

// Effectively a cache.addAll() that only creates the cache on all requests being successful (as a weak attempt at making it atomic)
// and can optionally cache-bypass with fetchWithBypass in every request
function CreateCacheFromFileList(cacheName, fileList, bypassCache)
{
	// Kick off all requests and wait for them all to complete
	return Promise.all(fileList.map(url => fetchWithBypass(url, bypassCache)))
	.then(responses =>
	{
		// Check if any request failed. If so don't move on to opening the cache.
		// This makes sure we only open a cache if all requests succeeded.
		let allOk = true;
		
		for (let response of responses)
		{
			if (!response.ok)
			{
				allOk = false;
				console.error(CONSOLE_PREFIX + "Error fetching '" + originalUrl + "' (" + response.status + " " + response.statusText + ")");
			}
		}
		
		if (!allOk)
			throw new Error("not all resources were fetched successfully");
		
		// Can now assume all responses are OK. Open a cache and write all responses there.
		// TODO: ideally we can do this transactionally to ensure a complete cache is written as one atomic operation.
		// This needs either new transactional features in the spec, or at the very least a way to rename a cache
		// (so we can write to a temporary name that won't be returned by GetAvailableCacheNames() and then rename it when ready).
		return caches.open(cacheName)
		.then(cache =>
		{
			return Promise.all(responses.map(
				(response, i) => cache.put(fileList[i], response)
			));
		})
		.catch(err =>
		{
			// Not sure why cache.put() would fail (maybe if storage quota exceeded?) but in case it does,
			// clean up the cache to try to avoid leaving behind an incomplete cache.
			console.error(CONSOLE_PREFIX + "Error writing cache entries: ", err);
			caches.delete(cacheName);
			throw err;
		});
	});
};

function UpdateCheck(isFirst)
{
	// Always bypass cache when requesting offline.js to make sure we find out about new versions.
	return fetchWithBypass(OFFLINE_DATA_FILE, true)
	.then(r => r.json())
	.then(data =>
	{
		const version = data.version;
		let fileList = data.fileList;
		const currentCacheName = GetCacheVersionName(version);
		
		return caches.has(currentCacheName)
		.then(cacheExists =>
		{
			// Don't recache if there is already a cache that exists for this version. Assume it is complete.
			if (cacheExists)
			{
				// Log whether we are up-to-date or pending an update.
				return IsUpdatePending()
				.then(isUpdatePending =>
				{
					if (isUpdatePending)
					{
						console.log(CONSOLE_PREFIX + "Update pending");
						Broadcast("update-pending");
					}
					else
					{
						console.log(CONSOLE_PREFIX + "Up to date");
						Broadcast("up-to-date");
					}
				});
			}
			
			// Implicitly add the main page URL to the file list, e.g. "index.html", so we don't have to assume a specific name.
			return GetMainPageUrl()
			.then(mainPageUrl =>
			{
				// Prepend the main page URL to the file list if we found one and it is not already in the list.
				// Also make sure we request the base / which should serve the main page.
				fileList.unshift("./");
				
				if (mainPageUrl && fileList.indexOf(mainPageUrl) === -1)
					fileList.unshift(mainPageUrl);
				
				console.log(CONSOLE_PREFIX + "Caching " + fileList.length + " files for offline use");
				
				if (isFirst)
					Broadcast("downloading");
				else
					BroadcastDownloadingUpdate(version);
				
				// Note we don't bypass the cache on the first update check. This is because SW installation and the following
				// update check caching will race with the normal page load requests. For any normal loading fetches that have already
				// completed or are in-flight, it is pointless and wasteful to cache-bust the request for offline caching, since that
				// forces a second network request to be issued when a response from the browser HTTP cache would be fine.
				return CreateCacheFromFileList(currentCacheName, fileList, !isFirst)
				.then(IsUpdatePending)
				.then(isUpdatePending =>
				{
					if (isUpdatePending)
					{
						console.log(CONSOLE_PREFIX + "All resources saved, update ready");
						BroadcastUpdateReady(version);
					}
					else
					{
						console.log(CONSOLE_PREFIX + "All resources saved, offline support ready");
						Broadcast("offline-ready");
					}
				});
			});
		});
	})
	.catch(err =>
	{
		// Update check fetches fail when we're offline, but in case there's any other kind of problem with it, log a warning.
		console.warn(CONSOLE_PREFIX + "Update check failed: ", err);
	});
};

self.addEventListener('install', event =>
{
	// On install kick off an update check to cache files on first use.
	// If it fails we can still complete the install event and leave the SW running, we'll just
	// retry on the next navigate.
	event.waitUntil(
		UpdateCheck(true)		// first update
		.catch(() => null)
	);
});

self.addEventListener('fetch', event =>
{
	const isNavigateRequest = (event.request.mode === "navigate");
	
	let responsePromise = GetAvailableCacheNames()
	.then(availableCacheNames =>
	{
		// No caches available: go to network
		if (!availableCacheNames.length)
			return fetch(event.request);
		
		// Resolve with the cache name to use.
		return Promise.resolve().then(() =>
		{
			// Prefer the oldest cache available. This avoids mixed-version responses by ensuring that if a new cache
			// is created and filled due to an update check while the page is running, we keep returning resources
			// from the original (oldest) cache only.
			if (availableCacheNames.length === 1 || !isNavigateRequest)
				return availableCacheNames[0];
			
			// We are making a navigate request with more than one cache available. Check if we can expire any old ones.
			return clients.matchAll().then(clients =>
			{
				// If there are other clients open, don't expire anything yet. We don't want to delete any caches they
				// might be using, which could cause mixed-version responses.
				// TODO: verify client count is as expected in navigate requests.
				// TODO: find a way to upgrade on reloading the only client. Chrome seems to think there are 2 clients in that case.
				if (clients.length > 1)
					return availableCacheNames[0];
				
				// Identify newest cache to use. Delete all the others.
				let latestCacheName = availableCacheNames[availableCacheNames.length - 1];
				console.log(CONSOLE_PREFIX + "Updating to new version");
				
				return Promise.all(availableCacheNames.slice(0, -1)
									.map(c => caches.delete(c)))
				.then(() => latestCacheName);
			});
		}).then(useCacheName =>
		{
			return caches.open(useCacheName)
			.then(c => c.match(event.request))
			.then(response => response || fetch(event.request));
		});
	});

	if (isNavigateRequest)
	{
		// allow the main request to complete, then check for updates
		event.waitUntil(responsePromise
		.then(() => UpdateCheck(false)));		// not first check
	}

	event.respondWith(responsePromise);
});