Service Worker: Make ServiceWorkerRegistration.update() return a promise. (Blink Layout 3/3)

LayoutTests/http/tests/serviceworker/resources/update-worker-error.php

Index: LayoutTests/http/tests/serviceworker/resources/update-worker-error.php
diff --git a/LayoutTests/http/tests/serviceworker/resources/update-worker-error.php b/LayoutTests/http/tests/serviceworker/resources/update-worker-error.php
new file mode 100644
index 0000000000000000000000000000000000000000..c8f43db36cfc9ca1d80d320c42c9dc0724a3fa3f
--- /dev/null
+++ b/LayoutTests/http/tests/serviceworker/resources/update-worker-error.php
@@ -0,0 +1,25 @@
+<?php

[nhiroki, 2015/08/06 09:57:28]

Can you rename this file to "update-error-worker.php"? Using "-worker" suffix
for worker scripts is our convention (sorry for our non-explicit rules...)

[jungkees, 2015/08/06 10:53:12]

Sure, I'll change it.

+if(!isset($_COOKIE['mode']))
+  $mode = 'plain'; // For the first fetch, set the mode to plain.
+else
+  $mode = 'error'; // For the second fetch, set the mode to error.
+
+// Plain mode sets the normal mimetype.
+if ($mode == 'plain') {
+  echo '// Plain mode.';
+  header("Cache-Control: no-cache, must-revalidate");
+  header("Pragma: no-cache");

[nhiroki, 2015/08/06 09:57:28]

Let me confirm my understanding... in the latest spec (after [1]), update() does
not always bypass the HTTP Cache, so we need to set these "no-cache" parameters
here. Is this correct?

(In the current implementation, update() always bypasses the HTTP Cache[2])

[1]
https://github.com/slightlyoff/ServiceWorker/commit/8ea4c1e2f0ebb612962fffed4...
[2]
https://code.google.com/p/chromium/codesearch#chromium/src/content/browser/se...

[jungkees, 2015/08/06 10:53:12]

Yes, that's my understanding and intention. I noticed that we might need to set
the |force_bypass_cache| to false there. I'll double check it with the other
editors before changing that.
Right, I'll take care of it.

[jungkees, 2015/08/07 05:18:30]

I've double-checked this with jakearchibald@. It's the decision from the latest
f2f that update() does not always bypass the HTTP cache. It should follow the
same rule (i.e. min(max-age, 86400) has passed since the last update) to bypass
the cache.

I guess that max-age change
(https://github.com/slightlyoff/ServiceWorker/commit/8ea4c1e2f0ebb612962fffed4...)
hasn't been tracked yet. If so and nobody's already working on it, I'm happy to
take on it as a separate CL.

+  header('Content-Type:application/javascript');
+  setcookie('mode', 'error'); // Set cookie for 2nd fetch to work in error mode.
+} 
+// Error mode sets a disallowed mimetype.
+else if ($mode == 'error') {
+  echo '// Error mode.';
+  header("Cache-Control: no-cache, must-revalidate");
+  header("Pragma: no-cache");
+  header('Content-Type:text/html');
+  unset($_COOKIE['mode']);
+  setcookie('mode', '', time() - 3600); // Delete cookie at the end of the test.
+}
+echo '// ' . microtime();
+?>