While measuring performance for bug 1391693 I realized that there is a measurable penalty for internal redirects at times. Its particularly noticeable in non-e10s mode. Our no-fetch optimization currently triggers an interception and then calls ResetInterception() which does an internal redirect to reset the channel. We should consider avoiding the interception altogether in this case. The one tricky thing to get right, though, is we still want to trigger a possible update check. Making sure that still happened was why we originally entered the interception path in this case.

Pushed by shindli@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/0b092833a6ef Avoid creating InterceptedHttpChannel if the service worker has no fetch event handler r=dom-workers-and-storage-reviewers,asuth

Backout by shindli@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/36a13a8545e4 Backed out changeset 0b092833a6ef for wpt14 permafailures CLOSED TREE

Pushed by csabou@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/d32be4ed8786 Avoid creating InterceptedHttpChannel if the service worker has no fetch event handler r=dom-workers-and-storage-reviewers,asuth

Pushed by csabou@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/0666d5b6f6a5 Avoid creating InterceptedHttpChannel if the service worker has no fetch event handler r=dom-workers-and-storage-reviewers,asuth