commit 392793004f49b07eb813b2c600ef438d4b176524
author David Bokan <bokan@chromium.org> Tue Aug 18 17:52:45 2020
committer Blink WPT Bot <blink-w3c-test-autoroller@chromium.org> Tue Aug 18 18:34:56 2020
tree b8337b0f96bd5279d1bd22594653be3ad1059dfc
parent cc1b58784ba7ab62924f0e81e062ceaec7834876

Enable text fragments across redirects

A text fragment occurs in a URL fragment and begins with ":~:text=...".
It is used to highlight and scroll the provided text into view when the
page is loaded. For user privacy reasons, we restrict scrolling the text
into view unless the navigation occurred via a user gesture. See:
https://github.com/WICG/scroll-to-text-fragment#security-considerations
for more details.

However, it is common (particularly on social and messaging services
where content is user-generated) for links to be served via a redirect.
A typical example (from chat.google.com) works like this:

 1. User receives and clicks a link to https://example.com#:~:text=foo"
 2. chat.google.com opens a new tab using window.open("", "_blank")
 3. chat.google.com calls document.write on the newly opened window to
   write a <meta> tag-based client redirect to
   google.com/url?url=https://example.com... which is the URL
   redirection service with the destination URL as a query param.
 4. google.com/url then calls window.location and writes
   "https://example.com#:~:text=foo" into it
 5. the new tab finally navigates to example.com

The only navigation that had a user gesture attached to it is the
initial empty document navigation in step 2. This means the example.com
page is navigated to without a user gesture and the text fragment is
blocked. A similar pattern is seen on many popular services: Twitter,
Instagram, Facebook Messenger, etc.

This CL solves the above scenario by introducing a "text fragment
token". This token grants its holder permission to invoke a text
fragment. The token can be used during load to invoke the text fragment,
or it can be passed into a navigation to grant permission to the next
page without requiring a user gesture. However, in either case, the
token is consumed so a page cannot both invoke a text fragment and pass
the token.

The token is created in only in DocumentLoader's constructor and while
processing a same-document navigation. For regular navigations, it is
only created if the current navigation was user initiated. For
same-document navigations, it's created only if browser-initiated and
the navigation has a text fragment. This mechanism can be thought of as
a user gesture that applies only to text fragment and whose lifetime
extends across navigations but cannot be copied and is always consumed
on use.

Bug: 1055455
Change-Id: Icddd849937d24b579bbeb5a4b9f87539d8339905
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2159324
Reviewed-by: Mike West <mkwst@chromium.org>
Reviewed-by: Avi Drissman <avi@chromium.org>
Commit-Queue: David Bokan <bokan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#799173}