Add feature policy for client hints, used for third-party subresources.

[chromium-wpt-export-bot]

This disables all client hints for third parties by default, except for
the simplified UA hint. Feature Policy can then be used to delegate
specific hints. Currently, only the hints which were previously being
sent to third parties on Android are available for delagation. A follow-
up CL adds the remaining ones.

Bug: 968201
Change-Id: Idea42e814078592f3bdaec67bd931a99cfaea046
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1567927
Commit-Queue: Ian Clelland <iclelland@chromium.org>
Reviewed-by: Ken Buchanan <kenrb@chromium.org>
Reviewed-by: Yoav Weiss <yoavweiss@chromium.org>
Cr-Commit-Position: refs/heads/master@{#664347}

---

Combined by @foolip with a follow-up change:

Add the remaining Client Hints to Feature Policy.

This allows all client hints to be potentially delegated to third-party
requests. By default, no hints (Except for the simplified UA hint) are
sent with any third party resource requests, but these can be enabled
for specific origins through the use of Feature Policy.

Bug: 968201
Change-Id: I60a94deb0e5553b85da351f08cbabfc8ae0f6e65
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1584400
Commit-Queue: Ian Clelland <iclelland@chromium.org>
Reviewed-by: Ken Buchanan <kenrb@chromium.org>
Reviewed-by: Yoav Weiss <yoavweiss@chromium.org>
Cr-Commit-Position: refs/heads/master@{#664498}

[wpt-pr-bot]

Already reviewed downstream.

[rakuco]

I've pinged the CL authors to check if the flakiness is something they could look at.

[foolip]

From the logs of https://tools.taskcluster.net/groups/av_5AV-US8GCEQVMLK22Dw/tasks/V3EPGIvXSHKYoK9Bb1ommA/details:

Unstable results

Test	Subtest	Results	Messages
/client-hints/accept_ch_lifetime_same_origin_iframe.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/accept_ch_lifetime_subresource.tentative.https.html		ERROR: 1/10, OK: 9/10
/client-hints/accept_ch_lifetime_subresource.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 1/10, PASS: 9/10	assert_false: device-memory-received expected false got true
/client-hints/accept_ch_lifetime_subresource.tentative.https.html	Loading of resources/do_not_expect_client_hints_headers.html did not finish.	PASS: 9/10, TIMEOUT: 1/10	Test timed out
/client-hints/http_equiv_accept_ch_lifetime_same_origin_iframe.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html		ERROR: 1/10, OK: 9/10
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 1/10, PASS: 9/10	assert_false: device-memory-received expected false got true
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html	Loading of resources/do_not_expect_client_hints_headers.html did not finish.	PASS: 9/10, TIMEOUT: 1/10	Test timed out

I suspect this is like #15036, where the test isn't written to be possible to run multiple times in a row.

[foolip]

https://crrev.com/c/1584400 also builds on these tests and might change the results here. I wouldn't bet on it, but I'll push that commit to this PR to see if it does have an impact.

[foolip]

The logs now say:

Unstable results

Test	Subtest	Results	Messages
/client-hints/accept_ch_lifetime_same_origin_iframe.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/accept_ch_lifetime_subresource.tentative.https.html		ERROR: 1/10, OK: 9/10
/client-hints/accept_ch_lifetime_subresource.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 1/10, PASS: 9/10	assert_false: device-memory-received expected false got true
/client-hints/accept_ch_lifetime_subresource.tentative.https.html	Loading of resources/do_not_expect_client_hints_headers.html did not finish.	PASS: 9/10, TIMEOUT: 1/10	Test timed out
/client-hints/http_equiv_accept_ch_lifetime_same_origin_iframe.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html		ERROR: 1/10, OK: 9/10
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 1/10, PASS: 9/10	assert_false: device-memory-received expected false got true
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html	Loading of resources/do_not_expect_client_hints_headers.html did not finish.	PASS: 9/10, TIMEOUT: 1/10	Test timed out

That is the same as before AFAICT.

I've tested /client-hints/accept_ch_lifetime_same_origin_iframe.tentative.https.html locally and it does begin to fail after a few reloads. In an incognito window it then passes again but fails on reload.

Some of the failures seem related to pre-existing async_tests that don't wrap event handlers in t.step_func and result in a harness error, so I'll fix those and see if it affects the results. Maybe they'll be flaky with that fix too, unclear.

[foolip]

Sent those change in #17164.

cc @clelland on this issue.

[foolip]

Now the logs say:

Unstable results

Test	Subtest	Results	Messages
/client-hints/accept_ch_lifetime_same_origin_iframe.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/accept_ch_lifetime_subresource.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 8/10, PASS: 2/10	assert_false: device-memory-received expected false got true
/client-hints/accept_ch_lifetime_subresource.tentative.https.html	Loading of resources/do_not_expect_client_hints_headers.html did not finish.	FAIL: 8/10, PASS: 2/10	assert_equals: expected "PASS" but got "FAIL"
/client-hints/http_equiv_accept_ch_lifetime_same_origin_iframe.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html	Precondition: Test that the browser does not have client hints preferences cached	FAIL: 9/10, PASS: 1/10	assert_false: device-memory-received expected false got true
/client-hints/http_equiv_accept_ch_lifetime_subresource.tentative.https.html	Loading of resources/do_not_expect_client_hints_headers.html did not finish.	FAIL: 8/10, PASS: 2/10	assert_equals: expected "PASS" but got "FAIL"

That removes two lines from the previous round.

In https://wpt.fyi/results/client-hints?diff&filter=ADC&run_id=250780005&run_id=230730003 the results of these tests are unchanged.

I'll submit a PR adding whitespace to the two flaky tests to see if they were already flaky.

[foolip]

Probing in #17166.

[foolip]

From #17166 (comment) it looks like accept_ch_lifetime_same_origin_iframe.tentative.https.html and accept_ch_lifetime_subresource.tentative.https.html were already flaky, which leaves http_equiv_accept_ch_lifetime_subresource.tentative.https.html as possibly newly flaky.

[foolip]

@clelland this resource is used by http_equiv_accept_ch_lifetime_subresource.tentative.https.html.

Do you think these changes could have made that test flaky?

[foolip]

Giving it another spin in https://tools.taskcluster.net/groups/ITD387cHQ4iP6gWIa-hugg.

[foolip]

OK, the flakiness affects only Chrome and so I'm going to admin merge this and see how it fares. I've filed https://bugs.chromium.org/p/chromium/issues/detail?id=972569 to enable the tests in Chromium.