OOPIF support for testRunner.dumpAsText and similar layout dumps.

OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

  - The synchronous dumps are still performed if the test didn't ask for
    recursing over all the frames.  Retaining the synchronous behavior
    in this case is needed, because in some tests the dump is captured
    while the frame is being detached (and would no longer be present
    after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Committed: https://crrev.com/a8960461e06e8478f79fde21531e237443b5629f
Cr-Commit-Position: refs/heads/master@{#371896}

[Łukasz Anforowicz, 2016-01-20 22:37:09]

Description was changed from

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

This CL moves responsibility for coordinating layout dumps (i.e. dump as
text, dump as markup, dump scroll positions with extra flavors like dump
as printed, dump line box trees, etc.) from the renderer process into
the browser process.  This means that layout tests that used to crash
when run with --site-per-process will actually complete and produce
valid comparison results.  Example:

  $ third_party/WebKit/Tools/Scripts/run-webkit-tests -v -t gn \
      http/tests/security/cross-origin-appcache-allowed.html \
      --additional-drt-flag=--site-per-process --no-retry-failures

This CL doesn't affect the following dump modes (which for now remain
incompatible with OOPIFs): dump as audio, dump as custom text, dump
pixels, dump back/forward list).  Additionally, setting and reading of
dump modes is done in a renderer process (which can be incompatible with
OOPIFs when testRunner.dumpAsText() and testRunner.notifyDone() are
called in cross-site frames running in different renderer processes).

This CL means that testRunner.notifyDone() no longer performs dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps,
    - the case where notifyDone is called from a secondary window.

  - Layout tests are passing with the changes.

BUG=477150
==========

to

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

    - The synchronous dumps are still performed if the test didn't ask for
      recursing over all the frames.  Retaining the synchronous behavior
      in this case is needed, because in some tests the dump is captured
      while the frame is being detached (and would no longer be present
      after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
==========

[Łukasz Anforowicz, 2016-01-21 00:26:49]

Description was changed from

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

    - The synchronous dumps are still performed if the test didn't ask for
      recursing over all the frames.  Retaining the synchronous behavior
      in this case is needed, because in some tests the dump is captured
      while the frame is being detached (and would no longer be present
      after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
==========

to

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

  - The synchronous dumps are still performed if the test didn't ask for
    recursing over all the frames.  Retaining the synchronous behavior
    in this case is needed, because in some tests the dump is captured
    while the frame is being detached (and would no longer be present
    after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
==========

[Łukasz Anforowicz, 2016-01-21 00:27:08]

lukasza@chromium.org changed reviewers:
+ mkwst@chromium.org

[Łukasz Anforowicz, 2016-01-21 00:27:08]

Mike, could you please take a look?

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
File components/test_runner/layout_dump.cc (right):

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
components/test_runner/layout_dump.cc:24: std::string
DumpFrameHeaderIfNeeded(WebFrame* frame) {
Copy&pasted from components/test_runner/web_test_proxy.cc.

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
components/test_runner/layout_dump.cc:37: std::string
DumpFrameScrollPosition(WebFrame* frame) {
Copy&pasted from components/test_runner/web_test_proxy.cc + recursion removed.

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
components/test_runner/layout_dump.cc:54: std::string DumpLayout(WebLocalFrame*
frame, const LayoutDumpFlags& flags) {
DumpLayout is a rough equivalent of WebTestProxyBase::CaptureTree from
components/test_runner/web_test_proxy.cc, except that:

- DumpLayout handles single frame only
  (i.e. doesn't recurse over child frames)

- DumpLayout doesn't handle the following dump modes:
  custom text, back-forward lists.

DumpLayout is called from

1) the old code (from WebTestProxyBase::CaptureTree)
2) the new code in
content/shell/renderer/layout_test/layout_test_render_frame_observer.cc

[Łukasz Anforowicz, 2016-01-21 18:30:07]

BTW: I have looked at the linux_blink_dbg trybot failures and I think it is ok
to ignore these failures:

- linux_blink_dbg is not part of CQ.

- linux_blink_dbg has been quite red on the waterfall recently - the same
failures can be seen in other people's CLs over

  -
https://build.chromium.org/p/tryserver.blink/builders/linux_blink_dbg/builds/...
  -
https://build.chromium.org/p/tryserver.blink/builders/linux_blink_dbg/builds/...

- I tried one remaining failure not covered by the 2 builds above, but I cannot
repro it locally with my changes:

  - http/tests/accessibility/slow-document-load.html

[Mike West, 2016-01-22 19:58:56]

Assuming the bots are happy, moving things around in this way looks pretty
reasonable to me. I've kicked off another pass for the bots that failed; LGTM if
they go green.

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
File components/test_runner/layout_dump_flags.h (right):

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
components/test_runner/layout_dump_flags.h:14: bool dump_scroll_positions;
If these are mutually-exclusive, perhaps you could model them as an enum?

[dcheng, 2016-01-23 05:17:33]

https://codereview.chromium.org/1589643003/diff/80001/third_party/WebKit/Layo...
File
third_party/WebKit/LayoutTests/canvas/philip/tests/2d.text-custom-font-load-crash.html
(right):

https://codereview.chromium.org/1589643003/diff/80001/third_party/WebKit/Layo...
third_party/WebKit/LayoutTests/canvas/philip/tests/2d.text-custom-font-load-crash.html:28:
_assert(true, "Test passes if it does not crash.")
Not sure if I missed this a comment about this, but what's the reason for this
test change?

[Łukasz Anforowicz, 2016-01-26 00:50:03]

On 2016/01/22 19:58:56, Mike West wrote:
> Assuming the bots are happy, moving things around in this way looks pretty
> reasonable to me. I've kicked off another pass for the bots that failed; LGTM
if
> they go green.

It seems that right now even linux_blink_rel bots are red - I need to take a
closer look at the failures before we can rereview and attempt to land... :-/

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
File components/test_runner/layout_dump_flags.h (right):

https://codereview.chromium.org/1589643003/diff/80001/components/test_runner/...
components/test_runner/layout_dump_flags.h:14: bool dump_scroll_positions;
On 2016/01/22 19:58:56, Mike West wrote:
> If these are mutually-exclusive, perhaps you could model them as an enum?

Done.

https://codereview.chromium.org/1589643003/diff/80001/third_party/WebKit/Layo...
File
third_party/WebKit/LayoutTests/canvas/philip/tests/2d.text-custom-font-load-crash.html
(right):

https://codereview.chromium.org/1589643003/diff/80001/third_party/WebKit/Layo...
third_party/WebKit/LayoutTests/canvas/philip/tests/2d.text-custom-font-load-crash.html:28:
_assert(true, "Test passes if it does not crash.")
On 2016/01/23 05:17:33, dcheng wrote:
> Not sure if I missed this a comment about this, but what's the reason for this
> test change?

This change is not necessary in the latest patchset - I've removed it.

When testing earlier patchsets I've noted that
LayoutTests/canvas/philip/tests.js:122 should fail a test if the test didn't
call any of the _assertFoo functions.  The modified test did fail with my
earlier changes, without the extra call to _assert.  I don't understand why the
test is passing without my changes and with my latest changes.

[Łukasz Anforowicz, 2016-01-26 19:51:15]

Description was changed from

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

  - The synchronous dumps are still performed if the test didn't ask for
    recursing over all the frames.  Retaining the synchronous behavior
    in this case is needed, because in some tests the dump is captured
    while the frame is being detached (and would no longer be present
    after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
==========

to

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

  - The synchronous dumps are still performed if the test didn't ask for
    recursing over all the frames.  Retaining the synchronous behavior
    in this case is needed, because in some tests the dump is captured
    while the frame is being detached (and would no longer be present
    after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[Łukasz Anforowicz, 2016-01-27 21:16:26]

Trybot failures in patch set 9 are not related to the changes in this CL (more
detailed notes below).  Based on this I'll rebase and try pushing to CQ as-is.

LINUX_BLINK_DBG (patch set 9)

  Regressions: Unexpected text-only failures (5)
    fast/events/touch/gesture/pad-gesture-fling.html [ Failure ] [ Fails w/o
patch on bot ]
    fast/forms/submit-to-blank-multiple-times.html [ Failure ] [ Fails w/o patch
on bot ]
    http/tests/accessibility/slow-document-load.html [ Failure ] [ Fails w/o
patch on bot ]
   
virtual/scroll_customization/fast/events/touch/gesture/pad-gesture-fling.html [
Failure ] [ Fails w/o patch ]
    virtual/smoothscrolling/fast/events/touch/gesture/pad-gesture-fling.html [
Failure ] [ Fails w/o patch ]

  Regressions: Unexpected image-only failures (1)
    svg/transforms/text-with-pattern-inside-transformed-html.xhtml [ Failure ] [
Fails w/o patch ]
      - A few pixel-sized dithering/shading changes?

  Regressions: Unexpected timeouts (6)
    fast/dom/StyleSheet/stylesheet-move-between-documents-crash.html [ Timeout ]
[ Fails w/o patch on bot ]
    fast/dom/Window/window-function-frame-getter-precedence.html [ Timeout ] [
Fails w/o patch on bot ]
    fast/dom/shadow/svg-style-in-shadow-tree-crash.html [ Timeout ] [ Fails w/o
patch on bot ]
    fast/encoding/char-encoding.html [ Timeout ] [ Fails w/o patch on bot ]
    http/tests/security/w3c/cross-origin-objects.html [ Timeout ] [ Fails w/o
patch on bot ]
    http/tests/security/xss-DENIED-cross-origin-stack-overflow.html [ Timeout ]
[ No local repro x 1000 ]

WIN_BLINK_REL (patch set 9)

  Regressions: Unexpected image and text failures (1)
    fast/forms/text/text-font-height-mismatch.html [ Failure ] [ Fails w/o patch
on bot ]

  Regressions: Unexpected text-only failures (1)
    http/tests/inspector/resource-tree/resource-tree-non-unique-url.html [
Failure ] [ Repros without my changes ~60% of time ]

MAC_BLINK_DBG (patch set 9)

  Regressions: Unexpected image-only failures (2)
    svg/transforms/animated-path-inside-transformed-html.xhtml [ Failure ] [ Not
investigated ]
    svg/transforms/text-with-pattern-inside-transformed-html.xhtml [ Failure ] [
Not investigated ]

[Łukasz Anforowicz, 2016-01-27 21:32:11]

The CQ bit was checked by lukasza@chromium.org

[Łukasz Anforowicz, 2016-01-27 21:32:12]

The patchset sent to the CQ was uploaded after l-g-t-m from mkwst@chromium.org
Link to the patchset: https://codereview.chromium.org/1589643003/#ps180001
(title: "Rebasing...")

[commit-bot: I haz the power, 2016-01-27 21:34:07]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1589643003/180001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1589643003/180001

[commit-bot: I haz the power, 2016-01-27 22:27:40]

Committed patchset #10 (id:180001)

[commit-bot: I haz the power, 2016-01-27 22:28:51]

Description was changed from

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

  - The synchronous dumps are still performed if the test didn't ask for
    recursing over all the frames.  Retaining the synchronous behavior
    in this case is needed, because in some tests the dump is captured
    while the frame is being detached (and would no longer be present
    after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
OOPIF support for testRunner.dumpAsText and similar layout dumps.

Layout Tests dump page contents (to compare against expected results).
The dump can include frame contents (i.e. dump as text, dump as markup,
dump scroll positions with extra flavors like dump as printed, dump line
box trees, etc.).  Since renderer process is (for security / by design)
not able to see frame contents of remote frames, it means that old
Layout Tests code is not able to dump frame contents when site isolation
is enabled (i.e. when running with --additional-drt-flag=--site-per-process).

This CL is a step toward making layout tests compatible with site
isolation.  After this CL, if recursing over all frames is required,
then BlinkTestRunner::CaptureDump will ask the browser process for
stiching together the frame contents, before continuing with the other
dump flavors in BlinkTestRunner::OnLayoutDumpCompleted.

The above means testRunner.notifyDone() might no longer perform dumps
synchronously.  This is okay, because:

  - The dumps were already performed asynchronously in some cases:
    - pixel dumps (i.e. see how dumping is resumed after
      BlinkTestRunner::CaptureDumpPixels aka OnPixelsDumpCompleted),
    - ShouldDumpBackForwardList (i.e. see how dumping is resumed after
      BlinkTestRunner::OnSessionHistory),
    - the case where notifyDone is called from a secondary window (i.e.
      see how BlinkTestRunner::TestFinished asks the browser to continue
      in the main window).

  - The synchronous dumps are still performed if the test didn't ask for
    recursing over all the frames.  Retaining the synchronous behavior
    in this case is needed, because in some tests the dump is captured
    while the frame is being detached (and would no longer be present
    after an extra hop to the browser process).

This CL doesn't affect the following dump modes (which for now remain
potentially incompatible with OOPIFs): dump as audio, dump as custom
text, dump pixels, dump back/forward list).  Additionally, setting and
reading of dump modes is done in a renderer process (which can be
incompatible with OOPIFs when testRunner.dumpAsText() and
testRunner.notifyDone() are called in cross-site frames running in
different renderer processes).

BUG=477150
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Committed: https://crrev.com/a8960461e06e8478f79fde21531e237443b5629f
Cr-Commit-Position: refs/heads/master@{#371896}
==========

[commit-bot: I haz the power, 2016-01-27 22:28:52]

Patchset 10 (id:??) landed as
https://crrev.com/a8960461e06e8478f79fde21531e237443b5629f
Cr-Commit-Position: refs/heads/master@{#371896}