To allow Marionette to capture the content of OOP iframes, the
new Fission compatible screenshot API has to be used.
For callers requesting a read-back of the pixel data as
currently in the compositor/window, the 'drawWindow()'
method still has to be called.
Differential Revision: https://phabricator.services.mozilla.com/D40654
--HG--
extra : moz-landing-system : lando
This moves all the screenshot related code from the framescript to
the parent process, so that canvas.js is no longer called from
within a content process.
The remaining code in the framescript is only needed to compute
the dimensions of the screenshot, and from all the to highlight
elements.
This move is necessary to allow switching to the new drawSnapshot
API which only works from within the parent process.
Differential Revision: https://phabricator.services.mozilla.com/D40209
--HG--
extra : moz-landing-system : lando
On Android there is no focus ring drawn, and in general it depends
on the visibility of the caret, which constantly blinks. As such
intermittent and perma failures can always occur. Best is to just
remove this flaky test.
Differential Revision: https://phabricator.services.mozilla.com/D42493
--HG--
extra : moz-landing-system : lando
The feature was used in the past to highlight broken elements for l10n
specific tests. Given that those tests don't exist anymore (for already
a long time) the highlight feature doesn't have to be kept alive.
Also it isn't covered by the WebDriver spec, and as such a custom feature
which is not worth keeping its code working. Especially with the Fission
work upcoming.
Differential Revision: https://phabricator.services.mozilla.com/D42823
--HG--
extra : moz-landing-system : lando
Temporarily skip on Windows to get the new screenshot API landed.
Given that this change makes this failure permanent it should be
easier for investigation after its landing.
Differential Revision: https://phabricator.services.mozilla.com/D42322
--HG--
extra : moz-landing-system : lando
We were incorrectly failing to put all the data onto the comparison stack
when handling nested reftests. Ensure that we do this correctly.
Differential Revision: https://phabricator.services.mozilla.com/D42284
--HG--
extra : moz-landing-system : lando
The patch makes the code for handling screenshots in the driver and
listener module easier to understand. Also it sets the default values
for optional arguments once, which doesn't require duplicated code
for the listener anymore.
Further there was a bug which treated the "full" parameter with a
higher priority as an element, if both are defined.
Differential Revision: https://phabricator.services.mozilla.com/D40818
--HG--
extra : moz-landing-system : lando
All dependencies are fixed, or no longer reproducible. As such both
flags can be re-enabled. Also reftests won't have to specify the
exactly same flags on its own anymore.
Differential Revision: https://phabricator.services.mozilla.com/D40395
--HG--
extra : moz-landing-system : lando
Instead of reporting the pixel differences in a logger info message,
put it in the text that's returned to the harness. This has a notable
advantage on android where this will cause it to be logged as part of
the harness logs rather than ending up in the logcat from the
device. It also makes these messages more accessible in other
consumers of the logs e.g. wpt.fyi.
Differential Revision: https://phabricator.services.mozilla.com/D40064
--HG--
extra : moz-landing-system : lando
Instead of reporting the pixel differences in a logger info message,
put it in the text that's returned to the harness. This has a notable
advantage on android where this will cause it to be logged as part of
the harness logs rather than ending up in the logcat from the
device. It also makes these messages more accessible in other
consumers of the logs e.g. wpt.fyi.
Differential Revision: https://phabricator.services.mozilla.com/D40064
--HG--
extra : moz-landing-system : lando
Internal marionette listeners should neither be visible to content
script, nor have their registation affected by changes made in
content. The evaluate method was breaking these constraints by
creating listeners in a sandbox with Xrays disabled, which is
appropriate to the injected script itself but not to the
harness-internal parts.
Use a different sandbox for the harness code compared to the injected
code, and move away from using onunload to using addEventListener for
the unload handler.
Differential Revision: https://phabricator.services.mozilla.com/D39522
--HG--
extra : moz-landing-system : lando
