In the WebDriver protocol, when a command fails because it can not
interact with the target element, an "element not interactable" error is
generated. It can be a transitive issue (for example, due to an
animation), so when the error is received the command should be tried
again, just like done, for example, with "ElementNotVisible" exceptions.
However, the last version of the "instaclick/php-webdriver" library
compatible with the Selenium Driver of Mink did not support yet that
WebDriver error. And even if Chrome is run using the old protocol an
unknown "element not interactable" error can be received anyway in some
cases. When an unknown error is received by the
"instaclick/php-webdriver" library it is thrown as a generic Exception
so, until the library can be updated, the message of generic exceptions
is checked and the command is retried if it matched.
For the time being "element not interactable" errors are handled like
"ElementNotVisible" exceptions; this may need to change once the error
is better understood.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The acceptance tests used the last Selenium 2 Docker container
available, which provides a rather old Firefox version (Firefox 47).
Nevertheless, despite some rendering issues, most things still worked as
expected due to the JavaScript files being built with support for older
browsers. However, now that support for Internet Explorer 11 and older
browsers will be dropped things could start to fail, so a newer browser
(and thus a newer Selenium version) should be used in the acceptance
tests.
Selenium has been standardized by the W3C, and the protocol to
communicate between the Selenium server and the browser has changed due
to that. Firefox >= 48 only supports the new W3C protocol, but the
Selenium driver for Mink does not support it yet.
The old protocol can still be used in recent Chromium/Chrome versions by
explicitly forcing it, so for the time being the acceptance tests will
need to be run on Chrome instead (although Firefox provides some
interesting features like the fake streams that would be needed to test
calls in Talk, so they should be moved again to Firefox once possible).
Finally, the default shm size of Docker is 64 MiB. This does not seem
enough to run newer Chrome releases and causes the browser to randomly
crash during the tests ("unknown error: session deleted because of page
crash" is shown in the logs). Due to this "disable-dev-shm-usage" needs
to be used so Chrome writes shared memory files into "/tmp" instead of
"/dev/shm" (the default shm size of Docker could have been increased
instead using "docker run --shm-size...", but that seems to be
problematic when the container is run in current Drone releases).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
When the value is set in some input fields a carriage return was sent to
simulate pressing the enter key and thus confirming the input. However,
different browsers use different keys (Firefox uses "\r", but Chrome
uses "\n"), so the carriage return was replaced with the WebDriver
"ENTER" constant which is common to both browsers.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Sending the "enter" key is not needed in those input fields that auto
save while the user is typing or when the focus is lost (which since
version 1.4.0 the Selenium driver for Mink is automatically done after
setting the value).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Since version 1.8.0 of Mink "Mink::getSession()" no longer starts the
session automatically (see https://github.com/minkphp/Mink/pull/705), so
it now needs to be explicitly started.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Since version 1.4.0 the Selenium driver for Mink uses again the element
on which the value was set (see
https://github.com/minkphp/MinkSelenium2Driver/pull/286). When creating
a new folder or renaming one sending a new line ("\r") caused the
element on which the value was set to be removed, so the element was no
longer attached to the DOM when the driver tried to use it again, and
thus a "StaleElementReference" exception was thrown.
Due to this now it is needed to explicitly click the confirm button when
creating a new folder. In the case of the renaming, on the other hand,
nothing else besides not sending the new line is needed, as the Selenium
driver now unfocuses the element (that is why it uses again the element
after setting the value) which triggers the renaming.
Besides that, the Selenium driver for Mink uses a library to simulate
certain events, bitovi/syn. In version 1.4.0 that library was updated to
version 0.0.3, which seems to somehow break pressing the "escape" key.
Due to this now the sharing menu has to be closed by pressing "enter" on
the share menu button instead.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The PHPUnit update also required an update of "symfony/yaml", so besides
the changes needed for PHPUnit the "behat.yml" file also had to be
adjusted.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
User "user1" is added when installing and configuring the server, so it
is already added in all tests.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
User "user1" is added when installing and configuring the server, so it
is already added in all tests. As the test verifies that just added
users can be searched in the contacts menu a new user should be actually
added.
The test did not fail because it assumed that "user1" did not exist and
just checked that it existed after "adding" it, but not whether adding
it failed.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The input element is always hidden, so the check always ended falling
back to the label. Moreover, the label is the element that the user
interacts with, so it must be the one used.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Although if the element could not be found an exception would be thrown
and the test aborted if an element is in the DOM but hidden it would be
found and the test would pass.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Instead of looking for the bundle button and then checking its value now
the expected value is included in the locator and the button is checked
similarly to other elements.
No "Disable all" locator was added as it was not currently needed
anywhere.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
"Actor::find" is a more robust way to look for elements, as it handles
some exceptions that may be thrown. Therefore, even if the elements are
not actually used and it is only checked whether they exist or not using
the actor is the preferred way when possible (and it also makes it
consistent with the rest of the acceptance tests).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Adding some missing asserts showed that the "delete user" acceptance
test was silently failing, as the deletion was not being confirmed in
the dialog and thus the user was not being deleted.
The dialog button contains a single quote ("user0's"), so the XPath
expression had to be adjusted (it seems that it is not possible to
escape a single quote in a string enclosed in single quotes in XPath
1.0).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
WaitFor::element... calls only perform the waiting and return whether
the condition succeeded or not, but that result needs to be explicitly
checked to prevent further steps from being executed if the wait failed.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
As no timeout was specified the elements were tried to be found just
once. This caused the steps to fail if the elements did not appear yet
in the page when they were tried to be found.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
As "composer.lock" was not versioned the dependencies had to be resolved
everytime that the acceptance tests run, which took some precious time.
Besides that the dependency versions were also tightened for better
control.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The menu button and the menu of a share are no longer direct childs of
the actions of the share row. The menu button is now a child of a
".trigger" element, while the menu is a direct child of the body and has
an id defined in the "aria-describedby" attribute of the ".trigger"
element.
In XPath 1.0 it does not seem possible to "backreference" a value or
create variables, so when the share menu or one of its item is needed
now the ".trigger" element is first found and then its XPath expression
is used to compose its "aria-describedby" attribute in the XPath
expression for the menu.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
When a row is added to the file list the opacity of the file row is
animated from transparent to fully opaque. As the file actions menu is a
descendant of the row but overflows it when the row is not fully opaque
clicks on the menu entries "fall-through" and are received instead by
the rows behind.
The opacity animation is a CSS animation, and it is not possible to know
if a row is appearing or not except from its opacity (the row will have
the "appear" CSS class even after the animation ended). Therefore it
should be waited until the row of the file is fully opaque before using
the menu.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The (old) Firefox version used in the acceptance tests does not properly
render the share link menu. As the menu is taller than it should
sometimes it covers the copy link button, so it is not possible to click
it without hiding the share link menu. Moreover, in those cases the
share menu button is also covered by the share menu, so the menu needs
to be closed by pressing the "Esc" key.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Bump p-limit from 2.2.2 to 2.3.0, @nextcloud/auth from 1.2.2 to 1.2.3, @babel/preset-env from 7.8.7 to 7.9.5, @nextcloud/event-bus from 1.1.3 to 1.1.4, @nextcloud/password-confirmation from 1.0.0 to 1.0.1, @nextcloud/l10n from 1.2.2 to 1.2.3, @nextcloud/initial-state from 1.1.1 to 1.1.2, core-js from 3.6.4 to 3.6.5 …
To continue this formatting madness, here's a tiny patch that adds
unified formatting for control structures like if and loops as well as
classes, their methods and anonymous functions. This basically forces
the constructs to start on the same line. This is not exactly what PSR2
wants, but I think we can have a few exceptions with "our" style. The
starting of braces on the same line is pracrically standard for our
code.
This also removes and empty lines from method/function bodies at the
beginning and end.
Signed-off-by: Christoph Wurst <christoph@winzerhof-wurst.at>
Simounet
Daniel Calviño Sánchez
Joas Schilling
kesselb
Morris Jobke
Julius Härtl
Georg Ehrke
Roeland Jago Douma
Arthur Schiwon
John Molakvoæ
John Molakvoæ (skjnldsv)
Christoph Wurst