Firefox and Chrome drivers for Selenium refuse to click on an element if
the point to be clicked is covered by a different element, throwing an
UnknownError exception with message "Element is not clickable at point
({x}, {y}). Other element would receive the click: {element}". Although
in general that would be a legit error (as the user would not be able to
click on the element) due to a bad layout, sometimes this can be just a
temporal issue caused by an animation, in which case there would be no
problem once the animation finished and the elements are all in their
final location.
Unfortunately, automatically handling those situations in which the
problem is caused by an animation by just retrying a few times if the
element to be clicked is covered before giving up would probably cause
confusion instead of easing test writing.
The reason is that if the center of the element is covered by another
one the Firefox driver for Selenium tries to click on the corners of the
element instead. The problem is that the coordinates used for the click
are integer values, but Firefox has sub-pixel accuracy, so sometimes
(depending on which corner is not covered and whether the left, top,
width or height properties of the element to be clicked have a decimal
component or not) the clicks silently land on a different HTML element
(and that is with squared borders; with round borders they always land
on a different HTML element. That was partially addressed for Selenium
3.0 by clicking first on the edges, but it would still fail if the
middle point of the edges is covered but not the corners).
It is not possible to fix or even detect all that from the tests (except
maybe with some extreme hacks involving accessing private PHP members
from Mink and bypassing or replacing the standard JavaScript executed by
the Firefox driver with a custom implementation...), so it is not
possible to ensure that clicks during an animation will land on the
right element (in fact it is not possible even on static elements,
although except when the layout is wrong there should be no problem);
sometimes retrying a click when the element is covered would solve the
problem, sometimes it would cause a different element to be clicked (and
sometimes there would be even no retry, as the first click would have
silently landed on a different element than the expected one).
Therefore, a different approach must be used. Instead of trying to
automatically handle clicks during animations the tests must be written
being aware of the problem and thus waiting somehow for the animations
that can cause a problem to end before performing the clicks.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
This is a preparatory step for a following commit in which the position
of the favorite icon and the checkbox will be swapped; in that new
design the favorite icon is no longer expected to be an action but just
a simple mark on whether the file is favorited or not (the action is
expected to be triggered then only from the file actions menu).
The favorite icon is now fully shown or completely hidden depending on
whether the file is favorited or not. As the icon is just informative
but no longer an action now it does not change when hovered or focus. In
the same way, the alternative text when the file is not favorited now it
is not "Favorite" (an action) but "Not favorited" instead.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Currently a file can be favorited either through the inline action or
through the file actions menu. However, the inline action will be
removed in a following commit and then it will be possible to do it only
through the file actions menu.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
In some cases the acceptance tests have to explicitly wait for something
to happen without using the "find" method from the actor; in those cases
the timeout multiplier needs to be taken into account too, so the test
cases must be able to retrieve it from the actor.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
By default "127.0.0.1:4444" is used, so nothing needs to be set when the
acceptance tests and the Selenium server share the same network (like
when called by "run.sh").
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
By default "127.0.0.1" is used, so nothing needs to be set when the
Selenium server and the Nextcloud test server share the same network
(like when called by "run.sh").
Besides passing the domain to the acceptance tests the Nextcloud test
server configuration must be modified to see the given domain as a
trusted domain; otherwise the access would be forbidden.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The NextcloudTestServerLocalHelper started the PHP built-in web server
for the Nextcloud test server at 127.0.0.1; as the Selenium server has
to access the Nextcloud test server they were forced to share the same
network. Now, the domain at which the PHP built-in web server is started
can be specified when the NextcloudTestServerLocalHelper is created,
which removes the need of sharing the same network, as the Selenium
server now can access the Nextcloud test server at an arbitrary domain.
However, by default "127.0.0.1" is still used if no domain is given.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The app navigation is not exclusive to the Files app but a generic
component used by other apps too, so its locators and steps should be in
its own context.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Acceptance tests opened the details view by clicking on the middle of
the file row, but due to the changes made in issue #4921 that now opens
the file instead; this commit updates the acceptance tests to open the
details view through the "Details" item in the file actions menu.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Commands executed on Mink elements may fail for several reasons.
ElementWrapper is introduced to automatically handle some of those
situations, like StaleElementReference exceptions and ElementNotVisible
exceptions.
StaleElementReference exceptions are thrown when the command is executed
on an element that is no longer attached to the DOM. When that happens
the ElementWrapper finds again the element and executes the command
again on the new element.
ElementNotVisible exceptions are thrown when the command requires the
element to be visible but the element is not. When that happens the
ElementWrapper waits for the element to be visible before executing the
command again.
These changes are totally compatible with the current acceptance tests.
They just make the tests more robust, but they do not change their
behaviour. In fact, this should minimize some of the sporadic failures
in the acceptance tests caused by their concurrent nature with respect
to the web browser executing the commands.
However, the ElementWrapper is not a silver bullet; it handles the most
common situations, but it does not handle every possible scenario. For
example, the acceptance tests would still fail sporadically if an
element can become staled several times in a row (uncommon) or if it
does not become visible before the timeout expires (which could still
happen in a loaded system even if the components under test work right,
but obviously it is not possible to wait indefinitely for them).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
Starting a session for an Actor can fail, typically, due to a timeout
connecting with the web browser. Now if the session fails to start it
will be tried again up to "actorTimeoutMultiplier" times in total before
giving up.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The timeout passed to the "find" method was multiplied by the
"findTimeoutMultiplier" attribute. However, as "find" used
"findAncestor" and "findAncestor", in turn, used "find" itself the
timeout was increased exponentially for ancestor elements. Now "find"
was split in "find" and "findInternal"; the first method is the public
one and modifies the given parameters as needed and then calls the
second method, private, that performs the find itself.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The "named" Mink selector first tries to find an exact match for its
locator and then, if not found, tries to find a partial match. Besides
other harder to track problems (see comment in the commit in which the
"content" locator was removed), this could cause, for example, finding
an action link titled "Favorited" when looking for the action link
titled "Favorite" (that is, one that conveys the opposite state to the
one found).
Although currently all the acceptance tests are compatible with both the
"named" and the "named_exact" Mink selectors the predefined locators are
modified to use the "named_exact" Mink selector to make them more
future-proof; the "named" Mink selector can still be used if needed
through the "customSelector" method in the builder object.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The "content" locator uses the "named" Mink selector and the "content"
Mink locator to find the element. The "named" Mink first tries to find
the elements whose content match exactly the given content but, if none
is found, then it tries to find elements that just contain the given
content.
This behaviour can lead to hard to track issues. Finding the exact match
and, if not found, finding the partial match is done in quick
succession. In most cases, when looking for an exact match the element
is already there, it is returned, and everything works as expected. Or
it may not be there, but then it is not there either when finding the
partial match, so no element is returned, and everything works as
expected (that is, the actor tries to find again the element after some
time).
However, it can also happen that when looking for an exact match there
is no element yet, but it appears after trying to find the exact match
but before trying to find the partial match. In that situation the
desired element would be returned along with its ancestors. However, as
only the first found element is taken into account and the ancestors
would appear first the find action would be successful, but the returned
element would not be the expected one. This is highly unlikely, yet
possible, and can cause sporadic failures in acceptance tests that,
apparently, work as expected.
Using a "named_exact" Mink selector instead of the "named" Mink selector
does not provide the desired behaviour in most cases either. As it finds
any element whose content matches exactly the given content, looking for
"Hello world" in "<div><p><a>Hello world</a></p></div>" would match the
"div", "p" and "a" elements; in that situation the "div" element would
be the one returned, when typically the "a" element would be the
expected one.
As it is error prone and easily replaceable by more robust locators the
"content" locator was removed from the predefined ones (although it can
still be used if needed through the "customSelector" method in the
builder object).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
The data storage (the "notebook") is shared between all the actors, so
the data can be stored and retrieved between different steps by any
actor in the same scenario.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
As requested by Morris Jobke, the passwords in the acceptance tests were
modified to make them valid both for a clean Nextcloud server and one
with the password_policy app enabled.
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
This tests whether a user can see navigation entries after enabling
apps. This includes the app's group restriction.
This currently expects that a group "group1" exists until we have code
to auto-generate groups.
This commit also provides a utility function
Page.multiselectSetSelection() to make it possible to select entries
inside a multiselect.
