FIXME: Reusing HelpDialog as a module entrypoint creates a circular
reference. It's harmless because the dependencies are added at
different times, but also easy to refactor away.
Change-Id: I3608a78baddf2376cc9eb4524625f4911c130c06
Each RL module comes with the cost of its name + hash being sent over every
request and in the startup module increasing the time to load the page.
Bug: T229791
Change-Id: I3ae230a84112a70ab369ffbff70e173b8f07ec1a
The patch adds a first package of node selenium tests including
test for the user and tag filters.
The classes for user- and tag-rows were re-added to have better
access via selectors.
Change-Id: I8c53d9c923820e177d83ee900cee08e93cd3f65b
This is a direct follow up to the change done in I885f8e7. I found the
code very hard to read after the change, and tried to rearrange it a
little bit. This is what I came up with. What the code does is entirely
untouched.
This also rearranges the replaceWith calls so the replacements are done
from top to bottom.
This patch also avoids calling mw.config.set multiple times.
Change-Id: I1cbce8e8aa030d6ba5cd7d19ab26fa363e13474b
I have seen this failing:
https://integration.wikimedia.org/ci/job/mwext-mw-selenium-jessie/9255/console
The idea is to pre-set this cookie so the help dialog never shows up
in the first place. The tests do not need to look for the element, click
it, wait for the dialog to close. This should even speed up the browser
tests, I believe.
Also this adjustes waiting conditions in several places to address
timing issues with failing browser tests.
Since these tests should be replaced by Node.js on the long run I
added "sleep" in the cases although it's not recommended.
Change-Id: Ida381faa390aa62ebaadd3713864676be80abd51
This patch mainly reintroduces the option to click on bars and move
the pointers with it. To do this, 'ghost' pointers are introduced
to show what would happen when bars are clicked. The pointers moved
differ depending on where the user clicks on a bar. Pointers are
still not allowed to change positions, so in some cases booth pointers
move with one click. See the task description.
The patch also includes some renaming and also refactoring of the
click handling in general. Furthermore bar hover mechanics are handled
by the RevisionListView class now.
Moving both pointers is not possible when it would push one of them of
limits.
Bug: T172092, T173566
Change-Id: I32a8256f7667e03081324d54accdf03a17454faf
Pointers can always be identified as the pointer for the "older" and
"newer" revisions. Class nameing was cleaned up and changed accordingly.
Change-Id: Id17878e780b5a405b196fdc5ec7429eeaef5bd4c
This patch fixes issues with the RevisonSlider assuming that the higher
revision id belongs to the newer revision. Min/Max methods to decide what
the diff and what the oldid is are removed and the usage of methods is
adjusted accordingly.
Also the test for switchover pointers is removed since this will not work
with the patch.
Bug: T164455
Change-Id: If5d9cbb8ebd872aee376d249942e6881c8edb984
The end border for the pointer lines can not be displayed because the
bars on the biggest revisions used all the height. This patch reduces
the max height of the bars slightly and refactores the calculating method.
Bug: T163436
Change-Id: Idfa6bdb9f85221063e13f3ec16b22fcfbe6cbd9c
The browser tests do not gain very much from the beta
feature tests and the default setting having the extension
only availible as a beta feature slows down the tests in
general.
Furthermore removing this will make it easier to let tests
pass for the temporary change in the slider logic.
Change-Id: I7377a6f11aa103af76d14fea3cbb34572495192a
The meaning of first in this context was totally unclear and not
very usefull. It turns out, that the "first" revision is always
the oldest and the last is always the newest revision.
Change-Id: I009438777908a5f4a8833f8f5cb7d3041057741c
This patch changes the way OOjs UI tooltips are constructed.
Until now we did not use the full potential of the widget.
Some strange behaviour remains and it seems we can not use
'horizontalPosition' here. Therefore the direction is passed
down to the RevisionsListView to center the tooltip manually
and margins are set for the help and arrow tooltips.
Bug: T159428
Change-Id: I0a67340da4d94f2f6b29e507f94fc695b44b71f9
The waitr click on the checkbox seems to be broken or is broken with
intend since the actual OOUI checkbox has an opacity of 0.
Additionally when_visible fails when the element is not present yet.
The the way to go here seems to wait for the element to be there
and then check for visibility.
Also made use of PageObject method for checkboxes, the name selector
and the link element.
Bug: T158074
Change-Id: I2c3d6d963c8bcdc420bfbc57c1c2380bce7ef2f1
This now attaches the revisionslider to the
older and new links that are by default in the
diff.
This means when clicking them the revision slider
will use an ajax call to reload the diff (while moving
its own pointes) rather than reloading the whole page
Change-Id: I11fa92828140f980ca8edee27403a4e2a5f5a6a7
Looks like custom settings used in a test for number separator
transformation added in Icc9bc09675151645055c7e7f706472a15cc81b0f
influenced RevisonSlider's tests resulting in test failure if
that core test was run before RS test.
Relevant RevisionSlider tests now are ignoring any custom value.
Any value before the test is restored after the test run, though,
in case other tests relied on this (which does not seem to be a case).
Bug: T153121
Change-Id: Ida81b93e4d081b6948e760995a92d182ad79fe0f
With the parent of this commit (current master)
Loading the RevisionSlider JS without it expanded increases
the request size by roughly 767KB on first request and 242KB
on subsequent requests.
The large size of the first request is mainly down the to
dependancies of the slider.
This lazy loading patch means the RevisionSlider JS
only causes an increase of 184 KB per request.
If the user has the bar expand by default the main JS will
be loaded straight away (and the lazy JS will not be).
This patch also means that when only the bar is loaded the
pin button to set auto expand will not be shown.
This will be added once the RevsionSlider is loaded.
Bug: T151668
Change-Id: I054a82e9ea2aa89326464632e744497239f7adba
- add Gemfile.lock
- use recent version of gems
- ditch README and link to wiki documentation
- clean-up environments
Change-Id: I76bb1d491911abbc0c0bb4387b436ab3ff2cebc2
- Applied the Given-When-Then rule
- Added separate accessors for disabled arrow elements
Bug: T146276
Change-Id: I0278ef607f96851ed69e7373b952f5d02139cb0c
This will avoid checking pointer data prematurely when
they are still moving and the diff was not reloaded yet.
Change-Id: I3c29a43940e40b7c338a00b3520e5aa6b237ea4a
- Applied the Given-When-Then rule
- Created alias for the "diff-has-loaded" step for readability
- Created higher-level step for loading the revisionslider
- Added assertions where missing
- Added ability to click the upper or lower area of revisionbars
Bug: T146276
Change-Id: Iad5abbb6e73a31afad1e2058b26ece3fbc7ad942
- Applied the Given-When-Then rule
- Created alias of "close-help-dialog" step for readability
Bug: T146276
Change-Id: Icc8aeb8720025bd3c6d6bc811713fac48a9919dd
- Applied the Given-When-Then rule to scenarios
- Added alias for one step for better readability of scenarios
Bug: T146276
Change-Id: I08aabda113d6ad9f1bbf06161484143549b13a7c
First I jumped on replacing both jscs and jshint with eslint but
it might be premature decision. Although linting with eslint
is possible (like in there is wikimedia config for eslint)
it is still not clear should it
But in case the change happens we will be ready.
Apart from config stuff this changes few bits spotted by eslint:
improves some indentation, removes weird spaces, completes some
doc blocks, changes IIFE forms in tests. These changes do not
seem controversial.
Change-Id: I9f8bf0f5745da8e662685f4cd879ea4baa609c01
This adds a button which allows a user to make revision slider
automatically expand on each diff page (disabled by default).
User's choice is stored as a hidden user preference.
This is a bit hacky as it squeezes a button on top
of another button (100% wide expand/collapse button).
This also adjusts styles of ToggleButtonWidget so
the button looks more like a frameless button, although
it is created as a framed one (to have inverted behaviour
when button is in its "on" state).
The button only gets visible when Resource Loader finishes
loading JS and CSS to avoid button jumping around the top
of slider window before "right" styles are applied.
Bug: T142196
Change-Id: Id561485344cba9b136666fe31b086151467de19e
* The jQuery for appending the 'darkness' mask was broken (not nesting)
* Use a more semantic name
* Simplify styling to just set opacity to 50%.
* Remove browser compatibility hacks as all MediaWiki's JS
supported browsers also support CSS opacity.
Change-Id: Id893a75bb90a4e6e2e8a26ebc3863de565d8a4ee
Localized format of date containing month name, day, year and time
is used instead of format enforcing the order of time and date,
and use of short month name.
Bug: T141167
Change-Id: I39b2844e777beb986cefb3097d52d481bb038a63
Instead of loading RevisionSlider only add a little button
to expand RevisionSlider on top of the diff page.
This makes RevisionSlider only steal a bit of space over the
diff, and only inserts quite a big slider to users that want
to have it visible for the particular diff.
API calls are only made once RevisionSlider has been expanded.
This is re-submit of b0f229d75f
that was reverted in I26427faaa00b38c2aa1377a66224c9062dcca302.
Bug: T141871
Change-Id: I879de5774b2cce7b908e73cbbe869fd48d6afa23
Instead of loading RevisionSlider only add a little button
to expand RevisionSlider on top of the diff page.
This makes RevisionSlider only steal a bit of space over the
diff, and only inserts quite a big slider to users that want
to have it visible for the particular diff.
API calls are only made once RevisionSlider has been expanded.
Bug: T141871
Change-Id: Ib312f6225b85b9ebdf4ac5d16e254a57d5cf6411
This changes the previous behaviour of fetching always up to
500 most recent revisions.
Now the extensions fetches N revisions including the newer
revision selected to diff as the most recent revision.
N is number of revisions that would fit in the current
window when rendered as bars.
When user is close to either "end" of the slider, extensions
fetches another batch of up to N older or newer revisions,
as long as user does not reach the oldest or the newest
revision of the page.
Among others, this removes the limitations of the previous
approach: showing only 500 revisions, and failing to show
anything when any of selected revisions was older than
500 recent revisions.
This change also simplifies usage of Api class.
Bug: T135005
Change-Id: Ib3f4a6ac57ff17008f9d8784c4716bd294443096
After fetching a batch of revision data, user names are extracted
and another API query is made to get gender preferences for users.
This change also moves a code responsible for MediaWiki API calls
to its own class.
Bug: T136367
Change-Id: Id11fe14e9ca37829141ae92b13b51f10f992eb96
In particular, link to a Meta page on the related community wish
should open in the new window/tab.
Bug: T140875
Change-Id: Ib0d7bf679c9ac8a3ee2fdf2289fc53cfdfdaa671
This adds tests for navigating the timeline using the back
and forward buttons.
It also makes sure that the buttons are disabled / enabled
at the correct times.
Bug: T133278
Change-Id: I9834dd3d3bf107557b15fdd105a946902d53b8f2
Despite its documentation the "constructor" of RevisionList tends
to expect array of revision data in a format returned by API instead
of array of Revision objects.
Due to different name of ID fields in Revision object and in API array,
if "real" Revision object is passed to RevisionList, its ID is lost.
This changes RevisionList so that it only accepts an array of Revision
objects. This provides better abstraction.
This is basically a revert of I147270f28381038d05f8bcfd2317e8c269b2e458
which aimed at the same problem but suggested solution doesn't seem right.
Change-Id: Ic45cdd3e7b707a8c6a19eecf0a84d4c11696cd1f
This allows an arbitrary number of revisions to be created
on a page very easily.
The page created will have a name relating to RevisionSlider
suffixed by a random float.
The base logic for this is taken from:
Ic165f108517c534a4b1c12883621b3624c7b0336
This patch also removes the @vagrant tags as RevisonSlider
is not yet in vragrant.
Please see T139296 for the ticket to add it.
Bug: T133278
Change-Id: I8a25db411bee2fa44555a23d1a840907a185c998
Testing the visibility / presence of things around
the dialog appears to be hard and the DOM for the dialog
remains unchanged when shown / hidden as far as I can see.
These tests test that the dialog is only shown once per
user.
And that the sequence of the dialog works in regards to the correct
buttons being present at the correct times.
Bug: T133278
Change-Id: Ia86cb69573da3e888c2897b8b50f1d2e5c61d8d4
There seems to be no good way to enforce language on the QUnit tests. It works
for the number formatting by setting the user language on the fly but for everything
else it fails. I simplified some tests and skip some tests or asserts when the language
is not 'en'.
I also added a helper to have conditions for skipping tests. This should be part of
QUnit IMO.
Still I dont know if this is something we should care about at all or just assume, that
devs will know what they do when they try to run tests.
Change-Id: I624ccf3984434e06200abffe7ccdd25b396251af
