When the user sets a different language than the projects content
language, ReferencePreviews were rendered LTR. By using the bdi
tag this behavior is fixed.
Note that PagePreviews gets language and direction from the
TextExtracts response and use them to achive the same. The latter
is not straight forward available for ReferencePreviews so I'm
using bdi to achieve the same.
Bug: T345833
Change-Id: I6554e6e1db3aac06c04c9185500e05de238adde2
The rate is unexpected: over 5 million in the last 7 days
I don't time to investigate this further so suggest disabling this
for now.
Change-Id: Ie55d1da1652239ad4f6ea94743d6efdd12de9974
Apparently caused by a mistake in I8697232, and the QA for T340256 not
covering the full functionality of what was changed in the patch.
Bug: T345829
Change-Id: I482d32183ef7f2226c7ad114be0fbaf97bc7b898
- Removes .mw-ui-icon and .mw-ui-icon-element
- Aligns markup/styles between generic and disambiguation previews
- Update padding for generic and disambiguation previews to be the same as other popups, results in minor visual changes
Bug: T341899
Change-Id: I9a58fc6a93160d07452ea6f903e1797dd9421d92
Dwell delay varies between popup type so should be part of the
definition. Since this is needed by reference tooltips we should
make this part of the Page preview model. This will allow us to
separate concerns of the architecture and reference tooltips
for better maintenance between WMDE and web team.
Bug: T326692
Change-Id: Ia021c8b90fd69d70c4b33744520bb963a65db036
Use the topic "error.web-team" - error prefix necessary
for it to be treated as an error, and 'web-team' moves
it off main channel.
Error logs tagName and nodeType which should be sufficient
for us to understand this issue. Can expand with other
information later if need be. Cannot add to stack trace
as any non-standard stack trace would be dropped by the intake
code.
Bug: T340081
Change-Id: Ia011aaf9f8b5b932695da3311f849682c0105cfe
- check if the element has the function in it before using it
- implementing a native code to get the same result based on polyfill examples
- fix Coverage Block
Bug: T340081
Change-Id: I1c2ce46a88cde6323ab862964062dd722ff5edcb
According to the docs:
The Element.append() method inserts a set of Node objects or string
objects after the last child of the Element.
String objects are inserted as equivalent Text nodes.
When given a string we want to treat these as HTML
Also fix the storybook which hinted at this being a problem
(incorrectly fixed in 504eefcc1)
Bug: T341318
Change-Id: I415c9821a6a37b45cf0f8a65f58a41bc7d0a042d
The footer link shows when a page preview has been disabled
by an anonymous user via the settings cog.
Change-Id: I236a620322716f42443fc3d90a0405352132a99f
Previously we did a try/catch when closest didn't exist. Now that's
gone we're flagging this issue.
Follow up to Ia45c03d1fd6949bf83ebed6d40075e453e42cdd7
Bug: T336650
Change-Id: I7a4f453ae1a4a8222863a8342a27d615883d339a
For fetch and AbortController we provide native polyfills (see
resources/src/skip-web2017-polyfills.js) so safe to use this here.
This will be empty for modern browsers.
Change-Id: Ic0f55eb0a0276be3587a4b866834bddff1124ad2
Rather than obscurely loading instrumentation code as a side
effect of loading UI code, run it explicitly inside the index.js
initalization code. Instrumentation is moved to its own file
and Popups modules now support an init function.
Change-Id: I9d2643ec8fb4e1dedc7ab9534b2965272f12335f
eventlogging reducer was removed in
I640ab367cd235ab8da7dd70dbef7ae9076712e84 so this is syncing
a non-existent value.
Change-Id: I58386bf9a0219f6344c08f757af3aaafbae92751
Certain browsers have the closest method but do not
support not selectors with multiple arguments. This
variant caters for both.
Bug: T325113
Change-Id: Ib5fc912bfe0f831fea4c9882c25b27541d83b66f
Follow up to
Iefe98c1f0422dbf034e385b1a41a859d030a2cf4 where we switched from the
jquery event delegation pattern to native methods. One thing that we
overlooked was that we also need to consider the case where the selector
matches the parent of an element, for example a span nested inside an
eligible link. I've rethought this logic, to first find the closest
eligible element to normalize the element passed to model methods
before running matchesSelector.
Bug: T325007
Change-Id: I4133751dc900a51829173e9c0d965cbb18e6a33e
Adjust the logic of elementMatchesSelector to fallback
to jQuery when the native matches fails.
Follow up to
Iefe98c1f0422dbf034e385b1a41a859d030a2cf4
Bug: T324514
Change-Id: I0e9e894e77e2eb29fee65853aa98b141bc2a11a3
Allow extensions to register new types of previews via
extension attributes.
Changes:
- The check for reference previews doesn't make sense
as $('a[ href*="#" ]' ) will match any elements with a hash
fragment, so the additional check to Title.getFragment
will not provide a different result. This was introduced in
I9ec57e0fbb0d21beaaa7b359c1c2bef64d2c14f5
- Links that point to themselves are marked with mw-selflink
in MediaWiki so this can use the not selector we already have.
- The new API is used internally and only available via extension
Attributes
- An example is provided in SkinJSON
(https://github.com/jdlrobson/mediawiki-skins-skinjson/pull/14)
Bug: T233099
Change-Id: Iefe98c1f0422dbf034e385b1a41a859d030a2cf4
eslint:
* Remove hardcoded glob paths that are (relatively slow) to expand
before linting could start, in favour of specifying directories
which can be iterated while linting happens.
* JSON files in i18n/ were skipped, unlike other repos.
JS code under .storybook/ was skipped.
JS code under resources/ext.popups/ was skipped.
Fix all these by doing what other repos do, which is to let
ESLint iterate the repository and tell it which directories not to
enter. This has the happy side-effect of making IDE integrations
for ESLint work correctly (as customisation in package.json has the
same problems as using Gruntfile, namely that invoking 'eslint'
directly can't be aware of this), as well as allowing things like
`eslint --fix` to be used.
```
/Popups/resources/ext.popups/index.js
4:2 error Unexpected var, use let or const instead no-var
```
nyc:
* Remove SPAWN_WRAP_SHIM_ROOT.
This was a hack for an early CI experiment that hasn't been
needed for several years.
storybook:
* Remove gitignore entries for additional npm and build output in the
storybook subdirectory, which appears to not exist anymore since
189b386a13.
Doxygen:
* Sync with cookiecutter example from
https://gerrit.wikimedia.org/g/mediawiki/tools/cookiecutter-library/+/HEAD/
- remove redundant stuff,
- enable quiet mode to hide verbose non-warning,
- include CoC.
Change-Id: If8f6b833067192aea96a87f04c7978c9af11f996
This was done while working on T277639 where we introduced
multiple "enabled" flags for individual popup types. This
change listener is one of very few places that work with
this flag. This patch is meant to make this code more robust
and easier to change.
A few unrelated but trivial changes are included that make
code shorter and hopefully easier to read.
Also fixes a bug with tooltips overlapping previews.
Bug: T287119
Change-Id: I7fb0a8d4bb9f5e78fe62cfca524cc157ea89a233
Provide containers for page previews examples
In storybook we avoid position absolute but this deviation from
how they behave in reality could lead to subtle differences that impact
the rendering of previews. It also doesn't allow to see the preview
in the context of the link which is an important part of visually
verifying the position of the pokey arrow.
This also allows us to rethink the broken RTL mode. We now scope
LTR rules to a LTR class that is present on the container, and use
the CSSJanus parsed stylesheet for RTL.
Change-Id: I189019824ddd6aa759790fd162ffcd543619a953
The requirement for landscape images is to be at least 320px
wide. The requirement for portrait images is more relaxed,
only 250px high. Images that fall between these two
requirements currently don't show a thumbnail, even if they
could.
This change affects a very specific group of images:
* Square images from 250 x 250 to 319 x 319.
* Landscape images from 251 x 250 to 319 x 250.
* Landscape images from 319 x 250 to 319 x 318.
The most extreme ratio is 319 x 250. This will be cut to
203 x 250. I.e. the absolute extreme are 58px missing left
and right, but never more.
Requested at https://www.mediawiki.org/wiki/Topic:Vwl97pm6as9fuf6k
Additional stories for testing more extreme small images:
* Small Tall - 300x1000px
* Small Short - 300x200px
Update tests accordingly.
Bug: T268999
Change-Id: I811f1c0e7e9b0c30280b36a61cc7831a5b9e58c8
Turns out this was only in place because the test was
(intentionally) incomplete. But it's dead code in
production. Let's get rid of it.
Change-Id: Ieeb145b6972dceb7eeda3a167d907b680d5c3ce4
For example: We know the HTML element we are dealing with is
always an <a> element.
I believe the small changes in this patch are all obvious and
non-controversial. Please let me know if you disagree.
Change-Id: I9fe15845affffdd0f5f0fd6ef7d6b607cb567ac7
This is not really an error. It does not really have
consequences for the user, except:
* It might waste bandwidth because the module is loaded, but
unused.
* It might waste a tiny bit of CPU time.
However, this can only happen for registered users, and only
with very specific combinations of settings (notably: a
conflicting gadget must be activated).
The reason it's marked as an error is that previously we could
give more guarantees about this. This changed recently. We will
check if the log message is still helpful and either remove it,
or restore the previous guarantees.
Bug: T271206
Change-Id: I218726c9c4879a405acef68710e79c6ac8d070fe
