We found that Popups was overcounted approximately 2.5x relative to Cite.
This patch attempts to nearly match the circumstances under which Cite
(and its tracking) is loaded.
Bug: T214493
Change-Id: Ib31df3c33879f4ea63d9808ffd260861069a8977
The Popups pageviews were suspiciously high, about 2.5x higher than
Cite counts which should have reported the same number. This patch
attempts to deduplicate pageview events.
Bug: T214493
Change-Id: I51bf6d1909d65ecd9d3ef28eda285852897343ec
We want to know the side of this population in pageviews, in order to
produce proportional metrics later.
Bug: T214493
Change-Id: I940872870754e85d19688f3855d6857e65b982b1
We're getting a few dozen records per day at 1:10, so let's stop sampling.
We need enough data to get significance on events happening at 0.1% or less.
Bug: T214493
Change-Id: I8a913fe1ee1e5b72d84914e183ac2386ddb20d84
Notice how this actually reduces the size of the final, compiled index.js.
It's not much, but still.
One issue I noticed is that the coverage reports for the JS code stopped
working. I have no idea why.
Bug: T208951
Change-Id: I2fe92579574b3b1ba4d2dd064899eee944045a96
The signature mw.RegExp.escape() is deprecated since MediaWiki 1.34.
By the way, this is the 4th or even 5th time in my short career this
tiny, single line (!) helper function is moved around and I need to
update all my code. This felt already ridiculous when it happened the
2nd time.
Change-Id: I4d05a49120aff64ebc316d0af7736c62385d9307
We've logged zero events after 12hr of deployment, so it should be
safe to increase sampling by 100x.
Bug: T214493
Change-Id: Icd67aed448269f603dd4465f7e46eac9a64bd1ab
Logs events to the ReferencePreviewsPopups EventLogging schema, in
order to understand whether Reference Previews is helpful for
end-users.
This will be removed along with the older tracking, as soon as our analysis
phase is finished.
Incidentally disables a lint rule for the generated JS, it's about
readability so irrelevant to the minified code.
Bug: T214493
Change-Id: I2638611ba67b785338f7e98a1c4b08a5e829812d
The eventlogging `isEnabled` function determines when to sample,
this patch removes the `isAnon` conditional so that we can measure
reference reading habits for all users.
Also guards against a non-function navigator.sendBeacon, which was
previously intended but incorrectly tested.
Bug: T214493
Change-Id: I42cb3082fb85c7900426a2055dfa3c2f6ecfd968
I was able to track the issue down to the changes made in this patch:
https://gerrit.wikimedia.org/r/331563
This patch is mostly a simplification of ce8a2d4
(I9a73b3086fc8fb0edd897a347b5497d5362e20ef):
- Don't make wait#wait() abortable. This adds complexity and isn't
needed since the token is rechecked in the .then() of
actions#linkDwell(). The request is permitted to continue and fetch if
that token still matches and is never issued otherwise.
Once a request has been issued, that request is still abortable.
However, note that calling XHR.abort() is just a request to abort and
may not be granted. Whether or not XHR.catch() is invoked is what
dispatches the FETCH_ABORTED action (or doesn't if the request to
abort was denied).
- Remove the abort tests for wait#wait() since the functionality is no
longer provided.
- Pass the preview token in the FETCH_ABORTED action and reduce
FETCH_ABORTED the same way as ABANDON_COMPLETE in reducers/preview.
The follow-up patch, I3597b8025f7a12db0cf5d83cce5a77abace9bae3, adds
integration tests for the specific bug fix. Note that these Selenium
tests are incompatible with the content proxy, so it is probably best to
simply unset $wgPopupsRestGatewayEndpoint and $wgPopupsGateway and allow
the defaults to be used. Also note that the tests are incompatible with
recent versions of Node.js (so use NVM) and emit many deprecation
warnings (so set deprecationWarnings to false in
tests/selenium/wdio.conf.js) An example run of the tests looks something
like:
chromedriver --url-base=wd/hub --port=4444 &
# If any changes are made locally, also run `npm -s start &`.
MW_SERVER=http://localhost:8181 \
MEDIAWIKI_USER=foo \
MEDIAWIKI_PASSWORD=bar \
DISPLAY= \
npm -s run selenium-test
Live testing may be performed as well. Remember that RESTBase requests
are incompatible with MobileFrontend's content proxy hack so ensure to
comment it out if $wgPopupsGateway is configured for RESTBase (see
T218159).
1. Open the DevTools network tab.
2. Disable the browser cache. Chromium, at least, won't abort requests
coming form the cache.
3. Hover back and forth quickly over a preview. In Chromium, canceled
requests are labeled and appear red. This is a good scenario to test.
With the patch, a preview should always be shown when ultimately
resting on a link. Without the patch, it is possible to rest upon the
link with no preview showing. This may require several attempts.
Bug: T219434
Change-Id: I9da84b0296dd14e9ce69cb35f1ca475272fb249a
As mentioned in T205744, EventLogging schema ResourceLoader modules have
been deprecated. This removes those modules from loader calls.
Bug: T221281
Change-Id: I1b7355c69e09756f50ccd1c1955b45cae4a64b9e
Removing wrong ARIA `role=tooltip`. Those are meant only for static
text content on short `title` like dynamic tooltips.
Bug: T223827
Change-Id: Ib4f490c7ad421e516fb0cf47eff4335bcaf26c43
We did a strict === comparison before. This works fine when the page
zoom is either 100% or 200%. Other zoom factors produce fractional
numbers. Comparing them with === is doomed to fail because these numbers
are IEEE representations, not necessarily being identical in situations
where you would expect them to be identical.
This applies two fixes: Change the comparison to a >=, and always ignore
1 extra pixel. If we are so close to the bottom, the gradient is not
helpful any more.
Change-Id: Idf4c604142de8d2a803e1d94f3093cec8a8abb72
This fixes page previews appearing when there is self link to a page
that has a file extension in it's name.
Bug: T222869
Change-Id: I5caf610fa76986026948a5b7b55537723752b755
This is relevant in case the HTML looks like this:
<… class="reference"><a><span>[</span>2<span>]</span></a></…>
The additional <span> cause many additional mouseover and mouseout
events. The code already tries to filter these duplicate events that
are all triggered on the same link, but gets confused, especially when
the multiple chains of events overlap each other in unexpected ways.
It's a timing issue. This change does not fix the fundamental issue,
but does make it much less painfull.
This patch also removes the > from the selector. This is in case the
HTML looks like this:
<… class="reference"><span><a>[2]</a></span></…>
The additional inner <span> would prevent any reference preview from
being shown.
Bug: T214693
Change-Id: If2554ba78072245c27a1f85c46f33e3c58582c1d
Introducing the REFERENCE_CLICK action that will fetch and show the
preview for the clicked reference right away without any delay.
The main goal of the new chain of events introduced with the reference
click is showing the reference preview right away. The actions triggered
by the dwelling include delays in multiple parts of the process.
If there's a dwell action-chain in progress when the click action is
executed, the related promise ( that might still include steps with
delays ) and the reference preview is retrieved and shown right away
re-using the token.
In the case where there's no dwell action running ( e.g. when the click
was triggered via touch ) we create a new token and start from scratch.
In either case we want to avoid, that multiple clicks trigger multiple
actions and abort early when there's already a click action in progress.
Bug: T218765
Change-Id: I073a93be2d17a21178aebe12267765f60a2811b9
I actually tried to bring the animation for the "bottom" property back.
When finished, I realized this animation is done on *top* of the
horizontal scrollbar. This looks fancy, but is not what we want. We
need to keep animating the opacity.
This patch here contains a bit of refactoring that was left after I
went back animating the opacity.
Bug: T220200
Change-Id: Icf613f72f3baa3de86f8aa319667c8e8f16593fd
The changes in the patch check if a horizontal scrollbar is present
and move the absolute positioned fade overlay above that scrollbar.
Since we cannot change the CSS of the :after element via JS a new div
element was introduced.
Bug: T220200
Change-Id: Ia69c9be0facaf3ecebdb9f76d36f7cb3412c0816
Pretty much all usages of this function do *not* use the second
parameter to pass a page name prefixed with a namespace, but pass the
page name only.
This patch here removes the redundancy. The namespace prefix is now a
generated one, saying "Namespace <number>:…". It turns out no test
relies on this so far.
Bug: T220097
Change-Id: Ibd45d49c91e86a2647afe676a5e3bb07dfeab6ed
config.get( 'wgTitle' ) returns the unnormalized title of the current
page while title.getName() gets the normalized title (e.g. with underscores ).
On pages with spaces in the check failed before this patch.
Bug: T220097
Change-Id: I58a532627bb27be030cbc553f1181a89109edd80
Named references may include non-ascii characters, so we decode the fragment before matching against reference IDs.
Bug: T220196
Change-Id: I63bba59fa8f0f6aa95aeadbb1f85745d480988bd
It is set based on the same conditional that loads the code,
thus checking it inside the loaded code is a no-op and adds
extra HTML to the <head> that blocks text/layout rendering and
delays fetching of Popups JS.
Bug: T219342
Change-Id: I9c1f4b3861ce2cecb654eb0a78469a616730a40b
I had to disable ESLint to be allowed to upload this patch. It starts
complaining about something in code I did not even touched. The error
message does not make any sense to me (something about globals being
forbidden in code where I can not spot anything that would be remotely
global).
Change-Id: I6d4b178a65126c4b81b87d99142a6cdc845ae5ee
When a thumbnail in portrait-mode is narrower than the 200px
expected width, the SVG clip-path should be shifted on the x-axis
in order to align with the thumbnail image.
Adds extra test-cases to validate this logic.
Bug: T204627
Change-Id: I9359c9fb335e5fad3f7d5ba33ee89d2a1f26b8b2
We run into this issue already one time, see Ifb5fe89 (T214710). The
exact same applies here.
The effect of this bug is that for certain references (typically ones
with a colon character in their name) the "jump down" link will be
broken.
Change-Id: Ic6723bd910cb5e5e1e1872ce39f2e271012245de
During story time on 2019-03-12 it was decided to consistently talk
about "References" in all messages. Main motivation is that this is
the term the community is most familiar with, and it is also the term
that is used in reference section headings most of the time.
Bug: T215063
Change-Id: Iaab8d2c0da1546a3c9d27bc8e2e1c784050ed135
The method itself has not much to do with gateways as such, it's
more about the general preview type selection. Since the preview
types "live" in the model, I thought it might good to move it there.
Doing that the "original" getPreviewType method in the model was renamed
to avoid conflicts. If I get this right, that method is quite specific to
page previews, since it processes the output from the TextExtracts API-
request. - Therefore I also removed the TYPE_REFERENCE there, because this
code path will never be reached with that type afaik.
Inspired by the comments in Id1fa7dad59d8fe80bc60c1e2d7c3fb4087e52d1f and
as preparation for that patch.
Bug: T215420
Change-Id: Ic9e24a73e945c7d56435c656ecfdb42b65601d22
This happened one time before and was fixed via T212419. It now came
back after I2670331, a patch for T214861 that made the title parser
accept more links than before.
This patch implements a decision tree that goes as follows:
1. If it's a link to another page, it can't be anything but a page
preview, as having a peek into another page is effectively the
definition of a page preview.
2. Otherwise it's probably a reference preview, but only if:
* It's a link with a jump mark.
* The link is actually marked as being a "reference".
* Reference previews are enabled.
3. If neither of these definitions fit, do nothing. Note this is not
"hiding errors" we would be able to fix in this code. There are many
ways to manually arrange wikitext in a way that it looks like one or
the other, without fulfilling all requirements. Unfortunately the
user who would see error message or dysfunctional popups would not
be the same as the user who wrote such wikitext.
Bug: T216683
Change-Id: I8d021f19ddc73a261e6a0c62959ddd0cb1d3182d
While reviewing the code of the reference endpoint I found it encodes
two CSS class names, "mw-reference-text" and "reference-text". So far
our scraping gateway only implemented one.
https://phabricator.wikimedia.org/diffusion/GMOA/repository/master/
This patch also extracts a piece of code to a named function. This makes
it much easier to read, I feel.
Bug: T214908
Change-Id: I9d1bb1f4c21eb9d57a6b763ca1f756e6cf7049e0
These icons are currently unused because the gateway does not deliver
the necessary information. This will be used starting with I6223cbb.
This patch aims to introduce all resources needed by the later.
Bug: T214908
Change-Id: Ie0c3c059222700169f2605c3123554c74d974256
The definition of "self-link" in this context is an <a href="…">
element that points to the exact same URL as the current document's
location, excluding a possibly different #… fragment. This is typically
the case when the <a> element does not contain a full URL, but something
like href="#fragment". JavaScript's HTMLAnchorElement.href getter
automatically expands this to be a full URL.
Example:
var a = document.createElement( 'A' );
a.href = '#test';
console.log( a.href );
Notes:
* This new code assumes the wgPageName setting properly reflects the
page name requested via the current document's location. Core does
give us this guarantee.
* The only URL element that is intentionally not compared is the
protocol.
* This accidentially fixes T215899 as well, because the namespace check
is now bypassed for self-links (as it should).
Bug: T214861
Bug: T215899
Change-Id: I2670331cbbdebf7dc9fc70d7342724534f9f54ec
According to the standard.
Note: In a previous patch I removed the <?xml …?> line. This conflicts
with what https://www.mediawiki.org/wiki/Manual:Coding_conventions/SVG
suggests. However, I think the removal is ok in this particular case
because this file will never be shown as an image, and never be shown
standalone. Instead, it will be inserted in the documents DOM. The XML
header doesn't matter anyway then.
Change-Id: If23ad54985abb30f8c92500546bd04eeca44fab3
I decided to keep the comments because they are sooo helpful, but
tried to shorten them a bit. The biggest change is the indention with
tabs and the much more compact <path> elements. The shapes are the
exact same. I manually confirmed this for all four.
Change-Id: I2d1294c9ae7e398dcbe2d111c42848d17be8a67e
I tested this with all 16 possible combinations:
* The pointer can show up in all 4 corners.
* The popup can contain a thumbnail or not. The code for the pointer
is very different then, because the SVG masks are only relevant in
this scenario.
* The thumbnail can be tall or not.
* I even tested tall popups without a thumbnail. This is a combination
that is impossible in production scenarios.
I found 3 issues. This patch fixes 2 of them:
* The pointer is misplaced in the bottom-right corner when the popup
does not have a thumbnail (as reported in T215194).
* The pointer is misplaced in the upper-right corner when the popup
shows a thumbnail.
* The pointer in the upper-right corner is gray instead of white when
the popup is tall, but does not show a thumbnail. As this is not
relevant in production, I did not fixed it.
It seems both misplacements are because of the same reason: For some
reason, calculations are done based on the assumption the popup would
be 300px wide, but it is 320px wide.
Note I did not just added 20 everywhere, but manually alligned the
pointer triangles so they are placed exactly the same distance from
the corner as in the three other corners.
Note I did not tested this (yet) in RTL scenarios.
Bug: T215194
Change-Id: If0ca63d4d4b6e8083c7de1517fe32f49671a40e6
As well as:
* Simplify the selectGatewayType() function a bit. Or was this
intentional?
* Remove a TODO we don't need any more, at least not in this file.
Change-Id: I5528f0012cbaf8b4e88e22c7e2a8d87bf027e8f1
We discovered a bunch of possible solutions (see previous patch sets),
including replacing the `$( document )` selector with a more specific
one. That idea does not pass the linter.
Very late I realized the original selector starts with
`#mw-content-text`. This heavily limits where popups are allowed to
appear: really only in the main text content area.
We should limit reference popups to the exact same scope.
This fixes the issue described in T215195. Before, the content of the
popup was covered by the selector. Reference links *inside* the popup
would trigger another popup, which makes the current popup disappear.
Now the popup itself is not covered by these event handlers any more.
Bug: T215195
Change-Id: I142aee68abbd57ca321873855fef9209e0db0bbf
There are a few cheap checks done before this regular expression is
even needed, most notably the check for a pretty URL (without query
parameters). Since the vast majority of links processed by this parser
are pretty, I believe this optimization is worth it.
Change-Id: I730b87dc010161e8bc3f311c517293c0ad553326
This is now possible since the render functions return jQuery objects.
All this code is exclusively used in the pagePreview.js file, and
doesn't need to make the already very big renderer.js file even
bigger.
Note the tests for all renderers have always been collected in a
single file. That's why the test case does not move.
Change-Id: I0c24638751c5f0e93d2bc0f3f4bb61fa0cf50d15
