Use `DOMCompat::getBody( ... )` as a nicer getter than
`->getElementsByTagName( 'body' )->item( 0 )`.
Remove overly defensive checks and redundant annotations on its
return value. Since we're dealing with HTML documents throughout,
the document body is guaranteed to exist.
We previously needed some of them to convince Phan when it thought
the body may be null, but this seems to no longer be needed.
Change-Id: If7aee7b6adbfa78269c7ba28b26a6eaa21fe935b
Use DOMCompat::querySelectorAll() instead.
CommentModifier::isHtmlSigned()
* Copied the CSS selector from the JS equivalent function.
CommentUtils::unwrapParsoidSections()
* Copied the CSS selector from the JS equivalent function (in VisualEditor).
CommentItem::getMentions()
* Trivial.
This causes Phan to report some more issues, which are also fixed.
Follow-up to 25272e7a4a.
Change-Id: Iaf1222f7114916f2eca19942c3686168899486fd
We can just use insertBefore() normally. There was never a PHP bug,
but rather a Phan bug, and it no longer affects us.
https://gerrit.wikimedia.org/r/c/mediawiki/extensions/DiscussionTools/+/596813/70/includes/ImmutableRange.php#373
This reveals a bug in CommentParser where it sometimes produces
incorrect ranges (we were incorrectly treating `false` like `null`,
hiding the issue). I'll fix it in a separate commit.
Follow-up to 25272e7a4a.
Change-Id: I4afba38f1d82ddbf8732bfe3e4d4f6ebe2f8de5d
In case 4 and case 6, no notifications are expected. In all other
cases we now get the expected notifications.
Bug: T285528
Change-Id: I9e813bb3a053bc1232783f9eae1ad75672b4fa7e
These changes ensure that DiscussionTools is independent of DOM
library choice, and will not break if/when Parsoid switches to an
alternate (more standards-compliant) DOM library.
We run `phan` against the Dodo standards-compliant DOM library,
so this ends up flagging uses of non-standard PHP extensions to
the DOM. These will be suppressed for now with a "Nonstandard DOM"
comment that can be grepped for, since they will eventually
will need to be rewritten or worked around.
Most frequent issues:
* Node::nodeValue and Node::textContent and Element::getAttribute()
can return null in a spec-compliant implementation. Add `?? ''` to
make spec-compliant results consistent w/ what PHP returns.
* DOMXPath doesn't accept anything except DOMDocument. These uses
should be replaced with DOMCompat::querySelectorAll() or similar
(which end up using DOMXPath under the covers for DOMDocument any way,
but are implemented more efficiently in a spec-compliant
implementation).
* A couple of times we have code like:
`while ($node->firstChild!==null) { $node = $node->firstChild; }`
and phan's analysis isn't strong enough to determine that $node is still
non-null after the while. This same issue should appear with DOMDocument
but phan doesn't complain for some reason.
One apparently legit issue:
* Node::insertBefore() is once called in a funny way which leans on
the fact that the second option is optional in PHP. This seems to be
a workaround for an ancient PHP bug, and can probably be safely
removed.
Bug: T287611
Bug: T217867
Change-Id: I3c4f41c3819770f85d68157c9f690d650b7266a3
For compatibility with Parsoid's document abstraction (Parsoid may
switch to an alternate DOM library in the future), don't explicitly
create a new document object using `new DOMDocument`; instead use
the Parsoid wrapper `DOMCompat::newDocument()`. This ensures that
the Document object created will be compatible with Parsoid.
There are a number of other subtle dependencies on the PHP `dom`
extension in DiscussionTools, like explicit `instanceof` tests; those
will be tweaked in a follow-up patch
(I3c4f41c3819770f85d68157c9f690d650b7266a3) since they do not affect
correctness so long as Parsoid is aliasing Document to a subclass of
the built-in DOMDocument. Similarly, the Phan warnings we suppress
do not cause runtime errors (because of the fixes included in
c5265341afd9efde6b54ba56dc009aab88eff83c) but phan will be happier
once the follow-up patch lands and aligns all the DOM types.
Bug: T287611
Depends-On: If0671255779571a91d3472a9d90d0f2d69dd1f7d
Change-Id: Ib98bd5b76de7a0d32a29840d1ce04379c72ef486
The user interface only allows you to subscribe to level 2 headings.
But we would generate events for whatever heading was the closest,
If it was e.g. level 3, no one would receive that notification.
Now we generate events for the closest level 2 heading, or we don't
generate the event at all if there isn't one (if the only headings are
of level 3 and below, or level 1, or if the comment is added before
the first heading on the page).
Bug: T286736
Change-Id: Iae99853070e353ab81c9cc29ef1d53c877adfc66
We used an internal API requests to fetch page content because it was
easy, but there's no way to guarantee that it returns data from the
primary database.
Use ParserOutputAccess::getParserOutput() to fetch from cache if
available. Also, use canonical output instead of user-specific,
not that it should matter.
Bug: T285895
Change-Id: I7dcd9659be77746dc2a0c4eeae2319887936b555
…without making the topic subscriptions feature available in user preferences.
Follow-up to these commits, which added these checks in ad-hoc ways:
* 9420f22e9d
* f3422f40a6
* 23a490deca
* a555db7892
Bug: T284491
Change-Id: If2e3fb1e06d1cc489fbca14796ed77c83bb52991
If the revision from which we generated the notification has been
deleted, we shouldn't include the content snippet, nor the direct link
to the comment (because the fragment ID is generated from the content).
This matches how Echo handles mention notifications.
Change-Id: Ica939f3a4efd39d0c295511d58280d3f9d584129
As it happens, most of Echo does not actually parse this message,
but it is for some reason parsed in HTML email notifications.
Change-Id: I414cd242d9bcc4d8b5a1c2a2a71be9e5f00ea8be
We don't display [subscribe] buttons on your user talk page,
but the API still allows those subscriptions.
Use the same approach as for mentions to ensure this doesn't cause
duplicate notifications.
Remove some code in SubscribedNewCommentPresentationModel,
now guaranteed to be unused.
Change-Id: I99a276a48d8562552ed2c54cc0323e8e428845fd
Otherwise, the global context is used (RequestContext::getMain()),
which is undesirable when you're building a rubegoldbergian
contraption and we're already inside an internal action API request
with a fake context.
Change-Id: I01daf8dc70b5751bc1e157fe598988cd5d3219e5
Per Manuel Arostegui in T263817#7033384. The limit is 5000.
(I picked it arbitrarily, there's no real rationale for it.)
Also log a warning when any user reaches half of the limit,
so that we might make a decision about changing this mechanism
before it starts affecting users. Maybe at that time we'll
have data to show that it's safe to remove the limit.
Bug: T263817
Change-Id: I18a8ee0ad7383759229c5721d5253fb591457d4d
Using `updateCacheExpiry()` in this way appears to be established
with examples of other use in WMF production such as:
- CategortyTree extension:
custom cache expiry for pages with `<categorytree>`.
- RSS extension:
custom cache expiry for pages with `<rss>`.
- intersection extension:
custom cache expiry for pages with `<DynamicPageList>`.
- Math extension:
custom cache expiry if `<math>` failed.
- Wikibase extension, Flow extension:
no caching for certain namespaces or content types.
- Graph extension, Kartographer extension:
via onParserAfterParse hook, no caching if on preview.
Bug: T280605
Change-Id: Iea41ab8599ffae4622c97d682258b1b64eaf9ba2
Previously we relied on the NamespaceInfo check below to reject
special pages, but after commit 07d885248bc54bdc0f12d9745916c794d45ec81c
in MediaWiki core, PageProps throws an exception when called with a
special page.
Bug: T281180
Depends-On: I32c94107fde96b9d6344c77b621be9b3b9b7faaf
Change-Id: I0fd893c63a6e92f6c84e7aa92270852e1137fcad
The issue occurred when replying to a comment consisting of multiple
list items, starting with a <dt> (instead of the expected <dd>), so
that the comment is considered to be unindented.
Modifier tried to add the reply directly inside the list (<dl>) rather
than inside the last list item (<dt>), which caused it to be confused
about indentation levels and try to un-indent more times than there
were indentations.
The simplest solution, given the existing code, is to add the reply
outside the list instead, in a new list. This results in a "list gap"
(<dl><dt>...</dt><dd>...</dd></dl><dl><dd>...</dd></dl>), but I think
it's acceptable for this rare case.
There are separate tests cases for old Parser and for Parsoid HTML,
because they parse the original wikitext differently (with the old
Parser producing HTML with a list gap too).
Bug: T279445
Change-Id: Ie0ee960e7090cf051ee547b480c980e9530eda51
The regexps needs to be non-greedy, otherwise it could swallow up a
large chunk of the page (up to the next comment).
I noticed this when adding tests for this code, in the 'unclosed-font'
test case (Ief9648b8805fadcc170c54b627eb669cc8b907b6).
Change-Id: I5f67a9599b0cb07bdd53abeebac9ada221181b66
If curLevel or desiredLevel are calculated incorrectly, this loop
could never end.
In JS, something would throw an exception before going infinite, but
PHP is happy to trot along despite accessing properties of null and
attaching multiple children to a document node.
Bug: T279445
Change-Id: I1784f550ec3a23dcded4f2b1def97e51cb414b7b
We added it because the initial designs for the subscribe action were
much easier to implement like this, and topic "containers" (T269950)
would have required it.
However, the latest design of the subscribe action will not need it
(T279149), and topic containers are still very far away, so let's
remove it for now.
Bug: T280433
Change-Id: I21a23e9bea43f24d265750926fbd62b99038d3f1
The 'dtenable' does not appear to take user-provided content that
requires language normalization. In general getRawVal should be used,
or if it's user input that needs normalization, use getText(). Perhaps
one day we'll alias or deprecate getVal (which is currently an odd
mid-way hybrid, most closely to getText; originally created for
EditPage.php textareas).
Change-Id: I8364c84f8c4f700da6e208df2e87c29bf254d685
We can't allow it, because the required database tables may not exist
yet (T280082).
This is meant to be temporary until we complete DBA review and the
tables are created.
Bug: T280082
Change-Id: I8f947b779c6829763d3413931c6d354e6f7aee4d
As of 7ad6328223, we also use this data
to check whether comments exist on the page, not only whether they're
transcluded.
Follow-up to 42ce942c86.
Bug: T275821
Bug: T273413
Change-Id: I95eb85354e7b84cc10ab703d28315d0667696f4c
The existing comment IDs can't be used to find the same comment on
a different revision or page (when it's transcluded), because they
depend on the comment's parent and its position on the page.
Comment names depend only on the author and timestamp. The trade-off
is that they can't distinguish comments posted within the same minute,
or in the same edit, so we will still need the IDs sometimes.
Prefer using comment names when replying, if they're not ambiguous.
This fixes T273413 and T275821.
Heading names depend on the author and timestamp of the oldest comment.
This way we don't have to detect changes to the heading text, but we
can't distinguish headings without any comments.
Bug: T274685
Bug: T273413
Bug: T275821
Change-Id: Id85c50ba38d1e532cec106708c077b908a3fcd49
Longer, but follows the style guide and less likely to conflict.
We need to account for init classes in the cache being around for
a while.
Change-Id: I738bc93393850db320fdbda2b003ca8ac40556da
Now it detect signatures generated by en.wp's {{Undated}} template,
and signatures of people who do weird stuff to the timestamps.
Bug: T275938
Change-Id: I27b07f6786ca5433a3c02a5fe68e4716d41401bb
In some cases it would return the parent node, instead of the siblings
it should return.
It's a private method only called by getFullyCoveredSiblings(), and
that method had a bug that cancelled out this one, so everything
worked correctly. But I want to use it elsewhere now and ran into it.
Change-Id: Ic12f007d57a8502a1bea5f0af17b29e9d59093d6
The horrendous 11-line if() condition did not correctly handle
signatures wrapped in inline formatting markup, like <small>.
Instead, implement this logic in the code for skipping to the end
of a paragraph, which didn't exist yet when that condition was
added, but seems like a much better place to check this now.
Bug: T275934
Change-Id: I5cccff889b5e15b5f8fde0538bf4bccb22e762cf
We may need this for topic subscriptions, and it seems
generally useful to include in the API response anyway.
Change-Id: If9522dc0c79a9a9ffb3a80f83fb17bf3c9399d6d
This code expected $container->firstChild to be a
<div class="mw-parser-output">, but that element is not present
when we're running on HTML to be saved in parser cache.
We ended up inserting the marker inside whatever node was the
first on the page, and if it was a <style> element, both our
marker and the styles would be lost when serializing, like in
6c7a0ca9a2.
When we're running on final HTML, the marker will now be outside
of <div class="mw-parser-output">, but that seems to be fine. Only
early versions of I4e60fdbc098c1a74757d6e60fec6bcf8e5db37c1 had
problems with that (see comments on patchset 41), but it works now.
The added test case also covers the fix for T274709.
Bug: T275440
Change-Id: I38d45dd8686919be51e1d307ded12b0afe185eb5
* HookUtils:FEATURES lists all features
* CommentFormatter::USE_WITH_FEATURES are all features
which require the comment formatter
Change-Id: Idbbe8bdd910b9c7b23c7fee76af7bb7ee13c2759
Going forward this will allow us to remove the parser
cache split, and toggle features just using CSS.
The CSS will be modified in a later commit to give the
anon caches time to clear.
Bug: T273072
Change-Id: I83c84b8bc63e1881e07b49acd8499b811adfccd4
I found this error in our logstash. I was not able to find an
existing Phabricator ticket.
Note how line #348 extracts the last element from the
$siblings array. It uses the function end() there, which
returns false in case the array is empty. $siblings[0] can't
do this but yields an error.
An alternative is to use reset(), which can return false as
well. But that's not really better. Especially not better
readable, I would argue.
Change-Id: Ic90cd2392ede15078ba0d5b4d67b8dc5d05f9bf7
Yes, this is still needed, removing it causes failures in tests
(and the old outputs look better).
Change-Id: I5bcedb0295a1f0ac4f6e51eaa9a9e072d8236f3c
Top-level comments that start or end with a list (inconsistent
indentation) would not have triggered the logic for detecting
wrappers.
Bug: T273692
Change-Id: Idcb4eed73e391f5f86eca2eb05cb3cea0d86f30a
* Ignore rendering-transparent nodes between discussion comments.
* Improve isRenderingTransparentNode() so that <link> nodes
representing TemplateStyles are not considered transparent,
otherwise this would undo ae920b831f.
Using a regexp from Parsoid.
Bug: T272746
Change-Id: I0b3c3251156ba6c4826abf5ba44ea93f80ebc01d
Splits the cache on the reply links feature being enabled
for a particular user and title.
An additional check is done after parsing in case the user
has the feature enabled via query string or cookie.
Bug: T267404
Depends-On: I883a37fd67108243e7a20683b1a5d59fd0f6e39f
Change-Id: I3bc06ca7d4aea7f0fe39eef0e77ad88d1f9c1043
Add yet another tree walking utility: CommentUtils::linearWalk().
Unlike TreeWalker, it allows handling the beginnings and ends of nodes
separately – kind of like parsing a XML token stream, or kind of like
VisualEditor's linear model.
(Add unit tests for this utility. The simple.html test case is copied
from [VisualEditor/VisualEditor]/demos/ve/pages/simple.html.)
Use this utility to stop skipping when we reach either a closing or
opening block node tag. Previously we'd skip over such tags inside
nested "transparent" nodes (like <a>, <del>, or apparently <font>).
Bug: T271385
Change-Id: I201a942eb3a56335e84d94e150ec2c33f8b4f4e0
The tool may go in and out of beta as new features
are release/graduated to opt-out. Users should only
need to opt in to the beta feature once to get all
future sub-features.
Bug: T272071
Change-Id: If6834b7fc07fc7e84757dc5fdcea814cd0d65936
Don't assume a feature is available because the code has
loaded and the user option is set. Export the logic from
Hooks.php to the client.
Change-Id: Ica0e58de7ed0d59e3b09645193eb2b691ae41c39
If DiscussionToolsABTest is enabled (set to `all` or a feature), logged
in users who have never used the tool before will be assigned to an a/b
test bucket. If they're in the test bucket, they get the feature
enabled.
If they manually set their beta feature preference, we don't override
that but do maintain their bucket for logging purposes.
Bug: T268191
Change-Id: I9c4d60e9f9aaef11afa7f8661b9c49130dde3ffa
1. Extend the JS modifier to allow adding top-level comments
(that is, replies to headings). PHP modifier doesn't do this
because we'll save the changes using paction=addtopic instead.
2. Subclass CommentController to allow adding a new heading and a
top-level comment underneath it at the same time.
3. A lot of ugly code in ReplyWidget to customize the interface
for this case. Much of it should probably be moved to
CommentController/NewTopicController.
Bug: T267595
Change-Id: I9c707bb7f7aae1b92c72fb4dee436490f8c8409b
Allows for multiple features in the near future.
Separate availability and enabled.
Separate User/Title/Output checks.
Change-Id: I454bd8407675749d93ff3d2b4c5d624b433204db
