When we mark an entire bundle as read, we don't care if it was opened
before or not - we call the API for the list of sources and their items
and build a list of IDs to mark as read in the remote source.
Then, we make sure that bundleIDs are expanded, added to that list.
Previously, we then sent those IDs to 'markCrossWikiItemsRead'
which marked the item IDs in the remote wikis as read.
markCrossWikiItemsRead is also expanding item bundle
IDs (correctly, because we also use it when we mark individual
xwiki items, and they should be expanded)
However, in cases where we mark individual xwiki items, the
model list is already filled, so markCrossWikiItemsRead trusts
the models to deliver the sub items (and then expand on their
IDs properly)
In the case of marking an entire xwiki bundle as read without
opening it first, however, that operation is not only redundant,
it produces a problem where the models were not yet filled
(because the xwiki bundle wasn't opened) and so we get empty
array of IDs, and the API has nothing to mark as read.
The solution is simple in this case - skip this method for working
with an entire bundle. It's both a redundant operation and the
wrong model to check. Send the information directly to the API
instead.
Bug: T142143
Change-Id: I4ed3bbc5c83290ed5791060b124840b1c3b12a75
Foreign notifications, though non-expandable, can represent
bundles (4 new topics on your talk page). When marking one
as read, it's important to provide all the IDs that are
part of that bundle. Otherwise, only the dynamic base is
marked as read and the bundle shows up again on reload
(3 new topics on your talk page).
For bundles that do not include the count
(The description on Flow_Page was edited multiple times)
it looks like the mark-as-read action did not work.
Tagging with T140309 as it may explain it but not 100% sure.
Bug: T140309
Change-Id: I19c9eb18669b7cad8173c966392a6e5977018121
Make it optional through the unpgrouppages parameter, so that
generic usage of the unreadnotificationpages API is still possible.
In the front-end, store which display title maps to what set of titles,
and pass in the full set rather than just the display title when
filtering by a page.
Bug: T137502
Change-Id: I443ca00ff5e5d36fd6910101226358942e6aa8ee
The fade in/out animation is asynchronous. This means that if we are
sorting multiple items one after the other, by the time the item faded
out, it will be reinserted back into the wrong position, breaking the
sorting.
This also broke the promise of OO.SortedEmitterList whereby all its items
are always in order.
The way to fix this was to force a better synchronization with the item
order while we hide and show the item in its new place. To do that,
a new widget is created as a fake clone of the old one, in the original
position of the old one. The original item is then reinserted (while hidden)
to the proper location -- preserving order. The fake clone is then faded
out, and the real item is then faded in.
For this to work properly, the cloned item had to preserve some of the
original item's information, like timestamp, foreigness and id. However,
since both the real item and the fake new clone have the same details,
the clone fakes its ID by adding a fraction to it - promising that the
fallback in case of equal timestamps (which happens on the real and
cloned items) will still resolve with some decision about the placement
of the items rather than (falsely but understandably) decide they are
both the same.
Since this whole animation is somewhat of a hack, the list now has a
configuration parameter to turn the animation on.
The animation is on in the popups, but off in the special page.
Bug: T141419
Change-Id: Ic7c35e5ddefc51bf7fde497eab36414b4dddcd9e
* Add ISO 8601 date format to notification output
This is actually supposed to be the only output date format used,
per https://www.mediawiki.org/wiki/API:Data_formats#Timestamps , but
I'm not doing anything to deprecate the others right now.
* Change wgEchoSeenTime to use ISO 8601. mwgrep and extension grep do
not show any usages. However, since it is a breaking change, to
minimize disruption, I'm also using this opportunity to change
'message' to 'notice'.
* Remove wgEchoInitialNotifCount. I was going to also change 'message'
to 'notice' here too, but then I saw it was totally unused.
(It was read in Echo to populate a JS variable, but then it was
unused.)
* Make sure the Special:Notifications page aggregation by days is
done by local days, even though the timestamp per item is still
UTC. This is to make sure the days are displayed correctly in
the local timezone.
* Change all reverse sorting callbacks to handle comparisons of
ISO 8601.
Bug: T141413
Change-Id: I20271345c7d350dc3e7f467288e5cdc98e6250cc
Add a global-wiki 'mark all as read' to the Special:Notifications page.
The 'mark all as read' will makr all notifications in the given
wiki. The context of the wiki changes when filters are chosen,
and so the message of the button changes as well.
Bug: T115528
Change-Id: Ibd9dcdf7072d6cbc1a268c18e558e6d0df28f929
When we are viewing a certain read state filter ('read' or 'unread')
the visibility of items should correspond to that state even when
the user marks a specific item as read/unread. That means that the
system should remove these items from view when the action is taken.
In this commit:
* The controller makes the judgment of whether to remove items when
read/unread action is taken, based on whether a filter is set.
* We clean up the terminology of discard - no more 'remove' - to
make sure we have consistency in the code.
* Related: The 'discard' event is now scoped within the hierarchy;
meaning, lists emit 'discard' when an item is removed, grouplist
emits 'discard' when a group is removed, and the manager emits
'discard' when an entire notification model is removed. This
means we can actually have proper hierarchy and organization with
a single event, and not worry about clashing between the intentional
'discard' action and the event 'remove' that is also used while
resorting happens.
* The model manager emits a discard event when a model is removed
so that the general list can listen to the manager and remove an
entire batch of items if needed.
* The pagination model now updates the count for the current page
rather than some vague notion of the last page. This is also
updated when the controller removes items, so we can get an
accurate count in the page for the number of notifications that
are displayed.
Bug: T136891
Change-Id: I247c618042ef256fadf09922f7b83bd1ad361f64
This organizes the operation of seenTime so we can store and
follow up on it based on different sources, as well as update
it correctly remotely when needed.
Change-Id: I629ecfc84999be998b45c9c7adb00ea7e3e51742
We need the API to return the notifications for the special page
in timestamp order and not read/unread order, so that the 'continue'
value is still correct.
On top of that, if we have many unread notifications, they should
still be placed according to the dates, so the API must bring back
proper result.
In this fix, we add 'unreadFirst' to filters, and only use that
filter when needed (namely, in the popup)
Bug: T136885
Change-Id: I3018d09b009d735402d83074a5ffcd14ea1c242a
When we fetch the pages per wiki, the API returns an object that
defines the local wiki we are in as its dbName; this then gets
stored into the model as the source, which forces us to check against
the dbName whenever we want to perform an operation so we can tell
the API layer to fetch and perform the actions locally, rather than
use a foreign API.
The term 'local' makes no sense for naming the model (and upcoming
tech debt work will fix this) but it makes sense in the context of
the API layer -- and hence, the source name should follow suit.
This fix makes sure that the local wiki objects always have their
sources defined to be 'local', and thus making us use the test for
dbName in very specific points (when we get the data) rather than
throughout the codebase, randonly.
This is also the first step to allow proper updating of things like
seenTime accross wikis, especially in cases where we view a remote
wiki as if it is local (in the Special:Notifications page)
Change-Id: I94633a1cd074580cbc5029d7c75d179e908e5c52
Marking local or xwiki notifications as read
are different operations that require different
information. Call different functions on the
controller.
Bug: T139114
Change-Id: I824ca668c31ab41d101c632442c91d3a8a1e5c5a
To allow individual notifications to be
marked as read/unread or moderated,
bundles are created by grouping associated
notifications when they are fetched for display
instead of when they are created.
From a product perspective, this change doesn't
introduce moderation or expandable bundles but
it counts each individual notifications.
For instance, the bundled notification
"3 new topics on PageA" now counts as 3
notifications.
Bug: T93673
Bug: T120153
Change-Id: Iacd098573efd92bb1e3fcd7da4cd40cea9522f15
Add a sidebar with cross-wiki sources and pages of unread notifications.
The filter allows the user to fetch notifications from a foreign source
and specific pages if those exist.
Bug: T129366
Change-Id: I57d827a47f80274d75364c2099a9624049a26834
- Remove duplicated counter adjustments.
- Differentiate between all IDs to mark as
read vs. all IDs that represent visible
notifications when adjusting the counter.
Change-Id: I1b3e52a929c6e920e3b782c9ef028539b3db0c99
This should have been done from the beginning; the model manager
pulls models by their symbolic names. So far, we've used the source
for that, but that assumes that two modules always have different
sources, and that is absolutely not necessarily the case.
For example, internal local bundles will each be a model, but have
the same ('local') source. They should still be differentiated in
the manager by their names, but the source should state clearly that
it is local.
For this, the models now have "getName" method and the name is
created separately from their source. Items also preserve a
reference to their parent's symbolic name so they can provide
that for items that require the controller to manipulate a specific
model.
Change-Id: I8c39d5d28383d11fb330addce21e07d5c424da6f
In the current bundling system, only the
bundle base is mark as read. It leaves all
the non-base bundled notifications without
a read_timestamp. They would all appear
as read in the new bundling system.
With this change, all notifications in
a bundled are update with a read_timestamp
when the bundle is read.
The implementation of this change is
somewhat temporary as the new bundling
system brings changes to the models
and controller.
Bug: T136368
Change-Id: I70b71d722d8d62cbdd1adc004293030ef900ac94
The zero results can either be because there are no notifications
at all for this user in the local wiki, or because there are no
results for the specific filter. Both messages are used for either
case.
Also, clean up the display of push/pop pending for the inbox widget
and hide the label in case the message count is 0 or 1 notifications
as it is unhelpful and irrelevant in these cases.
Bug: T136586
Bug: T136574
Bug: T129363
Change-Id: I1465f772bb9f5247df645d6612f951e5fd7d38cf
Allow for the widget to have a mark-as-read button to its individual
groups, as well as change the event listening from 'remove' to 'discard'
The problem with 'remove' event is that it is triggered when an item
is either intentionally removed from the list *and* when an item is
changing its position in the list (move event includes 'remove' and
then 'add' event)
If we listen to 'remove' events we will get both cases, which is
unhelpful. Instead, a new event - 'discard' - was introduced so
we are certain it is used with the clear intention of removing the
item completely.
Change-Id: Ia08720bf4c547fa41edf62331eeb1a45ff4965b7
Changing the way Echo's front-end architecture works to work with
model-view-controller methodology.
Change-Id: I97862402c41bc04dd41cd08d79f19ff677340249
