Misskey's code does the same, but our groups behave differently enough
that this may be not the best choice
for example, let's say we have:
- notifications 1-5 for reaction to note A
- notifications 6-8 for reaction to note B
- notifications 9-12 for reaction to note A
- notification 13-19 for non-groupable events
- notification 20 for reaction to note A
and that events happened one every minute (so the last notification is
from 20 minutes ago)
client requests the most recent 10 notifications; we fetch
notifications 1-10, and reply:
- grouped id 6 for reactions 6-8 to note B
- grouped id 10 for reactions 1-5, 9-10 to note A
then the client requests 10 more notifications, untilId=10; we fetch
notifications 11-20, and reply:
- non-grouped notifications 13-19
- grouped id 20 for reactions 11,12,20 to note A
because we sort by id, and also the `createdAt` marks the _newest_
event in each group, the client will then show:
6 reactions to note B, 6 minutes ago
4 reactions to note A, 1 minute ago
notifications 13-19, 13 minutes to 19 minutes ago
3 reactions to note A, 11 minutes ago
I don't know how to make this work better ☹
this will make it much easier to debug problems for instances that run
unreleased versions!
when run on a tagged commit, `git describe --tags` prints the tag name;
otherwise it prints something like `2025.4.3-32-ga4c0ef824c` which
means:
- the closest tag is 2025.4.3
- there are 32 commits between that tag and this commit
- this commit's id is `a4c0ef824c` (the `g` is just a prefix)
notice that the version as reported by the frontend (in
`/about-sharkey` for example) is _not_ changed, that one is still
sourced from `/package.json` (so, for example, you don't get a
"sharkey has been updated!" pop-up every time)
the top element must be a single <a> in order for verification to work, so MfmService has gained an 'inline' argument to remove the <p> added by default
we remove all duplicate items from the registry, keeping the
most-recently updated one (which presumably is the one that the users
expect to be used)
then we use `upsert` instead of the more complicated previous logic,
which was subject to races
