Whether the actor is suspended is already checked in the performOneActivity
function and does not need to be checked here again since the actor is still
the same as it was there.
The assumptions made in the comment is actually wrong. The comment says:
"If value is a string, this means it must have passed through the resolver,
which means it must be public." But this is not true because we do signed
GET requests which means we may well get non-public posts using the
resolver.
This deletes as many files as necessary to ensure the drive quota for
remote users is kept. Previously only one file would have been deleted
for each file added.
Changelog: Fixed
Co-authored-by: CGsama <CGsama@outlook.com>
Co-authored-by: tamaina <tamaina@hotmail.co.jp>
Instead of checking this configuration in the respective component (queue)
or not at all (server), the configuration can be checked when starting the
respective workers.
This log level replaces the "MK_QUIET" environment variable to unify
the interface in a sensible way. This also removes the "MK_VERBOSE"
environment variable which was unused.
To avoid a circular dependency this requires moving the log level
definitions. Also to avoid a circular dependency the env.ts file cannot
use a logger and instead uses plain `console.log`.
This resulted in endless ping-pong traffic on the websocket, happening
every interval of network latency to the server (e.g. for me, with 40ms
latency to my server, it was about every 40ms). On my server this ended
up taking about 20% of foundkey's CPU usage. Now, just send pings every
30s, and check if we have received any pong's in last 60 seconds to
check that the connection is still alive.
Changelog: Fixed
This makes the returned values `localFollowingCount`,
`remoteFollowingCount`, `followingCount`, `localFollowersCount`,
`remotefollowersCount`, `followersCount` optional on the API endpoint
`users/stats`.
Changelog: Fixed
The `data` field is not used anywhere in the logger. While it would
be possible to send structured data through syslog, it seems unnecessary
at present and also the way in which this structured data would have to be
provided sounds too cumbersome to implement for no real value.
The CacheableUser, CacheableLocalUser and CacheableRemoteUser are
identical types to User, ILocalUser and IRemoteUser so it seems
nonsensical to have different types for them.
When rendering the HTML for outgoing activities, the mentions are now
matched case insensitive and should also work properly for IDNs. The
username is also compared case insensitive. Mentions of local users
are also handled properly independed of whether the hostname was given
or omitted.
The query to get mentions is now also only executed once instead of
for each mention individually.
Changelog: Fixed
While refactoring the previous commit, it seemed like the previous
authors expected that a system account could be registered somehow
and that this would be an error condition. However, as now made
explicit with this, it is not possible to register a system account.
This means that any account by that name could only ever have been
created by the system itself so fetching them should be fine and not
an error condition.