We are getting closer to the next major release. This version will have many breaking changes, so we are listing them here for app and client developers to adjust their projects.
As we prepare for the release of Lemmy 0.19.0
, we’d like to provide any app or client developers ample time to upgrade their apps, as well as discover any problems, before we do the release. This will be at least 4 weeks from now (but likely longer).
Server admins can also upgrade to the latest release candidates for testing. Be aware that they are still unstable and shouldn’t be used in production. As with any upgrade it is important to have working backups in place.
It should be possible for clients to support both Lemmy 0.18 and 0.19 without major workarounds. If backwards compatibility is causing you trouble, comment below and we will help to find a solution.
To test, you can point your app to the following test instance running a release candidate of 0.19.0
: https://voyager.lemmy.ml
A diff of API changes is here: lemmy-js-client API changes from 0.18.3 -> 0.19.0-rc’s
Note for developers not using typescript or rust:
If you’d like to auto-generate an API client for your language, you can try out @MV-GH’s lemmy_openapi_spec, or (if in kotlin), use Jerboa’s script here.
Major Changes
Authentication
Previous Lemmy versions used to take authentication as query/post parameters. This is insecure and unnecessarily complicated. With 0.19
, the jwt
token can be passed either as cookie with name auth
, or as header in the form Authorization: Bearer
.
A major advantage is that this allows us to send proper cache-control headers, with responses to unauthenticated users being cacheable. It also prevents token leaks in web server logs. The login and registration endpoints attempt to set the cookie automatically. If that is supported on your platform, you don’t have to worry about the authentication token at all.
In order for your client to be compatible with both Lemmy 0.18
and 0.19
, you should send auth in both ways. Meaning with each API call, send the old auth
query/post parameter, as well as the new header or cookie.
A few PRs detailing these changes:
- https://github.com/LemmyNet/lemmy/pull/3725
- https://github.com/LemmyNet/lemmy/pull/3926
- https://github.com/LemmyNet/lemmy/pull/3946
- https://github.com/LemmyNet/lemmy/pull/3982
Users can block instances
Users can now block instances, so that their communities are hidden from listings. This is done via POST /api/v3/site/block
with parameters int instance_id, bool block
.
https://github.com/LemmyNet/lemmy/pull/3869
New sort options
A new scaled
sort option has been added. This sort is identical to the Hot
sort, but also takes into account the number of each community’s active monthly users, and so helps to boost posts from less active communities to the top.
https://github.com/LemmyNet/lemmy/pull/3907
2FA / TOTP Rework
Two-Factor-Authentication is now enabled in a two-step process to avoid locking yourself out. Now a secret needs to be generated first with POST /api/v3/user/totp/generate
(no parameters). The generated token needs to be added by the user to an authenticator app.
Once this is completed, 2FA can be enabled with POST /api/v3/user/totp/update
. This takes a string parameter totp_token
(generated by authenticator app), and boolean enabled
. 2FA can be disabled again with the same update
endpoint. Additionally, the 2FA algorithm has been changed to SHA1
for better compatibility.
The update disables 2FA for all accounts, so that users who are locked out can use their accounts again, and to ease the transition to the SHA1
algorithm.
https://github.com/LemmyNet/lemmy/pull/3959
Timestamps now include timezone
Previous Lemmy versions used timestamps without any timezone internally. This caused problems when federating with other software that uses timezones.
Going forward, all timestamps in the API are switching from timestamps without time zone (2023-09-27T12:29:59.113132
) to IS
timestamps (e.g. 2023-10-29T15:10:51.557399+01:00
or Z
suffix). In order to be compatible with both 0.18 and 0.19, parse the timestamp as IS
and add a Z
suffix if it fails (for older versions).
https://github.com/LemmyNet/lemmy/pull/3496
Cursor based pagination
0.19
adds support for cursor based pagination on the /api/v3/post/list
endpoint. This is more efficient for the database. Instead of a query parameter ?page=3
, listing responses now include a field "next_page": "Pa46c"
which needs to be passed as ?page_cursor=Pa46c
. The existing pagination method is still supported for backwards compatibility, but will be removed in the next version.
https://github.com/LemmyNet/lemmy/pull/3872
New endpoints for export/import of user settings data
Users can now export their profile settings data (including subscriptions and blocklists) via GET /api/v3/user/
. The returned JSON data should not be parsed by clients, but directly downloaded as a file. Backups can be imported via POST /api/v3/user/import
.
https://github.com/LemmyNet/lemmy/pull/3976
Make remove content optional during account deletion
When a user deletes their own account using POST /api/v3/user/delete_account
, there is a new parameter called delete_content
. If it is true, all posts, comments and other content created by the user are deleted (this is the previous default behaviour). If it is false, only the user profile will be marked as deleted.
https://github.com/LemmyNet/lemmy/pull/3817
Outgoing Federation Queue
The federation queue has been rewritten to be much more performant and reliable. This is irrelevant for client developers, but admins should look out for potential federation problems. If you run multiple Lemmy backends for horizontal scaling, be sure to read the updated documentation and set the new configuration parameters. The Troubleshooting section has information about how to find out the state of the federation queues.
While SHA1 might be considered problematic security-wise in terms of collision (using it for certs today would be very bad, for example), it is not problematic in terms of preimage attacks (even MD5 isn’t broken that way IIRC), which is what truly matters in the context of 2FA / TOTPs
As for “why not SHA256”, compatibility
That’s a huge understatement.
SHA-1 was cryptographically broken over 20 years ago and should not be used anymore.