Close#304.
Notes:
- This patch was made on top of Pleroma develop, so I created a separate cachex worker for request signature actions, instead of Akkoma's instance cache. If that is a merge blocker, I can attempt to move logic around for that.
- Regarding the `has_request_signatures: true -> false` state transition: I think that is a higher level thing (resetting instance state on new instance actor key) which is separate from the changes relevant to this one.
Co-authored-by: Luna <git@l4.pm>
Reviewed-on: https://akkoma.dev/AkkomaGang/akkoma/pulls/312
Co-authored-by: @luna@f.l4.pm <akkoma@l4.pm>
Co-committed-by: @luna@f.l4.pm <akkoma@l4.pm>
Changes follow_operation schema to use BooleanLike instead of :boolean so that strings like "0" and "1" (used by mastodon.py) can be accepted. Rest of file uses the same. For more info please see https://git.pleroma.social/pleroma/pleroma/-/issues/2999
(I'm also sending this here as I'm not hopeful about upstream not ignoring it)
Co-authored-by: ave <ave@ave.zone>
Reviewed-on: https://akkoma.dev/AkkomaGang/akkoma/pulls/301
Co-authored-by: ave <ave@noreply.akkoma>
Co-committed-by: ave <ave@noreply.akkoma>
- Drop Expect-CT
Expect-CT has been redundant since 2018 when Certificate Transparency became mandated and required for all CAs and browsers. This header is only implemented in Chrome and is now deprecated. HTTP header analysers do not check this anymore as this is enforced by default. See https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Expect-CT
- Raise HSTS to 2 years and explicitly preload
The longer age for HSTS, the better. Header analysers prefer 2 years over 1 year now as free TLS is very common using Let's Encrypt.
For HSTS to be fully effective, you need to submit your root domain (domain.tld) to https://hstspreload.org. However, a requirement for this is the "preload" directive in Strict-Transport-Security. If you do not have "preload", it will reject your domain.
- Drop X-Download-Options
This is an IE8-era header when Adobe products used to use the IE engine for making outbound web requests to embed webpages in things like Adobe Acrobat (PDFs). Modern apps are using Microsoft Edge WebView2 or Chromium Embedded Framework. No modern browser checks or header analyser check for this.
- Set base-uri to 'none'
This is to specify the domain for relative links (`<base>` HTML tag). pleroma-fe does not use this and it's an incredibly niche tag.
I use all of these myself on my instance by rewriting the headers with zero problems. No breakage observed.
I have not compiled my Elixr changes, but I don't see why they'd break.
Co-authored-by: r3g_5z <june@terezi.dev>
Reviewed-on: https://akkoma.dev/AkkomaGang/akkoma/pulls/294
Co-authored-by: @r3g_5z@plem.sapphic.site <june@terezi.dev>
Co-committed-by: @r3g_5z@plem.sapphic.site <june@terezi.dev>
The header name was Report-To, not Reply-To.
In any case, that's now being changed to the Reporting-Endpoints HTTP
Response Header.
https://w3c.github.io/reporting/#headerhttps://github.com/w3c/reporting/issues/177
CanIUse says the Report-To header is still supported by current Chrome
and friends.
https://caniuse.com/mdn-http_headers_report-to
It doesn't have any data for the Reporting-Endpoints HTTP header, but
this article says Chrome 96 supports it.
https://web.dev/reporting-api/
(Even though that's come out one year ago, that's not compatible with
Network Error Logging which's still using the Report-To version of the
API)
Signed-off-by: Thomas Citharel <tcit@tcit.fr>
There were async calls happening, so they weren't always finished when assert happened.
I also fixed some bugs in the erratic tests that were introduced when removing :shout.:shout is a key where restart is needed, and was changed in the test to use :rate_limit (which also requires a restart). But there was a bug in the syntax that didn't get caught because the test was tagged as erratic and therefor didn't fail. Here I fixed it.
During compilation, we had a warning `:logger is used by the current application but the current application does not depend on :logger` which is now fixed as well (see commit message for complete stacktrace).
Co-authored-by: Ilja <ilja@ilja.space>
Reviewed-on: https://akkoma.dev/AkkomaGang/akkoma/pulls/237
Co-authored-by: ilja <akkoma.dev@ilja.space>
Co-committed-by: ilja <akkoma.dev@ilja.space>
Fixes one of the 'erratic' tests
It used a timer to sleep.
But time also goes on when doing other things, so depending on hardware, the timings could be off.
I slightly changed the tests so we still test what we functionally want.
Instead of waiting until the cache expires I now have a function to expire the test and use that.
That means we're not testing any more if the cache really expires after a certain amount of time,
but that's the responsability of the dependency imo, so shouldn't be a problem.
I also changed `Pleroma.Web.Endpoint, :http, :ip` in the tests to `127.0.0.1`
Currently it was set to 8.8.8.8, but I see no reason for that and, while I assume that no calls
are made to it, it may come over as weird or suspicious to people.
Co-authored-by: Ilja <ilja@ilja.space>
Reviewed-on: https://akkoma.dev/AkkomaGang/akkoma/pulls/233
Co-authored-by: ilja <akkoma.dev@ilja.space>
Co-committed-by: ilja <akkoma.dev@ilja.space>
The problem was double. On the one hand, the function didn't actually return what was in the DB.
On the other hand the test was flaky because it used NaiveDateTime.utc_now() so test could fail or pass depending on a difference of microseconds.
Both are fixed now.
It was tested if the updated_at after marking as "read" was equal as the updated_at at insertion, but that seems wrong.
Firstly, if a record is updated, you expect the updated_at to also update.
Secondly, the insert and update happen almost at the same time, so it's flaky regardless.
Here I make sure it has a much older updated_at during insert so we can clealy see the effect after update.
I also check that the updated_at is actually updated because I expect that this is the expected behaviour and it's also the current behaviour.
This fixes a race condition bug where keys could be regenerated
post-federation, causing activities and HTTP signatures from an user to
be dropped due to key differences.
Non-Create/Listen activities had their associated object field
normalized and fetched, but only to use their `id` field, which is both
slow and redundant. This also failed on Undo activities, which delete
the associated object/activity in database.
Undo activities will now render properly and database loads should
improve ever so slightly.
Use Websockex to replace websocket_client
Test that server will disconnect websocket upon token revocation
Lint
Execute session disconnect in background
Refactor streamer test
allow multi-streams
rebase websocket change
This field replaces the now deprecated conversation_id field, and now
exposes the ActivityPub object `context` directly via the MastoAPI
instead of relying on StatusNet-era data concepts.
This field seems to be a left-over from the StatusNet era.
If your application uses `pleroma.conversation_id`: this field is
deprecated.
It is currently stubbed instead by doing a CRC32 of the context, and
clearing the MSB to avoid overflow exceptions with signed integers on
the different clients using this field (Java/Kotlin code, mostly; see
Husky and probably other mobile clients.)
This should be removed in a future version of Pleroma. Pleroma-FE
currently depends on this field, as well.
Incoming Pleroma replies to a Misskey thread were rejected due to a
broken context fix, which caused them to not be visible until a
non-Pleroma user interacted with the replies.
This fix properly sets the post-fix object context to its parent Create
activity as well, if it was changed.
* rejected_shortcodes is defined as a list of strings in the
configuration description. As such, database-based configuration was
led to handle those settings as strings, and not as the actually
expected type, Regex.
* This caused each message passing through this MRF, if a rejected
shortcode was set and the emoji did not exist already on the instance,
to fail federating, as an exception was raised, swiftly caught and
mostly silenced.
* This commit fixes the issue by introducing new behavior: strings are
now handled as perfect matches for an emoji shortcode (meaning that if
the emoji-to-be-pulled's shortcode is in the blacklist, it will be
rejected), while still supporting Regex types as before.
It retrieved two ReportNotes and then checked one of them. But the order isn't guaranteed, while the test tested on the content of the first ReportNote.
I made the test on the content more generic
elixir gettext current does not fully support fallback to another language [0].
But it might in the future. We adapt it so that all languages in Accept-Language
headers are received by Pleroma.Web.Gettext. User.languages is now a comma-separated
list.
[0]: https://github.com/elixir-gettext/gettext/issues/303
For an example, here, zh is not supported, but zh_Hans and zh_Hant
are. If the user asks for zh, we should choose a variant for them
instead of fallbacking to default.
Some browsers (e.g. Firefox) does not allow users to customize
their language codes. For example, there is no zh-Hans, but only
zh, zh-CN, zh-TW, zh-HK, etc. This provides a workaround for
those users suffering from bad design decisions.
When someone isn't a superuser any more, they shouldn't see the reporsts any more either.
Here we delete the report notifications from a user when that user gets updated from being a superuser to a non-superuser.
For some reason I had a test who suddenly failed, mix test test/pleroma/web/o_auth/app_test.exs:54. A user has a list of applications and this test adds them and then sees if the list it gets back is the same as the apps it added.
When I ran mix test a day before I didn't have this problem and when I pushed code today in a different MR, the pipeline succeeded (see https://git.pleroma.social/ilja/pleroma/-/jobs/205827), yet locally it failed. So it seems the test can sometimes succeed and sometimes fail, which makes it untrustworthy.
The failure I see is because the returned list is in reverse order. I assume that's not per sé wrong. You just want to know if the apps you added are actually there. I fixed the test by first ordering the lists before comparing.
AFAICT (and as far as that's relevant) the test got introduced in commit cb2a072e62
Mastodon uses the Reject activity also for the purpose of removing
a follower, in addition to reject a follow request. We should
also update the original Follow activity in this case.
