This makes the public key fields private to ensure that we don't
accidentally swap them out. It also moves the construction of the
subkeys into the master key type.
Since the master/self-signing/user-signing public key types are used for
public user identities as well as for the private key type we have, and
we'd like to sign the public key types it makes sense that the types
itself aren't using an Arc.
Let's instead put the Arc inside the user identity structs.
This will allow us later on to more easily sign the public key types.
This patch removes support for OpenTelemetry because it's not used
anymore by anybody. It also adds multiple duplicated dependencies (like
`reqwest`). Anyway. Farewell.
Preferring to use "reject" wording rather than "failed to
create/update". The latter can be easily misinterpreted as a failure of
the local client to create an entirely new device from scratch, rather
than refusal to instantiate a new local device representation of an
(invalid) device definition received from the server.
This patch makes the `LinkedChunk::update_history` field optional,
so that it doesn't require the user to drain it to avoid eating the
universe.
The `new` constructor disabled the update history, the
`new_with_update_history` enables it.
This patch is a turn around about the `LinkedChunkListener`. Many
patches have been removed because `LinkedChunkListener` needed to
support I/O, so errors and async code. The whole code was affected by
that, resulting in a complex API. The idea of this patch is to decoupled
this. Here is how.
First off, `LinkedChunkListener` is removed. So it's one less generic
parameter on `LinkedChunk`. It's also one less trait, so less
implementations.
Second, now `LinkedChunk` accumulates/collects all “updates” under the
form of a new enum `LinkedChunkUpdate`. These updates can be read with
`LinkedChunk::updates(&mut self) -> &Vec<LinkedChunkUpdate>`. The reader
can simply read them, or even drain them. The reader is responsible
to handle these updates and to dispatch them in a storage or whatever.
`LinkedChunk` is no longer responsible to do that, removing the need to
support errors and to be async.
Third, the simplification has led to an optimisation by introducing a
new type `LinkedChunkLinks`. The documentation explains what it does
and why it was needed. The benefit of this type is: it doesn't increase
the size of `LinkedChunk`, but it simplifies the code: no more `Arc`,
no more `Mutex` (it was required because with I/O and async), no more
borrow checker trick, and the code stays as safe as before.
This patch removes the `displaydoc` dependency. Why?
1. It creates a warning in rustc nightly:
```
warning: non-local `impl` definition, they should be avoided as they go against expectation
--> crates/matrix-sdk-store-encryption/src/lib.rs:49:17
|
49 | #[derive(Debug, Display, thiserror::Error)]
| ^^^^^^^
|
= help: move this `impl` block outside the of the current constant `_DERIVE_Display_FOR_Error`
= note: an `impl` definition is non-local if it is nested inside an item and may impact type checking outside of that item. This can be the case if neither the trait or the self type are at the same nesting level as the `impl`
= note: one exception to the rule are anon-const (`const _: () = { ... }`) at top-level module and anon-const at the same nesting as the trait or type
= note: this lint may become deny-by-default in the edition 2024 and higher, see the tracking issue <https://github.com/rust-lang/rust/issues/120363>
= note: the derive macro `Display` may come from an old version of the `displaydoc` crate, try updating your dependency with `cargo update -p displaydoc`
= note: `#[warn(non_local_definitions)]` on by default
= note: this warning originates in the derive macro `Display` (in Nightly builds, run with -Z macro-backtrace for more info)
```
2. `thiserror` is already used, which seems to provide a similar issue.
3. That's less dependency, and less proc-macro, which will improve the
compilation time in general.
Also:
- rename `display_name` to `computed_display_name` in several places,
and reflect that change into a few callers
- simplify slightly the `computed_display_name()` method
This is in line with what the other method using sliding sync does. This
wasn't tested before, because this required `filter_by_push_rules()` to
be enabled in the notification client; now that it's the default, the
test revealed the bug, and so it could be fixed.
The builder had only one meaningful method, `filter_by_push_rules`,
which was always called by the applications — and in fact should always
be true. It was designed as an extra method because it was experimental
at the time, but it's stabilized sufficiently that we can enable this
behavior by default now, considering that a notification that is not
wanted by the user shouldn't be kept, to respect their intent. (This is
in the UI crate, which is opinionated, so it's fine to assume such
intents by design.)
They need to be updated together
because the latters depend on the former.
matrix-authentication-service is still using http 0.2
so we need to add a conversion layer between both major versions
for OIDC requests.
We need to update vodozemac too because of a dependency resolution issue.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
This makes all the requests (/context and /messages) cancellation-safe
by making two changes:
- first, use sync locks instead of async locks for the prev/next batch
tokens. This ensures that we can't get cancelled after receiving a
response and managing internal state, i.e. we keep run-to-completion
semantics until the end of the function, once we got a response.
- second, introduce a RAII guard that will reset the state to a given
value when dropped. Then use this to reset the state to Initial (resp.
Idle) in /context (resp. /messages) when the async call is aborted. We
use `mem::forget` once the response has been returned, so as to not call
the `Drop` implementation of the guard later on.
A regression test has also been introduced.
