I have also changed the priorities so that manual updates are preferred.
This means that duplicate updates do not happen if the room was previously unknown.
Signed-off-by: Timo Kösters <timo@koesters.xyz>
Vectors grow in powers of two while reading, which is doubly wasteful:
* causes every byte to be copied in average once
* leaves the vector in average 25% larger than it needs to be, wasting memory.
For example, adding this test to
`crates/matrix-sdk-crypto/src/file_encryption/attachments.rs`:
```
fn encrypt_decrypt_minimize_memory() {
let data = std::iter::repeat("abcdefg").take(10000).collect::<String>();
let mut cursor = Cursor::new(data.clone());
let mut encryptor = AttachmentEncryptor::new(&mut cursor);
let mut encrypted = Vec::new();
encryptor.read_to_end(&mut encrypted).unwrap();
let key = encryptor.finish();
assert_ne!(encrypted.as_slice(), data.as_bytes());
let mut cursor = Cursor::new(encrypted);
let mut decryptor = AttachmentDecryptor::new(&mut cursor, key).unwrap();
let mut decrypted_data = Vec::new();
decryptor.read_to_end(&mut decrypted_data).unwrap();
assert_eq!(
decrypted_data.len(),
decrypted_data.capacity(),
"{} bytes wasted by decrypted_data",
decrypted_data.capacity() - decrypted_data.len()
);
}
```
errors with:
```
assertion `left == right` failed: 61072 bytes wasted by decrypted_data
left: 70000
right: 131072
```
By initially setting this capacity, the vector should be slightly larger
than needed from the start, avoiding both copying and leftover capacity
after decryption is done.
`matrix_sdk_ui::timeline::Message::msgtype` returns `&MessageType`,
and variants of `MessageType` implement `MediaEventContent`, so it is
allowing references here avoids cloning message content.
This didn't show up, and it's quite useful to know for which room we're trying to send a receipt, without having to look up the room based on the
target event id.
This patch rewrites all the integration tests for the notable tags. The
tests were good, but we could merge some together. The names were too
long, they have been shortened.
The workflow inside `Client::receive_sync_response` is a little
bit fragile. When `Client::handle_room_account_data` is called, it
modifies `RoomInfo`. The problem is that a current `RoomInfo` is being
computed before `handle_room_account_data` is called, and saved a
couple lines after, resulting in overwriting the modification made by
`handle_room_account_data`.
This patch ensures that the new `RoomInfo` is saved before
`handle_room_account_data` is called.
Now that `NotableTags` has replaced `RoomNotableTags` entirely, this
patch can rename `NotableTags` to `RoomNotableTags` as it's a better
name for it.
Note that this type is private to `matrix_sdk_base`, so it's fine to
rename it.
The Matrix specification uses the `m.favourite` orthography. Let's use
the same in our code. So `set_is_favorite` becomes `set_is_favourite`.
This patch updates this in various places for the sake of consistency.
The previous patch has introduced the new `NotableTags` type to replace
the `RoomNotableTags`. This patch removes the latter.
This patch keeps the `Room::set_is_favorite` and `::set_is_low_priority`
methods. However, this patch adds the `Room::is_favourite` and
`::is_low_priority` methods, with the consequence of entirely hiding the
notable tags type from the public API.
This patch is the first step to remove the [`RoomNotableTags`] API. The
idea is to compute the notable tags as a single 8-bit unsigned integer
(`u8`) and to store it inside `RoomInfo`. The benefits are numerous:
1. The type is smaller that `RoomNotableTags`,
2. It's part of `RoomInfo` so:
- We don't need an async API to fetch the tags from the store and to
compute the notable tags,
- It can be put in the store,
- The observable/subscribe mechanism is supported by `RoomInfo` behind
observable instead of introducing a new subscribe mechanism.
This patch exposes the 1-to-1 encryption method that is usually used to share a room key with a device. Users might want to send encrypted custom to-device events to a device directly, so let's expose this functionality.
Co-authored-by: Damir Jelić <poljar@termina.org.uk>
We don't want the rest of the method to abort executing because
automatic backup creation failed.
We also move the infallible event handler registration at the top of the
method.
The logic to automatically create a backup uses the following logic:
1. We check if we need to create a backup, this includes a check if a
backup exists on the server.
2. We conclude that we should create a backup, no backup exists on the
server.
3. The method to create the backup checks again if a backup exists, this
was an API change because the method is public and users misused the
method.
4. Finally, we create the backup.
A race exists between the first time we check if a backup exists and the
second time, i.e. step 1 and 3.
It seems that some users create two Client objects which then make this
race a common occurrence. Clients should not do that, but at least we
don't error out too soon anymore if the automatic creation of the backup
fails.
Quick performance improvement on `save_change` for indexeddb.
All the serialization/encryption is now done outside the db transaction
---
* do serialization/encryption before db transaction
* clippy
* add changelog for indexeddb crate
* clean comments
* fix typo
* Review: Fix typo in changelog
* Review: refactor, rename DbOperation to PendingOperation
* Review: rename variant PutKeyVal to Put
* Review: fix doc typo
* Review: rename perfrom_operation to apply
* Review: remove unneeded isEmpty checks
* Review: refactor better API for PendingStoreChanges
* Review: Prefer BTreeMap to HashMap
* Refactor: rename IndexeddbChangesKeyValue
* Refactoring: get the list of affected store from PendingIndexeddbChanges
* cleaning
* Review: Better names and comments
* Review: use filter_map instead of filter then map
fix: Preserve the event of any settings that are changed back to the default level.
sdk: Rename get_room_power_levels (drop the get).
chore: Refactor with more sensible naming.
sdk: Clean up the RoomPowerLevelChanges API.
The IndexMap::remove method has been deprecated, the documentation[1] on
the method tells us that we can replace the usage of it with
IndexMap::swap_remove:
> NOTE: This is equivalent to .swap_remove(key), replacing this entry’s
> position with the last element, and it is deprecated in favor of
> calling that explicitly.
[1]: https://docs.rs/indexmap/2.2.2/indexmap/map/struct.IndexMap.html#method.remove
This patch implements the `category` room list filter. It introduces a
new type: `RoomCategory`, to ensure that “group” and “people” are
mutually exclusives.
This patch implements `Room::direct_targets_length`. It avoids to call
`Room::is_direct` if and only if we don't care about the room's state
and we don't want an async call, and if we don't want to pay the cost of
`Room::direct_targets` which clones the `HashSet` as an alternative way
to get a similar information than `Room::is_direct`.
