Like other recursive operations on the dependency graph, this is
not done in a single transaction to prevent an attacker from creating
arbitrary large transactions.
So at startup, the `ValidationManager` finds and resumes any
unfinished operations, by looking for shared messages with unshared
dependencies.
The old behaviour was a leftover from the days of limited retention periods. The new behaviour will interact better with dependencies and message queues.
Database queries for metadata only returned it for messages that were delivered already.
However, there are cases (e.g. a pending message needs to be delivered) where
the validator needs to retrieve the metadata from the database.
For these cases, a special database query has been introduced.
This adds a new table to the database to hold message dependencies.
It introduces two more message states: pending and delivered
The valid column in the database was renamed to state to better reflect
its new extended meaning.
The DatabaseComponent was extended with three methods for:
* adding dependencies
* getting dependencies of a message
* getting messages that depend on a message (dependents)
* getting messages to be delivered (by startup hook)
* getting pending messages to be possibly delivered (by startup hook)
In order to reflect the new states, things that were previously true for
VALID messages have been changed to now be true for DELIVERED messages.
Since pending messages should not be available to clients, many database
queries have been modified to only return results for delivered
messages.
All added methods and changes should come with updated unit tests.
Please note that the database version was bumped in this commit.
* force decline when two of our own identities are introduced to each
other
* throw away introduction requests to the same identity
(impossible to trigger from UI)
Closes#284
Clean up tests
* Broke up ConstantsTest (#280) - the key encoding parts are now in KeyEncodingAndParsingTest, the message encoding parts are in MessageSizeIntegrationTest
* Renamed the other integration tests in briar-android-tests
* Moved the integration tests in briar-android-tests to the top-level package, as they all involve code from multiple packages
* Separated DatabaseExecutorModule from DatabaseModule so we can use a different @DatabaseExecutor in integration tests
* Merged AppModule with AndroidModule (@ernir, this touches code you're working on but I don't think there are any conflicts)
* Renamed some TestUtils methods for consistency
* Used TestUtils.getRandomBytes() where applicable
Fixes#280.
See merge request !133
Create local state for clients at startup. #279
Most of the clients we've written so far use private groups shared with individual contacts and/or a local group that's not shared with anyone. To make it easier to ensure that the necessary groups exist when we need them, this patch allows clients to register startup hooks for creating their local state.
Fixes#279.
See merge request !131
Don't allow reentrant transactions
The database's transaction lock is reentrant, meaning that a thread that's already holding the lock can acquire it again. This would allow a thread that already has a transaction in progress to start another transaction, which could cause transaction isolation issues and/or lock timeouts on the database's internal locks.
Check that the current thread isn't already holding the lock when starting a transaction.
See merge request !127
The database's transaction lock is reentrant, meaning that a thread that's already holding the lock can acquire it again. This would allow a thread that already has a transaction in progress to start another transaction, which could cause transaction isolation issues and/or lock timeouts on the database's internal locks.
Check that the current thread isn't already holding the lock when starting a transaction.
This requires exposing the `containsContact()` method to the `DatabaseComponent`
and is needed for finding out efficiently whether a contact already exists.
Don't keep message status rows for groups that are invisible to the contact - this avoids the need to join the groupVisibilities table when selecting messages to offer or send. Add or remove status rows when group visibility changes.
