Description

#264 implements sharing of all hypercore keys over extension messages. This introduces a small security risk: a bad actor who has a project key could make other clients add lots of cores that do not actually belong to the project and inject data.

Ideally we change this before the MVP because it avoids having to maintain backwards compatibility.

One solution to this is to verify whether cores belong to a device that is part of the project before indexing them, but this still means that disk space could be taken with cores that are not part of the project.

A more robust solution (but we should maybe also do the solution above) is to only add auth namespace cores via extension messages (as we were doing before), and to add other cores once we know that they are part of the project.

"being part of the project" means in this case that any capability record refers for a given device, because we still want to include cores that are from a device that was previously part of the project and is now "blocked" (if we want to exclude that data we should do so at index time, or avoid downloading their data at sync time, but we should still keep the cores in the project).

We will also need to check the coreOwnership records to get the actual core keys for a device.

Unfortunately our indexing is unordered, so we might get core ownership records before capability records, or vice-versa.

One easy way to solve this problem is to listen for the index-done events, and for each event read all the capability records, look up the corresponding coreOwnership records, and add any missing cores. However this could add a lot of overhead - index-done happens a lot, e.g. whenever a user adds or edits data, and on every sync, and in many cases there will be no cores to add.

A more efficient way would be to add an event whenever a new coreOwnership or capability record is indexed. We could then listen on both events:

1. coreOwnership record added -> look up capability record for device -> if it exists then add cores listed in coreOwnership record, otherwise ignore
2. capability record added -> look up core ownership for device -> if it exists add cores, otherwise ignore.

Tasks

• Wait for indexing to be done before returning data #228
• Add a onValue() method to DataType - I think this is better than just emitting a value event for every record, because that would add a lot of unnecessary overhead emitting thousands of events.

  There isn't really an overhead for emitting a value for every record, since if no listeners are attached this is just a no-op, at least there is no more overhead than doing a check for listeners ourselves. However attaching a listener to a datatype with lots of records, like the observation type, would have a huge overhead, because it would be called hundreds or thousands of times after an initial sync, so we should probably not expose this to the API and just use it internally for records like core ownership and capabilities where we know there are not many records.

• feat: dataType.on('updated-docs') when docs update #396
• Add a capabilities.on('update', updateCapabilities) event. Internally it needs to listen for both new role records, but also new coreOwnership records, because we only know the role of the project creator via their core ownership record (because they own the auth core with the key that matches the project key).
• Add cores once capabilities are read and a device is confirmed as having a role in the project
• Only send "pre-haves" to devices that have sync capability for a particular namespace