[backport 0.14] client: Fix AsNeededBacklog hang with old cores

[digitalcircuit]

In short

• Fix default Only fetch when needed hang with pre-0.13 cores
  • Fixes backlog fetch taking 3× time compared to 0.13 client
  • Impacts clients with Ubuntu 18.04 server (Quassel core 0.12.4)

Criteria	Rank	Reason
Impact	★★☆ 2/3	Fixes backlog fetch regression with cores 0.12.5 and older
Risk	★☆☆ 1/3	Mimics well-tested fixed backlog fetch method
Intrusiveness	★☆☆ 1/3	Small set of changes, shouldn't interfere with other pull requests

Details

For whatever reason, if the client doesn't call updateProgress(...); after initial backlog request, it hangs.

To fix this, AsNeededBacklogRequester has been changed to set buffering/isBuffering() to true on initialization, which causes the client to call updateProgress().

Why does the disabled GlobalUnreadBacklogRequester need to set buffering to false initially? Why does this cause the client to hang? Should buffering always be initialized to true..?

This should be investigated further in the future.

Testing

Steps

1. Set up pre-0.13 Quassel core
   • One option: install Ubuntu 18.04 LTS in a VM, then sudo apt install quassel-core
2. Restore an SQLite (or PostgreSQL?) database with many buffers
   • Tested with 156 buffers
3. Connect 0.14 client, do initial setup to use the restored database
4. Set client to Fetch only when needed backlog fetching mode
5. Time how long connecting takes
6. Apply this patch
7. Rebuild, time how long connecting takes

Test case

• 167 MB SQLite database from 2014 with 156 buffers
• 0.12.4 core (Ubuntu 18.04)
• 0.14-git client (Kubuntu 20.04)

Before

• Average 9 seconds to finish backlog fetch

After

• Average 2 seconds to finish backlog fetch