core: Fix Postgres messagesNewerThan wrong buffer

[digitalcircuit]

In short

• Fix the PostgreSQL select_messagesNewerThan statement limiting to the wrong buffer
  • Replace use of message ID $1 for bufferid in inner SELECT with the actual buffer ID, $2
  • Fixes using the wrong buffer for lastmsgid limit, resolving missing messages with unread fetching
  • Fixes regressions from pull request 273, missed in pull request 274

Criteria	Rank	Reason
Impact	★★★ 3/3	Fixes missing messages with Postgres unread backlog fetching
Risk	★☆☆ 1/3	Variable change shouldn't break more, risk accepted in previous PRs
Intrusiveness	★★☆ 2/3	Minor schema change, might interfere with other pull requests

Apologies for missing this in pull request 274. There, I tested that loading backlog worked for new setups. I did not test on a larger data-set.

Note: This does not fix the client-side issue of generating the invalid IDs. That still needs fixed in a follow-up.

Example

Before

Quassel with unread fetching misses new messages, e.g. Test message does not show, Freenode does not show activity in status buffer

After

Unread fetching grabs the new messages, e.g. Test message does show, Freenode shows activity in status buffer

Side-by-side (production)

Quasseldroid uses fixed fetching, Quassel desktop uses unread fetching, both on the git master core. Notice the #sandstorm channel and the Freenode status buffer

[Nevcairiel]

Does this automatically apply to cores that already updated to git master today?

[digitalcircuit]

@Nevcairiel If you upgraded to the latest commit (master as of today, 570fe57 ), got the schema upgrade, then upgrade to this, yes, it will fix the issue (no schema upgrade required). My testing involved swapping the binaries for with and without this PR, no database changes needed.

[digitalcircuit]

@egs-me Belated apology for the hassle over missing this in pull request #274. I'll be more careful with future testing.

[MrEgs]

OK, that's an obvious error and a clear fix.

@digitalcircuit No worries. It's not the first time that I've broken quassel:master. ;-) Unfortunately, I really do not have any time to do properly testing myself or even testing at all. So I have to rely on the testing of others and judge by the diffs, which, by and large, looked well.

Anyways: thanks again to everybody involved in this entire affair. All of you did valuable work. And although, the master branch should always be in a working state, I'm quite sure that it can overcome a small hickup.

Cheers,
Marcus