Bulk insert (+ a few simple regexp improvements)
This gives an enormous performance boost when requesting more backlog (ie. by scrolling up). The startup time remained pretty much the same for me though and I'm not sure why exactly.
My C++ coding might still be a bit rusty, but I hope it saves some of your time at least. I think it's pretty clean :)
#1 Updated by EgS over 14 years ago
I love the concept, but it is sadly faulty... The problem is, that it's not possible to insert a bulk of messages of one buffer into the message model, by just taking that bunch and insert it where the first (or last) message of that bunch belongs.
The reason behind this: the message model is sorted by message Id and message Ids are unique for a whole core (not even only a user). So imagine 2 buffers A, B receiving messages:
10:30 A: msgA1
10:35 A: msgA2
10:32 B: msgB1
10:37 B: msgB2
That would result in msgIds:
1 - > msgA1
2 -> msgB1
3 -> msgA2
4 -> msgB2
So the way your patch handles the bulk inserts would destroy the order. This is at the time being not a visible issue, but is crucial for some future things like logically merged buffers or things similar to the chat monitor that show chat from a configured bunch of channels.
But I think it's definitely something we can work with, I have some Ideas to fix that order issue without going back to single inserts and also some tweaks.
Thanks for investigating!
#2 Updated by sph over 14 years ago
The bulkinsert works properly now :)
Some benchmarks: http://lithitux.org/~sphilips/quassel/benchmark.txt
It won't make a difference until the code in qtuimessageprocessor calls insertMessages() instead of insertMessage() (which I am doing in the benchmarks)
(New diff: bulkinsert2.diff)