Manuel Pöter

Yes, the reason is that the `ramalhete_queue` can only store raw pointers. For `unique_ptr` we can simply use `release` in `push` to get the raw pointer and take over ownership,...

Hi @LimHyungTae, thank you for the kind words! It seems the problem is not MacOS but the ARM architecture. I always wanted to add support for ARM but unfortunately I...

@LimHyungTae I switched to CircleCI, made the adjustments for ARM support and added some Linux ARM CI jobs. I still need to add some Mac jobs, but perhaps you can...

@altras please note that the RocksDB compaction temporarily needs _more_ space. Compaction means that one or more sst files will be combined to a new one, but obviously the old...

Thank you, that is great! How urgent do you need it? I am currently working on some small fixes for ARM but need a bit more time. I would prefer...

Hi Daan! All good, thanks for asking. :) I guess I wasn't clear enough. I would like to be able to change some options before mimalloc is initialized, _but if...

Yes, I think that should work - thank you!

Thanks for reporting, this is an interesting one! The problem is the following - at some point the `foo` instance will be reclaimed which calls the destructor. There you are...

If you can safely and deterministically release the resources yourself (manually) this is usually always preferable. Most reclamation schemes have to defer reclamation at some later time when it is...

I originally found this bottleneck while profiling ArangoDB which uses RocksDB as its storage engine. It looks like the only benchmark in `db_bench` that uses snapshots is `randomwithverify`. I ran...