Christopher Berner

Ya, that's fairly straight forward, and is similar to the #271 issue I have filed already. The main thing I'll need to think through is the performance aspects: either the...

It's linear in the number of insertions/deletions, not the number of transactions, but ya that still seems better. Ya, the idea is that you can abort part of your transaction...

@casey I thought of a simpler and hopefully better implementation, which is to just save a complete copy of the allocator state in the checkpoint. How would you feel about...

Actually it's even easier than that. It doesn't free any pages that can be referenced by a checkpoint, just like when a read transaction is open. As an aside, I...

checkpoint is probably more common, but it means something different I think. SqlServer [says that a checkpoint](https://learn.microsoft.com/en-us/sql/relational-databases/logs/database-checkpoints-sql-server?view=sql-server-ver16) is a "known good point from which the SQL Server Database Engine can...

Oh ya, I should really rename that constant. It's actually the **default** max region size, and the user can even change it when creating the database by using `DatabaseBuilder::set_region_size`. The...

Oh, the main reason is because there was a compilation error: https://github.com/cberner/redb/issues/302 Then I reduced it to 128MiB, because I figured on a 32bit OS it's unlikely that someone would...

It has to be a multiple of the page size, and should really be a power of 2 since a tree allocator is used. I just [merged a change](https://github.com/cberner/redb/pull/333/files) to...

Hmm, I don't know much about PAE, but it sounds like it just allows the OS to run multiple processes that each use different physical address ranges, and the physical...

Ya, that's a reasonable concern. I think the best thing to do is to change that to a `u64` so that it's pointer width independent. Probably not a priority right...