Jamie Brandon

Oops. Do `git clone https://github.com/jamii/imp ../imp`. I added a note to the readme.

Tangentially related - you can remove autocomplete on return and just use it on tab: ``` :editor.keys.hinting.active {"tab" [:auto-complete] "enter" [:passthrough]} ``` That's probably why I haven't noticed that autocomplete...

Probably, although I don't recommend using focus at all :)

Everyone was :+1: @benesch is taking the first change, @jamii the second.

I hit possibly the same crash in tigerbeetle with zig 0.9.1. As a workaround, I found that the alloc api seems to work fine: ``` zig ___tracy_emit_zone_begin_alloc(___tracy_alloc_srcloc( src.line, src.file.ptr, src.file.len,...

This is probably not correct even with many linux filesystems, depending on the settings. Operations on directories (ie creating files) can be reordered arbitrarily except before/after an fsync on the...

I'd recommend writing the caching code so that it can be run against a mocked filesystem which reorders operations at random unless prevented by an fsync. Or even just throws...

Read/write iop usage during a single measure.  A lot of empty space as some compactions finish and just a few are left running.

Also noticed that: * In the first 31 ops we typically do one compaction tick per tree. * Every 32nd op we do the remaining ~2k ticks.

Being cpu-bound definitely explains other mysteries, like why increasing the cache size and slashing disk reads didn't produce any throughput improvement.