garrison
@garrison.corporate.fm
I had no trouble reading this at all, I'm honestly kinda shocked the models can't do it
January 28, 2026 at 8:32 PM
I had no trouble reading this at all, I'm honestly kinda shocked the models can't do it
dw I will just write them all into an ets table this is fine
January 28, 2026 at 6:39 PM
dw I will just write them all into an ets table this is fine
tbh what really tilts me about this stuff is it inevitably snipes at least one decent person into replying in good faith because they don't recognize what's really going on
January 27, 2026 at 8:56 PM
tbh what really tilts me about this stuff is it inevitably snipes at least one decent person into replying in good faith because they don't recognize what's really going on
going forward all design should be centered around me specifically
January 27, 2026 at 6:52 PM
going forward all design should be centered around me specifically
sealing my signed records in a bottle and throwing them into the ocean
January 27, 2026 at 5:14 PM
sealing my signed records in a bottle and throwing them into the ocean
crossing the gleam line rn
January 26, 2026 at 5:39 PM
crossing the gleam line rn
Also glad to see bsky has grown enough that "dude whose entire post history is answers to the niche technical question you asked" is a thing here.
thanks for replying, I will enjoy reading your posts
thanks for replying, I will enjoy reading your posts
January 25, 2026 at 10:02 PM
Also glad to see bsky has grown enough that "dude whose entire post history is answers to the niche technical question you asked" is a thing here.
thanks for replying, I will enjoy reading your posts
thanks for replying, I will enjoy reading your posts
I will 100% try it at some point but my concern is that it will differ a lot between hw and the impls in question so unless I do a huge survey it won't mean much
I did not mean to assert the hypothesis, I was actually asking
I did not mean to assert the hypothesis, I was actually asking
January 25, 2026 at 9:58 PM
I will 100% try it at some point but my concern is that it will differ a lot between hw and the impls in question so unless I do a huge survey it won't mean much
I did not mean to assert the hypothesis, I was actually asking
I did not mean to assert the hypothesis, I was actually asking
weirdly most of the time it's way too low but every so often it comes out just right, so I think the algo is just off
I think what I want is an algo that tries to limit the number of printed lines while remaining balanced with nested terms
I think what I want is an algo that tries to limit the number of printed lines while remaining balanced with nested terms
January 25, 2026 at 4:27 AM
weirdly most of the time it's way too low but every so often it comes out just right, so I think the algo is just off
I think what I want is an algo that tries to limit the number of printed lines while remaining balanced with nested terms
I think what I want is an algo that tries to limit the number of printed lines while remaining balanced with nested terms
typing everything twice is just straight up annoying
January 24, 2026 at 7:50 PM
typing everything twice is just straight up annoying
In the long term Atmosphere will be inevitable, only early adopter nerds are gonna care enough to capitalize the T
January 24, 2026 at 3:04 PM
In the long term Atmosphere will be inevitable, only early adopter nerds are gonna care enough to capitalize the T
strict serializability is a human right
January 23, 2026 at 7:21 PM
strict serializability is a human right
apparently there's a nice wiki article with dates. I guess AMD had them earlier
en.wikipedia.org/wiki/SHA_ins...
en.wikipedia.org/wiki/SHA_ins...
SHA instruction set - Wikipedia
en.wikipedia.org
January 23, 2026 at 7:11 PM
apparently there's a nice wiki article with dates. I guess AMD had them earlier
en.wikipedia.org/wiki/SHA_ins...
en.wikipedia.org/wiki/SHA_ins...
It depends on the CPU. IIRC intel/amd x86 have had the instruction since ~2020. Not sure about ARM.
Also depends if the impl you're using actually uses the hw accel of course.
Also depends if the impl you're using actually uses the hw accel of course.
January 23, 2026 at 7:05 PM
It depends on the CPU. IIRC intel/amd x86 have had the instruction since ~2020. Not sure about ARM.
Also depends if the impl you're using actually uses the hw accel of course.
Also depends if the impl you're using actually uses the hw accel of course.
Tigerbeetle switched from blake3 to AEGIS here: github.com/tigerbeetle/...
VSR: use Aegis128 over Blake3 for checksum by kprotty · Pull Request #651 · tigerbeetle/tigerbeetle
Ported from the stdlib's std.crypto.aegis.State128L with the output/cipher portions removed and only returning the mac. This version exists already as std.crypto.aegis.Aegis128LMac but it's...
github.com
January 23, 2026 at 5:15 PM
Tigerbeetle switched from blake3 to AEGIS here: github.com/tigerbeetle/...
Would blake3 actually be faster than sha256? My understanding is that sha256 is faster on modern CPUs due to some new instructions
Also, the cool (database) kids are using AEGIS which is *way* faster, e.g.
github.com/tursodatabas...
Also, the cool (database) kids are using AEGIS which is *way* faster, e.g.
github.com/tursodatabas...
Add support for AEGIS encryption algorithm by avinassh · Pull Request #2742 · tursodatabase/turso
Depends on #2722
This builds upon the #2722 PR which let me configure the encryption algorithm. So this adds AEGIS and uses it as default.
Note that choice of cipher at higher APIs is still not pos...
github.com
January 23, 2026 at 5:11 PM
Would blake3 actually be faster than sha256? My understanding is that sha256 is faster on modern CPUs due to some new instructions
Also, the cool (database) kids are using AEGIS which is *way* faster, e.g.
github.com/tursodatabas...
Also, the cool (database) kids are using AEGIS which is *way* faster, e.g.
github.com/tursodatabas...
redesigned from PDSLS to PDSls
January 23, 2026 at 1:04 AM
redesigned from PDSLS to PDSls