| Website | https://val.packett.cool |
| Nerdy Linktree | https://keyoxide.org/aspe:val.packett.cool:DV7YKMH5QMHF5ZVU5UUSIXXXMI |
| Pronouns | she|they |
| Pronombres | ella|elle |
| Cuenta alt (en argentino) | https://rebel.ar/@val |
| Pixelfed | https://luzeed.org/@valpackett |
if anyone's unsure as to how GoToSocial performs so well, it's simply by not having enterprise tech brain.
code for the longterm. refactor things if you want to. stop focusing on horizontal scaling bullshit. microservices are not the answer.
PSA: do not ifdef struct fields in the header that aren't actually conditional in the ABI! Maybe don't ifdef struct fields, like, ever.
(Yes, reported.)
Apple: “Only now are computers so powerful that they can display UIs with complex physics-simulating ‘liquid glass’ effects”
Future Crew on a potato PC in 1993:
#lazyfedi random #3dprinting slicer question: Orca & co support opening solid models directly from STEP files. Apparently not just for convenience, Orca's random tip thing mentions that it's "more precise".
So where does the precision come from? Does the slicing directly use solids? Or does the slicer have a super cool precise mesher that's so much preciser than your CAD software? Or is it just about doing in-slicer transforms like scaling in BRep before meshing?
To elaborate - Patreon is testing the waters via survey to see how much pushback they might get if they tried rolling out AI (again, IIRC). These were sent randomly to *some* creators, but not all. The link works for *anyone* that has a Patreon so….if you hate AI (uncommon as that is in creators) this is your chance to shine.
Please bump/boost/retoot
https://patreon.sjc1.qualtrics.com/jfe/form/SV_9sjIUT9LuK5gzoq
#Patreon #Writers #SupportIndieArtists #Comics #WebComics #NoAI https://woof.tech/@alahmnat/115035358945745644
https://gitlab.com/qemu-project/qemu/-/issues/2574
kernel's cursed? kvm's cursed? hardware's cursed?
the debian rocm team using full PCIe passthrough was getting a "VM crashes with EFAULT when accessing GPU memory"… and i'm seeing the same symptom with two different non-passthru setups (qemu+gfxstream and crosvm+native-context) on an integrated gpu?!
"hm, let me finally investigate that UHS SD card issue"
"is there anything about the modes in the other dt- OHHH"
Now we're cookin'.
(This is still very slow cookin' because SD LOL, but in the traditional "high" speed mode the top number was 20 MB/s.)
@valpackett hm could it be ur missing a BAM node in ur dts?
for msm8960, Sam helped me figure out that single byte write to the sd card/emmc were fine, just big writes would cause it to fail
https://github.com/torvalds/linux/commit/5ee449c75f49c9a6b0cbff7848f922183e7888c1
this finally solved the problem for me
@Logical_Error bam hasn't been used for sdhc for quiiiite a while xD By sdm845 there were only cryptobam and slimbam; in current socs only cryptobam is left it seems.
This is not about writes; this is actually about reads, specifically in SDR104 mode (well maybe other UHS modes but not HS); and seemingly specifically early reads (like reading GPT/FS headers). And it wasn't seen on reference platforms, only on production laptops. Actually I should test adding delays on probe xD
kimothy siddon
Thomas 🔭🕹️
CautionWIP (he/him) 🏳️🌈🇨🇦🕎
LogicalErzor