Time Nick Message 00:00 MTDiscord Could be a dependency changed? 00:00 MTDiscord Unlikely I know 00:54 fishmongler well, it looks like 5.5 is affected too it's just rare 00:55 fishmongler i'd say our problem is that we have two competing scene graphs in one engine 01:21 fishmongler i also suspect that we're processing too much of each entity at once sequentially, instead of doing the processing in clearly defined waves 01:22 fishmongler and we ought to sort the entities by their generation (depth) in the transform hierarchy 01:26 fishmongler apply pending hierarchy changes -> make a list of CAOs sorted by transform depth -> play all animations and scripted movement -> evaluate absolute transforms, applying bone overrides along the way -> estimate bounding boxes for early culling -> cull early -> skin -> refine bounding boxes based on skinning -> re-cull those -> render 12:50 MTDiscord That works great for now, but in the future when we start using vectorized processing, that will hurt performance. I'd say it's better to just fix the pipeline issue than to crutch with waves 13:39 fishmongler @exe_virus wait i'm confused, i thought doing a lot of the same operation in parallel is more vectorization and thread friendly than doing it sequentially like we're doing now; what i am proposing is fixing the pipeline issue 13:40 fishmongler we could make it fully modern and make it a generic job system that transforms buffers of data arranged in a huge dependency graph 14:17 MTDiscord Ah, if you are grouping like operations, then yes, if you are using sorting via distance, you will be separating like operations due to distance 14:21 fishmongler It's not a sort by distance, it's a sort by transform hierarchy depth. Traverse the entire scene graph noting the hierarchical depth of each scene node. Unattached entities are depth 0, their root bones or trivial attachments are depth 1, etc. Then sort the list of scene nodes by this depth and stick to this order for the rest of the frame for 14:21 fishmongler anything animation or physics related, that way you won't have race conditions where a dependent node is evaluated before its parent, using a matrix that's stale by one frame. 14:22 fishmongler In addition, having a data oriented architecture with data buffers and a pipeline of jobs processing them would make it impossible to use a stale frame's data, as it would live in a buffer that's invalid for the current frame. 14:26 MTDiscord Hello Did You received Google warning about minetest api for the 31 aug ? 14:27 [MTMatrix] yes, should be no problem since we're at 33 already 14:27 [MTMatrix] I mean it should be easy to update 14:31 ROllerozxa let's hope 5.9.0 will be done by then 14:40 MTDiscord Ok 15:33 rubenwardy FYI: F-Droid have concluded that the update detection doesn't require the "Tracking" anti-feature 15:43 pgimeno thanks for the update (pun unintended) 15:44 fishmongler On a technicality I'd say, or the guy didn't understand the issue too well 15:46 fishmongler You can only "customize this" by bricking CDB entirely at the moment 15:47 fishmongler and nobody tells you that downloading a single package signs a deal with the devil to ping a server from the singleplayer tab until all packages are uninstalled 15:51 fishmongler It's also a gigantic chore to download and install anything manually on Android, in general, so 100% of the users will use the tab 15:53 celeron55_ if nothing else, it would be at least polite to notify the user once they've installed their first cdb package that "updates will be checked on startup" 15:55 fishmongler @celeron55_ if it's just a notification and not even a question, that's alt+f4, uninstall, dunk pc in holy water territory 15:57 fishmongler the discussion on this has been so ridiculous i'm starting to believe that anon talking about coredevs being compromised and forced to cooperate 15:57 fishmongler if you've been served with a secret court order, blink twice. 16:03 fishmongler greetings user. you've just activated our trap feature, by having clicked this button you have hereby consented to have sad dark men come into your house and smelt your computer for mese ore. you have three seconds to yank the cat5 out of your wall 16:06 celeron55_ fishmongler: well i'm rather open about this, but i'm writing my recommendation in the github issue right now 16:23 celeron55_ "React is in the Longhorn phase now, MS has a huge incentive to break as much compat as they can get away with, as quickly as possible, because an actively supported clone of 7 might become a major threat." 16:23 celeron55_ i like this thinking 16:24 celeron55_ there may be reason to not dropping windows 7 support any time soon, as long as compilers and libraries permit doing so 16:24 rubenwardy ah yes I have nefarious uses for user's IP addresses and operating systems. I like to print them out and laugh at the obscure distros. Mw hahahaha! 16:25 rubenwardy not to mention the kick backs I get from the deep state and big tech 16:26 celeron55_ fishmongler: i liked the 4chan take on the subject. a fresh perspective! 16:28 MTDiscord that incident with gitgud.io devs is actually real 16:30 fishmongler so drybones has been glowed? that's very interesting, when did that happen 16:33 fishmongler not like i feel bad for him one bit, gods forbid 16:36 MTDiscord happened a few years ago, not going to mention details though. the project is allowed to exist 16:39 fishmongler sounds like a scary campfire story akin to all the rumors about moot back in the day 16:39 fishmongler "a few years ago" is very likely still after those people stopped being relevant in any way, and "is allowed to exist" is laughable because, well, they're irrelevant 16:40 fishmongler i don't see any reason why intelligence would want to compromise services adjacent to a community they've already destroyed and dispersed 16:41 rubenwardy I have no idea what you're talking about 16:42 MTDiscord whatever, then it's not true 16:43 rubenwardy jesse, what the f are you talking about 20:57 rubenwardy Merging #14837 20:57 rubenwardy In 10 20:57 ShadowBot https://github.com/minetest/minetest/issues/14837 -- Set ContentDB updates detection cache time to 24 hours by rubenwardy