01:51 <+bridge_> <gorp_tw> i think chillerdragon should have rights
01:51 <+bridge_> <gorp_tw> anyone else agree?
02:03 <+bridge_> <teero777> human ones? yes
02:34 <+bridge_> <pioooooo> Just go through the process ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯
06:54 <+bridge_> <manikoo> ofc
08:35 <+bridge_> <ryozuki> good morning humans and others
08:37 <+bridge_> <melon> good norming
08:37 <+bridge_> <melon> -# damn it felt good saying this again..
08:42 <+bridge_> <chillerdragon> I recently struggle to choose between struct and array because sometimes i want to access individual fields with a descriptive name and sometimes i want to iterate all
08:42 <+bridge_> <chillerdragon> i ended up creating arrays with named indecies
08:42 <+bridge_> <chillerdragon> but i had another crazy idea
08:42 <+bridge_> <chillerdragon> in scripting languages you have these hashes,objects,table thingies where you can just iterate over the key value pairs if you need it
08:42 <+bridge_> <chillerdragon> and a C++ struct basically is also a key value pair
08:43 <+bridge_> <chillerdragon> and if all members are the same type it could theoretically be iterated with sizeof() and pointer increment magic? does that make sense? does anyone do that? xd
08:43 <+bridge_> <melon> cursed C
08:45 <+bridge_> <milkeeycat> @learath2 After more thinking about this, I think the cleanest approach would be to have 2 versions of HIR and one target specific IR. First HIR would be used as input and it would have high level concepts like structs, and second HIR(SIR? Simple IR or LIR. PIR? Plan IR) wouldn't have structs but it would still have most of the instructions from the first HIR, and then this IR would be used for instruction selection.
08:45 <+bridge_> <milkeeycat> 
08:45 <+bridge_> <milkeeycat> So the pipeline would look like:
08:45 <+bridge_> <milkeeycat> HIR -> lower high level concepts such as structs to simple HIR instructions, build SIR/PIR -> do instruction selection on SIR/PIR which would build MIR
08:45 <+bridge_> <milkeeycat> 
08:45 <+bridge_> <milkeeycat> What do you think?
08:46 <+bridge_> <chillerdragon> this indeed prints 1 2 3 xd
08:46 <+bridge_> <chillerdragon> https://paste.zillyhuhn.com/M1
08:46 <+bridge_> <chillerdragon> no way this worked first try hahaha so cursed
08:47 <+bridge_> <chillerdragon> i might use this
08:47 <+bridge_> <melon> why on gods earth would you want to do this :kekw:
08:47 <+bridge_> <chillerdragon> if i want to iterate all values
08:47 <+bridge_> <chillerdragon> of a struct
08:48 <+bridge_> <melon> just write a proper iterator D:
08:48 <+bridge_> <chillerdragon> so i dont have to type out every field manually
08:48 <+bridge_> <chillerdragon> a what now
08:48 <+bridge_> <chillerdragon> you can write iterators for your own objects?
08:48 <+bridge_> <chillerdragon> yea i guess
08:48 <+bridge_> <chillerdragon> but bru that sounds so c++
08:50 <+bridge_> <melon> oh no. C++ in the big 26... :kekw:
09:38 <+bridge_> <gorp_tw> ew c**
10:07 <+bridge_> <learath2> Sounds sane-ish, but is it really worth having an entire extra IR that is basically exactly the same as HIR?
10:10 <+bridge_> <milkeeycat> Ye, that's a fair point, but I think it's better to use types to enforce invariants(not have structs) rather than having asserts everywhere, no? That's a tough choice :\
10:11 <+bridge_> <milkeeycat> Ye, that's a fair point, but I think it's better to use types to enforce invariants(not having structs) rather than having asserts everywhere, no? That's a tough choice :\
10:11 <+bridge_> <learath2> This assumes there is no padding between struct elements, which is not strictly true. The only guarantee C gives is that the start of a struct is not padded
10:14 <+bridge_> <learath2> Well, yeah it would be more "proper", then why can't you lose it when going to MIR instead? Are there some optimizations you want to do in HIR?
10:19 <+bridge_> <milkeeycat> Hhhhm, I didn't think it xddd. I always thought about these translations like "you have structs, you decompose them into primitive instructions, and then that simplified IR is used for instruction selection", maybe I could have some custom logic during lowering which would first decompose the instructions/operand and then lower it to MIR
10:19 <+bridge_> <milkeeycat> I kinda like it
10:20 <+bridge_> <milkeeycat> Hhhhm, I didn't think it xddd. I always thought about these translations like "you have structs, you decompose them into primitive instructions, and then that simplified IR is used for instruction selection", maybe I could have some custom logic during lowering which would first decompose the instructions/operands and then lower them to MIR
10:51 <+bridge_> <milkeeycat> Hhhhm, I didn't think about it xddd. I always thought about these translations like "you have structs, you decompose them into primitive instructions, and then that simplified IR is used for instruction selection", maybe I could have some custom logic during lowering which would first decompose the instructions/operands and then lower them to MIR
11:03 <+bridge_> <aegisub> chillerdragon: C++26 reflection will fix
11:04 <+bridge_> <aegisub> but until then consider either enums for named indices into your array or a helper function to iterate over all entries of your struct
11:49 <+bridge_> <milkeeycat> @learath2 I'm not sure if there will be any problems during lowering from HIR(with structs) to MIR, but I like this idea and will go with it. Thanks for the help :deen_star:
11:50 <+bridge_> <learath2> Hey, if there is a problem we didn't foresee, then you'll have a very good reason to introduce one more IR 🙃
11:54 <+bridge_> <aegisub> just one more ir guys I promise just one more
12:12 <+bridge_> <Assa> hey @learath2 my love, may I ask you a favor? 🙈
12:31 <+bridge_> <learath2> you can always ask, idk if I can deliver 😛
13:04 <+bridge_> <Assa> please review the envelope trigger tiles 🙈 I think this is huge, but it may have some technical debt
13:23 <+bridge_> <heinrich5991> @ryozuki I have data in one format (sqlite3) and want data in another format (json that another program consumes), converted with a python script. what is your setup for LLM coding? I think this may be a fairly well defined problem that LLMs might be able to solve to my satisfaction
13:26 <+bridge_> <ryozuki> i use claude with opus 4.5, though with "sub agents" you can use lower cost models for tasks like reviewing or exploring code, and then the costlier one to plan
13:26 <+bridge_> <ryozuki> 
13:26 <+bridge_> <ryozuki> afaik codex doesnt provide this yet
13:26 <+bridge_> <ryozuki> 
13:26 <+bridge_> <ryozuki> iguess i would put the sqlite db in a repo/folder, tell it to analyze it, (mention the sqlite3 tool and it will do it automatically) and write a python script
13:26 <+bridge_> <7.5i> omg hi
13:27 <+bridge_> <ryozuki> generally if you tell it to make a implementation plan / explore it and get context first it does way better
13:27 <+bridge_> <ryozuki> explore -> plan -> implement
13:30 <+bridge_> <heinrich5991> I'm not going to spend 200$ for this experiment. so I either find someone with such a subscription or I could try the 20$ subscription on claude.com, unless you tell me that's worthless trying
13:30 <+bridge_> <heinrich5991> I'll get back to you when I have more specifics or need help 🙂
13:33 <+bridge_> <heinrich5991> thanks for the hints so far 🙂
13:34 <+bridge_> <ryozuki> @heinrich5991 i can do it for you if you want
13:34 <+bridge_> <ryozuki> but for the task u mentioned, the 20$ will work
13:34 <+bridge_> <ryozuki> you can use opus 4.5 it just will use your "usage" faster
13:34 <+bridge_> <ryozuki> but sonnet should do the trick for this
13:34 <+bridge_> <ryozuki> i dont even need to tell your comments to the ai
13:34 <+bridge_> <ryozuki> https://cdn.discordapp.com/attachments/293493549758939136/1468947936967921768/image.png?ex=6985dfdf&is=69848e5f&hm=d12b152551f3ab94db8ee22777abcedc3a2e8777cb908a5fd873bf591f559a5b&
13:37 <+bridge_> <ryozuki> we cooked
13:37 <+bridge_> <ryozuki> https://cdn.discordapp.com/attachments/293493549758939136/1468948623092875359/image.png?ex=6985e083&is=69848f03&hm=9963807b89aeb437b2dcafd779f4b762186ba645e59c989fb513c197a073d4ff&
13:39 <+bridge_> <ryozuki> idk if u meant only the changes in the pr or previous code is already repeated
13:39 <+bridge_> <ryozuki> if its the latest i can change that
13:40 <+bridge_> <remakepower> I question something in #7666 just now:
13:40 <+bridge_> <remakepower> https://github.com/ddnet/ddnet/issues/7666#issuecomment-3853364593
13:40 <+bridge_> <remakepower> :thonk:
13:40 <+bridge_> <DDNet> https://github.com/ddnet/ddnet/issues/7666
13:43 <+bridge_> <remakepower> wait a minute.
13:44 <+bridge_> <ryozuki> btw i wanted to share a app i made in flutter and to share on ios you need to pay 99$, you cant easily share a apk like in android and sideload it
13:45 <+bridge_> <ryozuki> it just makes me hate apple more
13:45 <+bridge_> <ryozuki> btw i wanted to share a app i made in flutter and to share on ios you need to pay 99$/yearly, you cant easily share a apk like in android and sideload it
13:46 <+bridge_> <aegisub> the funniest part is when you try to register and *want* to give them money but they lock you out of the developer program for reasons they cannot even tell you
13:46 <+bridge_> <aegisub> (this has happened in one of my projects)
13:46 <+bridge_> <ryozuki> F
13:47 <+bridge_> <ryozuki> the funniet part is my gf has a iphone and i "sideloaded" it on her phone via xcode, aka compiling it and "running" it, but apparently that has a expire time too
13:47 <+bridge_> <ryozuki> its crazy
13:57 <+bridge_> <heinrich5991> your comment on that issue doesn't contain a question and it's also quite terse. can you elaborate in that comment?
13:57 <+bridge_> <heinrich5991> war on general computing 🎉
13:59 <+bridge_> <Assa> AI didn't impress me yet with your PRs, but it's close. Still required human intervention
14:01 <+bridge_> <remakepower> ok, I did.
14:05 <+bridge_> <heinrich5991> @remakepower you ask why are we blocked by demo skipping. then you answer your own question with a quote of my message
14:05 <+bridge_> <heinrich5991> I still don't understand what you want to say… it seems to me that you see some contradiction, but you don't point it out
14:05 <+bridge_> <remakepower> > we already parse every single message from the start to where want to seek to seek.
14:06 <+bridge_> <heinrich5991> ah, you're saying that we already parse the demo from start
14:06 <+bridge_> <heinrich5991> I don't think that's the case
14:08 <+bridge_> <remakepower> If that's not, then why the client could display 0.7 skins correctly
14:08 <+bridge_> <remakepower> *after they skip a part of demo.
14:08 <+bridge_> <ryozuki> yeah, tho those are simple prs, and i could do better setup, like making it analyze the codebase for style, conventiosn and save it to a file to reference it, etc
14:08 <+bridge_> <heinrich5991> because the client artificially adjusts the snapshot in demos to contain player information
14:08 <+bridge_> <heinrich5991> because the client artificially adjusts the snapshot in demos to contain player information in 0.7 demos
14:08 <+bridge_> <ryozuki> if u give me a pr idea thats not making accounts i can try
14:09 <+bridge_> <remakepower> ah
14:09 <+bridge_> <remakepower> no
14:09 <+bridge_> <remakepower> there's nothing in the snapshot related to player skin,.
14:09 <+bridge_> <melon> add localization to all strings in the editor :WICKED:  /s (please dont, no translator would ever do this)
14:09 <+bridge_> <ryozuki> xD
14:10 <+bridge_> <remakepower> :thonk:
14:10 <+bridge_> <remakepower> wait
14:10 <+bridge_> <ryozuki> i saw you changed the translation system right
14:10 <+bridge_> <melon> yeah, we're using weblate now
14:10 <+bridge_> <remakepower> ah yeah we have
14:10 <+bridge_> <heinrich5991> not really, we just use weblate as a translator UI
14:10 <+bridge_> <heinrich5991> the translation format hasn't changed (so far)
14:11 <+bridge_> <remakepower> But we can remove those things.
14:11 <+bridge_> <remakepower> idk why
14:11 <+bridge_> <melon> > (so far)
14:11 <+bridge_> <melon> 
14:11 <+bridge_> <melon> are we planning to? :o
14:11 <+bridge_> <remakepower> I tested CNetMsg_Sv_KillMsg
14:11 <+bridge_> <heinrich5991> I'd love to get away from our home-rolled translation formats
14:11 <+bridge_> <ryozuki> what format you thinking?
14:11 <+bridge_> <heinrich5991> I'd love to get away from our hand-rolled translation format
14:11 <+bridge_> <melon> a lot of projects use JSON
14:12 <+bridge_> <heinrich5991> that's the hard question
14:12 <+bridge_> <ryozuki> did you port to sdl3 yet
14:12 <+bridge_> <heinrich5991> that's not a translation format on its own, you'd need to specify the schema
14:12 <+bridge_> <pioooooo> I think even if they don't, nothing gets worse :)
14:14 <+bridge_> <heinrich5991> https://github.com/ddnet/ddnet/pull/10246
14:14 <+bridge_> <heinrich5991> current PR
14:14 <+bridge_> <heinrich5991> I'd guess the hard work is the testing, not producing the code
14:15 <+bridge_> <ryozuki> yeah
14:15 <+bridge_> <heinrich5991> translations can sometimes be worse than no translations, e.g. if the translations are wrong
14:15 <+bridge_> <ryozuki> last time i tried sdl3 wasnt on gentoo anyway
14:15 <+bridge_> <ryozuki> but thats probs a long time ago
14:16 <+bridge_> <ryozuki> it exists now
14:20 <+bridge_> <remakepower> :thonk:
14:21 <+bridge_> <remakepower> ah ok
14:21 <+bridge_> <remakepower> I got it.
14:25 <+bridge_> <aegisub> is there even anything other than gettext (and frameworks like qt and wx that implement their own gettext) available for c++
14:25 <+bridge_> <aegisub> ignoring self-rolled solutions
14:28 <+bridge_> <aegisub> in the javascr*pt world there's i18next but that's not for c++
14:28 <+bridge_> <learath2> would be nice if whatever we use can be exported by weblate directly, it's annoying to have a transform in the middle
15:06 <+bridge_> <Assa> idk, do the material layer i.g. 😄 just for fun
15:07 <+bridge_> <Assa> or water
15:25 <+bridge_> <heinrich5991> gettext or maybe something like fluent (which will power JS internationalization, AFAIUI)?
15:25 <+bridge_> <heinrich5991> gettext is annoying because we'd probably have to use a third party library to be able to change language during runtime
15:29 <+bridge_> <ryozuki> xD
15:29 <+bridge_> <ryozuki> 54k lines is crazy
15:31 <+bridge_> <ryozuki> i dont have clang-format 20
15:31 <+bridge_> <ryozuki> only 21
15:40 <+bridge_> <melon> :OMEGALUL: :OMEGALUL: :OMEGALUL: :OMEGALUL:
15:41 <+bridge_> <ryozuki> :owo:
15:43 <+bridge_> <01000111g> I don't even know if localizing the editor is a good idea. This way noone can help in the mapping channel anymore, if you dont even have the same word for obscure things like paralax
15:44 <+bridge_> <ryozuki> ¯\_(ツ)_/¯
15:45 <+bridge_> <ryozuki> feel free to close the pr if its unwanted, i just left some heavy work done
15:45 <+bridge_> <ryozuki> i agree its a double edged sword
15:45 <+bridge_> <ryozuki> but usually accesibility is better ig
15:45 <+bridge_> <ryozuki> maybe translating it makes them ask less
15:46 <+bridge_> <ryozuki> also idk an spanish translation of paralax so maybe some words are just same
16:14 <+bridge_> <teero777> yo is there any discussion about gameskins being shared to the server&client? i mean it would be cool to see other peoples gameskins
16:16 <+bridge_> <teero777> this wouldn't introduce any backwards incompatibility and they can be treated just as skins with an external db and reference by name
16:18 <+bridge_> <Assa> it probably wont work in some cases, where it needs `Localizable` as well, so it's not even working
16:19 <+bridge_> <Assa> yeah, make that a feature so I can confuse my enemies by switching the grenade and laser sprite
16:19 <+bridge_> <teero777> that would be amazing
16:19 <+bridge_> <Assa> when we're at it, add a skin shop or loot boxes
16:19 <+bridge_> <teero777> huh?
16:20 <+bridge_> <teero777> why the sarcasm tho? it would be the same as normal tee skins
16:37 <+bridge_> <melon> ddnet-rs allows that
16:39 <+bridge_> <louis> theres a lot less good gameskins unfortunately
19:34 <+bridge_> <syrinok> Plenty of good ones, just have a channel to submit new ones and be more strict when adding them
19:37 <+bridge_> <syrinok> At the very least, having a reliable source of gameskins would be nice, one that doesn't just vanish because someone decided to shut it all down
21:17 <+bridge_> <ryozuki> i would fix it but its not wanted anyway
22:03 <+bridge_> <learath2> Yeeeah, I think keeping the editor one language isn't the worst idea
22:24 <+bridge_> <pioooooo> Those understand English decide for those who don't
22:26 <+bridge_> <kebscs> just some important parts would be enough
22:26 <+bridge_> <kebscs> at least tile explainations and the most basic buttons