reworked my compilation logic to work on expressions created from both syntax and values
much cleaner
and now this finally works:

(tget :hello '{:hello "hello, world!"})

OMG THANK YOU <3

>>109992750
could you at least use non-slop maid pictures, internet is full of them

>>109992737
anything where there is an optimal path is competitive retard

>still programming in 2026
lol lmao even

File: mid_img_0554.jpg (208 KB, 900x675)

208 KB JPG

>>109992750
fucking maid shitters. I was koding in the temple. Fuck you, I wanted to make a real thread for once

File: crying maid.jpg (211 KB, 850x963)

211 KB JPG

>>109992779
I used this one because it was in the last thread and I thought this maid is really pretty. I didn't notice that one of her hands has six fingers before I posted it. If you have a maid you'd like to see in the next OP, please post her.

>>109992802
the maids more diligent then you

>>109992802
The maid thread is the real thread.

>>109992811
>diligent
>doesn't kode but only waits for bump limit
K R I N G E
https://i.4cdn.org/wsg/1790273653398767.mp4

>>109992858
just a small part of being a maid. how many seconds are there in a day again?

Have you fuckers ever used the Fifth Normal Form (5NF)? Or it's just CS larp?

>>109992983
I just use AI

>>109992983
I thought there are only 3? I probably learned 3 times only 3

>>109992983
I just humanvibe my databases

>>109992991
>>109992993
1NF Atomic values
2NF Depends on the whole key
3NF Depends only on the key
4NF No independent multivalued dependencies
5NF No unnecessary join dependencies

>>109993013
I get the 5th, but I thought the 3rd implies that there is a one-to-one mapping of key-value. So the 4th wouldn't even exist, no?? That's why you get all the key tables

File: merged-image-1791261474325.jpg (1.05 MB, 1536x2223)

1.05 MB JPG

protocol(Foo, {
 rel p(x)
  start x=2
  y=1 until y=2 and x<2 or always x=1
  eventually...
})

So I've this crazy idea that since the language has goals then they could be used for a proofs or a model checker or some such without changing the base syntax ('x>2' itself is already a constraint) and I saw mentions of TLA+ in some website. basically you write a relation inside

protocol

that gets recognized as such, maybe macros if they get added. So in addition to x>2 there are temporal operators such as

until/always/init/next

, or even a forall mechanism, which have meaning and the AST could be re-used for checks.

Cosmos source
     
     ▼
 typed AST
     
     ▼
  Cosmos IR
     
     ───────────────► normal compiler
     
     ───────────────► TLA+      (protocol/temporal checking)
     
     ───────────────► SMT       (constraints/ARITH)
     
     ───────────────► Rocq      (deep formal proofs)

                 Cosmos temporal IR
                        
              ──────────────────
              ▼                   ▼
             TLA+                 SMT
                                 
             TLC             SMT solver
                                 
       state-space check     bounded/property check

So I got these suggestion and I'm only wondering if someone is familiar with such tools and if they might find it useful for the language to have it.
>>109992991
yeah me too

109993031
>yeah me too
ok didn't read you're wall of text. Was kinda obvious by how the text is formatted. Glad I skipped

>>109993031
>and I'm only wondering if someone is familiar with such tools and if they might find it useful for the language to have it.
why are you making something for someone else?

>>109993075
>>109993031
oh i just figured it out. cuz u dont program and dont know what to make for yourself. how silly of me

>>109993031
you acting like the rust and c++ comitee making things for people bloating the language fucking nigger make it for yourself

a comitee of fucking faggots

>>109993075
uh, hello? It's a language. There are other people who might use it and usecases even if I don't use it. Plus, I might end up finding it useful when I hear their opinion, really this is the reason I'm asking, I want to know if that'll be awesome to use in my videogame, after all somebody was saying it's useful.
>>109993097
Heh, that's where you're wrong. There's no bloat.

protocol(Foo, ...)

is completely valid syntax whether or not anything else is done. The language simply lends itself to it naturally. A single syntax does all of this.This is a result only possible thanks to the language being the design of a single person-me. There'd be 23905 more syntaxes had this been the result of a committee, but thanks to me it doesn't.

so were you saying cosmos is a language that isnt a natural language (programming language not a prompt) compiles into a prompt then the ai turns that into asm machine language w.e? i wasnt paying attention

explain it in language to someone who doesnt understand programming languages (me)

nvm read your past posts i get it now you prompt and the ai will use your language

i think not sure :P

>>109993275
The full course is $100. Pay or leave. It's a tough world, buddy.

>>109992983
No, I use a Datomic-like DB, no need for normal form shittery

File: yakui-the-maid-anime-dance.gif (1.89 MB, 592x640)

1.89 MB GIF

Hi Maids! I haven't gotten up to much recently. I've just been studying leet code, but its been a while since I had a solid side project. My last full stack MS interview was building a Rate Limiter and then a excel table thing in React / TypeScript. Completed it, but got rejected. I think its just my soft skills are trash. Anyways, I'm looking forward to doing everybody.codes and Advent of Code soon with you all. Minasan ganbatte ne!

>>109993353
Why'd you get rejected?

>this guy is unemployed
it's unironically over for me
https://www.youtube.com/watch?v=y8PLpDgZc0E

for how long the little kids are going to be forcefully expossed to this degenerate trying to push propaganda that you have to be a massive homosex to be interested in programming?

>maidniggers hijack and spam yet another thread
mods really are fags for letting this continue

>>109993629
he is unemployed by choice

File: smug.jpg (33 KB, 657x527)

33 KB JPG

>>109992750
my PR for a major FOSS project just got merged

>>109993749
There are no more active admins in this forum here. Or do you see anyone with a badge???

>>109994067
>see deprecated stuff in github repo I use
>fix it and make pull request
>'oh wow, thank you!'
>it gets merged without any complaint whatsoever (2k+ stars)
feel good

>>109993749
You're getting mad because of a picture of a maid?

Sometimes I feel a bit like Homer in the kitchen setting everything on fire.

>>109994526
hes probably mad because youre a bunch of rotten pieces of shit beneath a thin veneer of appearances

Hey all I'm trying to learn python but what the hell is an integer? I get its a number but why not say number. It's like they make it confusing on purpose and no one ever tells how just to make like, a normal program with buttons and a mouse.

>>109994574
integers are represented differently from floating point numbers
thats why calling both "number" just doesnt cut it.

>>109994574
An integer is a number without any decimal component.

>>109994574
Because the distinction between and not an integer is actually pretty important.
Maybe you're not ready for the full IEEE 754 talk yet if you're asking that question.

>>109994542
tell us how you really feel

>>109994526
I'm mad at some shithead insisting a delusion not only onto himself but everyone in the thread
imagine some dog shitting day in and day out in front of your front door and you can't do anything to stop it

>>109994628
between an integer and not an integer*

>>109994629
hard to describe it in words
kinda like a de-clutched engine, running on minimum rpm

its a lazy, sunny tuesday morning where i live and i didnt have my coffee yet

File: Capture.png (18 KB, 777x236)

18 KB PNG

>>109994574
>select a word in a page
>right-click
>"Search the web for <word>"
>get explanation

>>109994589
>>109994617
>>109994628
Hmm I guess that does make sense if it needs to be stored differently in memory. I know IEEE and scientific notation in base 2 but why not just unpack it and treat as a normal number? Like 5 = 5.000 why can't it just interchange them

>>109994658
because the hardware has special pathways to deal with floating point numbers.
the 1's ad 0's follow a physically different path in the chips where theyre dealt with in a way thats engraved into the silicon itself
doing otherwise would be MUCH slower

>>109994658
>Like 5 = 5.000 why can't it just interchange them
Lua and a few others do that https://www.lua.org/pil/contents.html

>>109994671
you also have casts in C but i dont think thats what anon meant

File: nanami.jpg (331 KB, 850x1235)

331 KB JPG

>>109994634
This isn't your house and nobody is near your front door. This is a maid cafe.

>>109994658
>Like 5 = 5.000 why can't it just interchange them
that's weak typing and it causes a lot of bullshit like in Javascript

>>109994669
Does every computer have that? I heard it can do four floating point operations simultaneously on the 128 bit media registers in modern x86 but obviously you can't opening it up and see what exactly it has

>>109994681
youre mentally retarded

>>109994679
>>109994671
I think until a relatively recent version (notably not in LuaJIT), Lua just used doubles for basically everything.

>>109994686
>Does every computer have that?
i havent looked up the specs for everything that classifies as a computer, but its a safe assumption everything you would be working with does that
maaaaaybe some exotic shit doesnt, but otherwise its pretty certain

>>109994689
ops on floats are ~4x slower than ops on integers on x86 :/

>>109994658
In python you can just interchange between them, but you still need terminology for talking about specific cases.
Saying that something takes and integer tells you not to pass floats to it. Just saying it takes a number is not descriptive.

>>109994658
Basically you don't want to slip into floating point accidentally if you really don't mean to.
There is a performance cost converting between floats and integers, while incredibly minor, it is there. But a better reason is you don't want to accidentally introduce any floating point errors to fuck all of your shit up.
If you just do integer math but in floats, it's fine. 1.0 + 2.0 really equals 3.0 exactly, but if you're interfacing with code not like that, that's when you get into the 10 != 10.000000004 kind of stuff.

>>109992781
No, something has to be a competition for it to be competitive. Nobody is competing with your hobby project. There can be programming contests, and those are competitive, but programming overall is not.

>>109994686
These days registers are multipurpose for the most part.
If we go back a number of years, the PS2, for example, had a set of registers specifically for floating point operations (and they were 32bit in length as opposed to the 128bit general purpose registers).

>>109994542
Return to Reddit. There is someone there who will upvote your enormous collection of Funko Pop dolls and then you will feel better.

File: curren.jpg (203 KB, 850x1384)

203 KB JPG

>>109993353
Interviews require people to do free work now? Did they keep your code? Are they using it in their system? Why would anyone want to be a wagie? They just get abused and exploited all day to make wealthy people even wealthier.

>>109994765
case in point
post code, miserable faggot

Lets kode!!!

>>109994796
>no coder calls other people no coders

>>109994835
>mfw nocoder

theres only one maidshitter whos not a sad sack and hes the osposter

the rest of you are *entirely worthless

File: Screenshot from 2026-10-0(...).png (301 KB, 1339x706)

301 KB PNG

>>109994835
>>109994842
aw forgor picrel
this is what no kofi does to a mfer

File: maidsheet in MAIDS.png (114 KB, 1920x1080)

114 KB PNG

>>109992750
Maid character sheet generator, written in MAIDS, running on MAIDS OS which is also written in MAIDS.

File: Screenshot from 2026-10-0(...).png (216 KB, 1339x706)

216 KB PNG

domination asserted successfully, then

ill even throw in a makefile, just for good measure

Hmm, I seem to have found something that doesn't work as it should in RPCS3.

File: quadragram.png (65 KB, 1920x1080)

65 KB PNG

>>109994893
Quadragram trained on Shakespeare, running on MAIDS OS. The emitted model is pure MAIDS. It is not doing statistical nonsense when it runs.

>>109994986
quadragram as in 4-gram?
how do you make it work without stats?
an n-gram (4-gram in this case) is the results of stats afaik

>>109994986
ackshually
im ngl, your quadgram sounds like something i could have fun doing

immediately i thought about how to improve upon the idea

i might actually turn my stat engine into a text generator by generating a complex ngram of sorts
where the properties constituting such an advanced ngram are no longer just the probability of occurence, but any transformation i would encode into the engine
i think the key upgrade to the concept of an ngram would be to attach some mathematical properties to a token, which would describe its relation to other tokens, as part of the identifier
this should increase precision alot

this is off the top of my head but ill remember that. this sounds thke theres potential there

File: mount.jpg (2.24 MB, 3000x2143)

2.24 MB JPG

Fixed the IDE hard drive detection.

>>109994634
homie's right

>>109995128 cont
this actually sounds like an alternative architecture to llms and even large world models now that i gave it a fink

its just an idea though, im not good enough to actually model the thing and estimate how it would stack up, computationally
and what is the impact of having to manually encode every transformation you'd need to get a coherent model, instead of using an universal function approximator

theres also interesting sub-question i have even now-
do i chain small transformations to express more complex ones?
or should i encode more various transformations

kinda like
>do i encode a moving average, or do i just observe a ratio over multiple samples?
and how far can i push it?
also i have no idea of the transformations i would have to do on text even if, in an abstract way, text is just numbers so maybe theres generalist transformations that could get me there. maybe the generalisation is to use atomic transformations and let the stats guide the inference

im gonna need empirical data to figure that out.

File: 1601063667755.jpg (48 KB, 563x651)

48 KB JPG

>>109992983
Possibly. It's been too many years since I last brushed up on them.
I just do it how I find it sensible.
I even challenge myself with a "no nullable fields" rule,
which comes either from the first or sixth NF.

File: quadragram_in_interpreter.png (532 KB, 1920x1040)

532 KB PNG

>>109995028
I should have been more precise. The model is technically statistical. Context determines which rule to look up, then a character is randomly pulled from that rule. So it contains statistical information, just represented as a lot of strings, rather than anything using a probability table. The emitted model is 11.5k lines of these rules. I had to open the code for the screenshot in the interpreter's IDE, because currently, though MAIDS OS can compile and run it, the editor truncates it (currently working on that).

The original goal was to see if this representation makes the model more human readable. It does not.

>>109995469
ok, now im impressed.

about the legibility, you might wanna try different representation methods

youre still reading maidcode, and... yeah
its not the most legible thing in the world

i would be tempted to build a gui and run the training on a smaller body of text to make as much of the model fit on the screen as possible
because youre talking about context, that means tons of information on top of the ngram itself

but i think a json-like output interspersed with strategically placed newlines could go a long way by itself

>because youre talking about context, that means tons of information on top of the ngram itself
(and so i would be tempted to work with all various context-variant ngrams of the same token collapsed into a single node and the contextual relations represented as edges in a graph. thence gui, or rather just a graphical projection. maybe some color highlights to make better sense of the relations)

>>109995469
kino edit btw, made me kek audibly + saved

>>109995469
>but i think a json-like output interspersed with strategically placed newlines could go a long way by itself
oh, and i would also sort the output grouped by tokens, obligatorily
even though you work at the character level
so it wouldnt be too informative

on another note-
how bout trying to run the thing at the level of a whole word?
the output should make much more sense then, no?

even if arguably working at a character level and still getting coherent output is more technically impressive
i think the output and the structure could become much more apparent
kinda like staining biological stuff before you put it under a microscope

>cant you work at the level of a word
>´V(vrrrrrvrrfuunutodurstrdu)V`

you actually create arrays of the letters that have been seen and pick one at random from em, dont you?
word-level-tokens would then explode the output's size, wouldnt it?

if thats what it is, the ngram part is actually maximally explicit and representative, youre just looking for the wrong thing
you see the statistical distribution of the next token, raw

File: maid-question.gif (1.89 MB, 462x427)

1.89 MB GIF

>>109993353
>Why'd you get rejected?
I completed the challenge, but it could be my code was bad? Most likely I'm too autistic to do the HRs correctly. "Whats your greatest weakness? Tell me about a time when you had a disagreement with a coworker."
>>109994779
Yes, I'd venture most jobs (that are actually hiring) put you to completing a leet code challenge and / or some sort "practicle" non trivial software implementation like a URL shortener or recommendation system. I doubt they use it in their systems.

Maid sister, I could not agree more. If we lived in a socialism (food and shelter provided), I'd happily learn and write algorithms all day.

It seems like Rust is the goto language for AI rewrites, so should people even bother learning it? I'll stick to C++ and ask claude to rewrite it to rust if need be.

>>109992983
No. The DB admin fags won't let you touch shit and the manager wants the data NOW NOW NOW so all you get is either a bunch of extra redundant columns or a repurposed nvarchar(max), be grateful.

File: tohru4.jpg (308 KB, 850x1294)

308 KB JPG

>>109995469
>ok, now im impressed.
>mfw <3

>but i think a json-like output interspersed with strategically placed newlines could go a long way by itself
The problem is that the strings themselves are not really readable, just looking like a pile of characters. If you stored them in JSON instead of MAIDS code, they would still not be meaningful at a glance. If you stored in lexicographical order, you would get something like "aaaaabbccc..." and at that point, I might as well use a weighted alternation like

seen:: 5 "a" | 2 "b" | 3 "c";

which I might do later, because it would make the code much less verbose.

>how bout trying to run the thing at the level of a whole word?
The problem is that on such a small training set, a quadragram would probably just reproduce the text it was trained on.

>>109995712
They're strings rather than arrays.

`some_identifier(

and

)some_identifier`

are string delimiters. MAIDS lets you define your own delimiters to avoid having to escape strings. Normal strings have the single escape character in the language which is \". I am going to remove it, because making your own delimiters solves the same problem in a cleaner way. That way, strings can always be WYSIWYG. I can't remember if I stole this idea from C++ or Rust. I am very pro-stealing useful language features if they make string processing better.

>>109995729
The ultimate goal is to try to train a small LLM without needing floating point math, because I can't afford a GPU, but I can afford to try to use an FPGA as a hardware acceleration for strings. I think I might be able to do this? There is something called a bag of words vector, but the problem is the order of words is lost if you use this.

>>109996008
Eventually the Cybernetic Socialist Science Foundation for Maids will solve this by releasing advanced Mathematics and Computer Science research to Public Domain. You can contribute by making Public Domain research.

>Read out of bounds in GPU memory
>Sound on my youtube video stops playing
I can only imagine what a clusterfuck of a security nightmare these drivers are
WebGPU probably exposes thousands of avenues of getting ring 0 from browser javascript

graphics programming looks fucking terrifying honestly I'm lost on how these people are deriving equations to express where vertices, edges and faces go and then putting it into 3D like I'm sorry for being this inferior, man.

File: 86e518bdb7fd1eafaf4cdc86e(...).jpg (114 KB, 675x988)

114 KB JPG

>months or years without using this thing
>still the same maid schizophrenia

>>109997319
sex with whoever that is and TMD

>>109996884
>WebGPU probably exposes thousands of avenues of getting ring 0 from browser javascript
Hmmmmm

>>109996188
Larper nocoder.

>>109997319
Maidposting is one of the oldest traditions on /g/

>>109997653
So is maldposting, though

C shat itself and segfaulted, so I added a missing forward declaration and the segfault went away. Does C shite itself if I pet it the wrong way too?

File: 1787430552803121.jpg (24 KB, 549x275)

24 KB JPG

>>109974243
Abandoning this because you're basically forced to do a bunch of preprocessing on the model to make it even approachable and I don't want to burn another 100GB of disk space on it.
We should be moving away from highly detailed baked models anyway and towards proceduralism. I'm gonna work on a glTF extension for defining procedural geometry.

>>109998435
just enable address sanitizer, if you are not on msvc, enable undefined behavior sanitizer too.

>>109998491
If you want to do something really challenging then try a hybrid model of static and dynamic.

>>109996188
I laughed at Ginger Bill seriously engaging with a Rusttroon that turned out to be an actual troon.

The Rust community being F'd really isn't an exaggeration or a meme lmfao.

>>109994847
boilerplate
>>109994896
makefile

no koder

what is the solution to prop drilling? I know theres context but that is untyped and not very good practice to use as its like a global thing

File: texttopic.png (28 KB, 816x1380)

28 KB PNG

Okay, I'm finally finished with this. It converts ascii text to a picture without any corruption. Even if there are non-ascii characters, they're stored without corruption. The original text data is inside the image in the form of specially coloured pixels, so the text can be recovered from the picture with the use of another script.

As you can tell by reading the source code, this was made entirely by Claude - it has many telltale signs of Claude's programming style.

File: back.png (4 KB, 474x234)

4 KB PNG

>>109998772
And this is the script that recovers the text data from the picture. So now you have everything you need to take the source code and do whatever you want with it.

Strings are the most powerful data type, they can be turned into anything

>>109998783
except a working program

File: shit.png (29 KB, 3600x138)

29 KB PNG

>>109998782
It can even encode binary files and the file converts back 1:1 without corruption. This is the image from that post run through the encoder.

File: wide.gif (91 KB, 640x360)

91 KB GIF

>>109998878
>3600x138

>>109994658
>Like 5 = 5.000 why can't it just interchange them
Because floating point numbers, no matter the size, are approximations. Computers don’t have the luxury of handling an infinitely large number; floats are usually only ever 32 or 64 bit in regular hardware (or even lower in GPUs for AI slop), or they are software emulated in math software to have higher precision. There is only so much information that can be stored in the mantissa before you get bad information loss and the loss is compounded the larger the exponent. If you’ve ever played minecraft and ran from the center of the world to somewhere as far away as possible, you’d notice your character actually start to jitter around more and more because of this information loss (notch just used a raw vector global position like a retard instead of localizing the vector to each world chunk to avoid this happening). All kinds of software that work around floats require things like epsilons to logically cover edge cases correctly because the model itself is ultimately flawed.

>>109998783
Booleans too

File: zoo.jpg (34 KB, 680x659)

34 KB JPG

>not a single post on the last 5 hours
This forum is so fucking dead. Holy fuck

File: blue archive maid.jpg (282 KB, 850x1133)

282 KB JPG

>>109998783
This is correct.

>>109998832
>t. never heard of string processing languages

>>109998984
Booleans are bloat. Embrace goal-directed programming.

Tabby now compiles fast, runs fast, and produces small binary sizes. I also successfully ported it to Linux.

>>109998727
Either a local or a global store

A local store is basically a context,
A global store is something like Redux or Zustand

>>109999386
>>109998727
also, Context is typed...

>>109999193
k, I'll share what I've got.

I saw https://higesho.github.io/ and I'm lusting after its look and feel so I'm gonna use its graphic approach to make a more sophisticated procgen tile-based roguelike game, voxel

mimalloc is a beast got dayum

Integer multiplication and FFT are no longer O(nlogn) thanks to AI. They're going to have to rewrite all textbooks.

>>110000000
who got the nice integer

Project Euler problem 1 in Tabby

euler1(\=
    result = 0
    i = -1; while(; i = i + 1; i < 1000) : (
        | !(i % 3) || !(i % 5) : result = result + i _
    )
    result
)

>>109999832
arenas > any and all generic heap implementations

>>110000252
the fuck is tabby? your language?

-fsanitize=undefined

enables everything right? I don't need to specify bounds etc?

>>110000165
>Integer multiplication [is] no longer O(nlogn)
You're gonna have to be more specific.

File: 1767085280672753.jpg (43 KB, 1007x129)

43 KB JPG

>>110000504

>>110000475
Yes

>>110000513
I want to get off mr bones wild ride

File: 1781426660252809.png (103 KB, 1920x1107)

103 KB PNG

>>110000576
It also proved that nxn matrix multiplication can be done in O(n^2.25). Humans have been struggling to get under O(n^2.37) for the past 30 years.

>>110000513
>1-2^-182
KEK

>>110000606
>>110000513
I wonder if the guys that shared their ideas about these tiny improvements on some ancient html only blog or obscure academic website regret it now

>>110000513
https://en.wikipedia.org/wiki/Galactic_algorithm

>>110000606
>Humans have been struggling
Humans = 2 people
Struggling = thinking about sometimes after work

>>110000513
Sounds useless.
>>110000606
>that fucking Y axis

Apache spark?

File: apu-tongue.jpg (446 KB, 4000x4000)

446 KB JPG

re

File: screenshot_2026-10-07_16:51:30.png (237 KB, 1968x1351)

237 KB PNG

I've been reworking my timer.
Input is now just minutes taken as a floating point number because I am sick of typing 1:30, 10:00, 15:00, 1:30:00 and would rather just do 1.5, 10, 15, 90. Most of my timers are for cooking so sub 1 hour and rounded to the nearest minute.
No longer using massive ascii art to display the time remaining.
Output displays the actually time the timer will elapse.

File: 4g0.9.5.png (34 KB, 1142x324)

34 KB PNG

I updated my 4chan extension for emacs
https://github.com/eNotchy/4g

>>110000420
good luck writing a concurrent arena that beats mimalloc and doesnt explode

>>110003457
what makes you think that would be hard?

>>110003457
you don't need it to be concurrent, you should have many different per thread arenas

>>110003475
because that's most of what jemalloc / mimalloc / etc provide, concurrency

>>110003518
and now you've invented a shitty jemalloc, congrats

>>110003537
yeah, with a completely different allocation strategy...

>>110003537
the best way to deal with thread synchronization is to not do it. it makes no sense have synchronization in an allocator that the memory allocated from will still need its own synchronization for anyway. why are we doing this?

arenas are already supposed to be granular. do them per thread and when you need to share something you wrap the thing in a mutex, do barrier sync it or do atomic stuff. why do you think you only need a single allocator for all threads with all the additional complexity that involves from free list synchonization to per thread caching the allocator will have to do to be fast anyway? it's just a dumb way to frame the problem. you have many terabytes of virtual memory to abuse, plus arenas have the added benefit that after finishing a final barrier sync, freeing all the corresponding arenas across all the threads is just trivial and blazingly fast