It even supports 5 level paging
#MINTIA (not vibecoded)
dense vigil
What bochs supports heavily depends on what config options are used
I recommend building from source
Bochs is also amazing for USB debugging
vast valve
I’m not sure wtf was going on when I built it from source then?
dense vigil
Did you tick all the options
coarse current
My os fully works on it, never had any problems
vast valve
Must have been a severe skill issue then 
Or just that one commit on master had issues when I tried it?
digital pivot
a bit interesting opinion on vad structs from reactos chat
i was looking at timers code and noticed that https://github.com/xrarch/mintia2/blob/4cdd2e9a0af52062117f2a5d54c6cf09c3260f35/OS/Executive/Ke/KeTimer.jkl#L201 doesnt indicate success ?
hollow musk
pairing heap seems like a good choice for timers
hollow musk
in windows, timers are organized into linked lists in timer tables
and it uses bitmaps to find pending hands
apparently, the windows approach is based on the timing wheels paper https://www.cs.columbia.edu/~nahum/w6998/papers/sosp87-timing-wheels.pdf (1987)
coarse current
pairing heap seems like a good choice for timers
Deletion is slow though
digital pivot
i remember you criticized nt's object manager namespace
have you seen any comments from the nt people regarding its design and implementation
twilit smelt
mintia's name was on a slide in the ceo's presentation at a major silicon valley startup's all hands meeting today
(under my name on the list of new hires and their backgrounds)
shadow ridge
congratulations
junior pivot
congrats!!!!!!!
shadow ridge
i hope mintia 2 doesnt stagnate now that will has a real j*b
junior pivot
i hope mintia 2 doesnt stagnate now that will has a real j*b
its Mintia© 2®™ now 💔
vast valve
its Mintia© 2®™ now 💔
minta/2
shadow ridge
limntia/2
no
it's MintAI now
fully vibe coded neocapitalist NT-tech operating system that will revolutionize low level bit concurrencies
get the vibecoded crypto concurrency $MINTAI RIGHT NOW!
dense vigil
(under my name on the list of new hires and their backgrounds)
Congrats that's pretty cool
Is your job related to osdev/sysdev?
hollow musk
i remember you criticized nt's object manager namespace
what about it?
twilit smelt
Is your job related to osdev/sysdev?
Ya firmware
digital pivot
what about it?
as per @twilit smelt the problem is name caching which is centralized in unix likes
in nt object namspace each fs driver operates with its own cache
for example, a path like devices\harddisk0\partition0\ could exist and any part of the path string after that is passed to be parsed to the fs layer
nocturne nymph
I don't think it's really that big of a deal.
nocturne nymph
Ya firmware
Congrats
You've successfully overcome challenges in the past, and there's no reason to think it will be any different now.
digital pivot
I don't think it's really that big of a deal.
i think mintia2 implements a separate namespace tree subsystem which provides an explicit representation of the namespace as a tree
nocturne nymph
i think mintia2 implements a separate namespace tree subsystem which provides an...
I haven't touched OB in a long time (actually, that's not entirely true, given the recent RCU changes... but I definitely haven't thought much about namespace caches in a long time). I think the problem with having a central cache that includes the sub-namespaces (e.g., within filesystems) is that there are semantic issues that would create inter-dependencies that would limit the flexibility of the implementations. Character set rules, symbolic links, hard links, compound files (e.g. NTFS streams) would have to be understood by the global caching system. And I don't think that is really necessary.
As long as sub-namespaces are self-contained (i.e., not allowed to reference backwards into the containing namespace), then having namespace lookups be two-step (locate the target namespace and then call into that namespaces implementation to resolve the rest of the pathname) should work just fine.
In kernels the second part should just be a function call (though NT does complicate things with its IRP-based IO system).
Hard/soft links and use of the namespace to implement a security hierarchy (i.e., protecting files by securing parts of the namespace) create additional complexity (and subtle vulnerabilities). Plus the optimization of relative pathnames creates some more issues (i.e., when you hold a handle on part of the namespace and access relative to that point).
And generally trees are (probably) a bad idea and don't promote much flexibility in searching (thus the cross links). You have to get the cache invalidation to work securely and efficiently if changes in the sub-namespace are to be correctly reflected back into the global name cache.
short owl
Ya firmware
Please don't make life to osdevvers shit by writing broken fw 🙏
Don't become what you swore to destroy
shadow ridge
write very quirky firmware
the kind that requires extensive debugging to figure out why it doesnt work in some random edge case
coarse current
No, just write correct firmware that breaks the operating systems that assume too much
devout geyser
Sleep(1000000)
cursive charm
https://devblogs.microsoft.com/oldnewthing/20150814-00/?p=91811:
Originally, there was no
TerminateThreadfunction. The original designers felt strongly that no such function should exist because there was no safe way to terminate a thread, and there’s no point having a function that cannot be called safely. But people screamed that they needed theTerminateThreadfunction, even though it wasn’t safe, so the operating system designers caved and added the function because people demanded it. Of course, those people who insisted that they neededTerminateThreadnow regret having been given it.
is there any truth to this?
iirc, the headers from NT 3.1 build 196 declare it
cursive charm
i've checked proc.doc before, and its revision history doesn't seem to mention adding NtTerminateThread(). (the NtTerminateThread() system service was defined in that document.)
checking ke.doc now
cursive charm
why do i keep reading this paper
that hyperlink just leads to raymond's blog post on the allocation granularity: https://devblogs.microsoft.com/oldnewthing/20031008-00/?p=42223
lucid umbra
Will implement*
will implement? yeah he usually does that
digital pivot
And generally trees are (probably) a bad idea and don't promote much flexibility...
do you think the way how its done in nt is a better way?
nocturne nymph
do you think the way how its done in nt is a better way?
It's just a matter of opinion that it's not that bad and worth changing.
quiet mauve
why do i keep reading this paper
Yeah, I still feel bad about my tussle with @twilit smelt on that whole thing. I wasn't being clear enough about the situation and it caused a misunderstanding. Although I am still a little ticked that Raymond's blog wasn't on point. Maybe it's common, but I always thought his anecdotal stories were beyond reproach.
To be clear I don't mean the allocation unit size (or the effects), just the history behind the choice.
nocturne nymph
It's just a matter of opinion that it's not that bad and worth changing.
There is another view. Is the idea of using namespaces to organize storage a good one?
Human beings find pathnames useful, but it is potentially much more efficient to identify stores by tuples of persistent GUIDs <namespace, file, version>. Mapping names to files is due to the historic file cabinet abstraction for organizing data, but this abstraction implies key hierarchies and secondary indices to aid in searching which limits the flexibility.
Even the web has evolved away from the file cabinet model. Search is done by search which resolves into links which can then be captured and used rather than re-running the search. But a lot of work is required to make this work, especially in the old system.
shadow ridge
that might be more efficient but isnt it very limiting
it's like taking a file system and removing the "file" part
now it's just a binary blob database
nocturne nymph
One other note about namespaces and semantics. The NTOS OB was more like /dev than anything else. There is also a separate namespace for the registry which had lots of other semantics and features which led to lots of complexity due to abuse and failure to consider the information lifecycle of the contents.
It was supposed to provide a configuration database, but really just became a cautionary tale about how ingenious people can be at using wrenches as hammers and screwdrivers as chisels.
warm pine
Human beings find pathnames useful, but it is potentially much more efficient to...
this has been spoken about for a long time but doesn't seem to have arrived on desktops
and mobiles are providing a thin and leaky veneer over the hierarchical filesystem
nocturne nymph
Plan 9 contributed some interesting ideas in this area. But I don't know that anybody has worked this out completely.
The organic evolution of the web and search probably provides much better insight into how data should be organized and managed.
blissful smelt
what is mintia's plan with the split cache model of "buffer + page cache" for block devices when it comes to what it will do when the write hits metadata? is the whole write stopped and rejected?
i'm planning out that strategy and am running into a predicament with pages that are partially metadata and partially data that are mmap'd in
blissful smelt
i'm planning out that strategy and am running into a predicament with pages that...
i'm thinking either full denial of the write or a shadow page would be what I'd do (with the latter, the fault into the write would cause that page to be seen as partially metadata, resulting in a new page with the same data being created, and the write going there, and that data being filtered out into the real page), though idk
twilit smelt
what is mintia's plan with the split cache model of "buffer + page cache" for bl...
The buffers are backed by pages of the page cache
And the read logic is exactly that of the page cache
The only difference is that if you access a page through a metadata buffer and dirty the metadata buffer, the underlying page will be written out block-wise by a special buffer writer instead of page-wise
Or perhaps they'll share a writer which will generalize
Dunno
blissful smelt
The only difference is that if you access a page through a metadata buffer and d...
what's the plan for running into a partially metadata page through an mmap write though? does it simply use the different write-back machinery, i.e. any page containing a metadata buffer uses different machinery to write-back?
oh wait that's kind of interesting 🤔
the other concern i had with this strategy was with filesystems that vary in metadata and data extents like btrfs
i was considering some kind of global interval tree for the whole FS but that seems like it could make unhappy scalability
twilit smelt
what's the plan for running into a partially metadata page through an mmap write...
Well this can't happen because file data pages are virtual and only contain file data blocks
blissful smelt
Well this can't happen because file data pages are virtual and only contain file...
what about the block device page cache
twilit smelt
Coherency is usually not guaranteed when directly accessing a block device that is also mounted as a filesystem
It could be accomplished here though by causing writes to mapped block device pages to cause all the contained buffers to be dirtied rather than the page in the page cache
But this still doesn't obtain coherency with file data
Which again is not usually offered anyway
blissful smelt
Coherency is usually not guaranteed when directly accessing a block device that ...
ohh ok, I thought the split cache model was about not corrupting metadata through the block device page cache, and that things like altering file pages through the block device would cause the relevant files to invalidate those pages
so it's more about having a private buffer cache that just holds the filesystem metadata, and maintaining coherence between that and the various file page caches?
twilit smelt
It's about fixing aliasing problems where file pages and metadata pages can contain some of the same physical sectors
The buffer cache provides a per sector (or block) structure to track each metadata block
So that modified metadata blocks will only write back those sectors to the disk and not also write back other sectors that lie within the same page but belong to files
twilit smelt
mintia2 no longer belongs to me it has been purchased from me for $10,000,000 by a company who wants to use parts of it for their router firmware.
goodbye all osdevers
shadow ridge
mintia2 no longer belongs to me it has been purchased from me for $10,000,000 by...
did you forget to take your meds again
no grandpa, it's not the 90s
ok wait the router firmware thing is oddly specific
is there a chance this is like half-real
keen stone
tfw the larp becomes reality
shadow ridge
theres a universe where he signed a contract saying all of his work (even outside of work) belongs to the company or some shit like that without thinking about it
twilit smelt
It's funny because I almost did do that
I had to list mintia as my pre existing IP when I signed my offer letter otherwise it would transfer ownership to the company
If I didn't read it I wouldn't have known that
shadow ridge
I fucking knew it
these corporate bastards are just waiting for you to drop the soap at every turn
you should trademark mintia
like actually
together with xr station
warm mural
afaik it's a startup so it's probably not super corpo yet
twilit smelt
There are no walls just a lot of people under a roof in rows in what is essentially a very large room
twilit smelt
I had to list mintia as my pre existing IP when I signed my offer letter otherwi...
This is standard language for offer letters apparently
dense vigil
There are no walls just a lot of people under a roof in rows in what is essentia...
thats fairly common
open space stuff
dense vigil
yeah its supposed to make everyone feel like a team or whatever
twilit smelt
twilit smelt
Cool thing about a job is that there's no homework on the weekends unless you want there to be or you are being specially depended upon for a specific highly time sensitive thing which usually doesn't happen to interns
Which means i should be able to work a bit on mintia
nocturne nymph
That's good, because I remember having to work on the weekends to finish my internship project.
twilit smelt
maybe thatll come later
acoustic sparrow
Cool thing about a job is that there's no homework on the weekends unless you wa...
idk you say that
ah yeah i guess its mostly "unless you want to be there"
digital pivot
maybe thatll come later
what kind of task did they give you
do you work on it alone or in team
dense vigil
Hopefully make NT 2
nocturne nymph
Microsoft is already working on NT 10...
tame phoenix
I don't think id ever take a dev job where there's "homework on the weekend"
queen torrent
Hopefully make NT 2
will, generate nt 2 make no mistakes
mortal thunder
NT 2, now with better numa support
tame phoenix
was there never an nt 2? did they go straight to 3?
acoustic sparrow
This is standard language for offer letters apparently
in the us? or like in general
twilit smelt
was there never an nt 2? did they go straight to 3?
it started at 3.1 to be in sync with dos windows' current version number at the time
NT never would have released with a 1.0 version number because the original plan was for it to release as "OS/2 3.0"
twilit smelt
in the us? or like in general
I don't know
I was just surprised so i looked it up and it said that's in basically every offer letter you'll ever sign in tech
A page where you have to list your pre existing IP and anything you don't list transfers ownership to the company
twilit smelt
what country
twilit smelt
Might not be legal for them to claim all your pre existing IP like that
acoustic sparrow
i would imagine, yes
twilit smelt
It is in the US
acoustic sparrow
that feels dumb
it would make sense if you could implicitly grant a license to do anything with it, but not actually transfer ownership
but i guess its the us lol
twilit smelt
Offer letters here are like "we own all facets of your entire existence mind and soul that have ever existed that you do not explicitly list in the box below, excepting things already owned by previous businesses you worked for. Along with your mind and soul we also own your body starting on the day of employment until the last day of employment. Anything new you begin to do during that time is also ours. Please sign below if you don't want to be unemployable forever:"
twilit smelt
I could work on mintia without them owning it bc it's my pre existing IP i listed
As long as I don't do so during work hours or using work computers
If i started a new thing though they might have a claim to it, regardless of when or how i worked on it
warm pine
ireland
you often see random projects with 'Copyright such-and-such Google, inc.' in their licence headers these days on account of this being a normal practice in america
acoustic sparrow
you often see random projects with 'Copyright such-and-such Google, inc.' in the...
huh
i guess makes sense
warm pine
in britain and ireland, and i assume europe and probably the rest of the world, it's unheard of
random example i know about https://github.com/jmmv/sysupgrade/blob/master/COPYING a little script for netbsd is owned by google because the netbsd-contributor author worked for them in those days
acoustic sparrow
apparantly google has a procedure to assign the copyright to you according to this
icy bridge
wow that all sounds horrible
shadow ridge
I could work on mintia without them owning it bc it's my pre existing IP i liste...
you listed all of your projects right
like
not just "mintia" and nothing else
warm mural
Might not be legal for them to claim all your pre existing IP like that
Some countries straight up make it illegal to give up copyright so
shadow ridge
i would write down every xr station buzzword on that paper
like
maybe they dont have mintia but they have xr station new sdk now
acoustic sparrow
Some countries straight up make it illegal to give up copyright so
its different to give it up vs sell it
warm mural
Maybe yeah
What kinda internship did you get pitust
You're probably a really good candidate so I'm not surprised you got one lol
twilit smelt
maybe they dont have mintia but they have xr station new sdk now
I did write it down
acoustic sparrow
What kinda internship did you get pitust
its a company called arista
they sell switches mostly
twilit smelt
i wrote "XR/station project and all subprojects thereof including a3x, a4x, mintia, mintia2, aisix, newsdk, ..."
Or something
acoustic sparrow
You're probably a really good candidate so I'm not surprised you got one lol
well no bc im a first year uni student lol
that automatically makes me a worse candidate
but not worse enough ig
warm mural
IOI is overpowered on resume
If you position well on IOI it's like guaranteed entry into MIT
unironically
warm mural
tbf all the IOI winners have rich parents or are extremely cracked
acoustic sparrow
i dont think so?
warm mural
tourist is in the latter category
acoustic sparrow
tbf all the IOI winners have rich parents or are extremely cracked
all the ioi winners are chinese
acoustic sparrow
(this is only half true)
shadow ridge
I did write it down
great
acoustic sparrow
yeah poland did badly last year tho iirc
warm mural
I checked the winners of my province and they're all asian
shadow ridge
@twilit smelt was there a clause about work you do while under the contract
they dont automatically own everything you do while working there right
acoustic sparrow
who is tourist?
warm mural
look it up
shadow ridge
who is tourist?
a person who goes from their country
to another country
to visit it
and have fun
acoustic sparrow
ah
shadow ridge
theres this hyperactive fly that wont get out of the way of the fucking screen and it's making me go insane
warm mural
That is crazy talent
short owl
Offer letters here are like "we own all facets of your entire existence mind and...
Sounds like hell
And an extremely shitty practice
There's no limit to the greed
shadow ridge
theres this hyperactive fly that wont get out of the way of the fucking screen a...
i instinctively reached my hand out at the perfect time and smeared it across the screen
good riddance
twilit smelt
they dont automatically own everything you do while working there right
They do
twilit smelt
There's no limit to the greed
It's not even indicative of special greed it's just completely and utterly standard here
twilit smelt
Mintia is not new
So i could work on it outside work hours on personal devices without them owning it
But anything i start would be theirs
twilit smelt
It's a form of intellectual slavery or imprisonment kind of
But it's not special to this company
It's just standard in America which is a corporatocracy
acoustic sparrow
this is like vaguely reasonable tbh
shadow ridge
you could argue it's a form of anti compete
but it's not
i think it would be a huuuge reach to classify it as that
mortal thunder
Offer letters here are like "we own all facets of your entire existence mind and...
this is super idiotic
what the fuck are they thinking
does the US not respect human dignity
(don't answer that)
acoustic sparrow
the argument is that its to make sure that if you work for a company and you have some genius idea during work hours, you cant just leave and take it somewhere else
i think
mortal thunder
thats bullshit
shadow ridge
why does it matter if i think of it during work hours or outside of them
this isnt the military
icy bridge
yeah why would that be reasonable
shadow ridge
i dont sell my body and brain
mortal thunder
it's not
acoustic sparrow
yeah why would that be reasonable
well if you are paid to come up with genius ideas...
lol
mortal thunder
nobody at google is paid to come up with genius ideas
acoustic sparrow
ehh
short owl
It's not even indicative of special greed it's just completely and utterly stand...
Location: America
Checks out lmao
acoustic sparrow
thats not really true though
mortal thunder
they just bought most of the genius ideas
shadow ridge
a normal anti-compete would be much better than "i own your brain"
acoustic sparrow
google is actually the origin of a bunch of good ideas
mortal thunder
nobody at google is paid to come up with genius ideas
i was mostly referring to the fact that the executives come up with the good ideas and the engineers implement them
acoustic sparrow
i was mostly referring to the fact that the executives come up with the good ide...
this is not really how it works btw
mortal thunder
and the people experimenting on company time i guess
acoustic sparrow
like at all
twilit smelt
A famous example of the "they own everything you do" clause is when Steve Wozniak designed his first homebrew computer kit he had to show it to his bosses at HP at the time and they basically said "who would ever want a computer in their home? No we're not interested" and they released the copyright to him (per their policy) and it went on to be the Apple I i believe lol
mortal thunder
this is not really how it works btw
google does this differently?
acoustic sparrow
google does this differently?
this is not how software engineering works at all???
mortal thunder
i guess the executives just sleep on the job and rake in all the cash then
acoustic sparrow
literally
executives in really big companies dont even have that much power in practice
bc its very hard to get a big company to do what you want it to
mortal thunder
okay i guess i need to learn more about bigtech and how it works
acoustic sparrow
this isnt big tech only
mortal thunder
bc its very hard to get a big company to do what you want it to
it depends on how the company is set up
if the company is privately owned (like valve, or openai, or anthropic) it's much easier
acoustic sparrow
not really?
mortal thunder
publicly traded companies like google are beholden to shareholders
private companies are only beholden to their own interests
is that right?
acoustic sparrow
theoretically
acoustic sparrow
publicly traded companies like google are beholden to shareholders
obvious counterexample
spacex is not beholden to the shareholders at all
mortal thunder
spacex literally just issued their IPO like yesterday
mortal thunder
and also elon still owns like most of the shares
twilit smelt
mortal thunder
(which is why people claim he's the first trillionaire)
acoustic sparrow
even if you bought every share not owned by elon youd only have 20% of the voting power
and you arent allowed to sue
mortal thunder
is this favoritism or what
acoustic sparrow
no
mortal thunder
thats not usually how it goes
acoustic sparrow
yeah i know thats part of why the whole spacex ipo is bs
digital pivot
It is in the US
does this work for projects created using your personal hardware
acoustic sparrow
anyway with most of the big tech companies, original founders have a controlling share
twilit smelt
does this work for projects created using your personal hardware
Yes
digital pivot
and its this way in all american corps?
twilit smelt
Prob
digital pivot
this is cancer
twilit smelt
That's just part of employment here
twilit smelt
A famous example of the "they own everything you do" clause is when Steve Woznia...
.
see historical example of this
If HP didn't have a policy of releasing copyright for their employees personal projects upon meeting with them about it and deciding they didn't care, then there would be no Apple
acoustic sparrow
IP you come up with while you are employed being owned by the company is not just a thing in the usa btw
warm pine
IP you come up with while you are employed being owned by the company is not jus...
that one is reasonable enough
acoustic sparrow
yeah
warm pine
we say "in the course of employment" or words to that effect
twilit smelt
mortal thunder
If HP didn't have a policy of releasing copyright for their employees personal p...
wouldn't that have made Wozniak more free to release the Apple 1?
clearly he wanted to release it
so if there wasn't HP as a roadblock then he would've done it right away
twilit smelt
wouldn't that have made Wozniak *more* free to release the Apple 1?
The policy of releasing copyright upon deciding they didn't care about an employee's side project is an extra thing they were doing
If they had no such policy then they would have hoarded the copyright
And he would have had no way to sell it commercially without them
And they wouldn't have sold it
Bc they didn't care
So
Would have just died
mortal thunder
that one is reasonable enough
are you sure?
mortal thunder
IP you come up with while you are employed being owned by the company is not jus...
where else is it a thing?
acoustic sparrow
everywhere i think?
warm pine
are you sure?
it seems a not unreasonable rule to me if you write a program while at work then it's the employers
mortal thunder
while at work, yes
warm pine
it's when this applies to your work at home that it becomes outrageous
mortal thunder
i don't disagree
mortal thunder
it's when this applies to your work at home that it becomes outrageous
exactly
but pitust didn't mention anything about "work performed on company time"
warm pine
the americans however have this silly rule for some reason
it makes no sense
but note a dangerous implication of an argument that some free software advocates make about the GPL, that any exposure to it means you can't write software of a similar class
we can infer from the case with american firms "your code is ours" rule that the GPL claim must be excessive, because if so, american firms would effectively be traps that, once you work for them, you can never again work on a similar class of software at any other firm, since you knew about the first firm's proprietary work
twilit smelt
me when im skimming youtube comments on a video and theres a comment from 2 hours ago where someone says "Windows is not portable."
im a seasoned youtube comment arguer
tame phoenix
A page where you have to list your pre existing IP and anything you don't list t...
this smells
luckily I'll never work in the US
of course if you attribute all your IP to a company in your control, that makes it very hard for them to take it
warm pine
luckily I'll never work in the US
it's just normalised there
i think they only have something like 2 weeks holiday a year as well
tame phoenix
i think they only have something like 2 weeks holiday a year as well
there is no law saying how much you are entitled to iirc
and people are expected to just not take holiday anyway or be contacted on holiday
vast valve
Wait where did you get a job at? And what do you do lol
tame phoenix
from what I've read and heard US at will employment is a cess pool
and they say European employment law is communism
it's daft
warm mural
private companies are only beholden to their own interests
This is only partially true since VCs exist and other investors
For example openAI still has a bunch of private investors, and it would be in their favor to please them
twilit smelt
twilit smelt
from what I've read and heard US at will employment is a cess pool
its funny how here the list of reasons that you can fire someone is a blacklist rather than a whitelist
that is to say, there are banned reasons to fire someone
but if its for any reason outside that list its fine
for example if you fire them bc they are a specific skin color or gender or sexuality thats illegal
but if you say its because they looked at you funny thats completely legal
so you can just say that instead
the person would have to sue and find like written or recorded evidence that you had some kind of like racist prejudice against them or something to prove that u most likely fired them for that rather than for looking at them funny (which is a legal reason to fire someone)
which costs a LOT of money
and has low odds of success
its so funny how many self proclaimed libertarians here call the government bloated and say corporations are over-regulated and whatever
do we live in the same country
blissful smelt
its funny how here the list of reasons that you can fire someone is a blacklist ...
when "fire at will" becomes "fire will"
twilit smelt
no pls
shadow ridge
mr. will mint
come to my office please
we have to talk about your recent performance
twilit smelt
there arent even offices
tame phoenix
they don't need a reason do they?
that's why it sucks
best to not give a reason for redundancy because if you do it may be one of the forbidden reasons
mortal thunder
that's why it sucks
it does
it really does
you could theoretically take it to court if you suspect that they fired you for a forbidden reason but just say "no reason", but it's expensive and risky
tame phoenix
in the UK you can't just fire someone just because, and in some EU countries it's even harder. I hear also in Japan it's basically impossible legally to fire someone unless they've broken the law so they sideways reshuffle them into boring positions to make them voluntarily quit
random heath
I had to list mintia as my pre existing IP when I signed my offer letter otherwi...
That's some crazy laws over there in the US WTF
I think that's not even legal in germany
shadow ridge
i hope he listed $MINTIACOIN as his pre existing IP
imagine pulling a pump and dump
and your company takes the profits
twilit smelt
I would deserve it
golden meteor
@twilit smelt i am formally requesting cat pics
twilit smelt
which cat
twilit smelt
<@783460822525149196> i am formally requesting cat pics
Which cat
twilit smelt
what orange one
twilit smelt
lucid umbra
i used to have the exact same fountain
blissful smelt
mintia mascot
twilit smelt
the platonic ideal of mintia2 that exists in my mind is the most advanced hobby kernel and that should be good enough. i refuse to build it im a sophisticated being far beyond trivialities like tangible existence. thank you all for respecting me in this matter
shadow ridge
my samsung ssd
shat itself
and i almost shat myself with it
it wouldnt respond to any write commands and could not mount at all
had to use debugfs to pull whatever data i could salvage and now im on a fresh arch install
queen torrent
Also this is a selfie of me
you're not a hyena that lives in the sky...
vast valve
you're not a hyena that lives in the sky...
maybe the hyena in the sky is thier spirit animal
golden meteor
The hyena in the sky is the alternative hatman I see when I not just take 50 Benadryl, but also haven't slept in 3 days
Don't question why there's a sky inside
shadow ridge
it's very ironic that it wasn't depression, injury or death that ended this project
but a simple j*b
twilit smelt
it's not ended
I just haven't had time to work on it because I'm also behind in my 1 summer class and in my free time I'm too busy stressing about that while still not doing it to work on mintia
twilit smelt
The latest atrocity of the tech right: they have entombed the SJWs under the streets of San Jose
twilit smelt
every single character in my project's source is a manual keystroke or a copypaste of my own other code
which was manual keystrokes
this will sound insane to those who have only known vibecoding one day
even if it would genuinely improve my productivity to use it i kind of just dont want to just to be able to keep saying that forever
queen torrent
I will open a pr with code all written using dip switches. Lets see how the keyboard warriors will handle knowing its not all keystrokes anymore
tame phoenix
this will sound insane to those who have only known vibecoding one day
dramatic much lol
I don't think human made code is going away in our lifetimes
I also think we are going to get to a point where saying "I refuse to use any generative AI, ever" would actually sound like "I refuse to use keyboards"
not next week, but in 5 to 10 years
it's a tool in the toolbox, like a compiler or a linter and to be principled and never ever use it will become very niche, kind of a hot take, once the companies who made it stop shoving it down our throats
mortal thunder
I also think we are going to get to a point where saying "I refuse to use any ge...
absolutely not
we've never had a comparable technology
tame phoenix
we did. the internet
shadow ridge
stackoverflow
mortal thunder
the closest comparison I can think of is "I refuse to use machines for the entire process of making my pots, I prefer to make them by hand on a turntable"
tame phoenix
I'm more confident than an LLM that I'm right on this...
shadow ridge
i mean stackoverflow basically did what AI is doing rn
mortal thunder
and pottery is a hobby, and a respectable one at that
tame phoenix
the closest comparison I can think of is "I refuse to use machines for the entir...
yeah, this sounds right
shadow ridge
- cant solve problem
- copy paste from mystery source that may or may not be right for your use case
- you dont understand shit and your codebase barely worsk
mortal thunder
i mean stackoverflow basically did what AI is doing rn
kinda but also not really
shadow ridge
well yeah AI is doing it better
mortal thunder
StackOverflow doesn't really give shitty coders the feeling that they can do whatever they want
There are plenty of questions unanswered on SO
tame phoenix
but once the fuss dies down and a few people lose significant money in the AI space, it will settle into one of those things everyone uses for a few tasks. writing your own boilerplate code would be seen as antique and kinda dumb when there's a tool that does it, ya know? like expecting someone to hand roll an entire program in asm instead of a high level language.
mortal thunder
We had a tool for that already, it's called using IDE templates
Or github examples
shadow ridge
but once the fuss dies down and a few people lose significant money in the AI sp...
AI bros like to pretend that boilerplate code is more than what it actually is
tame phoenix
yeah, there's a lot of wrong assumptions right now like:
- AI will replace developers
- AI can't make good code
both wrong.
it can't replace developers because to make good code with it is a complicated iterative process with a competent developer in the loop, think of it like pair programming
digital pivot
stackoverflow
except that sooner or later llm's gonna cost a bunch
shadow ridge
except that sooner or later llm's gonna cost a bunch
thats just analogous to "your issue has been marked as duplicate and closed"
tame phoenix
except that sooner or later llm's gonna cost a bunch
yup, and then nobody will pay, and that's when it's crunch time, you'll find people self hosting it instead.
they'll either optimise their models and become affordable or go bust
look at that Chinese one for example
which the tech bros said was impossible
thousands of times more efficient to train than openAI
while the money is flowing, they have to incentive to optimise
shadow ridge
honestly i never really write programs with basically any amount of boilerplate so i dont think AI will ever help me
in my most recent codebase there are like 2 places where it wouldve helped
and theyre basic functions i wrote in seconds
tame phoenix
you don't? here's someone who's never worked on a big commercial project then I guess
shadow ridge
you don't? here's someone who's never worked on a big commercial project then I ...
thats actually true
tame phoenix
that explains it then
shadow ridge
i only do little hobby projects for myself (ig you could call my game commercial)
im not gonna pretend that AI wont be useful for others
it just doesnt help me in any way currently
but i know exactly how i would use it if i were to work with a big and unyielding codebase
tame phoenix
out of all places I use AI at work most now. I wasn't told to, haven't been given tokens or anything, it isn't mandatory. I chose to and use my personal gpt account. it makes boilerplate for me and helps resolve simple cards faster. but I don't deliver the project faster. I use the extra time to polish what I make and add more tests, refactor old stuff,.etc. it gives me more time to make stuff nice.
so rather than "this change will take 4 hours" the change is done in an hour and then 3 hours of system improvement
but that's also the advantage of small business procedures and not some huge corporation with 6 layers of management, a QA team, project managers and Jira.
digital pivot
out of all places I use AI at work most now. I wasn't told to, haven't been give...
sounds like a good place
tame phoenix
I'm given the leeway to produce as I see fit as senior dev
shadow ridge
thats pretty cool
tame phoenix
sounds like a good place
it is, pretty chill place
the only downside if you aren't that way inclined is, it's all php and js
mostly laravel
but i love php 
shadow ridge
i always wondered if AI gives better output in web dev tasks
like
that shit is ABUNDANT
mortal thunder
yeah it probably does
shadow ridge
and relatively simple
tame phoenix
i always wondered if AI gives better output in web dev tasks
like all tasks, it can be good and other times it can be shit
shadow ridge
ooh also i did use AI one time in the past few months
when i was writing the mandelbrot set test for mintia
i was doing bit math
to get every even bit into one int
and every odd bit into another
i realized it was tedious
so i had AI do it
that helped because i was already short on time and didnt want it to take multiple days
tame phoenix
when I do use it I find the spec I give it has to be absolute, and you have to continually drum in don't repeat yourself, don't get cute with style and "optimisation", readable code is king
if it's readable you can review, understand and iterate
the main thing I use it for though outside of work is docs
it's really good at docs
shadow ridge
honestly i cant feel right in a codebase if i dont know every single edge case and how every part of the design interacts with the others
so i would not feel comfortable making AI do anything other than "hey, this is a function. this is the inputs, this is the outputs. go do it"
tame phoenix
paste in C code... "add doxygen comments to this header", boom, hours of work saved
shadow ridge
the main thing I use it for though outside of work is docs
is it good in generating docs for code that has little comments?
cuz i thought that would be a good use case
to make it explain the general design of a project
and the file system structure
so i dont have to read it
like it's just boring work that nobody wants to do anyways
tame phoenix
yes so long as you give the implementation too and perhaps explain the edge cases of why the code does it the way it does up front, the ai can't know why you chose a specific approach
it's also good for translation, one of my projects is translated to like 10 languages via an automated script that burns gpt tokens
but that's a one off script I run every so often because it's relatively expensive (about £3 per run)
tame phoenix
I mean it is translating hundreds of thousands of words of prose
tame phoenix
how do you tell if the translation is any good
tbh for that project I just wing it and run on user feedback
but it's far superior to Google translate
shadow ridge
also why the fuck does arch linux have zip and unzip as different packages?? and they have different maintainers too
tame phoenix
lmao, check that it's not been hacked considering what happened to aur 😭
shadow ridge
hmm i dont think it is
i think theyre literally just two different projects judging by the urls
tame phoenix
I don't think I'd ever try arch
based on my past expriences with gentoo and the freebsd ports tree i dont think it would be my cup of tea
twilit smelt
it's a tool in the toolbox, like a compiler or a linter and to be principled and...
Not interested in arguing with this goofy opinion I've heard 500 times before
shadow ridge
Not interested in arguing with this goofy opinion I've heard 500 times before
im interested in hearing your opinion on it
because it's not an instantly indefensible position
twilit smelt
It obviously is though
shadow ridge
i dont really see it
it is very expensive (or about to be, at least)
but that would just put it in the same category as like a paid IDE subscription or something
dense vigil
Not interested in arguing with this goofy opinion I've heard 500 times before
Well Nadia said Claude was just vim sooo 
shadow ridge
who is nadia
dense vigil
A person who vibecoded their entire os (200k lines) and said that vibecoding was just a vim alternative
twilit smelt
I understand how someone can hold the "LLMs are just another tool in the toolbox! this is like being anti compiler!" opinion but only if they're being incredibly short sighted and refusing to think about it for longer than 5 seconds
shadow ridge
if we're going off of definitions
you could say that a tool helps you do something
it doesnt do it for you
shadow ridge
why do you think it's so absurd though
i think LLMs have more in common with compilers than vim
i bet you could train a neural network on huge C codebases and their binary outputs to make a weird AI "compiler"
dense vigil
they can already do that
not on like, the entire linux kernel but
simple programs they can transpile to machine code directly
queen torrent
Not interested in arguing with this goofy opinion I've heard 500 times before
Luddite!!! Already replaced ngmi etc etc etc etc etc
swe extinction event 18 months code monkey obsolete prompt engineer was the new plumber ai slop flood 10,000x efficiency bloodbath white collar massacre senior dev coping "i'll just learn prompting" too late agent swarm devours entire sprints autonomous codebase rewrite overnight mcp replace ceo cto cfo in one cycle google deepmind meta openai xai anthropic all racing to 10^6x inference scale datacenter cities glowing 24/7 sucking the entire grid dry energy armageddon power rationing for humans while gpus feast
shadow ridge
the one thing I don't get is people who're like
"coding is solved. we dont need programmers anymore"
like
wut
raven drift
Well Nadia said Claude was just vim sooo <:chad:815144781842743334>
well if you don't care about code quality and maintainablity and just results, you can do that
the main thing that agents can't do today is write "clean" / maintainable code
if you just care about time to demo, LLMs are great
warm pine
I don't think human made code is going away in our lifetimes
it will certainly persist and i dare say it will become the educational standard
everyone is taught arithmetic the traditional way at school regardless of the fact that they will go on to use an electronic calculator later
tame phoenix
exactly
shadow ridge
everyone is taught arithmetic the traditional way at school regardless of the fa...
i was very surprised when i first heard that in some countries people are allowed to use calculators in exams and stuff
tame phoenix
LLMs will replace some programming tasks, like calculators replaced some mental arithmetic. but you must still have the base understanding of the concept to be able to succeed
this is why 100% vibe coded projects often fail
ask an LLM to convert between two programming languages for example, or refactor code. and i dont mean agents
tame phoenix
i was very surprised when i first heard that in some countries people are allowe...
depends on the question, in our exams some questions require a calculator others forbid it, generally not in the same paper
or at least thats how it was when i sat GCSE maths in the 90s
shadow ridge
i find it odd that a question would require lots of mental arithmetic tbh
like
thats just the throwaway "lets get this over with" part after youve already done the work
twilit smelt
shadow ridge
my cat always wants treats
he'll meow in front of my door
and when i open it
he runs past me
into the kitchen
waiting for a treat
blissful smelt
every single character in my project's source is a manual keystroke or a copypas...
the no lsp lifestyle
twilit smelt
Little silly person
queen torrent
lsp is just one tiny guy in your computer giving you suggestions
dense vigil
Little silly person
As opposed to a large, smart person
twilit smelt
As opposed to a large, smart person
Yeah that one doesn't nitpick your code because he's the bigger man
shadow ridge
the no lsp lifestyle
this reminded me that
for like the past however many years ive been programming
ive just been rawdogging vim
with no lsp
night needle
lsp is just one tiny guy in your computer giving you suggestions
ai
blissful smelt
lsp:
twilit smelt
twilit smelt
Mintia is being renamed copperclad
sharp igloo
Mintia is being renamed copperclad
Bomboclat
twilit smelt
shadow ridge
help pls i ordered my XR/Phone by paying 0.12 $MINTCOINs ($6000+ dollars)
and when it arrived
and i opened the box
it was a pile of dog shit
i think there has been a mix up
do i mail this back or should i donate it to an animal shelter
shadow ridge
@twilit smelt do you prefer an incremental approach to lexical analysis
or tokenizing an entire file first
and then operating on it
lucid umbra
the common frontend, used by all the different newsdk tools that need lexical analysis, does the incremental approach
but that approach is the main thing that makes the hygenic macro system work
since the preprocessor/lexer is live while semantic analysis is happening, macro instantiations can create their own scopes
lucid umbra
my jackal frontend does preprocessing/lexing all at once, but it has to insert some special marker tokens in order to accomodate the macro scope stuff
i do this because its useful to be able to re-trace the token buffer for nice error messages (particularly involving the preprocessor) and the like
lucid umbra
it is nicer imo
especially if you can crunch your tokens down to be super thin
i got each token in coyote to be 8 bytes
shadow ridge
8 bytes?????
howw
thats line + column info
token type info
token data (lexed number constants etc.)
all in 8 bytes??
lucid umbra
you dont need line and column info
shadow ridge
you dont need line and column info
how come
lucid umbra
you can just reconstruct that when you need it
you only need it during error messages
shadow ridge
what about generating debug data?
lucid umbra
still trivial to generate if you need it later
the point is that you don't need it for most of the compilation pipeline, so you shouldn't keep it around for most of the compilation pipeline
typedef struct {
// token kind
u64 kind : 7;
// was this token "generated" by the preprocessor?
// this means that the span might not correspond to source text.
u64 generated : 1;
// textual content
u64 len : 8;
i64 raw : 48; // pointer trick! only the first 48 bits of a pointer are significant
} Token;
static_assert(sizeof(Token) == 8);
technically you dont even need the token length but whatever i think its useful to keep around anyway
you also don't need to keep a token's computed data, like a numeric token's value
that can just be computed in semantic analysis, right when its needed, you can just generate it and put it right into your AST or whatever
shadow ridge
how does this handle large strings
lucid umbra
jackal doesn't allow tokens over a certain length so it isn't a problem in coyote, but there are several ways to go about it
shadow ridge
ig you could just have C style string merging ("AAA" "BBB" = "AAABBB")
and just split it up to multiple string tokens
lucid umbra
the way that i would do it, and the way that the zig compiler does it, is to just not store the token length
most of the time its redundant information
short owl
Who's the man with the sandwich
lucid umbra
most of the time you only need a token's length in a single place in semantic analysis, and you can write a function to trivially recompute the length based on the token kind and the index/pointer in to the source text
for most token types actually, the length is static
for example, for the keyword int, you always know tokens with kind TOK_KEYWORD_INT will have length 3
and so on
shadow ridge
that can just be computed in semantic analysis, right when its needed, you can j...
i dont like that we need to do redundant operations here tbh
lucid umbra
wdym redundant? you don't need to compute a token's value in the lexer
shadow ridge
yes but you still need to iterate over it to validate it
lucid umbra
you just have to scan it according to the grammar rules of the token
lucid umbra
Who's the man with the sandwich
tis me
shadow ridge
my old compiler has 72 byte tokens
lucid umbra
i dont like that we need to do redundant operations here tbh
believe it or not the slight redundancy is faster than bloating the token structure with that value
shadow ridge
lucid umbra
wow
shadow ridge
this also precomputes hashed strings
lucid umbra
no im sandwichman
different
i work on a third-party jackal compiler and my own stuff
but im familiar with the newsdk internals so i answered jack's question abt it
shadow ridge
i wish i had more hours per day to work on all of my hobbies and education at the same time :\
lucid umbra
thats real
if you don't care about performance then you can store as much as you want in your token structure
im just saying that these tradeoffs are tried and tested
shadow ridge
i thought it was about memory footprint
lucid umbra
in modern compiler frontends
shadow ridge
not performance
lucid umbra
those go hand-in-hand
if you can fit 8 tokens into a cache line it's fantastic
zip zoom all the live long day
shadow ridge
i do the iterative approach in this compiler
shadow ridge
so does the cacheline really matter
lucid umbra
if you're not optimizing for performance, then no
it doesn't
i'm just saying that thats what i do and thats what other modern, performance-optimized frontends do
lucid umbra
i'm not actually sure
shadow ridge
hmm let me check
lucid umbra
carbon and zig are the big examples of hyper-performance-optimized frontends
that ive seen
this is rustc's token ```rs
pub struct Token {
pub kind: TokenKind,
pub span: Span,
}
nah their tokens are pretty big
lucid umbra
i suspect because of the preprocessor and all the c++ bullshit
but yeah thats why i make the decisions i do in coyote
it's all about what your goals are
my goals are performance and memory usage
if it makes the compiler harder to write in some places, that's a price i'm willing to pay
jagged marsh
but yeah thats why i make the decisions i do in coyote
coyote is not a public archive?
jagged marsh
oh yeah
lucid umbra
thanks
i threw it together in like 5 minutes and the generator itself is pretty slow lmao
but it does work
shadow ridge
language design wise
im still debating on
whether to have strings as a data type
or just a struct defined by the compiler
in my old compiler theyre just char * and 64 bit size in a struct str
lucid umbra
is your language manually memory managed?
shadow ridge
yes
lucid umbra
interesting
rust does a nice thing where str is basically identical to a slice of u8 with the additional guarantee that the content is utf-8
lucid umbra
i dont have arrays as types
does it have slice/array reference types
lucid umbra
inchresting
well then i lean towards compiler-defined struct
if you don't have either of those
shadow ridge
yeah ig thats the best
lucid umbra
slice types are pretty useful i recommend them
just pointer and length in a convenient package
twilit smelt
coyotl
shadow ridge
i just dont really see how to make a good type system without having it really minimal
a lot of stuff feels redundant or hard to parse / gen code for
lucid umbra
wont know til you try + i like to answer questions about compilers so feel free to hit me up
twilit smelt
coyotl
this was the original nahuatl (aztec) word for coyote but the Spanish were stupid so they kept hearing coyotl as coyote and now we say coyote
shadow ridge
with slice types my problem is
a lot of the time you want differently typed size field
otherwise you just get 2 words
lucid umbra
that's an interesting thing to consider in your syntax
but also, wouldn't it always take up two words anyways?
because the size of a type should always be a multiple of its alignment
and the alignment has to be word size, because pointers are word sized
so it would be the size of two words anyways
shadow ridge
but also, wouldn't it always take up two words anyways?
im considering cases where you have an upper bound already
like your token struct for jackal for example
lucid umbra
in case where you're optimizing for layout like that, it's probably beneficial to split it up into just separate data and length fields inside your container type
something that i like that zig does is that you can extract a slice's pointer and length just like a struct with data and len fields
so going from slice <-> ptr, len is trivial
the main advantage of slice types is for things like parameters and return values, or plain variables
imo
shadow ridge
in case where you're optimizing for layout like that, it's probably beneficial t...
i suppose i could also use my struct member offsets to overwrite a part of the slice with other data
twilit smelt
HEEEEEEELP!!!!!! HEEEEEEEEEEEEEELP
lucid umbra
i suppose i could also use my struct member offsets to overwrite a part of the s...
i definitely wouldn't suggest that but it is certainly an option
lucid umbra
HEEEEEEELP!!!!!! HEEEEEEEEEEEEEELP
🛟
devout geyser
HEEEEEEELP!!!!!! HEEEEEEEEEEEEEELP
i click on this thread and this is the first thing i see
queen torrent
HEEEEEEELP!!!!!! HEEEEEEEEEEEEEELP
twilit smelt
shadow ridge
this is probably older than you at this point
shadow ridge
i definitely wouldn't suggest that but it is certainly an option
struct thing
{
u8 'a; // u8 *data; u64 size;
$-4 u32 b; // overwrites half of size
};
hmm i could maybe use a backtick for spans
u8 `a
or nah idk
lucid umbra
this is probably bad syntax, but something like this ```rs
struct thing {
ptr: [^.len]u8,
len: u32,
};
ah..
lucid umbra
a "slice" type that is just a pointer with the static information that the length is contained somewhere nearby
this is possible
but
it makes your type system much more complex
lucid umbra
if you want a type where the length is part of that type, but you don't want to store it in the same storage as the type's value, then i don't see another way to do it
this doesn't add any runtime information btw
its purely static
it does make taking that slice type anywhere else very tricky
because you have to make sure that the length always accompanies it
shadow ridge
i think i have a better idea on how to make this work
lucid umbra
(using my own language's syntax as an example) what i would suggest is having a slice type []T, but also have an indexable pointer type [^]T which is distinct from a regular pointer type ^T
lucid umbra
use the slice type when you can, but otherwise just keep the indexable pointer and length separate if you need that
shadow ridge
struct thing
{
u8 *ptr;
u32 len;
const u8 ^span = ^(ptr, len); // this is purely an alias
};
like a macro that belongs to the struct ig
lucid umbra
interesting, afaik this is basically an alternate syntax to having like an inline method that constructs and returns a slice from the struct
lucid umbra
https://youtu.be/2FspwOKfGPY
the silly
lucid umbra
```c
struct thing
{
u8 *ptr;
u32 len;
const u8 ^span = ^(ptr, len); ...
this is a very interesting way to express it
i kinda like it
shadow ridge
use the slice type when you can, but otherwise just keep the indexable pointer a...
if i already have a slice type id rather not have any pointer arithmetic at all tbh
shadow ridge
this is a very interesting way to express it
you could also treat a span as a struct definition
and do
u8 ^span {
u8 *ptr;
u32 len;
}
i'm not sure what this is expressing
shadow ridge
the ptr and len still belong to the struct
the { } just mean the span is aliasing its own members as them or smth ig
shadow ridge
```c
struct thing
{
u8 *ptr;
u32 len;
const u8 ^span = ^(ptr, len); ...
yeah this is probably better but i dont like the concept of struct members having any logic
lucid umbra
the way i would write this in my in-progress language is: ```rs
struct Thing {
ptr: [^]u8,
len: u32,
pub inline fun span(^self) -> []u8 {
return self.ptr[..len];
}
};
and you would do thing.span() and use that as a slice
and because it's inline it's guaranteed to be inlined and optimized away
like if you were just using the ptr and len fields by themselves
lucid umbra
you could also implement the compiler's Index trait and then just use the thing[idx] operator directly
but i think a generics/trait system is out of the scope of your language given that you want it to be minimal
the lesson is that unfortunately i dont think there is a simple solution for this problem
lucid umbra
(using my own language's syntax as an example) what i would suggest is having a ...
the thing i would recommend for your language is this
a [^]T is just a ^T that allows you to use the index operator
but if you dont wanna separate those concepts then its okay
shadow ridge
i just want there to be 1 solution for every problem and it feels redundant to have all those things at the same time
lets say i have both spans and pointers
well pointers are just spans of size 1
so thats reundant
and feels shit
lucid umbra
unfortunately there is never one way to do things that will satisfy all use cases well
shadow ridge
this is my old type system
lucid umbra
you could do what C does and just have one pointer type with no length, but that means that you have to manually supply the length every time you want to pass it somewhere
which if you dont mind that, then that might work for you
shadow ridge
unfortunately there is never one way to do things that will satisfy all use case...
ig ill just keep my old system
lucid umbra
i personally prefer the ergonomics of being able to reference an array of values and keep the length around
sure thing
shadow ridge
this is my old type system
i think the most interesting thing here is the moff field in struct members
basically you can have the offset of a struct member
be the value of another struct member
lucid umbra
that sounds like dependent types to me lmao
shadow ridge
i think the most interesting thing here is the `moff` field in struct members
i think i intended for it to make it easier to parse file formats
and have multiple FAMs in one struct
lucid umbra
it's not a property of the type though
not of the field type, but the struct type
shadow ridge
i guess
lucid umbra
its size is dependent on a value contained in it
thats a pretty canonical example of a dependent type if i understand them correctly
that makes sizeof a runtime calculation then? for structs that have that?
shadow ridge
its size is dependent on a value contained in it
oh nah, it has a constant size no matter what
my structs are like little assembly programs ig
the size is just the offset of the last variable + its size
a union is literally a struct but it disables the offset increments
lucid umbra
the size is just the offset of the last variable + its size
but that offset might be dynamically determined, right?
shadow ridge
you can make a union without the union keyword by just making a struct with every offset as just @ 0
shadow ridge
but that offset might be dynamically determined, right?
no, if you use moff, it just doesnt affect the offset at all
so it's like a FAM (the field has effectively 0 size)
lucid umbra
ah
shadow ridge
struct Thing
{
u64 fam0_size;
u64 fam1_size;
u64 fam2_size;
u8 fam0[];
u8 fam1[] @fam0_size;
u8 fam2[] @fam1_size;
};
this is a good use case i think
lucid umbra
interesting
shadow ridge
but this struct would have a sizeof of 24
shadow ridge
struct Thing
{
u64 thing @-8;
};
``` ok i actually havent thought about this use case because it's totally possible and it would create a struct with negative sizehahaha this totally breaks pointer arithmetic
lucid umbra
i get the vibe that this language isn't big on safety checks
shadow ridge
i get the vibe that this language isn't big on safety checks
lol it does 0 checks
it's just C but with my additions
and removals
where i saw fit
i think i showed this to you before but u forgot ig
in like 2024
lucid umbra
hmmmge i remember vaguely
shadow ridge
this is an operator
shadow ridge
it also works on magic constants so you can make a bit enum and do var@FIELD0
lucid umbra
is just `(var >> 5) & 7`
is it & 7 or & ((1 << 7) - 1)
shadow ridge
& 7
lucid umbra
interesting
shadow ridge
so it extracts bits 5, 6 and 7
shadow ridge
what's the type of FIELD0 here?
it's an int where first 6 bits are offset
and later 6 bits are size-1
theres a special enum type that constructs these values
you could build them at runtime but it would generate extra code
here's an example usage in the lexer
where it's building that value at runtime
lucid umbra
what's the @@@ there
warm mural
yeah wtf is that lol
lucid umbra
is a weird editor thing
shadow ridge
ooh those are the spaces
lucid umbra
interesting
shadow ridge
i like to know if whitespace is tab or a space
lucid umbra
what editor is this?
lucid umbra
interesting
in vscode if you highlight it whitespace it will show you what the characters are
this is what spaces looks like
shadow ridge
yeah i should pick a better char
lucid umbra
sorry will would you like us to talk abt this somewhere else
twilit smelt
What will they look like in xrcode the native jackal ide running on mintia2
lucid umbra
What will they look like in xrcode the native jackal ide running on mintia2
fox emoji
shadow ridge
What will they look like in xrcode the native jackal ide running on mintia2
idk let the people in the next century think aboujt that
lucid umbra
or jackal emoji
we'll make one
and add it to unicode in some unused/vendor-reserved spot
shadow ridge
im returning to my project because will is employed so my only competition is aphelion now
shadow ridge
im not and i still have basically no time to work on it
damn you 24 hour days
cant the earth just revolve a lil slower
lucid umbra
if it did they would just make you work longer
shadow ridge
true
golden meteor
@twilit smelt (@hidden shore wants to ask) are you the only person in utah
keen stone
golden meteor
there are like 3.54 million people in utah
I've only ever heard of will, so I think this is false
shadow ridge
personally i havent seen anyone younger than 4 so that checks out
golden meteor
never known anyone else except him from utah
shadow ridge
what about that bricks and minifigs guy
he also lives in utah i think
not the reckless ben guy
the guy who stole the lego
twilit smelt
<@783460822525149196> (<@1482737888390021284> wants to ask) are you the only pe...
No, I am not in Utah
twilit smelt
I'm glad I got continuations done before I left
It'd really bother me if that were still half done
I mean I didn't fully implement the direct return to usermode continuation
Gonna work on that soon
It's the last piece of the puzzle
Fun stuff is needed for that
Specifically upon a syscall (and only a syscall) exception I need to save all the callee-saved registers in a special location that is not the kernel stack bc it's disposable now
Probably a block in the thread struct
The return to usermode continuation will restore those callee saved regs before jumping back to userspace
(And probably clear the caller saved regs to avoid leaking kernel info)
shadow ridge
i thought continuations were meant to lose information
why do you need to save any registers at all?
twilit smelt
Because if usermode does a syscall it expects the callee saved regs to be saved across it bc it's a function call like event
One example of a callee saved reg that's pretty important is the program counter for example
shadow ridge
Because if usermode does a syscall it expects the callee saved regs to be saved ...
but if you're specifying a continuation then it's not like a function call at all
twilit smelt
Continuations are a kernel mode implementation detail
shadow ridge
then why can the user specify a continuation on a syscall??
twilit smelt
Never said it could
shadow ridge
so if the user cant specify a continuation then why do you need to handle it
twilit smelt
Because the implementation of a syscall can use continuations
shadow ridge
ooooohh
twilit smelt
And it will otherwise lose the user context that was saved on the kernel stack
Unless I save it elsewhere
shadow ridge
ok nvm i misunderstood the issue
shadow ridge
Specifically upon a syscall (and only a syscall) exception I need to save all th...
hmm so why have any callee saved regs except a stack pointer and the PC btw
realistically those 2 should be enough
twilit smelt
Mainly it's to have a consistent user context to refer to while the thread is in kernel mode regardless of whether it got there via syscall or via exception or what
short owl
https://tenor.com/view/you-are-the-youngest-person-ever-how-old-are-you-im-4-yea...
He's 20 years old
keen stone
bruh
twilit smelt
teal trench
https://tenor.com/view/kanepixels-kaneparsons-white-people-when-pechomigurt-back...
youngest person ever
twilit smelt
@dense vigil remember our pact that when mintia2 was ported to amd64 then we would combine our efforts
twilit smelt
u don't remember?
dense vigil
I remember talking about it but iirc i said i didnt want to deal with it since it wasnt in C
twilit smelt
I said since u seemed to like working on everything other than kernel proper and I only liked kernel proper it would be cool if we combined efforts later and u agreed but this was like 3 yrs ago
I said ur parts would be in C
dense vigil
Which means I'm maybe halfway there
ex hyena would be 500% there
twilit smelt
Weren't you doing like uacpi and some gpu drivers and stuff for ur own hobby os
dense vigil
its in the plans yes
warm mural
u guys should instead focus on the obviously cooler project (mine) that runs on a REAL computer
twilit smelt
ex hyena would be 500% there
He died when I stopped being able to respect myself as a hyena.