based on my reading of the current announcement message id say it having been reviewed by you and the general structure architected by you makes it fit here
#Ultra
ornate jasper
the announcement doesnt (currently, im too lazy to try and look at edit history) make any mention of requiring things here to be 0% AI, just not full vibecoding
sinful river
I have 0% AI generated code apart from tests too
I'd say the no-ai distinction is a bit useless, and it'd be pointless to move the thread and lose history
ornate jasper
idk
i have a firm no-ai policy even in issues/docs/descriptions so moving there is right for my project
sinful river
Why
ornate jasper
ai shit is harder to read. it adds a lot of fluff that frankly makes it harder to get to the point of things, and it cannot be trusted to correctly interpret intention
haughty notch
I'd say the no-ai distinction is a bit useless, and it'd be pointless to move th...
it feels like that category is kinda invalidating all of my work
lost raptor
could always just do a complete rewrite and not use ai moving forward if its a big thing for you
haughty notch
could always just do a complete rewrite and not use ai moving forward if its a b...
It took me 4 years to write what I have, complete rewrite would be insane at this point
sinful river
could always just do a complete rewrite and not use ai moving forward if its a b...
That's not feasible for large projects
haughty notch
it's 70k LoC I think
sinful river
ai shit is harder to read. it adds a lot of fluff that frankly makes it harder t...
You can always write most of it, then refine it with the help of AI
haughty notch
it's 70k LoC I think
and a lot of planning and descisions and things I've been slowly iterating upon
ornate jasper
if the existence of a no-ai category is invalidating you rather than seeming like a place to stick the weirdo hardliners like me then uh maybe examine that feeling ig
sinful river
I'm not a native English speaker and it sure helps with writing technical documentation
willow fern
my experience of AI is that it can often turn out things where there don't have to be too many subtleties and considerations, but unfortunately kernels don't have much of that
ornate jasper
You can always write most of it, then refine it with the help of AI
im also extremely anti-ai-in-all-forms-no-exceptions
sinful river
why?
haughty notch
and a lot of planning and descisions and things I've been slowly iterating upon
Like sure, there are things I don't quite like, but I don't know if I would have energy to rewrite my OS
lost raptor
It took me 4 years to write what I have, complete rewrite would be insane at thi...
do you not have places in your code that you felt like you could do better?
sinful river
my experience of AI is that it can often turn out things where there don't have ...
Yeah, in a kernel there is too much context that is needed for it to be useful
willow fern
and if you try to get them to do anything where there are these considerations, you end up with them adding more and more flags and states and you end up with a combinatorial explosion
haughty notch
do you not have places in your code that you felt like you could do better?
wdym, like all of them
ornate jasper
why?
its a plagiarism machine that takes as much effort to edit into something actually useful as it wouldve been to just do shit myself in the first place
haughty notch
wdym, like all of them
But like I wasn't really telling AI to write code for me
willow fern
Yeah, in a kernel there is too much context that is needed for it to be useful
it's the worst for kernels but i was quite shocked at how badly they did on other tasks too
haughty notch
(except for a couple of places in 2023 I think, when the AI was all the hype, which I'm rewriting/not really using anyway because it's an unmaintainable mess)
willow fern
just today i had heard that claude 4.7 is a genius so i asked him how he would add auto-restart to the service manager i am writing for keyronex and the proposal was horrible
sinful river
it's the worst for kernels but i was quite shocked at how badly they did on othe...
I'm not sure, I've never tried giving it complex tasks, it's always been for writing tests or docs
ornate jasper
its a plagiarism machine that takes as much effort to edit into something actual...
like i ported the ultra log ring to zig for my project, and in the process learned how it works (up to the limits of my understanding of atomic order), and was able to bang that out fully by hand in less than an afternoon
ornate jasper
I'm not sure, I've never tried giving it complex tasks, it's always been for wri...
the thing for me is (apart from my own personal anti-ai bias), if i cant trust something (whether ai or a human assistant or whatever) to write the code right i cant trust it to understand it well enough to write docs or tests
sinful river
There's a difference between writing complex code and reading it
And even then, it's not like you can't just review and edit lol
willow fern
it was a mess and maybe it would have worked but i couldn't understand it, i have a single state variable to drive the service lifecycle and it proposed 4 new variables which all have subtle effects and turned my state machine with clearly defined inputs like "PID exited" or "Stop requested" into a monstrosity
ornate jasper
And even then, it's not like you can't just review and edit lol
that again would take me a lot longer than just doing it
sinful river
I doubt that
haughty notch
I haven't been using AI for documentation because I don't really have documentation 
Also, I think AI models kinda get confused by microkernels?
ornate jasper
There's a difference between writing complex code and reading it
maybe but docs are also meant to capture intent and ideas imo not just the raw what the code does. i can look at some weird linux code and understand what its doing and maybe ai could quickly whip up docs about what its doing, but if thats true than why do the docs need to exist for what its doing, the docs are for WHY and THAT is what ai cant be trusted to figure out imo
willow fern
this exactly reflects my last experience where i asked claude to make some improvements to my TCP stack. he wanted to introduce all kinds of new code paths while i deliberately had stayed to the course set by the BSD stack, where you have a single output function and output is programmatically determined based on the current states. just obnoxious work, i think AI seems to be allergic to that style of programming
sinful river
maybe but docs are also meant to capture intent and ideas imo not just the raw w...
Because id rather skim through a large documentation comment rather than try to understand the details and intricacies of implementation
willow fern
on the other hand they are alright for documenting and testing for the most part
ornate jasper
Because id rather skim through a large documentation comment rather than try to ...
doc comments take like 30 fuckin seconds to write itll take longer for an ai to fuckin load it up and process it than for me to write them, i was talking the bigger deal docs
haughty notch
Oh, I write doc comments myself
If that's what people mean by documentation, then I try to comment my stuff
ornate jasper
If that's what people mean by documentation, then I try to comment my stuff
it wasnt what i meant lol but ig thats part of the confusion here is that wasnt clear lol
haughty notch
I mean I let copilot autocomplete it, if that's more or less I'm writing anyway
ornate jasper
everyone saw docs and thought something different
haughty notch
I wanted to write a high level overview of what I have, eventually
Because the code comments are not a good documentation in my eyes, I guess
sinful river
doc comments take like 30 fuckin seconds to write itll take longer for an ai to ...
they dont take 30 seconds to write lol
by doc comments I mean large top-level comments of like > 30 lines
haughty notch
Because the code comments are not a good documentation in my eyes, I guess
With a microkernel, which makes it a collection of programs
sinful river
by doc comments I mean large top-level comments of like > 30 lines
can easily take 30 minutes to write properly
ornate jasper
by doc comments I mean large top-level comments of like > 30 lines
ok so 5-10 minutes but itd still take me longer to edit something down
prime wraith
the announcement doesnt (currently, im too lazy to try and look at edit history)...
0% ai is the other thread
this one is ai assisted
sinful river
idk for me specifically it helps, I am not a native english speaker and I would rather have something be slightly ai generated that was written in a better manner and faster than something all written by myself that takes longer to do and is less well written
ornate jasper
then leave your thread in this forum and its fine
sinful river
my stance is "no stance" basically
ornate jasper
as much as im anti-ai for my own stuff i wont really begrudge anyone else as long as its not in the place where bsky is rn where its fully 100% vibe-coded and breaking every 5 seconds because of it lmao
(in which case i just wont use it and still wont care)
yall do your own things lmao its fine
haughty notch
I think the thing that doesn't help is that I kinda dislike AI, but have tainted my stuff with it
ornate jasper
yeah thats your own issue to unpack
sinful river
I wouldn't say I like AI, but I do find it useful in some cases
I have been using it as a studying method (generating quizzes etc) and it works really well
prime wraith
idk for me specifically it helps, I am not a native english speaker and I would ...
I thought u were from Canada
prime wraith
I did but I didn't pay attention lmfao
prime wraith
So you're from french Canada?
sinful river
quebec yeah
prime wraith
Makes sense
sinful river
I literally have the least english name possible which is a bit of a shame
its hard to pronounce
prime wraith
I think the thing that doesn't help is that I kinda dislike AI, but have tainted...
Why do u care so much lol
prime wraith
x86 and PA-RISC are the only two architectures that select HAVE_UNSTABLE_SCHED_CLOCK 
prime wraith
What does it do, and why does Linux need it? <:trl:829316641191886888>
it adds code that works around jittery sched clock
aka unstable tsc on x86
haughty notch
Is it because of tsc offset firmware mess?
haughty notch
then why use tsc in the first place?
prime wraith
let me explain
sinful river
if no invariant tsc I fallback to hpet
haughty notch
me too, I use hpet/pmtmr
sinful river
yeah if no hpet then pm timer
haughty notch
I decided not to support PIT in any form
with kvmclock you would read the kvm struct anyway
prime wraith
linux has the following clock types:
struct clocksource, an abstraction over a counter (this can potentially be very expensive to read but it provides precise timestamps, client code usesktime_get()to access the best clocksourcesched_clock(), this is designed to be a very fast reentrant and nmi safe clock used on very hot paths, linux uses TSC even if its not constant for this purpose, since the scheduler just needs to know the amount of ticks that have elapsed on this specific CPU and doesnt care much if it drifts a bitstruct clock_event_device, an abstraction over an event generating clock, e.g. deadline lapic etcjiffies, just a global variable but generic code makes sure to increment it with ticks, this is the fastest but most coarse clock source, used as fallback etcwall_clock, a thing that returnsstruct timespec64, actual irl time
sinful river
I have recently renamed my TimeCounter to ClockSource actually
slowly turning into linux 
prime wraith
i kinda hate the name clocksource
its way too broad
ill probably call it a counter
sinful river
a counter is way more broad lol
sinful river
so TimeCounter was a better name maybe 
prime wraith
clocksource sounds like something that gives u a wall clock
prime wraith
so TimeCounter was a better name maybe <:meme:575445094589661215>
yeah maybe
sinful river
timecounter is a BSD name
prime wraith
i want to maybe have a similar prefix/suffix for event and counter clock types
sinful river
I renamed it because I liked clock.zig better than time.zig
prime wraith
clock sounds like it should return a timespec
a monotonic counter is definitely not a clock
sinful river
ok maybe I should revert the commit
prime wraith
based
sinful river
I think you're right that clock is a weird name
prime wraith
yeah
at least according to this definition
kvm has a separate msr that returns an actual wall clock
its u32 sec, u32 nsec
sinful river
like what? Also what about vmware
prime wraith
prime wraith
like what? Also what about vmware
vmware has even more
and hyperv has even more
most of arch specific stuff like that will be entirely offloaded
sinful river
actually idk if I want to rely on cool features more
because then it's not as portable
idk
sinful river
no
prime wraith
by overriding arch callbacks
sinful river
but then you get super fast tlb shootdown or IPI or something on kvm but not on other platforms
prime wraith
arch.tlb_flush = vmware_tlb_flush
and by default its just arch.tlb_flush = native_tlb_flush
sinful river
that is slow though
prime wraith
what is?
prime wraith
ur joking right
prime wraith
yes its phenomenal overhead
but yeah in reality since linux has a static_call mechanism and they didnt bother using it for this
sinful river
I don't really need that tho since my tlb shootdown wont be synchronous
prime wraith
that means it doesnt matter
prime wraith
I don't really need that tho since my tlb shootdown wont be synchronous
arch.tlb_flush_begin
sinful river
what would that do
prime wraith
start async work
the cool part is like, the vcpu you're trying flush the tlb of might be preempted on the host, so u can just skip flushing stuff on it
and kvm allows you to detect that
sinful river
I dont think I need architectural support at all for async shootdowns
only a way to invalidate a page
sinful river
the cool part is like, the vcpu you're trying flush the tlb of might be preempte...
and when will it flush then
prime wraith
on vcpu resume
u cmpxchg a flag that tells the kvm to do it
also pvspinlocks are insanely cool
you can force kick vcpus to schedule them over yourself on the host etc
u can tell kvm that you're try to acquire a spinlock currently held by a preempted vcpu so it will schedule it
sinful river
u can tell kvm that you're try to acquire a spinlock currently held by a preempt...
ah yeah I saw that in qspinlock code
prime wraith
ye
and hyperv is gigantic https://elixir.bootlin.com/linux/v7.0/source/arch/x86/hyperv
in terms of features
tlb flushing is just a hypercall
There's also XEN ofc
which offloads even more
but thats a bare metal hypervisor
and pvclock was invented by XEN
haughty notch
are you out of your mind
Meanwhile my kernel has virtual functions spammed all over it
prime wraith
fuck i could really use kconfig integration now
ornate jasper
and hyperv is gigantic https://elixir.bootlin.com/linux/v7.0/source/arch/x86/hyp...
I imagine this is partly because windows wants to run under hyperv itself if possible nowadays
so they can do secure vs nonsecure kernel memes
carmine token
but yeah in reality since linux has a static_call mechanism and they didnt bothe...
static_call is for spectre mitigation
not for perf
isn't it?
ornate jasper
yup
carmine token
yeah
Static calls use code patching to hard-code function pointers into
direct branch instructions. They give the flexibility of function
pointers, but with improved performance. This is especially important
for cases where retpolines would otherwise be used, as retpolines can
significantly impact performance.
prime wraith
I imagine this is partly because windows wants to run under hyperv itself if pos...
yep
carmine token
well it is to recover perf lost to retpolines
carmine token
it literally says so in the cover letter that added them
ornate jasper
it was originally as a less-shit-perf replacement for retpolines, but now can get used for other stuff because its also just better perf sometimes
ornate jasper
yep
the secure/nonsecure kernel memes are actually really cool tbh, maybe once my kernel is fully working for normal shit i can look into making that sort of thing (esp since aarch64 has hardware support for that too)
carmine token
All other usecases are bonus, but it would certainly be interesting to
investigate the impact of using these APIs for tracing: that too is a
feature enabled everywhere but utilized only by a small fraction of Linux
users - so literally every single cycle or instruction saved or hot-path
shortened is a major win.
static calls are still slower than the alternatives mechanism that linux has
prime wraith
alternatives are for patching instructions, not call targets
carmine token
can't you patch the whole call?
carmine token
apparently on some archs static calls compile to two calls
because these archs cannot patch call isns
ornate jasper
the one use of alternatives ive actually seen looking at their code is that the mmio read is changed by it to use load-acquire on some stupid early apple hardware lol
carmine token
so they patch a trampoline instead
prime wraith
like the original reason was to patch hot repolined stuff, but these days im pretty sure most uses that are arch specific are for avoiding indirection on hot paths
prime wraith
static calls are still slower than the alternatives mechanism that linux has
btw where did u get that info from
carmine token
apparently only x86 and ppc have STATIC_CALL_INLINE
carmine token
btw where did u get that info from
discussion in the original ml thread
prime wraith
slower for other arches or?
carmine token
since static calls still need to prepare the calling conv
while patching out stuff entirely doesn't
prime wraith
not sure what u mean by that
carmine token
for example, for tracing it is preferable to have no call at all (only nops)
if it's off
prime wraith
oh ok
carmine token
(and also no mov rdi, #first_arg etc)
prime wraith
but the most common use case is just replacing the call target
so even if u did alternatives its just the offset u patch
carmine token
$ grep -r 'select HAVE_STATIC_CALL'
arch/Kconfig: Where an architecture does not select HAVE_STATIC_CALL_INLINE, any
arch/powerpc/Kconfig: select HAVE_STATIC_CALL if PPC32
arch/powerpc/Kconfig: select HAVE_STATIC_CALL_INLINE if PPC32
arch/x86/Kconfig: select HAVE_STATIC_CALL
arch/x86/Kconfig: select HAVE_STATIC_CALL_INLINE if HAVE_OBJTOOL
why cant u patch these on arm btw
carmine token
you can probably but ARM doesn't have retpolines either
but also, on ARM the call isns immediates are smaller
carmine token
according to Kconfig, arm does not have special support for static calls and it just compiles them to function pointers
prime wraith
perhaps
carmine token
yeah this was added in 7.0
also HAVE_STATIC_CALL is the variant with trampolines
Unlike on x86, there is no pressing need on arm64 to avoid indirect
calls at all cost, but hiding it from the compiler as is done here does
have some benefits:
- the literal is located in .text, which gives us the same robustness
advantage that code patching does;- no performance hit on CFI enabled Clang builds that decorate compiler
emitted indirect calls with branch target validity checks.
this is the rationale for adding it
sinful river
no runtime tricks and is perfectly predictable
prime wraith
> Unlike on x86, there is no pressing need on arm64 to avoid indirect
> calls at...
What are these two points even trying to say
Especially the first one
carmine token
if (x2apic) x2apic_write
yeah this is better than an indirect call
sinful river
especially on hot paths
I noticed this when I was benchmarking a triangle rasterization function
carmine token
yeah
prime wraith
yeah this is better than an indirect call
How come, cpus can predict indirect calls also lol
sinful river
the cpu is better at predicting a branch even if that branch is in a hot loop than doing an indirect call
prime wraith
Cpus have branch target buffers for predicting this stuff
carmine token
they are fairly small and even then there are other costs
prime wraith
Anyway u have to profile to know for sure
carmine token
yes
doing any optimization w/o profiling is a bad idea
but it's true that check + branch is usually better than indirect branch
prime wraith
no runtime tricks and is perfectly predictable
Bro called vtables a runtime trick
sinful river
patching code is a runtime trick
prime wraith
Don't think there's a single kernel on here that supports code patching
sinful river
we were talking about linux
prime wraith
It wouldn't work on Linux since it would have many more branches than just x2 and x apics
Due to pv support etc
Also conditionally enabled with kernel configs
ornate jasper
static call trampoline code patching is actually something im considering trying to implement specifically for the apic/x2apic split even though thats basically my only use case
sinful river
Also conditionally enabled with kernel configs
ultra kconfig when
prime wraith
ultra kconfig when
Really want one
prime wraith
Yeah
prime wraith
static call trampoline code patching is actually something im considering trying...
Im thinking alternatives for now, they're super useful
prime wraith
So alternatives?
carmine token
Not sure how alternatives work
sinful river
for the apic use case this is really overkill
prime wraith
Not sure how alternatives work
Patch code based on cpuid
carmine token
But I'd just let the prekernel patch either isn sequences or call targets
prime wraith
Although alternatives support runtime patching
carmine token
runtime patching seems like a bad idea
sinful river
actually we should come up with a saner thing
or @ornate jasper we could make one in zig that parses a zon file, but i kinda dont wanna write all the code to handle terminals portably
prime wraith
Modules are the primary use case
carmine token
well, module load time != runtime, right?
prime wraith
or <@121733282676211714> we could make one in zig that parses a zon file, but i ...
Im not sure we need a saner thing, kconfig is already pretty sane
ornate jasper
Im thinking alternatives for now, they're super useful
x2apic for me is a compilation feature flag in addition to the requirement for hardware support so the loader cant know whether to enable x2apic and thus the selection of which impl is fully runtime
sinful river
Im not sure we need a saner thing, kconfig is already pretty sane
yes but I wanna NIH
prime wraith
well, module load time != runtime, right?
Ig
sinful river
I think KConfig is also like insane to parse
prime wraith
x2apic for me is a compilation feature flag in addition to the requirement for h...
Why not just vtables for this?
sinful river
there is python kconfiglib
prime wraith
Yeah that
ornate jasper
Why not just vtables for this?
because its two functions so runtime if seems better. idk how a runtime if to switch two fixed function calls compares to vtables perf wise tho
carmine token
x2apic for me is a compilation feature flag in addition to the requirement for h...
the loader can just check the kernel for supported features
sinful river
what I do is I try running kconfig-nconf first, or else I fallback to a python script using kconfiglib
carmine token
that's what Managarm does
sinful river
or just pass the kernel config to the loader
ornate jasper
the loader can just check the kernel for supported features
yeah but then i need to come up with a way to communicate that lol
prime wraith
because its two functions so runtime if seems better. idk how a runtime if to sw...
Its not just two, the way you send ipis differes a lot, also EOI and other stuff
carmine token
we use ELF notes
sinful river
or just pass the kernel config to the loader
ah no you cant without recompiling
prime wraith
Yeah elf notes seems perfect for config detection
ornate jasper
if i figure out how to actually write elf notes ill def do something like that for my feature tags
sinful river
you should add kconfig too
ornate jasper
because i hate external deps
ornate jasper
yeah that part is fine its the utility to manage them that i dont want to bring in, and then id need to write my own parser for them in zig too, and that just increasingly sounds like more work than i want to do
ornate jasper
ok then that wouldnt be a problem
prime wraith
And kconfig itself is configured via a frontend like python or anything, there's tons of them
ornate jasper
yeah but that frontend is an external dep which rubs me the wrong way lol
esp if its python
i hate python more than anything else in software lmao
prime wraith
Yeah the frontend is external, but it can be of users choice
ornate jasper
idk
ornate jasper
i also prefer to keep build option selection tightly integrated with the build process itself
prime wraith
Oh well
I think later down the line you inevitably have to reinvent something similar
carmine token
just enable everything unconditionally 
ornate jasper
yeah if i need more config options in future ill consider it
carmine token
and use load time patching to remove the stuff that you don't need
prime wraith
just enable everything unconditionally <:halfmemeleft:715831625233203250>
Thats the way of serenity
The 30 meg kernel
carmine token
that sounds more like a serenity skill issue
prime wraith
A lot of it is c++ template bloat also
wide ether
The 30 meg kernel
jesus what
prime wraith
Yeah lol
carmine token
Managarm is also C++ but nowhere near 30M
prime wraith
Well its a microkernel lol
ornate jasper
i actually dont know how big my kernel is
maybe ill check that today
i know its more than a 16-bit number of sectors because thats why i swapped off of bootelf to limine (before eventually from limine to my own bootloader)
or maybe it was 16 bit bytes i dont remember
sinful river
wtf even with releasesmall mine is 1.2mb
ornate jasper
esp if you use zig's formatting that aint bad
sinful river
1.2M hello world
tf you mean hello world it has cool fireworks
ornate jasper
the one downside to zig string formatting is it template explodes
sinful river
esp if you use zig's formatting that aint bad
formatting also uses a lot of stack
I think
carmine token
the NVIDIA driver just takes 90M 💀
ornate jasper
formatting also uses a lot of stack
for storage yeah but i meant for the template explosion from different format strings
carmine token
but this is with debugging symbols
sinful river
idk but like I was running formatting in some functions for debugging and the stack would overflow
ornate jasper
i had that more with debug symbols than formatting
carmine token
it's 16M without and that's the largest driver
sinful river
actually maybe it was unrelated and it didnt overflow
it was probably because of my pmm sharing pages bug
but it appeared less often when making stacks bigger
ornate jasper
if its in debugging code and youre using zig's dwarf unwind that may be part of it
sinful river
I don't
ornate jasper
that stuff uses a ton of stack and ive not been able to pin down why
ornate jasper
then idk
sinful river
it was probably because of my pmm sharing pages bug
this
ornate jasper
mhm
haughty notch
wtf even with releasesmall mine is 1.2mb
Mine is 2.3M with debug stuff and unwind tables (which I need to get rid off tbh, but my kernel init is hardcoded to my linker's layout atm)
prime wraith
tf you mean hello world it has cool fireworks
Anything without KDE is a hello world 
haughty notch
did managarm get kde in the end?
latent geode
Well its a microkernel lol
mine is like 1 mib :^)
doesn't say much as it barely does anything though
(most of it is .text iirc)
wide ether
latent geode
<:thinkong:815589724424241152>
needs gzip bro
wide ether
trust
ornate jasper
(releasesafe with full debug info)
lets try a stripped releasesmall build
stripped releasesmall (does still have eh frame just not full dwarf)
sinful river
stripped releasesmall (does still have eh frame just not full dwarf)
is that 3mb?
obtuse sparrow
A little bit more than that
sinful river
that's pretty big
ornate jasper
yeah im a bit surprised
wonder how much of that is log template explosion
but also cant be assed to actually check
carmine token
did managarm get kde in the end?
we are very, very close
stable orchid
kde in managarm before gta 6 is real
carmine token
there's probably no major stuff left, just bug fixing
we get to the point where plasma/kwin start and run but for some reason output doesn't work yet
prime wraith
Clanker suggested some solutions for uploading .config into cmake, lets see if it works
macro(load_kconfig config_file)
if (NOT EXISTS "${config_file}")
message(FATAL_ERROR "Invalid config file path: ${config_file}")
endif()
set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS "${config_file}")
file(STRINGS "${config_file}" config_lines)
foreach(line IN LISTS config_lines)
# 1. Match 'n' (Unset Booleans/Tristates)
if(line MATCHES "^# (CONFIG_[A-Za-z0-9_]+) is not set")
set(${CMAKE_MATCH_1} OFF)
# 2. Match 'y' (Boolean/Tristate True)
elseif(line MATCHES "^(CONFIG_[A-Za-z0-9_]+)=y$")
set(${CMAKE_MATCH_1} ON)
# 3. Match 'm' (Tristate Module)
# CMake doesn't have a native tristate, so we use the string "m"
elseif(line MATCHES "^(CONFIG_[A-Za-z0-9_]+)=m$")
set(${CMAKE_MATCH_1} "m")
# 4. Match Strings
# We strip the quotes for CMake, but set a hidden _IS_STRING
# variable so we know to put the quotes back for the C compiler!
elseif(line MATCHES "^(CONFIG_[A-Za-z0-9_]+)=\"(.*)\"$")
set(${CMAKE_MATCH_1} "${CMAKE_MATCH_2}")
set(${CMAKE_MATCH_1}_IS_STRING TRUE)
# 5. Match Numbers / Hex / Anything else
elseif(line MATCHES "^(CONFIG_[A-Za-z0-9_]+)=(.*)$")
set(${CMAKE_MATCH_1} "${CMAKE_MATCH_2}")
endif()
endforeach()
endmacro()
unironically works great
and sanitization is handled transparently by olddefconfig
thats insanely convenient
clanker is truly great at scripting
and this is stuff i hate working on so its a W for me
latent geode
straight to openslopware :^)
prime wraith
lol
nah the actual kernel will be almost 0 ai
clanker is not yet at the level where i would let it touch any c code thats not like something that has been solved a billion times before
latent geode
some parts of my build system is also llm generated/translated
prime wraith
like write a strstr impl etc
prime wraith
some parts of my build system is also llm generated/translated
yeah fair
most of mine is handwritten also
since it predates llms
being useful
lol
neat ig
haughty notch
straight to openslopware :^)
Ultra bootloader is in there as an alternative to limine
latent geode
mint added it
prime wraith
why do u need a bash script?
prime wraith
oh yeah i have my python wrapper for initial configuration like that Ill just add a warning there probably
sinful river
or do mean u like tui more
I like kconfig-nconf more
than the python -m menuconfig one
prime wraith
i get that one if i just do menuconfig
$ which menuconfig
/home/infy/.local/bin/menuconfig
all these tools are from kconfiglib i guess
but olddefconfig is truly the goat
sinful river
what is that
prime wraith
ah
prime wraith
what is that
infy@DESKTOP-IGT40G0:/mnt/d/Ultra$ cat .config
CONFIG_MODULES=m
CONFIG_ARCH_X86=y
# CONFIG_ARCH_ARM is not set
# CONFIG_ARCH_RISCV is not set
CONFIG_STUFF=m
# CONFIG_X86_STUF is not set
infy@DESKTOP-IGT40G0:/mnt/d/Ultra$ olddefconfig
.config:1: warning: 'm' is not a valid value for the bool symbol MODULES (defined at Kconfig:3). Assignment ignored.
Loaded configuration '.config'
Configuration saved to '.config'
infy@DESKTOP-IGT40G0:/mnt/d/Ultra$ cat .config
CONFIG_MODULES=y
CONFIG_ARCH_X86=y
# CONFIG_ARCH_ARM is not set
# CONFIG_ARCH_RISCV is not set
CONFIG_STUFF=m
# CONFIG_X86_STUF is not set
it also converts =n into is not set, so its always the same to parse
basically this thing is amazing
prime wraith
yeah so u cant insert bullshit values or incorrect values or mess up dependencies etc
if u manually edit it
sinful river
it's a weird name tho lol
prime wraith
it is indeed
ornate jasper
do you have static_call in ultra btw?
wide ether
do you have static_call in ultra btw?
what's that
obtuse sparrow
what's that
Rewrites a function pointer call into a normal call or something like that iirc
wide ether
wat
obtuse sparrow
To remove overhead
wide ether
you mean inlining?
wide ether
that sounds awful
prime wraith
do you have static_call in ultra btw?
no lol
if im doing something like that ill do alternatives first
prime wraith
that sounds awful
it rewrites a call address into another address
it does that for hot paths that would otherwise have to be turned into a retpoline
which is slow af
its a vulnerability workaround
on those cpus u cant use function pointers at all
(without it being a vulnerability)
prime wraith
In CMake, if you pass multiple separate strings to set(), it creates a list.
When CMake converts a list back into a single string (like when you write it to a file), it automatically joins the list elements with semicolons (;).
Look closely at how you defined the variable:
set(
CONFIG_HDR
"/*\n" # Element 1
"* Auto-generated kernel configuration header\n" # Element 2
"* Do not edit manually!\n" # Element 3
"*/\n\n" # Element 4
)
Because they are separate strings, CMake treats this as an array of 4 items. When it writes them out, it glues them together: Element 1 + ; + Element 2 + ; etc., resulting in those unwanted semicolons.
How to fix it
There are two ways to fix this.
Method 1: Use a single quoted string
Wrap the entire block of text in a single pair of double quotes, rather than four separate strings:
set(CONFIG_HDR
"/*\n* Auto-generated kernel configuration header\n* Do not edit manually!\n*/\n\n"
)
Method 2: Use CMake's "Bracket Syntax" (Recommended)
If you have CMake 3.0 or newer, there is a much cleaner way to write multi-line strings without needing \n or quotes. You use [=[ and ]=].
This tells CMake "Treat everything in here as one exact string, preserving all newlines."
set(CONFIG_HDR [=[
/*
* Auto-generated kernel configuration header
* Do not edit manually!
*/
]=])
This makes writing C-style templates inside CMake much easier to read, and it completely avoids the semicolon/list problem!
cmake is a fucking joke lol
but on the bright side kconfig integration works way better than expected
automatically notices changes + regenerates config.h + detects variable changes and lazily rebuilds only stuff thats actually needed
ornate jasper
cmake is great for like. most things. this is the one where it isn't lol
prime wraith
somehow they absolutely fucked up multiline strings
ornate jasper
yeah
ornate jasper
zig did em good yeah
for cmake big strings like that I always preferred template files where possible but idk offhand how well that integrates with variables
but also I've not touched cmake in like 8 years now
prime wraith
yeah, or ill just write a series of appends
ornate jasper
oof
prime wraith
yeah turns out you shouldnt let it glob recurse itself
sinful river
zig did em good yeah
would you be down to work on a kconfig thing in zig with me
using zon and ncurses
because parsing kconfig is actually ass
ornate jasper
maybe sometime yeah
sinful river
so it'd use its own syntax based on zon
ornate jasper
oh if own syntax im def in
(also could easily output zon too as an option, that can then be @imported by zig later)
is there good zig binds for ncurses?
sinful river
is there good zig binds for ncurses?
id just use the raw C honestly
I'm aware there are zig libraries for TUIs but I'm scared to depend on zig stuff
ornate jasper
mhm
sinful river
(also could easily output zon too as an option, that can then be `@import`ed by ...
exactly, it would output something like:
.{
.enable_logging = true,
.log_level = "info",
.output_dir = "./out",
.thread_count = 8,
.enable_tls = false,
}
which build.zig could import
ornate jasper
cool
sinful river
idk just a thought I had
ornate jasper
yeah no i do like it
sinful river
the only annoying part will be the TUI
ornate jasper
yeah for sure
sinful river
and probably managing dependencies
but that shouldnt be too hard (build a DAG)
zconfig it should be called
ornate jasper
mhm
sinful river
idk if I actually want to start work on this right now though
ornate jasper
yeah me neither
sinful river
but in the future that could be a really cool thing to work on
sinful river
wouldnt take too long to get an MVP and we can get community backing easily I'm sure
ornate jasper
my thing is rn my kernel doesnt need any non-bool options so my existing wacky config setup is fine
sinful river
in fact we could get an MVP tonight if I vibecode it but I won't do that
ornate jasper
so its a low priority for me
sinful river
my thing is rn my kernel doesnt need any non-bool options so my existing wacky c...
yeah but the idea would be to make generally available to the zig community
prime wraith
my thing is rn my kernel doesnt need any non-bool options so my existing wacky c...
one use case i found instantly is CONFIG_OPTIMIZATION_LEVEL
sinful river
not just for kernel stuff
sinful river
one use case i found instantly is CONFIG_OPTIMIZATION_LEVEL
yeah that too
prime wraith
range 0 3
ornate jasper
compiler optimization is already built into build.zig as a thing to pass as an option
ornate jasper
loading from a config could be nice ig yeah
ornate jasper
one use case i found instantly is CONFIG_OPTIMIZATION_LEVEL
rn im also using build cli flags for all my options anyway so the builtin option for build.zig for optimize level is fine
sinful river
the idea is more for larger projects with large configurations
ornate jasper
the idea is more for larger projects with large configurations
yeah no 100%, itd def be a useful thing for the future™
sinful river
- make this popular
- tigerbeetle adopts it (VC funded zig startup)
- we get sponsored
- profit
ornate jasper
just not a priority for me atm if noone else is also working on it
ornate jasper
mhm
sinful river
cant take more than like a week right
sinful river
I also thought about that
we can detect if ncurses is present during build
and if its not then it can be a console prompt
linux has that too
but it sucks obviously for large configs
ornate jasper
oh is ncurses a dynamic library? that kinda sucks
sinful river
?
I meant more like in the build script we check if we can find ncurses on the system
ornate jasper
oh
sinful river
and if we can't we fallback to a cli
ornate jasper
ah ok
sinful river
idk I'll try something and I'll ping you if I can come up with something good
ornate jasper
👍
sinful river
might not be tonight
prime wraith
are u gonna vibe code it
sinful river
i have homework
ornate jasper
yeah no worries
prime wraith
sounds like the perfect thing to vibe code
sinful river
are u gonna vibe code it
honestly kinda tempted to vibecode it
prime wraith
khitiara might not like that
ornate jasper
if its vibe coded then im not going to be involved or use it personally
sinful river
but anyway id need to modify most of it since AIs dont work well with zig
sinful river
if its vibe coded then im not going to be involved or use it personally
I'm not going to vibecode it but I might or might not use AI to avoid going through ncurses documentation, is that fine with you?
i dont really have time to invest learning how ncurses works
ornate jasper
eh? idk how i feel about that. would be more willing to consume that but idk if id want to be involved then
ornate jasper
i dont really have time to invest learning how ncurses works
this is part of why my instinct would be to use a zig library if theres a good one
or just like. maybe a better library than ncurses lmao
sinful river
ncurses isnt bad
prime wraith
ncurses cant be that hard
sinful river
but my problem like I said is mostly portability
ncurses works everywhere
and is present on most distros by default even
I don't wanna vibecode it because I would be ashamed to release something I vibecoded
I dont mind vibecoding useless stuff only I play with
prime wraith
tbh this json like stuff it can one shot
but yeah
depends how q&d u want it to be
sinful river
tbh this json like stuff it can one shot
zig has a parser built in
it's literally zig syntax
ornate jasper
zig has a builtin parser for zon yeah
ornate jasper
yeah using the zig parser
the zon stuff in std actually works as a (de)serialiser already
prime wraith
the only downside of using cmake stuff is its really damn slow, especially over wsl with its crappy p9 impl
ornate jasper
fwiw when i said console prompts i wasnt talking about cli args i meant readline-type stuff
ornate jasper
the only downside of using cmake stuff is its really damn slow, especially over ...
can always just use mingw or something, cmake does exist for windows
sinful river
the only downside of using cmake stuff is its really damn slow, especially over ...
meson
sinful river
fwiw when i said console prompts i wasnt talking about cli args i meant readline...
yes me too
prime wraith
i could actually build it on windows using clang --target i guess
sinful river
linux has that
prime wraith
but im too lazy
ornate jasper
yes me too
cool just wasnt sure if i was clear about it
sinful river
where it goes through every config option individually and asks you
prime wraith
yeah this definitely is not gonna brick my pc :clueless:
prime wraith
nah im not installing it
they definitely have a billion regressions there
it already has a problem waking up from suspend every now and again
i have to do a literal power cycle
ornate jasper
oof
prime wraith
suspend is an unsolved problem in computing
ornate jasper
yeah
prime wraith
and they made it even more fragile with s2idle
to appease microsoft wanting to do stupid background updates
anyway im loving how this ended up
ornate jasper
hell yeah
prime wraith
very happy with the end result and tbh i expected it to be way harder to integrate
ornate jasper
yeah config headers are super nice
ornate jasper
build.zig can generate headers too actually, theres a ConfigHeader step that can take a number of different input formats
prime wraith
i just want a bit more control i guess
since i have these variables in cmake anyway
prime wraith
build.zig can generate headers too actually, theres a ConfigHeader step that can...
cool
ornate jasper
since i have these variables in cmake anyway
you using the cmakedefine config header stuff?
prime wraith
Nah it just regex parses the .config to transform that into in local cmake variables and also at the same time appends them into the config.h header
ornate jasper
ah
(cmakedefine is a special syntax you can use to have cmake do config header processing using cmake variables)
ornate jasper
yeah
prime wraith
i know you wouldnt like my parsing logic since the regex was developed by clanker 😭
ornate jasper
F
sinful river
honestly i dont see why someone would be against AI for something like that
ornate jasper
yeah regex is hell on earth i can absolutely understand that case lmao
sinful river
didnt humans keep inventing stuff so we have to do less boring work
ornate jasper
i wouldnt do it but i can kinda understand using it for regex
prime wraith
regex is justifiable i guess
but everyone has their own line that they draw for that stuff
ornate jasper
my line is different for things ill silently judge, things i wont use, and things i wont do lmao
prime wraith
like i wouldnt feel bad if i had vibecoded the entire parser in zig, but i would feel bad if i actually vibecoded something important in the kernel
tools/buildsystems/tests is where my line mostly is
ornate jasper
my lines are that i wont use them for anything myself, but my i wont use something someone else made using it line is about where yours is for your own use
ornate jasper
yeah ive not been very good at words today lmao
sinful river
kinda funny we're three people all not talking in their native language
I think
unless khitiara you're english native idk
prime wraith
isnt khitiara from the us
sinful river
idk
sinful river
americans
ornate jasper
(i know german grammar almost completely but my ability to remember vocabulary is approximately null so i dont really speak it)
ornate jasper
yeah lmao
eternal sparrow
My English isn’t even good
inland wave
I'll have you know I know one and know a little bit of another 
prime wraith
I like kconfig-nconf more
that one doesnt even know what rsource is
prime wraith
this is so much more cleaner than random cmake variables scattered everywhere
haughty notch
I just have a bunch of makefile targets...
prime wraith
L
carmine token
Why does the build system not auto detect gcc vs clang?
haughty notch
How do you compile your kernel in the first place?
prime wraith
olddefconfig && mkdir build && cd build && cmake -DCONFIG_FILE=../.config && cmake --build .
but i have a python script for covenience that does image generation etc
if u select clang it just uses the host clang with --target
for gcc it builds one from source
haughty notch
I just have a bunch of makefile targets...
And jinx kinda takes care of architectures in the first place
prime wraith
fair
i dont use any of that stuff
I did have to add this check into cmake because it really doesnt like changing the toolchain after configure
but i usually just let my build.py create separate directories so i can test different toolchains in parallel
haughty notch
My makefiles do that
Also, I'm using my patched toolchain for both userspace and kernel, so I "can't" use host's toolchain anyway
sinful river
UX final boss
Cool but not NIHed so less cool
prime wraith
proper tulip
Cool but not NIHed so less cool
you're using zig build, that's not NIH either
haughty notch
at least the bootloader is NIHed
proper tulip
so is Hyper
sinful river
you're using zig build, that's not NIH either <:trl:829316641191886888>
But the kconfig is NIHed
ornate jasper
what does NIH even mean?
latent geode
what does NIH even mean?
not invented here I think
ornate jasper
ah ok
latent geode
skrill tissue
ornate jasper
lol
eternal sparrow
skrill tissue
prime wraith
Finished improving my python script for Kconfig support
It's so convenient now
./build.py -> makes a build directory that's build-{toolchain}-{arch}, makes a default config inside the build directory, builds it
./build.py --arch aarch64 -> same thing but replaces defaults with CONFIG_AARCH64
./build.py --config .config -> creates a build-user-config, builds using it
./build.py {--menuconfig,--guiconfig} -> convenience helpers for editing the autogenerated config, also works with --config
not to mention it already had --make-image/--run/--debug/--ide-debug etc
wait wtf did i do
ah nvm
also my aarch64 bitrotted 😭
prime wraith
fuck, i had the vscode setting that auto trims whitespace off
i have to go through a shitload of commits and check all code now
prime wraith
try:
from kconfiglib import Kconfig
except ImportError:
print("Could not import kconfiglib, trying to install in-place...")
subprocess.check_call(
[sys.executable, "-m", "pip", "install", "--user", "kconfiglib"]
)
user_site_packages = site.getusersitepackages()
if user_site_packages not in sys.path:
sys.path.append(user_site_packages)
importlib.invalidate_caches()
from kconfiglib import Kconfig
Least cursed build script
it might be cursed but i like when my build stuff does dependency shit itself
prime wraith
vendor how?
edgy remnant
like just put the kconfiglib there
edgy remnant
it‘s a single file, that‘s just what I mean
prime wraith
its definitetly not a single file
prime wraith
i need the menus also
otherwise how tf am i supposed to edit it lol
and if i have to install menus separately why would I vendor a 7k loc file
edgy remnant
well that‘s true then
prime wraith
whats interesting is the pip package maintenance was overtaken by zephyr
because the guy that originally made the repo abandoned it
this is the current repo
thanks to zephyr otherwise idk what i would do
edgy remnant
I took the kconfig stuff from zephyr too lol
prime wraith
whats your os
edgy remnant
I‘m redoing a bunch of things and just started reading some os books again so not much to show currently
edgy remnant
I had it setup on an old kernel and copied the build system for my new one
prime wraith
do u use make?
edgy remnant
cmake
edgy remnant
iirc it even was a bit of zephyrs cmake, i found the cmake integration on a github repository
prime wraith
oh ok
i had to do a shit ton of work to make sure it picked up changes to .config and all Kconfig files + sanitized it so that all options are set correctly etc
edgy remnant
do you have some sort of fixdeps.c equivalent
prime wraith
what is that?
edgy remnant
as in the linux tool
prime wraith
yeah but what does it do
edgy remnant
yes
edgy remnant
I found it interesting
edgy remnant
did u have one?
No, but I saw some talk about kconfig here and wanted to look into something like that
prime wraith
ah ok
i wanted to optimize it by just using cmake definitions for most common options
edgy remnant
though IDK how feasible it is with cmake
prime wraith
nah it has a diff purpose
some options are basically never changed but needed by a lot of code
e.g. the current arch, or the current arch bitness
or endianness
prime wraith
but yeah let me know if u figure out how to do that fixdep thing
prime wraith
iirc I just set arch as a cmake option and add a C define
i did that but it just kinda nicely fell into place with kconfig so i left it there
this allows me to include the respective kconfig of the arch subdir etc
pretty covenient
edgy remnant
true
prime wraith
fixdep would probably require digging into cmake internals
edgy remnant
but yeah let me know if u figure out how to do that fixdep thing
I'll have to look into cmake first, like if you can change and remove a dependency for certain files
prime wraith
to figure out how it does that depepndency stuff
edgy remnant
yup
edgy remnant
kconfiglib includes a function similar to the linux kconfig
something something sync directory
prime wraith
what does that do?
edgy remnant
like this functionality
* kconfig populates a tree in include/config/ with an empty file
* for each config symbol and when the configuration is updated
* the files representing changed config options are touched
* which then let make pick up the changes and the files that use
* the config symbols are rebuilt.
so i think the major challenge is getting the cmake side working
prime wraith
interesting
yeah that definitely sounds doable
so do u vendor kconfiglib or do u expect the user to install it
edgy remnant
vendor
prime wraith
damn
edgy remnant
i pasted that whole library in lol
prime wraith
why not?
edgy remnant
I'd want to ensure it doesn't break or depend on the network further, though I mean it's possibly cleaner
prime wraith
hmm, but don't u have other network dependency like the compiler sources etc?
edgy remnant
that's not really part of my kernel build process
sinful river
You can always require deps but still have your build fully offline
edgy remnant
sure
sinful river
Installing network stuff in the build script is a bit stupid tho I agree
edgy remnant
i just did it like zephyr :p
edgy remnant
well
sinful river
just copy paste a 30k loc library <:trl:829316641191886888>
No write your own
prime wraith
i just did it like zephyr :p
does it submodule or copy paste?
sinful river
prime wraith
interesting
edgy remnant
kconfiglib doesn't really change much anways
sinful river
yes but mine is like 2.5k
edgy remnant
kconfiglib is like so much comment
sinful river
for now
edgy remnant
kconfiglib is like so much comment
nvm that lol
prime wraith
nvm that lol
unironically it's very difficult to parse since options can recursively depend on other options for their values and ranges etc
you probably have to make a multi-pass option resolver
and run it in a loop as long as you make progress
prime wraith
kconfig syntax is whatever linux parser accepts
sinful river
zonfig with a shitton of { ftw
prime wraith
let me know when it supports recursive options
sinful river
let me know when it supports recursive options <:trl:829316641191886888>
What is that
prime wraith
What is that
config FOO_4
bool
config FOO_8
bool
config FOO
int
default 4 if FOO_4
default 8 if FOO_8
config BAR
int
range 0 FOO
config X
def_bool y
select FOO_4
U can do stuff like that
BAR depends on FOO for its value, which is selected only later by X (which may depend on other options for its values)
sinful river
idk if thats useful
prime wraith
idk if thats useful
heres an example
literally all properties of an option are dynamic
edgy remnant
you can't remove a dependency via cmake (as far as I found out)... however I can remove and could also inject dependencies with https://cmake.org/cmake/help/latest/prop_tgt/LANG_COMPILER_LAUNCHER.html#prop_tgt:<LANG>_COMPILER_LAUNCHER from a small python script
I'll have to look into how linux exactly modifies the dep files
also this would only work for makefiles/ninja
prime wraith
you can't remove a dependency via cmake (as far as I found out)... however I can...
this would allow to wrap the compiler in a script that changes options and stuff right?
edgy remnant
yes
I atleast removed the generated config header dep already
now would just need to scan the file for CONFIG_* stuff, I think that's how linux does it
prime wraith
damn
prime wraith
now would just need to scan the file for CONFIG_* stuff, I think that's how linu...
yeah i think it scans not only the file but also all of the headers that it includes
edgy remnant
yeah that makes sense of course
prime wraith
and the headers it gets from the dep file that the compiler produced
edgy remnant
wait right, the compiler generates the file list for you
prime wraith
yeah, so this whole thing only starts working after build 1
edgy remnant
before build 1 this wouldn't even make sense
as the config.h would've just been generated the first time
prime wraith
ye
prime wraith
@edgy remnant i decided to follow your advice but with a submodule instead of direct copy paste
i think thats cleaner
prime wraith
File "/mnt/d/Ultra/scripts/kconfiglib/guiconfig.py", line 263, in _detect_system_dark_mode
if sys.platform == "win32":
^^^
NameError: name 'sys' is not defined. Did you forget to import 'sys'?
wait do they even test their code
they use sys without importing it
edgy remnant
I used the sync_deps from kconfiglib as a template as ninja NEEDS the files to be there:
https://github.com/lretq/dep-file-cmake/blob/main/scripts/kconfig/kconfig.py#L293-L330
and then I just scan all files from the dep file and modify:
https://github.com/lretq/dep-file-cmake/blob/main/fixdep.py
I just added a new argument to the zephyr kconfig.py and call my sync_deps from there:
https://github.com/lretq/dep-file-cmake/blob/main/cmake/kconfig.cmake#L203
and I reuse that cmake var and pass it to the fixdep.py script as first arg
prime wraith
wait so how do u intercept/overwrite cmake dep files?
edgy remnant
via fixdep.py
set_target_properties(program PROPERTIES C_COMPILER_LAUNCHER
"python3;${CMAKE_SOURCE_DIR}/fixdep.py;${KCONFIG_SYNC_DIR}")
and this one
It works only for make and ninja as this LAUNCHER thing is only supported for those two, but I only use ninja anyways
prime wraith
interesting
edgy remnant
the script removes the config.h from the dep list and adds for every CONFIG_* symbol the stamp file to the dep list
it parses the call the CC and if it finds an argument -MF it takes the next arg as <dep file>, if it doesn't find anything I just do nothing
prime wraith
makes sense
for whatever reason if i call menuconfig not via subprocess but via menuconfig.menuconfig() it starts in some half corrupted state
if args.menuconfig or args.guiconfig:
os.environ["KCONFIG_CONFIG"] = args.config
module = mc if args.menuconfig else gc
kconfig = root_kconfig()
with enter_work_dir(pg.project_root()):
module.menuconfig(kconfig)
sys.exit(0)
like what does it want from me
it parses the options but they're not modifyable
edgy remnant
lol what
prime wraith
yeah lmao
ill just keep using subprocess lmao
and ill have to send a pr to fix their crappy guiconfig
edgy remnant
which repo do u use
prime wraith
u use that repo as well right?
edgy remnant
https://github.com/ulfalizer/kconfiglib this one?
edgy remnant
ah
prime wraith
u use the old one?
ah nvm someone already made that fix but they arent reviewing it https://github.com/zephyrproject-rtos/Kconfiglib/pull/41
edgy remnant
i use what zephyr uses lmao
prime wraith
why do they have two copies of the same repo??
edgy remnant
zephyr doesnt seem to check the changes they do, the old repo does it correctly?
edgy remnant
why do they have two copies of the same repo??
what huh
prime wraith
zephyr doesnt seem to check the changes they do, the old repo does it correctly?
this repo is part of the zephyr project
but for some reason the actual kernel has some copy pasted code
instead of submoduling this one
edgy remnant
yes
weird setup idk
anyways I checked and my guiconfig still is the version that supports python2
i have the version from 2019
prime wraith
they do pull in commits tho https://github.com/zephyrproject-rtos/zephyr/commit/745e0b28bd85668d66a5bedc2839de797adfca04
but they didnt pull in the commit that dropped python2 support
prime wraith
lol what
btw that was a python funny
menuconfig.py does from Kconfiglib import ..., that was importing Kconfiglib from pip instead of kconfiglib.py, which was causing isinstance checks to fail
from .Kconfiglib fixed it lol
edgy remnant
ah wow
prime wraith
this seems to be the more general fix:
import scripts.kconfiglib.kconfiglib as kc
# Make sure the rest of "import kconfiglib" always references this
sys.modules["kconfiglib"] = kc
also something I noticed with my dep file solution is that a missing kconfig symbol makes it so that the file is recompiled every time lol
was wondering why it didn't work but I never added that option to Kconfig
edgy remnant
well I used it from actual code
I guess I could warn when I encounter non-existant kconfig variables or something like that
edgy remnant
not if I do something like
#if CONFIG_OPTION
#endif
prime wraith
right right
its definitely possible to check that it exists yeah
although from python it might be slow af
edgy remnant
ig this could be helpful to catch typos
prime wraith
to run for every source file
edgy remnant
although from python it might be slow af
it's a single os.exists
as I create a stamp file for every option anyway
prime wraith
is it a flat hirerachy?
edgy remnant
it turns _ into /
prime wraith
wdym
edgy remnant
iirc the reasoning from linux was that some filesystems dont like thousands of small files
prime wraith
like whats the conversion path from CONFIG_FOO to its location on the fs?
edgy remnant
yes
prime wraith
linux has flat include/config these days
edgy remnant
well I based myself on the 2019 kconfiglib code idk
though it's not that big of a difference
as in I can easily just change the paths
prime wraith
yeah it seems zephyr's one still does that
edgy remnant
as in I can easily just change the paths
i'll just do that rq, this seems easier to understand
prime wraith
seems like only this function would have to be patched https://github.com/zephyrproject-rtos/Kconfiglib/blob/b8535af53f0650a69e52db4bf8585c09d55852e7/kconfiglib.py#L6479
edgy remnant
yes
I extracted the sync_deps stuff and did it seperatly from the kconfiglib.py so that integrating is easier and doesn't require a patched lib
prime wraith
ill probably just fork it
and fix the sys import + patch to allow configuring deps layout
prime wraith
I extracted the sync_deps stuff and did it seperatly from the kconfiglib.py so t...
where did u extract it to? Don't u have the full code locally anyway
u could just import the function from kconfiglib
edgy remnant
where did u extract it to? Don't u have the full code locally anyway
yes, just the script that calls into kconfig (the one called from cmake)
prime wraith
like u do from kconfliglib import sync_deps right?
edgy remnant
I might want to be able to update kconfiglib or smth idk, seemed better to me
edgy remnant
like u do `from kconfliglib import sync_deps` right?
no I added a check to sync_deps
cmake+ninja requires that the file actually exists
linux did some sort of makefile wildcard thing that would not require a rebuild if the dep stamp file is not there
as would be the case if the option is =n
prime wraith
so u copied basically the entire sync_deps code into the script that patches cmake deps?
edgy remnant
yes