Jj

Sorry, hit enter by mistake

There is talk about git being replaced by jj, which uses the Apache license (BSD friendly)

Apparently it's git-compatible or it's own alternative...i think. Gits internals really are garbage.

TIL jj - any relation to got?

farhan: Who's planning this?

Hm, Rust.

Planning? I don't follow?

"< farhan> There is talk about git being replaced by jj,"

Who's talking about it?

Not sure I want to encourage the Rust community.

Oh...umm..the most credible source: YouTube!!

I'm on team rust, but know the community is largely against it.

I'm also on team "yes, chase shiny new things"

I remain curious about gameoftrees.org

usually reinventing the wheel produces more money, specially when you must promote the use of whatever language...

lets do GIT again! we have some funds to expend somehow!@#!!

i've never heard of "jj" but there is definitely talk of git being replaced by got

but aiui, got is binary compatible with git, so this doesn't actually matter

ivy: by who lmao? theo? XD

badkat: by freebsd committers

:O

Seems reasonable to me.

i ran into some logistics issues when i was trying to write a script to clone a repo with got

it really doesn't like the thought of keeping the repo contents as a subdir of a worktree

I don't think that's necessarily a deal-breaker, you could keep the .git dir elsewhere and provide some tooling to checkout a .git tree at /usr/src to bridge the gap, but it's still a bit ugly

but the worktree layout's not necessarily compatible

so if you checkout /usr/src with got, there's no quick and dirty way to conver that to a git worktree; you have to blow it away and setup a new worktree from the repo

(as far as I've found)

I'm having trouble finding where in net80211 a ieee80211_rateset struct is set. Anyone have any insights?

Looking at you kevans (:

farhan: looks like it's setup via ieee80211_setmode -> ieee80211_setbasicrates, typically

ahh, setbasicrates in ieee80211_proto.c

yeah, there-

just remember.. you can just put some yacht rock on.. and try to use the telescope (neovim plugin) to search and then almost want to throw your monitor across the room.. when you can not find something

kevans: Hrm, I thought they were supposed to be interchangeable by design. TIL.

mason: you can at least reuse the same repository, so it's not a total loss if you try to switch between them

you just can't seem to setup a layout wherein both git and got will both Just Work(TM) without a bit of fun

That's the kind of thing I'd want to do. :P I'm glad you've experimented with it.

I posted to freebsd-git about it just a bit ago. I don't know if this is really a problem, as long as we can provide something to do the conversion to a git repo

otherwise we'd probably look at a script that can do your standard checkout and update along a branch, just enough to, e.g., get a copy of the ports tree and move it forward in case the version you pulled was tragically broken in some fashion

just an idea, not really a solid plan that anyone's agreed to

got is that the openbsd "game"

i see it popup a few times on discoverbsd.com and X.com

This uncovers a bit of ignorance on my part. If you git push, does that invoke a git (or got?) binary on the other end? I wonder if migration could be made to happen via pushes or clones.

voy4g3r2: gameoftrees, yes

mason: yes, a push would typically invoke a git on the other side to receive the push

but afaik that wouldn't strictly be necessary, the concern is we don't necessarily want to expose got

kk

ah, thanks kevans .. i am over here just dealing with git branches on small projects..

and as i learn more about git.. i wish client would STOP using bitbucket.. blargh

i find bitbucket to be less bad than some alternatives, at least

this is true.. the workflow that is "standard" makes managing releases and documentation horrible

i am also the person receiving the output.. i asked for access to it and then people started asking me.. to do code reviews..

i quicly asked to have my access revoked.. i just wanted to see the "signature" and the release details and someone was on vacation..

"standard" == customized and configured to some odd committee based metric.. that does not make sense..

i had one developer.. using a local version of git.. doing all the branches/merges/management of code.. then just throwing up a branch to merge in bitbucket..

and we were like.. why do you have so many commits.. but we do not see the work in bitbucket.. "oh well i use my own copy of git, on my machine.. because i do not like bitbucket seeing my work..."

oh well, nothing to do with this conversation.. just a baabble

g'evening

what would be the best way to move several ZFS datasets from one machine to another ?

is zfs snapshot -r zroot_jupiler/data/jails@migrate ; zfs send -Rv zroot_jupiler/data/jails@migrate | ssh root@myhost zfs recv -eu zroot_stout/data/jails enough ?

mage: it's been some time since i used zfs commands, so i can't confirm the flags but the structure of the command (send -> ssh pipe -> recv) looks right to me

that should be enough indeed

if the dataset is big, you may want to pipe it to a file instead on the source machine, rsync it to the destination machine and then pipe it to zfs recv

^ so that rsync could resume the transfer if your connection broke

hi, when installing freebsd 14.2 into qemu guest vm, does it make any sense to use ZFS (which ends up in qemu raw image) or also does it not have any disadvantage vs the UFS system? thanks.

clemens3: i suggest using zfs so you have the benefit of snapshots and boot environments, which makes upgrades easier and less risky. however if you have low memory (< 4GB) you may find UFS works better

clemens3: it's worth learning ZFS, even if it's only a VM to play wiht FreeBSD

(i run ZFS on systems with 2GB RAM and it's fine, but this is definitely not a super recommended configuration)

good, thanks all!

ZFS with low RAM isn't a problem unless you're using de-dup. ZFS will be fine on low-memory systems (for the most part.)

So, just don't use de-duplication. :)

ok, always wondered what they did back in the sparc days with ZFS when 4GB memory where gigantic

if it was ok then, should be ok now

they told people not to use ZFS on small systems :-)

4GB was basically the minimum you'd expect in a server or workstation back then

it didn't work on i386 without proper tuning

(iirc, it was also recommended not to use it on 32-bit systems, but i think that's been improved a bit... not that you should really be using 32-bit systems anymore anymore)

AFIK it doesn't work now on arm (32-bit), but it used to work in the past (2-3 years ago)

freebsd still doesn't have ARC integration with the memory manager

no, but there were some significant improvements to vm reclamation of memory from arc a few months back so that should be less of a problem than it used to be

i still set arc_max in a few places

yes, it's first step when you suffer from memory starvation

and make sure your filesystem doesn't get *too* full, might start losing files...

no worries

so, why is it that the x86 definition of struct trapframe has amd64 registers?

Back in the day, it was designed for Sparc64 systems. It was kinda hacky in the beginning to use it on i384/x86_64 (the latter being more supported, obviously)

putting it all in the same file and then surrounding it with #ifdef __amd64__ defeats the purpose of the machine-specific include directories

ring0_starr: Because hacky?

are recent improvements in 802.11 allowing to utilise 802.11ac now ?

mzar: I believe so? I tested out some changes a while back and I could've sworn *ac was working just fine.

thanks for feedback ek

mzar: Sure thing. Actually, now you have me curious and second guessing. I'm gonna re-install on one of these laptops and check it out. :)

Cure some of my boredom.

let me wish you a fun and fruitful installation then

om'

Haha. Thank you!

the code is subtly wrong for cpu_fetch_syscall_args on x86 and amd64, at least

argp starts at &frame->tf_rdi and then iterates through the struct as though it were an array

the assumption being that the next member of the struct starts at sizeof(syscallarg_t) bytes after, which is perfectly valid for subsequent array elements, but not for struct members due to potential alignment issues

maybe in a parallel universe this code is fine because the rules around struct member alignment have been relaxed by having stronger guarantees in the standard

but here, it is not, because the C standard allows for too many gotchas in using code like this

ring0_starr: I guess we could add some static assertions about the alignment to make you feel better, but generally we won't be doing anything to increase the alignment of the frame

nah it's fine - if there was an architecture encountered where the assumptions made here lead to what is technically UB, you'd notice it immediately by syscalls not working

sure, but otoh it's nice to be able to document these kinds of assumptions in a way that we can detect hijinkery at compile-time instead of runtime

nicer for the human debugger-to-be

you'd need to depend upon _Alignof() which I think is C23

mm no nevermind that's C11, and then changed to alignof() as of C23

hernan604: that command only fetches updates based on the currently installed world but doesn't tell me which upgrades (new releases) are available.

we use _Alignof in some places, which may be shimmed out to __alignof or alignof

our current CSTD in the kernel is gnu17

why is it following a gnu extension when you switched to clang like 12 years ago now?

because gnu added some niceties that were apparently appealing enough for clang to adopt. there is no clang17 that I'm aware of, just gnu17 and c17

ring0_starr: ac wifi depends mostly on the driver implementation btw not in the OS stack itself

what??

badkat: I'm quite sure ring0_starr and kevans are talking about something else completely. :)

lol i wanted to quote mzar :P

but.. please improve wifi ac in freebsd driver development overlords :(!

i don't even have any 802.11ac hardware to test with

cgit.freebsd.org/src/tree/sys/arm64/arm64/trap.c#n166 ????? what is up with this, how do return values work in arm64

err nevermind, i mean why does td_retval get set to anything in this function in any architecture?

the retval is going to just be garbage and overwritten after it actually executes

ring0_starr: might be a good one for -hackers@ if nobody responds

I could speculate, but better to hear from someone with confidence

what is -hackers@?

it seems like void syscalls have their second retval set to the second arg. except for arm64 which just gets 0.

ring0_starr: -hackers@ is a freebsd-hackers@ mailing list

i thought so, just wanted to make sure since there is an irc network by that name too