a topological sort would be nice but I'm not really sure how best to do that with elasticsearch

satanist: i thought you could only expand raidz, not mirrors

damnit satrapes stay in th channel bit longer

Freaky: I guess there's no way to order them maintaining the parent relationship / ordering by `git rev-list --count --first-parent <commit>` ?

this'll likely come up a lot with third party contributions in particular

kevans: there's surely something that can be done that'll be reasonably efficient, I just need to set aside the time to try stuff

if I leave it long enough maybe someone will make a nice search engine with graph database capabilities ;)

Freaky: you seen Hillel Wayne's lasted newsletters?

buttondown.email/hillelwayne/archiv…asis-types-in-programming-languages (things missing) and buttondown.email/hillelwayne/archiv…k-on-a-big-language-id-like-to-talk

Title: Unusual basis types in programming languages • Buttondown

oh no, more trees

I can't wait for civilization to collapse so we have to go back to basic linked lists

meena: thanks, these were fun and not too overwhelming

oh, wait, this one has Raku, I take it back

Raku is weird

it looks like Perl from another dimension

it's like, what are we spending our language's weirdness budget on? everything

Raku was a push towards larrys goal but it got complicated :)

that wasn't the goal?

When Perl 7? :P

higher version number = better

With Rust and C++, yes, yes it is :P

and browsers

Getting ready to put FreeBSD in a virtual machine to get a feel for it

Most of you also daily drive FreeBSD for your Desktop type computing?

Yep

Cool, I plan to do it with KDE Plasma. And will be debating replacing my Linux Host OS with it (Fedora)

perl is excelent. every time i install git, it depends on a bunch of perl modules

same with many oses

but i can understand many people prefer not to understand what is perl

I can say the same with python

and not trying to start a language war either

pretty similar.. except perl is independent.. no BIG company promoting it.. python has google behind

let the language wars begin

I often wonder why Rust gets so much attention and Golang isn't even in the conversation

Because there is a whole lot of zealotry behind it along with an intense motivation to blindly rewrite everything in it

thats a good sign... independent for the win

big players dont need my blessing

i use firefox :)

so does Firefox on FreeBSD use pulseaudio? this guide I was looking at regarding bluetooth audio makes it seem like getting audio from pulse to the bluetooth audio virtual_oss device involves a janky hack that introduces noticeable latency

but I dunno if I really want to use chromium

It can, but it can also use FreeBSDs own audio stack, which can do things no Linux audio stack can do.

I'm not sure I understand why you need to introduce pulseaudio, when bluetooth is handled by netgraph and the audio device should be present in /dev/sndstat.

Oh wait, no I'm misremembering - you use virtual_oss to create a sink.

For what it's worth, virtual_oss is used by at least one musician (specifically, meka in this channel) for low-latency audio monitoring.

See the -i switch on virtual_oss(8).

debdrup: I'm talking about cases where an application pulls in pulseaudio, regardless of what I'd want to use. apparently pulseaudio won't play through a virtual_oss sink without a hack where you use some pulse command-line utility to record audio and redirect it

nowadays I wouldn't willingly touch pulseaudio proper since pipewire exists, though while I see ports has a port of pipewire, I think I read something about applications still pulling in actual pulse as a dependency

(hell, even on Linux, before switching to pipewire, I slogged through using bare ALSA due to my prior experiences with pulse)

The only use-case I've had for bluetooth audio is using it for WebRTC based conferencing, which works via oss(4) on FreeBSD so doesn't involve pulseaudio - so I've not had to install it.

I was under the impression that some applications, ones that don't natively support OSS, or take the sane route of using some kind of portable abstraction layer such as libao, SDL, OpenAL, could pull in pulseaudio as that "abstraction layer"

like they use pulseaudio as a roundabout way of getting sound through OSS in lieu of native support

what I read made it sound like firefox was one of those applications with a hard dependency on pulse, and that's what I remember from using firefox with ALSA, which forces (or at least forced) firefox to use this library called apulse that basically provides the bare minimum of the pulse API for sound to function

tm512: Firefox uses a library called cubeb which, technically, supports a large number of audio backends including OSS (iirc). the catch is that Mozilla only supports the PulseAudio backend.

are we meant to remove *all* sccsid/rcsid tags now, or just $FreeBSD$ ones?

why does ipsec_set_policy take a char* instead of a char const *?

looks like there is no actual reason for it, should be fixed

lw: it's probably one of those cases where it changed something at some point, but then didn't anymore and the type stayed

wouldn't be the first one

... or last :)

it eventually gets passed to yacc (yy_scan_string), so perhaps that was changed to be const correct at some point

ipsec_dump_policy(char *buf, char *delim); ... "The buf argument points to an IPsec policy structure, struct sadb_x_policy." what?

this code must be like 20-30 years old if it comes from KAME i guess, but still, that's weird

hello. what is /usr/local/etc/namedb/dynamic?

is that some freebsd innovation?

first, i have never seen *namedb* before

V-T60: the default to write bind's dynamic zones. that specific path might be freebsd-specific but most distributions of bind have something similar. 'namedb' is the traditional name for the named configuration directory

okay

i have never seen such a terminology as *dynamic zone* unfortunately

i only refered to it as just *zone*

no, not frebsd magic

what is that for?

dynamic dns?

well yes, technically a dynamic zone is a type of zone, but dynamic zones need different file permissions so named can write to them, similar to secondary zones

so it's useful to put them in a separate directory

V-T60: it's a zone stored in an opaque database file rather than a text file

V-T60: for nsupdate(1), and also DNSSEC uses it in some configurations, and maybe a couple of other things

rtprio: hmhmhm, i can see how people store there just simple text files

Title: 6. Advanced Configurations — BIND 9 9.18.13 documentation

i wonder how much difference witness+invariants makes to compile times. should be relatively little, since that's all cpu-bound, i would have thought

hi, im trying to install freebsd, i have 2 disks, one failed DST check (i dont wna to u use it) freebsd install fines, i have another that freebsd gives me alot of errors saying "root mount waiting for graid" then it proceeds to install

but i cant install using the guided option because it stays "no disks"

the auto ufs shows the disk, but when i try to do anything it says "operation not permitted geom ada0"

dstolfa: freebsd/freebsd-src #1099

Title: libipsec: make const-correct by llfw · Pull Request #1099 · freebsd/freebsd-src · GitHub

lw, I'm working on mediaelch right now

came to the office on a saturday just for you :*

jbo: i feel bad now because it's really not *that* critical :-P but thanks :-)

nah all good - I wanted to have this done 10 days ago.

but $life

i actually found a weird (minor) bug in mediaelch the other day that i suspect might be a qt6-wayland issue... should probably investigate that

and I found a python software that I want to use on freebsd but I don't feel like maintaining python stuff

hmm, it has a pypi entry, so maybe not that difficult

that's what i thought when i wanted to port sublime-music and became toot maintainer

run, run while you still can

(although tbf the other ~4 python ports i submitted at the same time went fine, so...)

lw, f*cking hell have to build lang/rust again to test the mediaelch changes - and that probably for more than one jail :<<<<

jbo: yeah iirc it somehow depends on gcc, llvm and rust? although that seems to be normal for all ports nowadays

yeah it's stupidly ridiculous

i'm sure a lot of these gcc dependencies could be fixed with a bit of upstream work

AFAIK (but I am really not even remotely sure), rust uses GCC's standard libs under the hood so you can always build that first...

rust not so much... but did you see the rust binary port? i'm not sure if it's in-tree yet

it uses the binaries from the rust people instead of building it

yeah I have, rust-bin but as a committer I don't think I should rely on that. I need to make sure that things will build under a common/reasonable scenario which includes the official pkg builders

dstolfa, which one?

jbo: rust's

nobody seems to agree tho

gotta have your cargo cult

rust would be okay if it didn't seem to update so often, literally every time i build ports it's building rust

that might just be cognitive bias since i find it so annoying though

jbo: most people who had to work with it in the context of distributing software that i've talked to do

AFAIK microsoft is heavily jumping on the rust wagon - looking forward to see what happens to rust in the next 5 to 10 years

(i don't even dislike rust as language, but it's quite silly we have to compile so many compilers for these trivial packages like mediaelch)

rust's issues are almost entirely in the ecosystem

yep. and companies like microsoft do not have a track record of making those better

i'm sure their internal stuff is good, it's just not going to help anyone else

and on top of that I personally (!) still think that a lot of what people see in rust as a language is totally what modern C++ allows for too.

jbo: imo C++ still has the problem of being "accidentally" unsafe, but you can address that by judicious application of clang-tidy and the c++ core guidelines...

(put all your unsafe code in one module, apply a strict clang-tidy ruleset to the rest of the code)

(which is actually how you're meant to use the core guidelines, just no one does this)

a few years ago I had to work with libraries written by google quite often (and with that, their own tooling ofcourse such as gtest, abseil, bazil, ...). A lot of people are "whooo, it's a library written by google, it must be amazing". but what they fail to understand is that google, like any other company or person, makes stuff to solve a problem that they have. the overall problem is that google has problems that nobody else has. therefore, their solutions are

IMHO often not easy to employ in other real world scenarios

and I feel like rust has potentially the same problem, but I don't know enough about it for making a point/argument there.

jbo: but re: -social, you know it's going to be easier to get rust in base than to actually support C++

lw, yes, but IMHO for non-technical reasons.

oh yes

C++ is there, for free.

lw: i'm not so sure. the technical issues to be addressed with rust are pretty monumental at the moment

someone needs to do the work. until then, it's pretty much a no-go

i'd like to see a proof of concept

dstolfa: yeah, but people are working on them, while certain people will die on the hill of (for example) not correctly supporting C++ in libc

my proof-of-concept is building lang/rust probably three times today just to be able to commit a port of a media player written in python.

lw: people have worked on rust pocs for a while, but none yielded any actually useful results yet

there are prototypes sitting around, but not ones we can use

I should think of getting a PC hardware upgrade/replacement...

dstolfa: as jbo said, this is going to end up being a political issue, not a technical issue. the technical issues can be fixed (with enough work), but C++ will never be politically acceptable while rust is trendy

I suggest we'd have an easier time rewriting parts of freebsd in javascript than accepting the benefits that C++ can bring over C if one is willing to accept that 2024 C++ is not 1998 C++

i don't really care about what people find trendy, i care more about what the code actually does and how it works. right now, C++ is something i can use in base to solve useful problems. rust is still a long way away from being in an acceptable state to be in base

Windows and Darwin have C++ *in the kernel*, FreeBSD can't even support C++ in libc

lw, can you elaborate on that? most of my work is C++ but always outside of base. are talking about lack of libc++ in base or that libc in base is C++ "incompatible" ?

jbo: my libc patch was reverted here: lists.freebsd.org/archives/dev-comm…-src-main/2024-February/021484.html - which is fine, it's wrong - but see that the person who requested the revert said they would object strongly to any attempt to fix the problem requiring the revert

Title: Re: git: c27a89971805 - main - stdlib.h: add __noexcept to prototypes

from the point of view of a _USER_, I'd just like to point out that ever since firefox started transitioning to rust I did not get a better firefox experience. Obviously one can argue that this can be due to a lot of web crap but my point stands.

(meaning this problem can never be fixed)

as long as that same person also strongly objects against getting rust into base I'm fine :D

I am mainly upset about how one can rust fanboy but deny anything C++

but I guess the term "fanboy" explains why.

i haven't really done anything with Rust but people i know in the C++ community seem fairly pragmatic about both languages. the "you can never write secure code if it's not in rust" people are quite annoying though

yeah

any bhyve gurus here?

i'm somewhat conflicted with what kind of C++ should be in base and what we should handle. i don't think we should have any exceptions in base, and i don't think any of our libraries should support them or RTTI (for base, anyway). this isn't so much about C++ as it is about how exceptions work in general, pretty much in any language

although, it's also not untrue, if your only options are C and Rust... which they are in Unix, because see above re: C++ being politically unacceptable

dstolfa, not having exception handling or RTTI in base is a fine thing to do IMHO.

dstolfa: the problem is without libc knowing about exception, it's impossible for a C++ compiler running on FreeBSD to conform to the C++ standard

someone in #c++-social argued earlier that std::basic_view was implemented incorrectly

that doesn't mean strlcpy() should be throwing exceptions or anything, it just needs the basic support to provide conformance

you shouldn't use strlcpy() in C++ anyway

(i personally dislike exceptions and would be strongly opposed to base libraries throwing an exception anywhere - for one thing they are terrible for performance, and once you put them in libc, the user can't elect to avoid that overhead)

If anyone can explain WHY bhyve is stuck, I will be very happy :) antranigv.am/misc/bhyve_procstat_kstack.png

lw, quick question on all of this (because I think my understanding might be wrong here): you would use std::memcpy() instaed of C lib memcpy() in C++ code, right? and std::memcpy() would/could be implemented by libc++ so you can leave libc as is?

jbo: i would writes std::memcpy() in the source code but it's just an alias for ::memcpy() so it doesn't really make any difference, neither libc nor libc++ need to know about std::memcpy

(except that libc++ has to provide the alias in <cstring>)

lw: did kib elaborate on the reasons why he opposed a patch that would ensure we call std::terminate?

lw, but does the standard require it to alias C's memcpy()? your libc++ could just provide a "C++ implementation" of memcpy()?

dstolfa: no and i didn't ask since this isn't a discussion i want to get into right now

the longer we don't get into these discussions the harder it becomes down the road tho :<

jbo: i don't know the answer to that. the standard says that the contents of <cstring> should be "the same as" the contents of <string.h>. does that mean ::memcpy and std::memcpy need to be the same function? i have no idea

but i'm not sure why you'd want to provide a separate implementation of std::memcpy anyway

lw, that was with regards to not modifying libc at all

as in, that was my thinking - but I didn't know - hence I asked

iirc you aren't allowed to take the address of standard library functions so i don't think an application could tell if std::memcpy was a separate function or not

jbo: i'm not suggesting "C++ support in base" means providing std::memcpy() though. libc++ already does that

so, reading the email thread, it reads to me that kib is concerned about adding C++ ABI semantics in general into libc. i'd be interested to know his reasoning and i think this is a discussion worth having because it sounds like he has specific technical reasons for it, but didn't elaborate further on the mailing list

[00:36:28] [06] [00:35:39] Finished lang/rust | rust-1.75.0: Success

[00:27:19] [02] [00:00:00] Building devel/llvm15@default | llvm15-15.0.7_10

q__q

for a media player.

written in python.

seriously?

yes, ask lw

he makes me do this :p

oh yeah, that sublime player

ah wait, that is indeed incorrect. this is not for the sublime player, this is for mediaelch

even worse then :p

i'm pretty sure you actually offered to do this :-P although you might be regetting that now

nah, all good - I'm just teasing

I feel mainly bad that I couldn't do it the last two weeks because real life

I've just updated from 13.2 to 14. I followed the instructions that involved running freebsd-update a number of times and rebuilding all the packages I had installed from ports. Now, after the upgrade, unbound (from ports) refuses to start complaining about missing SSL libraries. I've rebuilt it twice but the problem persists. Does anyone have any suggestions on how to approach the problem?

njdoyle: which ssl library is it looking for?

Sorry, my network is being a pain. I may have missed some messages.

This is what I see when I try to start the service: unbound[32533:0] error: error in SSL_CTX verify crypto error:80000002:system library::No such file or directory

that's when I do: doas service unbound start

ldd says unbound is linked against /usr/lib/lib{ssl,crypto}.so.30 which seems correct to me post upgrade.

or, that is /usr/lib/libssl.so.30 and /lib/libcrypto.so.30

njdoyle: did you run the final freebsd-update step that removes the old libraries? i wonder if it's somehow picking them up during install. never saw this error though (and unbound is working fine for me on 14/15)

lw: I did run the final freebsd-update that removes old libraries.

lw, can I ask you sanoid related stuff?

the only thing i can think of is that openssl 3.0 removed some legacy cyphers and perhaps it's failing because of that? ... but i would expect a different error then, any in any case i don't think TSIG/DNSSEC uses those cyphers

jbo: sure but i promise no answers

Is there a way I might be able to figure out what file unbound thinks is missing?

njdoyle: you could try starting it under 'truss' which will log open() calls

(generates a lot of output, redirect 2> to a file)

lw, assume a dataset hierarchy like A/B/C1, A/B/C2, A/B/C3 ... A/B/C17 In sanoid.conf I have a rule for A/B which uses my production template. however, I want to exclude A/B/C7 because that is a gigantic dataset (poudriere VM image). I came across this: jimsalterjrs/sanoid #627

Title: How to exclude child datasets from snapshots? · Issue #627 · jimsalterjrs/sanoid · GitHub

627 – vidcontrol causes crash when booted with sc0 disabled bugs.freebsd.org/bugzilla/show_bug.cgi?id=627

so I created a rule for A/B/C7 using a template with everything set to no/0 but it's still making the snapshot

do you have that "exclusion" mechanism working?

let me see

lw, this is my actual config: termbin.com/zwgo

jbo: this is what i use and it works for me (i just checked) le-fay.org/tmp/30d/i75Eme.txt

jbo: i think recursive = zfs won't work for this because zfs snapshot -r doesn't care about your sanoid config

lw, any idea why mine wouldn't work? looks very similar except that I have more hierarchy in the dataset paths and I use recursive=zfs (not sure what the difference is)

lw, aah

try recursive = yes instead

lw, I couldn't find sanoid docs on the difference between recursive=yes and recursive=zfs. do you know the difference?

Okay, I figurd out my problem. It looks like security/ca_root_nss got uninstalled in package cleanup. It must have been a dependency of something and not a thing I installed specifically. Reinstalling that fixed things.

jbo: recursive = yes manually snapshots every descendent filesystem, recursive = yes uses zfs snapshot -r which is theoretically more consistent but it always snapshots everything

Okay, I figurd out my problem. It looks like security/ca_root_nss got uninstalled in package cleanup. It must have been a dependency of something and not a thing I installed specifically. Reinstalling that fixed things.

er, the second recursive = yes should be = zfs there

lw, but... what

njdoyle: odd that it caused that error, which says nothing about the actual problem

openssl doing openssl things i suppose

so recursive=zfs is proper -r flag in zfs. but recursive=yes means that somebody is doing the recursive handling in perl?

lw: agreed. The message could have been way more helpful.

jbo: yes, with =yes sanoid enumerates the filesystems by itself and snapshots them one at a time

lw, so you use =yes instead of =zfs everywhere?

jbo: yes

lw, so if I were to change that now.... this is a production server :D

i only have 73 filesystems so i very much doubt this causes a performance issue, maybe if you had thousands it would matter

nah I have low tripple digits

jbo: well, i've never tried changing it... but... i think it should create the same snapshot names so it *should* be fine. and you have backups right? :-d

:3

it just means the hourly_whatever timestamp on the child filesystem might be different from its parent, but syncoid doesn't care about that

lw, is there a sane way of deleting all snapshots of a dataset (without modifying the current state of the data on the dataset, similar to doing it manually, just "remove all the snapshots from this dataset but don't screw up"?

jbo: not that i know of, i usually do: zfs list -Honame -rt snapshot data/set/name | xargs -n1 zfs destroy # i STRONGLY suggest using 'echo zfs destroy' to make sure this does what you expect

i really, really wish the command to destroy a snapshot was different from the command to destroy a dataset

hell yes. 100% that

would be even fine to leave zfs destroy as is, but having a separate subcommand just for snapshots

zfs snapdestroy or whatever

zfs destroy-snapshot

you literally mistype an @ and you can be whooped :D

or even zfs destroy -t snapshot or something

yeah

like, say i want to destroy foo/bar@baz, i type "zfs destroy foo/bar" and my cat walks across the keyboard... whoops

lw, so, when I fix my config and sanoid no longer takes snapshots - will syncoid also not push them to the backup destination host?

jbo: if the snapahots exists they will be pushed, i don't believe syncoid cares about your sanoid config at all

yeah

I really want a proper --exclude option :<<<<<

Title: added ability to skip datasets... simply set syncoid:no-sync=true by attie · Pull Request #189 · jimsalterjrs/sanoid · GitHub

that sounds fuckly

oh i use that all the time, but isn't it called syncoid:sync=false?

so you set custom properties?

attributes

yeah

there's syncoid --exclude

Does Steam Gaming work with FreeBSD?

yes

SponiX: somewhat yes, there is one person (i think) who maintains Steam on FreeBSD

as to how many games actually work, no idea

Okay, thanks for the replies lw and jbo

lw, can you tell me about your reasoning/opinion on the zfs property vs. the syncoid --exclude=REGEX option?

lw, there's an outdated list

Title: Compatibility · shkhln/linuxulator-steam-utils Wiki · GitHub

SponiX, ^

jbo: to me whether a filesystem is backed up is a property of the filesystem itself and not something that should be centrally configured where i'll forget to update it. this way i can create a new filesystem and tell it not to sync at the same time, for example

jbo: analogous to other operating systems which have a "backup" file flag

and if you use zfs delegated administration, this lets users configure backupiness by themselves

does anyone here get some segfaults with these? paste.mozilla.org/preuMHCB paste.mozilla.org/MJB1rc32

Title: Mozilla Community Pastebin/preuMHCB (C)

lw, that sounds reasonable. so now to figure out whether it is syncoid:no-sync or syncoid:sync

jbo: syncoid:sync definitely works and i'm sure that's what i found in the documentation

yeah github.com/jimsalterjrs/sanoid#syncoid-dataset-properties

Title: GitHub - jimsalterjrs/sanoid: These are policy-driven snapshot management and replication tools which use OpenZFS for underlying next-gen storage. (Btrfs support plans are shelved unless and until btrfs becomes reliable.)

oh interesting, you can set it to a list of hosts, never knew about that

lw, that is very, very nice.

lw, thank you

meanwhile... [01:14:07] [02] [00:46:48] Finished devel/llvm15@default | llvm15-15.0.7_10: Success

do you want to ask more syncoid questions while gcc builds? i'm waiting on make packages (which is surprisingly slow)

also i think it's putting the packages in the wrong place so i should probably stop it and fix that

lw, I'll take a rain check on that :p

lw, so it's zfs set syncoid:sync=false A/B/C7 right?

jbo: y

don't do what i always do and set 'sanoid:sync=false' instead because that doesn't work

heh :)

I do stupid things myself every now and then, don't worry

ah it's ${EXTRA_REVISION} which is causing my bad package names, but you can override this with make PKG_VERSION=15 i guess

jbo: if you're worried, just use 'echo zfs destroy', review the commands and copy+paste them into your shell

that's what i did the first time i wanted to destroy a load of snapshots

that doesn't keep me from soiling my pants as we speak

lw, stupid question, but does something | xargs work when using doas?

something | doas xargs ...

i.e. doas something | xargs cmd

my shell foo is very, very poor.

unless something needs to run as root as well in which case doas something | doas xargs

jbo: i want to mention right now that if you delete all your datasets this was definitely not my fault and you still have to commit mediaelch

aaaaand it's gone

lw, of course, yes :D

:-p

lw, I honestly like that zfs property solution for syncoid:sync

that seems very, very nice.

yeah, i didn't even realise zfs had custom properties but i can see lots of uses for them

ah my packages are fine, it put a latest symlink in the package directory: lrwxr-xr-x 1 root wheel 21 Feb 3 17:46 latest -> 15.snap20240203170857

lw, so are you typically setting up your sanoid.conf to snapshot the entire filesystemn (including /etc, /usr/local/etc and so on) and then use the zfs property to prevent it from syncing or what's your normal operation mode here?

jbo: yes, i sync everything by default because i'd rather waste a bit of bandwidth than forget to back something up. i check on the backup server every couple of weeks to make sure it hasn't synced something it shouldn't

lw, would you mind sharing your typical sanoid config for that?

jbo: it's pretty basic, this is my desktop: le-fay.org/tmp/30d/WANqRJ.txt and this is our fileserver: le-fay.org/tmp/30d/4E3eJc.txt and this is what ends up on the backup server: le-fay.org/tmp/30d/KXHyRt.txt

lw, that brings me to my next question: I once asked you how you prune snapshots on the backup host. AFAIK you told me that you just run the same sanoid.conf of the source host on the backup destination host. However, I have a dataset for each host similar to what you just showed (/backup/${HOSTNAME}) how do you deal with that?

or do you just have a sanoid.conf that only does prune on the entire /backup dataset?

jbo: i just looked and apparently i didn't actually get around to setting this up yet, but my plan is to just make a sanoid.conf with one entry for all of [backups] that just needs a reasonable selection of snapshots (so autoprune = yes, autosnap = no)

lw, and then values for monthly, hourly etc?

did not realize that freebsd can use ubuntu binaries the things you learn when you read documentation :)

lw, but that means that it's indeed corect that you do not need to have THE SAME config as the one running on the source host, right? because that bugged me.

bsdbandit, +1 on always reading docs

jbo: yeah, i don't think the config can be the same in this situation

if only because the paths / dataset names are different

indeed

I have hdd/backups/hosts/${hostname}/zfs as the prefix :D

"prefix"

since you're asking i'm going to set this up now and see how it goes

and I'll wait for a single-core build of qt6-base to complete

lw, that syncoid:sync=false property seems to work :)

lw, what's your reason to bother explicitly disabling backups on zroot/var/crash ?

that would not be more than a few MB anyway, right?

/var/crash contains a memory dump of every kernel panic, it can easily end up being many gigabytes

aye

lw, let me know how that pruning works out

jbo: so i just created a sanoid.conf on the backup host like this: le-fay.org/tmp/30d/mxtrPM.txt - and it's now destroying all the snapshots as expected. however, syncoid will immediately push them again next time it runs. so i think what i'm going to do is set daily = 90, monthly = 12 and then see if i need to exclude hourly backups from syncoid somehow

hmm, actually, i don't think syncoid will push them again because it should only push snapshots created since the last syncoid snapshot, if i understand zfs send -I correctly. so that might not be a problem

are you sure about that? ._.

no, but i'm going to test it :-)

i might also add some specific config for filesystems like prometheus because i very much doubt i'll ever need to restore a specific prometheus tsdb from 3 weeks ago or whatever

(and if i do, i still have the snapshot on the source system...)

meena: make update-packages doesn't seem to work, it's just creating a new 15.snapwhatever with a new timestamp

lw, why tf do you have zroot/swap

jbo: because i thought 4GB of swap would be enough when i installed and it turns out it really isn't, i had to add another 16GB on a zvol

lol

you would think 32GB would be enough to run poudriere (with tmpfs disabled!) but no

it's not poudriere's fault

electron.... chromium....

it's not those either, when i asked on -current someone suggested it might be the zfs arc prune bug

basically something ends up wrongly using loads of memory (it's stuck in Inact) and eventually poudriere runs out of memory and the kernel complains about 'out of swap space'

is that -current only?

i've only noticed it on -current but i think the bug exists in older versions

b

c

oh, sorry

otoh i thought the bug was only about cpu usage, not memory use, so i'm not convinced it's actually this

i wanted to do ctrl+b+d

swap usage does seem to have gone down though so maybe it's fixed: Swap: 16G Total, 405M Used, 16G Free, 2% Inuse

in the past i'd end up with several GB of swap used after a (non-tmpfs!) poudriere run

(non-tmpfs! is important because whenever i mention this someone tells me to turn off tmpfs, like you can even use tmpfs on a system with less than 128GB of RAM)

lw, I remember that there were a couple of forum posts/threads from people that suddenly couldn't build stuff like lang/rust and similar on their poudriere sets. I don't recall the details but wondering...

Good, Good…

i'm just testing on my build VM for now, so if it breaks everything i can just rebuild and... wait... this may not have been the best idea

wow, pkgbase overwrote all my config files, that's a bit rude

they should still be in foo → foo.pkgsave

yeah but still:

lily# head /etc/master.passwd

root::0:0::0:0:Charlie &:/root:/bin/sh

that's what you get for using stupid hostnames. everybody knows that hostnames should be elements of the periodic table.

jbo: i'll call my next vm 'jbo'

:3

lw, I just wanted to tell you that testport for 14.0 mediaelch is still running when it finally dawned on my why this is taking HOURS

I'm an idiot.

i'm glad you're the idiot this time and not me

I was building on/for the stable jail rather than the release jail

whoo pkgbase FreeBSD lily.eden.le-fay.org 15.0-CURRENT FreeBSD 15.0-CURRENT #0: Sat Feb 3 16:12:37 GMT 2024 root⊙lelo:/builds/obj/data/public/src/freebsd/src/main/amd64.amd64/sys/GENERIC amd64

now to find out how to switch to MINIMAL

Tobuild: 205

fml

ok, i already like that i can 'pkg remove' all the useless crap i don't want and it will never come back

i don't like uname doesn't show the git revision though, is it ignoring my make.conf/src.conf for some reason? i have WITHOUT_REPRODUCIBLE_BUILD=yes set...

it does? FreeBSD tyrone.17es.net 15.0-CURRENT FreeBSD 15.0-CURRENT #4 n266291-2ffd30f7ee15: Mon Nov 6 20:12:25 UTC 2023 ...

rtprio: it does on my non-pkgbase system, it doesn't on my pkgbase system with identical make.conf/src.conf

FreeBSD 15.0-CURRENT #0 main-n268031-49619f73151a-dirty: Sat Feb 3 14:42:48 GMT 2024 lexi⊙ielo:/src/obj/src/freebsd/src/main/amd64.amd64/sys/ILYTHIA

FreeBSD 15.0-CURRENT #0: Sat Feb 3 16:12:37 GMT 2024 root⊙lelo:/builds/obj/data/public/src/freebsd/src/main/amd64.amd64/sys/GENERIC

oh, didn't catch it was pkgbase.

i notice it also built a bunch of pkgbase packages for things i have disabled in src.conf, which is fine, but makes me wonder if it ignores these files for pkgbase build

lw ping

hello the jbo

can you quickly sanity check this (visually): termbin.com/q7jo

jbo: looks good at first glance

you don't like pkg-plist?

lw, ack, testports almost done

> If the port installs just a handful of files, list them in PLIST_FILES, within the port’s Makefile.

Title: FreeBSD Porter's Handbook | FreeBSD Documentation Portal

hmm, i saw that also but i thought 'handful' would be <= 3, like 'a few'. it's fine though, i don't care either way

well the example there has 6 so... :D

ah

hence I changed it

handful is a terrible quantifier anyway

a handful of grains of salt vs a hand full of excavators...

this reminds me of the German habit of talking about how 'heavy' a file is

which i always quite liked, tbh

personally, I hate that convention anyway. I'd be all pro always having pkg-plist no matter what.

the handbook even tells you that it's a PITA

like 'this package weighs too much, you must make it smaller'

Im following the jail guide trying to set up vnet. It says for hosts to add this to the rc.conf:

ifconfig_bridge0="inet 192.168.1.150/24 addm em0 up"

Where does that ip address come from?

somebody just typed it as an example

Yeah i know

Where do I find the ip im supposed to use.

matthewp: if you're adding a new bridge containing the vnet interface and your existing public interface, it should be the address of your existing public interface

and remove the address from the public interface at the same time

(so the address of em0 in this case, i assume)

then just set ifconfig_em0="up" with no address

I don't have an em0

I have a vtnet0

ok, replace em0 with vtnet0 then

if this is on a VM, i highly recommend checking your serial console works before rebooting...

That's an external, public facing ip address, this is what i put into the rc.conf?

well, what do you have in rc.conf right now, for ifconfig_vtnet0?

I don't have that in the rc.conf at all

that's curious, do you know how it gets an address then?

even if it's DHCP i would expect it to have ifconfig_vtnet0="DHCP"...

im a novice in networking, i have route_linklocal in the config

can you paste your entire rc.conf somewhere?

yes

it's possible i've removed that... i don't recall doing so but if i did and haven't rebooted, could that be the reason?

oh wait

i'm so sorry, i do have a ifconfig_vtnet0

possibly yes, that will definitely break if you reboot though

ifconfig_vtnet0="DHCP -rxcsum -tso"

ah ok

hang on, let me check something in the manual

matthewp: ok, i *think* what you want is this: set ifconfig_vtnet0="-rxcsum -tso up", and set ifconfig_bridge0="DHCP addm0 vtnet0" ... however i've never tried a bridge with DHCP before so i can't guarantee this will work, either make sure you have serial console access, or take a snapshot of the VM first or something in case it doesn't come up after you reboot

ok, i will try that, thanks for your help

sorry, that's 'addm vtnet0' not addm0

And i'm removing DHCP from the ifconfig_vtnet0 part?

yes, because you want the address to be added to the bridge interface, which requires running dhcp on the bridge instead

this post seems interesting: forums.freebsd.org/threads/trying-t…-up-a-network-bridge-for-dhcp.20287

Title: Solved - trying to set up a network bridge for dhcp | The FreeBSD Forums

lw, how easily could you run a -current testport for me right now (not related to mediaelch)?

jbo: i'd say 87% easily, what do you need

lw, bugs.freebsd.org/bugzilla/show_bug.cgi?id=276228

Title: 276228 – audio/cava: update to 0.10.0

matthewp: yeah their config is wrong in OP because they're setting ifconfig_bridge0 twice, only the last one will be used

lw, but without the .if ${OPSYS} == FreeBSD && ( ${OSVERSION} >= 1400079 || ( ${OSVERSION} >= 1302505 && ${OSVERSION} < 1400000 ))

jbo: this is the makefile you want testported? le-fay.org/tmp/30d/FgNbP5.txt

i removed the .if and .endif

but left CFLAGS+=

lw, also drop the CFLAGS+= that was incapsulated in that block

ah ok

god pkg.freebsd.org is so slow, or at least my local mirror is

getting 5MB/s download speeds

lw where are you at?

antranigv: UK

deathstar #7

jbo: running now [main-audio_cava] [2024-02-03_19h51m56s] [parallel_build] Queued: 179 Built: 0 Failed: 0 Skipped: 0 Ignored: 0 Fetched: 142 Tobuild: 37 Time: 00:02:53

lw sometimes geoDNS might be configured wrongly, just use a specific one in FreeBSD.conf

lw, thanks

lw for some reason Armenia is set to South African server, should be changed to one in Europe, so I just change it manually.

antranigv: i wonder if the geodns doesn't work properly for ipv6

lw oh, that's also possible!

i'm not even sure if poudriere lets you change it, i should investigate that

jbo: can you PM an email address to send the invoice for electricity usage to?

(this is a joke, don't do that. although before long, who knows.)

lw, it should be a fairly light port. it is truly only if it's no inconvenience to you at all. I can handle it on my end otherwise.

no it's fine, i'm just editing shell scripts anyway

jbo: oh testport finished already, no problems, bulk is running now. do you want me to test the port itself? (i have no idea what it is/does)

lw, no need to runtime test. just whether testport succeeds on 15-CURRENT

jbo: ok all done then, no errors

thanks!

Warning: you might not need LIB_DEPENDS on libEGL.so

Warning: you might not need LIB_DEPENDS on libSDL2.so

fyi

ack

also cava.c:409:36: warning: passing arguments to 'getPulseDefaultSink' without a prototype is deprecated in all versions of C and is not supported in C2x [-Wdeprecated-non-prototype]

but it was not an error

yep

can someone help explain why i can ssh to a host on the lan, but not curl to it?

over nat

ports are open, can test that locally;

rtprio: tcpdump

i can tcpdump but not decipher it

show the tcpdump output while you're curling

on the destination

Title: 7ugV

which system is the server and which is the client?

24 is the server, the web server is port 3000

18 is the client, which is nat'ing wireguard

that's weird

from 18 directly works fine

also i noticed the sendq for that tcp connection increases until it's reset

it seems like the reply to the syn is getting lost

Proto Recv-Q Send-Q Local Address Foreign Address (state)

tcp4 0 32061 10.0.17.24.3000 10.0.17.18.58686 FIN_WAIT_1

i suppose i could try ipfw but now it's basically a "nat on $ext_if inet from !($ext_if) -> ($ext_if:0)"

how does curl behave, connect timeout?

connection reset

oh that's weird. there is no RST in that tcpdump

it feels mor like it times out

19:22 <lw> i notice it also built a bunch of pkgbase packages for things i have disabled in src.conf, which is fine, but makes me wonder if it ignores these files for pkgbase build ⬅️ weird. I would think this is a bug

after sending GET / and all that

rtprio: you're using pf? try adding log to your block rules (if you don't have them already) and tcpdump on pflog0 to see if it's dropping anything

meena: maybe i will ask pkgbase@, although i could do without someone telling me i'm an idiot today :-)

maybe i need just pass all; i didn't have any other rules at first and the same symptoms

rtprio: if you can start with pass all i'd do that and see if it works at least

or even set skip on wg0

(although that will break nat ofc)

lw: given PkgBase@'s response times, you should be safe

lw: do the packages contain anything sensible?

meena: yeah, the packages themselves are working fine

*weird*

lily# pkg info -l FreeBSD-kernel-generic|grep /kernel/kernel

/boot/kernel/kernel

/quit

ok, so new question, i created a git repository by pushing to it then put it on http, but my git client won't find it because info/refs doesn't exist: 2024/02/03 20:16:08 [error] 65926#117500: *43 open() "/data/public/src/freebsd/src.git/info/refs" failed (2: No such file or directory), client: 2001:8b0:aab5:106::20, server: git.eden.le-fay.org, request: "GET /freebsd/src.git/info/refs?service=git-upload-pack HTTP/1.1", host: "git.eden.le-fay.org"

am i missing something obvious here?

src.git/info only has 'exclude'

my local git repositories don't have a info/refs either so i'm not sure why it expects this

ah... i need something called git-http-backend

ah no, i just need to run git update-server-info... wonder why this is so complicated

lw, ping

jbo: YO

lw, I want to port this: github.com/gaphor/gaphor

Title: GitHub - gaphor/gaphor: Gaphor is the simple modeling tool

lw, any idea how much pain that will be?

jbo: well it's got a pyproject.toml so you can use PEP-517 which makes it easier. probably depends on how many of the dependencies already exist

lw, what is PEP-517 exactly?

i have never used poetry though so i don't know if that makes it more difficult

jbo: the replacement for setup.cfg/setup.py from distutils

Title: Python/PEP-517 - FreeBSD Wiki

ah, so it's the typical: "here is the 17th solution to the problem everybody has" type thing

yeah

like the python version of make/bmake/autotools/cmake/xmake/meson/...

you just write this in Makefile and it magically works though: USE_PYTHON= autoplist concurrent pep517

but like i say i don't know anything about poetry (that's the new version of pip/setuptools or something? not really sure) so...

lw, I still need to list RUN_DEPENDS and BUILD_DEPENDS manually by copying from the pyproject.toml?

yes

oh and it needs python 3.11, so have fun testporting it

-__-

jbo: although i'm not sure if you saw the comment on the sublime-music PR but apparently ports is switching to 3.11 as default RSN

not sure why we skipped 3.10 tbh

mumble mumble

apparently style(9) says nothing about where const should go, so i'm going to use 'char const *' as the Gods intended

daring

this might cause almost as much bikeshedding as C++ support in libc

but i am nothing if not a massive a troll, so let's find out

doesn't C support this now?

"now"? for literal decades.

top of any block, for decades. Anywhere at all, yes just recent

jgh, C99 allows that, no?

I was wondering if it was even c89

i thought it was in C99 but i'm not really sure without checking the standard

don't think so, but I could be wrong... I'm stuck in the 80's

i'm also not sure src/ is allowed to use anything later than C99... isn't the C11 update still in progress?

lw: yes, still not done

but it might also just be style(9) to prefer variables only at function / block start

rtprio: looks like an MTU issue

wheeeee freebsd/freebsd-src #1100

Title: move traceroute from contrib to base by llfw · Pull Request #1100 · freebsd/freebsd-src · GitHub

Question for the pf experts out there. I have a few Asterisk PBXs that I’m looking to secure behind a separate FreeBSD box with pf rules as the firewall. In iptables on the CentOS Asterisk PBX’s today I use a rule of “-A INPUT -p udp -m string -i eth0 --dport 5060 -j ACCEPT --string "REGISTER sip:[FQDN HERE]" --algo bm” to look for if you know the FQDN, you’re allowed. Is there an equivalent way to do this in pf?

hi all, is this article valid with freebsd13 calomel.org/freebsd_network_tuning.html

Title: FreeBSD Network Performance Tuning @ Calomel.org

use zfs on the host

ufs on the client

daemon: do you have any numbers to support that?

no and I do not have a server handy to prove it, but double caching is generally a bad idea in any case

well, UFS also caches, unless you mean to mount it with directio or something? (does UFS support that?)