we have two Lenovo X1 Yoga ThinkPad 20s (FR,UB). The FR used to mount windows (hate it but need some software occasionally) in FreeBSD fine, but lately is having trouble loading fusefs... wondering if I need it in loader.conf or only rc.conf and which is the correct way? I read a post about it but seems they maybe got the instructions switched around?

works fine manually, just wondering how to automate (so it doesn't start asking for input and without network)

if you can do stuff via rc.conf and loader.conf, then prefer rc.conf and only use loader.conf if you really need it to boot the machine

maybe names of some things to load have changed? If I've used *BSD since 1997 and on these laptops since maybe FreeBSD 11, and upgraded, how can I check everything in these two .confs is up-to-date or needs changing?

i read fuse was renamed to fusefs

On those occasions I needed fusefs, I loaded it from a shell as root, so rc.conf should be fine.

is it fuse_enable="YES" or fusefs_enable="YES" in rc.conf? Neither worked for me, though it works manually...

or maybe that's the format for loader.conf? I don't want it in there rather than rc.conf because this isn't essential and should boot even if fails

See "kld_list" in the rc.conf manual page.

Minor heads up, but the links that contain underscores on this (and I assume others?) page are broken...

Title: mac(4)

okay, that worked, but what about checking the rest of my loader.conf & rc.conf for changed syntax?

*_enable goes in rc.conf for activating services started by rc(8). *_load="YES" goes in loader.conf to load modules vert early.

s/vert/very/

Changed syntax between...?

I have a script snippet that I saved which lists which ones belong where.

comm -23 <(sysctl -T -a -N | sort) <(sysctl -W -a -N | sort)

Oh wait, those are for tunables, not modules.

already stated--old versions--like fuse was renamed fusefs... I want to check every line to see what else might've changed

i think i have some mistakes in those

Ah, no idea then.

I don't know of any facility that will inspect loader.conf and dump that info. I learned the hard way that's one reason release notes should be read between updates. I just backed up my existing loader.conf + rc.conf, moved basically everything into rc.conf, and then rebooted to see what loaded and what didn't. Whatever didn't load, I checked the

man pages to make sure it's still called that and then moved it into loader.conf. Reboot, rinse, repeat.

alright

instead of altering /etc/profile is it best to add in /etc/profile.d stuff like aliases.sh, environment.sh, functions.sh ?

if you want them system-wide (like for personal desktop not server)?

darwin, It's strange to put personal configuration into /etc anywhere, that would go into $HOME. But it is typical for universities with campus workstations to put global configuration there.

i'm just a person, not at a university--I said personal desktop!

And it would be quite unusual to be storing personal settings in /etc. However if it is your machine then it is your machine and you can do anything you wish.

But you were asking an "is it best" question. I respond with a no, it is not best to put personal configuration in /etc. Best to keep that in $HOME.

"Fun" would be when system-wide directories get new files to be in conflict with personal files

so what's the answer?

i was talking about /etc--which I already use for this--the whole time... thought that was clear. With UNIX/*BSD people are always assuming it's a server or public workstation but most people I know use it for their personal PC

I tried to say things very clearly. Sorry if I failed at that task. When you talk about /etc we are going to think it is about global configuration. Keep personal configuration in $HOME. But it is your machine and you are allowed to break the rules if you want. But then if you break it you are the one to debug it.

okay, what's /etc/profile.d for?

The /etc/profile.d/* is a place that global configuration files may be placed whole without needing to edit the /etc/profile file.

Installing a file and removing a file is easy to automate with a script. Editing a file is less easy and more troublesome.

I don't have any global configuration on my desktop and my /etc/profile.d directory is empty.

i don't think editing is hard; I've been doing that since 1993

Have you ever seen people write a script where they ,,sed -i '' s/abc/xyz/g somefile'' and then they realize they didn't want to replace all of abcdef but only abc and screw things up? It happens.

Sometimes people will install things like bash_completion globally in a /etc/profile.d/bash_completion file. That's not totally uncommon.

But generally if it is your personal configuration like an alias for some command, or setting email address, or setting less options, then those would go into $HOME in your .profile and/or .bashrc and/or your $ENV file, and/or your .cshrc, all depending upon your shell.

That's also a good thing because then it is easier to transfer yourself over to another system. Just copy $HOME over and then you are set and customized on the other system.

Editing files in $HOME is also something you can do as yourself and no need to su or sudo to root is needed.

i've edited /etc/profile since around 1997 or soon after and for my PC (no one else uses) am only asking if it's better to leave that unaltered from default and move my changes to /etc/profile.d or somewhere else global like if there's a /usr/local/etc

i keep files I copy to servers separately

because I don't use all the same aliases, environment, functions on those

I've been adding my global env vars to /etc/login.conf as a shell agnostic approach but maybe I'm weird. Although that doesn't apply to aliases and other things.

anyone have experience with zfs-periodic? i installed the package pkg install zfs-periodic and when i try to find the location through which zfs-periodic or with find / -name "zfs-peri*" -type f it returns NO results

voy4g3r2: pkg query '%Fp' zfs-periodic

uhh duh i was expectiung zfs-periodic and not zfs-snapshot

thank you stl

now we wait to see what snapshots it makes to then figure out how to use zfs-zend and zfs-recv to move off machine :)

stl: `pkg info` has -l or --list-files btw

Remilia: thanks. old habit

well, the periodic part in the name is about the periodic system, as the message on install tells you too, that executable seems more like an implementation detail

it also explains why man zfs-periodic does not work but man zfs-snapshot does

nimaje: makes sense to me, just habit.. in my limited experience, doing pkg install <whatever package> usually results in a file in bin dirs fo <whatever package>

hrm.. can someone help me understand what a root zfs is? is that root in filesystem or root of the whole zpool? i ask as i am trying to snapshot storage, which has multiple datasets underneath it.. i take it best practice is to explicitly name EACH and EVER dataset under the "root" /storage such as /storage/music /storage/movies /storage/timemachine /storage/fda_aers

when i try to run zfs-snapshot, each hour it is NOT doing it and i think this is the reason but not sure what i am missing.. figured, as long as not mounted as / that is fine.. but clearly not

does periodic run hourly, as the install message indicates? periodic is run via cron and then checks what it should run, if I understand the install message correct, hourly runs aren't in the default crontab

nimaje: yeah i have a cron job added, as per the instructions

for 2 past the hour

i see that the cronjobs are running but defaults to no output, that i can find

voy4g3r2: snapshots are not recursive by default

you *can* take recursive snapshots using the -r flag

the script does not use -r

okay, may have to make this a little more robust as i do not even see an /etc/periodic/hourly folder

so you need to explicitly list every dataset you want snapshots of

voy4g3r2: ports install to /usr/local/etc

are you sure you are looking in the right location?

well now i am, thank you

The zfs-periodic hourlies call for adding to /etc/periodic.conf or one of the add-ons to that.

If you change hourly_show_success to YES you'll get reports of success logged or emailed.

(it force-replaces paths in the Makefile: post-patch: \ .for _file in ${PERIODIC_FILES} \ @${REINPLACE_CMD} -e "s|/etc/periodic/zfs-snapshot|${PREFIX}/bin/zfs-snapshot|" \ \ ${WRKSRC}/${_file} .endfor)

I think I found gnome solution for servers, more heavy but still

personally I think hourly emails are a bit much

yeah but it is a good idea for at least testing.. to make sure it is operational.. i will have to go logs and i do not have mail services setup on this beaste

weechat-4.1.2 - Lightweight and user friendly ncurses based IRC client !LIGHTWEIGHT!

"The process will require 117 MiB more space."

They probably are! But if you're just checking for success a few times, it doesn't need to be permanent.

morena: I tried weechat and went back to irssi ha ha

Remilia ye I use it too, wanted just tried it again after years

but that piece require to install ruby, perl , python, lua and probably every programming language it found on the internet

(but I stopped using irssi years ago after setting up znc)

excellent packaging ;/

you could build your own without

Remilia znc with?

with something you would not use

ye, not yet familiar much with it, as it tried to build all these dependencies anyway

my desktop PC runs Windows because I have to do work

oh that's sad story

I should probably first install all tools needed for buiilding, then build it without these option

also I do not have anything against Windows :D

otherwise it will build all cmake and other tools

nobody has ;/

these are the defaults: i.koumakan.jp/2024-01-03/1704286714.png

aren't you happy JS and PHP are off

ye, but even when unchecking it has many build dependencies, so it will start to building everything

not a problem for me, cmake etc. are required by a ton of other stuff so my poudriere builds those

I canceled it when it started to build cmake or similar packages

I assume I can first regulary install it with pkg so it will not be builded?

and then deleted?

as long as the port does not require a newer version

will try to install all 'requires' in desc

morena: there is a better way

bc. building whole day some nonsense to just build other thing ...

ye?

make build-depends-list

it will show everything needed, so I can install it?

yes

very good dear Remilia

everything needed for your current configuration

to be honest I was still positively surprised about that building thing

check ports(7) for more

with that pop up dialog option

you mean the config target?

ye that make config

so I did not need to play with makefile

and do typos, mistakes and ques

you are not expected to modify anything inside the ports tree

good

ye I am getting it

unless you are updating/patching a port, but for that you really would prefer something like poudriere

for convenience

poudriere testport is super nice

-still need some more time to get more familiar with freebsd tools and way

I was using OpenBSD for like a decade but was never able to build anything it never works for me, here it was pretty easy to follow doc

as often it's like just turn on/off some options

doesn't openbsd also have ports

ye

but if I wanted to change something it was pain and never worked for me somehow

always throwed errors and whatanot

usually it started with some permission and so ;/

so I ditched it very early

that's why I was positively surprised how similar steps worked well here

at first try

and if I don't want to download and install half of the internet in freebsd, building will be necessary

as everything install everything it can ;/

dunno I have been using ports since probably 2000 or something

never really had issues

(on freebsd)

even a bit of confused with , porstnap, git, poudiere

one have no idea what is 'the way'

<morena> as everything install everything it can ;/ <- the repositories use defaults that are suitable for most users

117 MB terminal irc client is suitable for most users?

I do not think 117 MB is weechat alone

no it's not

but all these languages and nonsense it install

most users will have Python and Perl

most GUI users will have Tcl and often Lua

for reason, one may need some pointless script or so

and weechat users also ruby ;/

portsnap is used to update the ports tree via snapshot servers

git is git

poudriere is unrelated to these

poudriere is a package build system that allows you to build your own repository for pkg

now what I did that I locked builded/customized packag

I assume it is the way if port is mixed with regular packages

so it does not get overwritten

with upgrade and different option

morena: portsnap is obsolete, so you can forget about that. (the recommended way to fetch ports is via git now.)

ye, I hate git

no idea why everyone encourage to use it

but understand that on other hand

ye, just read that it is obsolete, did not get it at first try

also you know you can use portmaster?

no

it has flags like --packages-build

that's also other tool?

obsolete or not yet?

'use packages for all build dependencies'

oh

I don't know?

that sounds like solution for me

Title: FreshPorts -- ports-mgmt/portmaster: Manage your ports without external databases or languages

Last Update: 2023-11-30 10:16:07

there's also synth, which is like poudriere but it can automatically use packages instead of building them where possible

unlike poudriere which wants to build everything itself

unixwitch: you can tell poudriere to use packages for certain ports

synth is more designed for desktop users, poudriere is (at least originally) for running bulk builds and tests, although it works for desktop too

oh interesting

but it will forgo that if there is a new version not in repositories

so it is somewhat useless for llvm/gcc/etc.

what I also found hard

to get info about non installed package

like no simple command for that?

Remilia: perhaps more useful for quarterly where those don't get updated often?

what kind of 'info'

you cannot really get information about a package outside its description without fetching the file

at least description

try `pkg search weechat`

ye, but does not give info

just short line

weechat-4.1.2 Lightweight and user friendly ncurses based IRC client

that's a description

anything detailed requires the package file itself

no

oh okay

I found some way to get something with weird pkg query and some shenaninagns

if you want details just go to freshports

but

that website is unreadable

sure

morena: pkg rquery '%e' weechat may be what you want?

ye something like that

can I get even more info?

or that is max

like what specifically?

morena: have you tried `pkg search -f weechat` btw

like, it is straight in the man page for pkg-search

Remilia I think yes, but the issue is, that for example I want to info about one package

and?

but that will shows info for many wich include search term inside

use the full package name

again in almost unreadable format

I guess you should switch to Windows :\

ye but it will shows like py-name whatever-name and similar

11? ;/

I used it for a month

are you using the right command? this seems readable to me: bsd.to/pmBs - it's basically the same as tools on other operating systems like 'apt-cache search'

Title: dpaste/pmBs (Plain Code)

had in one computer, wanted to sell it, to kept win 11 there, then kept the machine, so used it for a month

it was fun I wouls day

unixwitch: I think they want pretty tables with mouse navigation

and pictures

terminal text is boring

unixwitch yes, with 'unreadable' I mean, soem space, empty line or so that would make it easer to at least get the name of the package or so

not just bunch of text

Remilia no that I don't want, never wanted

you window user! don't expect the rest of the world is also like you ;/

terminal plain text is the best

but I am perfectly fine with pkg output

We got someone working on pkg integration into gui: github.com/freebsd/pkg/pulls?q=author%3Aarrowd

Title: Pull requests · freebsd/pkg · GitHub

though I get the recent fixation on uhhh

huge spacing between UI elements

maye one issue is also, that if it shows more packages with all options, it will not even give a space between packages

I mean empty line

hello

so how do you run ipv4-only apps on an ipv6-only kernel? CLAT in libc?

Soni: ipfw supports CLAT, but when i looked at it, it seems to require allocating a /96 to a client which is not how i normally expect clien CLAT to work, so i'm not entire sure what the right way to use it is

(ipfw is kernel, not libc)

unixwitch: wouldn't that require a dual-stack kernel?

definitely

oh ye, that is another issue

right, if you don't have 'options INET' you can never run IPv4 application

ipv6

ntpd and ping complains bc. of that

what does 'ping complains because of ipv6' even mean

ping is not ipv6-compatible

Soni: do you mean you want to remove IPv4 sockets API but still run applications that use IPv4 sockets? ... i don't think libc is right place to support that. you could perhaps write an LD_PRELOAD library to do it

I think I fixed ping taht enabled ipv6, then put there option 'prefer ip4 or something like that

Remilia: ping6 and ping have been merged into one binary

if I disabled some ipv6 thing

I think ping tried ipv6 anyway

meena: oh

Remilia: ping does IPv6 now. traceroute still doesn't

so failed all the time

unixwitch: where else would you implement it?

to get hostname

Soni: me personally? nowhere, i'd just enable 'options INET' to get the existing IPv4 sockets API.

ip from hostname, dns

it should not try ipv6 if you did not enable ipv6

ye, it should

not only it try, it does only that

Soni: this seems like a very niche feature, so i don't think the increased size of libc from adding it it would be worth it. but i am not a freebsd developer, so my opinion doesn't really matter, i guess

unixwitch: well it's more that having the ipv4 stack is a potential vulnerability on ipv6-only networks

no idea why, if there is ping6

but okay that is fixed, now probably have some nonsense in ntpd which try ipv6 on a server

morena: once again, nothing will try ipv6 resolution first if you do not enable IPv6 on your interfaces

morena: someone else had this problem earlier and it turned out to be caused by unusual ipv6-related settings in rc.conf - have you set anything like that? ip6addrctl_policy for example

Remilia it does

ping

unixwitch ye some setting,w hen I disabled one ipv6 related option

so not only I have to enable it, but then also add some option to sys I think

morena: i would suggest just removing that. if you don't have IPv6 (= no v6 addresses assigned to interfaces) everything will just use v4 by default, including ping

ip6addrctl_enable="YES" when I disable, then ping does not work

as it fail withi ipv6

so when I enabled, I probably had to add also ip6addrctl_policy="ipv4_prefer"

ip6addrctl_enable="YES" is the default

that first option was enabled by default

ye

so I disabled it, then ping did not work

ip6addrctl_policy defaults to AUTO which is ipv4_prefer if no v6 interfaces are configured

or part of it, to get dns

i tested this last time it can up and could not reproduce it, removing routable IPv6 addresses made ping default to IPv4 and it worked fine. could you share the contents of your rc.conf and the output of 'ifconfig -a', as well as the ping command and the error output?

unixwitch okay, so should I try again to ip6addrctl_enable="NO" if ping will work?

you should remove that variable altogether

okay

morena: remove that option entirely, and ip6addrctl_policy and any other ipv6-related options, then reboot and try again

okay

may I boot again ;/

there should be no 'ifconfig_..._ipv6' or 'ifconfig...="inet6 ..."'

yes, works now well

no idea what I had before

probably I just removed service or so and did not reboot or so

you're the second person to have this issue and although i think the correct fix is to not set these unnecessary options, i don't actually understand why it breaks things. i might have a poke at that

but that ip6addrctl_enable= was enabled I think by default

at least I saw it in bsdconfig I played it

maybe I just hit that optoin by mistake

ye, in bsdconfig it showing it's enabled

as I said ip6addrctl_enable is enabled by default and policy is set to AUTO

and it will use v4 only if you do not have ipv6 configured

okay

oh is this an option in bsdinstall, that explains why people keep setting it.

so it is not in rc.conf but it is enabled

it is in /etc/defaults/rc.conf

which you should not touch

/etc/rc.conf overrides defaults

cgit.freebsd.org/src/commit/?id=df53ae0fdd98 never seen "X-MFC" before

Title: src - FreeBSD source tree

Remilia thanks dear Remilia ;/

that explain it

but when disabled, then it can make trouble

I assume

instead of assuming you could take a look at /etc/rc.d/ip6addrctl

true

but luckily I don't need more to check files

I will just come and ask Remilia

Dont remember who the kind soles from yesterday was. But big thanks for the crash course intro to building the kernel. Managed to add support for the serialport device now! So big thanks, this realy helped my OCD =D

is 'poudriere ports -u -Bwhatever' supports to work? it seems like it just silently does nothing, i wonder if -B is only for initial checkout

s/supports/supposed

only initial checkout

Are Intel Atom processors x86-64?

CrtxReavr: usually but not always, i think. you can check on ark.intel.com if you know the model number

Looking at the C3758

intel hasn't made what they call a Intel 64 processor since.. some time in the early 2010s

I'm not talking about IA64.

neither am i

Intel 64 is the brand name for x86_64

which was originally called EMT64

In FreeBSD speak, I'm talking about amd64.

CrtxReavr: ark.intel.com/content/www/us/en/ark…c3758-16m-cache-up-to-2-20-ghz.html says it does 64-bit. (there's another part with the same marketing name that also does 64-bit)

Title: Intel Atom Processor C3758 16M Cache up to 2.20 GHz Product Specifications

right, anything you're likely to be able to buy will be a 64bit x86 architecture and freebsd amd64 images will work

ah, the other part is C3758R

Looking at this guy: servethehome.com/the-everything-fan…-nas-appliance-qotom-qnap-teamgroup

Title: The Everything Fanless Home Server Firewall Router and NAS Appliance

peripheral device support might be less stellar

at least if the NIC is supported it's probably not going to be a realtek :-)

for example, what's the Ethernet PHY/MAC

net/realtek-re-kmod supports everything shipped by Realtek, it's just a question of getting it installed

copper ports are Intel i225, SFPs are Intel X553, according to that article

most people got scared off by the C2000-series developing a fault that caused the CPU to fuse itself

it was pretty widespread (not sure it was every C2000 ever sold, but a _lot_ of them were faulty)

Well, I'm sure "Intel" and "Fanless" was a painful engineering challenge to overcome.

CrtxReavr: what about something like aliexpress.com/item/1005006017166158.html ?

"ALI Express" does not inpspire confidence.

there are reputable sellers on aliexpress, although no idea about that particular one

router for the price of desktop ;/

I'd need something with at last one SPF+ port.

(To upgrade my Google Fiber to 2 Gbit.)

can you use your own SFP with Google Fibre? i wish we could do that, but none of the FTTP providers support that over here, you have to use the copper port on the ONT

unixwitch, not exactly. . .

The 1 Gbit service fully supports "BYOR," but the 2 Gbit requires you to use their router. . . *BUT*.

IT's jsut a custom SFP+, so people have had success pulling the Google SFP from the router, and plugging it into their card.

Some have even managed to get 2.5 Gbit peformance out of it.

There's also an ONT upgrade involved with the 2 Gbit, of course.

ah that sounds even worse than here, even if we have to use copper it's at least supported to use your own router

They're even supposed to be offering 20 Gbit here soon, though at $200/mo, I'll prolly not be jumping on that.

Plus, it's so very rare that my 1 Gbit pipe gets saturated. . . what am I going to saturate a 20 Gbit pipe with?

you can offer connection to whole village ;/

Actually, I could talk to my adjacent neighbors about sharing the 20 Gbit, so long as I control the router and the public v4 IP.

20Gbpa and you only get one IP address? that seems stingy

It's a residential connection.

we have a /28 on our residential connection :-d

Hell, most ISPs are doing the cgNAT bullshit, where you don't even get a public v4 IP.

unixwitch, is that in the 100.64.0.0/10 range?

no, our ISP doesn't use CG-NAT at all

(they are more business-oriented, but our service is explicitly sold as 'home' package... it's not uncommon to be able to get at least a /29 from residential ISPs here)

All I've ever *NEEDED* at home, is one public v4 IP.

'Course. . . I love me some v6.

Despite Google's warts, Google Fiber really is a great ISP.

You get the bandwidth you're promised, it's crazy reliable, they don't block shit.

did i see they've stopped deploying it in new locations now? i wondered why that was

I could run SMTP on 25/tcp if I was inclined.

unixwitch, they paused new deployments a few years ago, but they've resumed.

[15:56:46] <morena> router for the price of desktop ;/ <- yes that is pretty cheap for a router

They had a lessons to learn when they first got started.

Remilia, well, if it says "Cisco," "Juniper," Et al. on it, then yes, a router for the price of a desktop is quite cheap.

<CrtxReavr> I could run SMTP on 25/tcp if I was inclined. <- and yet you would be unable to deliver mail to a whole lot of hosts, unless you use smart relay

Though. .. $400 for a home router is. . . pretty pricy, depending on your needs.

oh cgnat nonsense, that's reason I can't host home, sad

I had some mikrotik hap ac2, so at least was able to put openwrt there

does Google Fiber let you set your own PTR / does it delegate v6 reverse?

Remilia, no but it does assign you a fairly static /56

means you can only receive over SMTP

not send

Remilia, depends, plus. . . I'm not looking to. . . I just know that I can open and receive connections on 25/tcp, which is bloody rare for a residential connection.

Remilia, though. . . I can (and have) run an HE tunnel on my GF connection, which gives me full DNS delegation.

The only issue I ever had with an HE tunnel is with Netflix.

They block HE v6 space, as they consider it a means to bypass country content restrictions.

it does not really matter because 90% of the email traffic goes through v4

and most SMTP servers will not receive from a host on an assumed-residential line without a PTR record

hmm, deskutils/xdg-desktop-portal seems to be broken, wonder if it's missing a dependency: meson.build:106:17: ERROR: Dependency "gdk-pixbuf-2.0" not found, tried pkgconfig

unixwitch: bugs.freebsd.org/bugzilla/buglist.c…rtal&short_desc_type=allwordssubstr zarro boogs

Title: Bug List

meena: bugs.freebsd.org/bugzilla/show_bug.cgi?id=276100

Title: 276100 – deskutils/xdg-desktop-portal: add required gdkpixbuf2 to USE_GNOME

strangely, i'm sure i built this port last week without this patch... perhaps i changed the build options on something else that stopped it from being pulled in properly

it does seem to be needed though, /usr/local/libexec/xdg-desktop-portal-validate-icon links to it

Anyone know how to set pourdriere to use an n amount of CPU cores? I can set the max amount of jobs, but for ports to use all available cores I want to limit it to n cores.

weust: i'm not sure there's a way to do that. you could use -J1 and set ALLOW_MAKE_JOBS=yes (so each package uses all available ports), then set MAKE_JOBS_NUMBER= in make.conf if needed, but some ports are still broken with ALLOW_MAKE_JOBS

weust: did I understand you right you found some port that doesn't respect MAKE_JOBS_NUMBER and just starts more jobs instead? that sounds like a problem that should get reported

unixwitch: Can try that. Thing is, I have set PARALLEL_JOBS=8 (8 core CPU with HT) and I don't mind when (for example) 1 job is running is can take 8 cores. I just want the other cores to be "free" so the system is still responsive

nimaje: no, no. not that

weust: what you want is for poudriere to know that one porting is using 7 cores, so another should only use 1, or whatever, right?

s/one porting/one port

(this is the thing that i don't think is possible)

i use PARALLEL_JOBS=4 and MAKE_JOBS_NUMBER=4 which seems to be a reasonable compromise on an 8-core desktop, but it does mean the actual number of running processes varies a lot over the build

but the default for poudriere is to build one port per core and limit that build to one job, so a build shouldn't take mor than one core (?)

for example, I have ALLOW_MAKE_JOBS_PACKAGES=electron*. it will use all CPU cores it can that are free. So say it's the only job running, it will use 16 CPU cores.

yeah. i think it *should* be possible to have to do something like this by looking at ALLOW_MAKE_JOBS_PACKAGES, but currently i don't think it can do that

So even when there are 8 parallel jobs, it still in that case still use 16 cores

OK

set MAKE_JOBS_NUMBER

nimaje: the problem is if you do a build with 1 core per package and you're building electron, or firefox, or something, your build ends up taking 10 hours for that one package while everything else finishes in 1 hour

exactly

so the fix for that is to use ALLOW_MAKE_JOBS_PACKAGES, but poudriere will consider that one 'job' and run it alongisde the other 8 packages, which can result in a *lot* of running cc processes

but the use case was limiting the jobs of some port in ALLOW_MAKE_JOBS_PACKAGES? MAKE_JOBS_NUMBER should do that

i'm trying to install immich in a jail. the logger is spitting escape and color codes into the log rending it very difficult to read. is there magically a way to tell the daemon to ignore them? i'm guessing no, but asking in case there's some magic i'm unaware of also, i will be afk so i will read any replies later. many thanks

the issue is limiting the concurrency across all running package builds, not inside a single build, which is what MAKE_JOBS_NUMBER does

nimaje: no, I have limited the amount of parallel jobs already, to 8.

nimaje: for example, say you set ALLOW_MAKE_JOBS_PACKAGES="firefox thunderbird electron* llvm*", and you set MAKE_JOBS_NUMBER=4, then poudriere ends up building those 4 packages at the same time, you now have 4*4 = 16 jobs of concurrent when you only wanted 4-8

(assumimg PARALLEL_JOBS=4)

yeah, basically I want an option to limit poudriere to use 8 cores. more as a global option.

if that makes sense

so the fix i can see for that is, if it starts building firefox, it should know that build will use 4 cores, and not schedule any other jobs until it's finished... but i have no idea how hard that would be to implement

ah, yeah, poudriere can't limit that, that's why I let it build two ports and have MAKE_JOBS_NUMBER=3 which leaves my desktop somewhat useable even if two heavy ports build in parallel

Where is MAKE_JOBS_NUMBER= set? I don't see it in my pourdiere.conf. perhaps it's an older version?

iirc in a -make.conf

aah

yes, in poudriere's make conf, so /usr/local/etc/poudriere.d/make.conf usually

(i think there are per-jail make.confs as well, but that one applies to everything)

I don't have a make.conf there, only the one for the jail. currently I only build for 14.0 anyway, so that's fine

you can put it in the jail-specific make.conf too

just doesn't work in /etc/make.conf (afair) because poudriere ignores that

understandable

I've set MAKE_JOBS_NUMBER=8 and will see what happens

I had a jail running on this server, when I first build it, with a DNS server. I couldn't even query the DNS server when it was using 16 cores at 100% :-)

hadn't seen Ctrl-T show TMPF, CPU% and MEM% now. cool

it grew those in one of the recently updates, it's quite handy

indeed

still used all 16 cores at 100% or close too it

it'll be using up to PARALLEL_JOBS * MAKE_NUMBERS_NUMBER jobs, so you probably want to reduce the former if you set the latter to 8

ok

if you like living on the edge, you could try something like PARALLEL_JOBS=1; PREPARE_PARALLEL_JOBS=2; MAKE_JOBS_NUMBER=8; ALLOW_MAKE_JOBS=yes but this has never resulted in a functional build for me. i wonder if there's a way to disable make jobs for only some packages...

just for my sanity, PARALLEL_JOBS= means JOB = CPU core? or CPU thread?

it's just the number of concurrent jails to run (so, roughly, number of processes), i don't know if it's better to consider number of cores or number of threads when setting it

iirc, and this has been several years since I set it up, parallel_jobs is the amount of ports to build at the same time. because, again iirc, I had it set to 16 at first but then set it it 8.

so yeah, amount of jails

right, that's what i intended to say :-)

you mentioned jobs before which sounded like CPU core to me :=_)

:-)

maybe i should go back to using ALLOW_MAKE_JOBS=yes and submit patches for the things that don't build

Now it's only doing 1 port/job at a time

and using only 1 core during build

if you used the settings i mentioned above ^ then yes, but it'll use make -j for all of them. it's probably slower for small ports due to the overhead of configure, etc

but MAKE_JOBS_NUMBER= = make -j, right?

yes

Then it's ignoring the jails make.conf option for it

but if the compiler processes are very fast (which is common for C code) it may not be able to spawn them fast enough to use 8 cores at once. it's more helpful with c++/rust

aah ok

so something like Electron or Firefox

yeah

Then I will experiment with parallel_jobs=4 and make_jobs_number=4

just to get a feel for it

I have my poudriere bulk run once a week from midnight onwards, but if decide to move some jails to it from my other server it would be nice if those jails were actually still usable

it just did ffmpeg, which is in allow_make_jobs_packages, and with the three other jobs running CPU core usage was alright

hi all! Anyone know if FreeBSD exports libfetch to MacOS or Linux?

MacPorts has libfetch

hm..need to figure out how to install it :)

new'ish to mac

Have you checked macports.org ? Using `ports` on Mac is kind of like using `pkg`. Except different.

Title: The MacPorts Project -- Home

At least, once installed.

(No built-in bootstrap.)

Also, NetBSD packages includes libfetch and that installs on just about any OS.

s/`ports`/`port`/

Got it installed!

Wondering how to proceed here...

`cc file.c -o file -lfetch` tells me `ld: library 'fetch' not found`

fikran: use '-L/opt/local/lib -lfetch', but you want to be using something like autoconf, cmake or meson to find dependencies like this automatically

thank you unixwitch !

first time booting Windows (client) in vm-bhyve, surprisingly hassle-free other than needing a VNC client

now all I need is a Makefile that will build a library...

unixwitch, /opt/local is an odd prefix.

Why not /usr/opt?

somewhat, yes, but it's where MacPorts installs by default for whatever reason

Strange.

/opt/macports would seem like a better choice, rather than /usr, although /opt is more of a System V thing and not often found on BSD...

/opt is normally for commercial stuff you're paying for, whereas /usr/opt is common for stuff built/maintained by hand.

I commonly maintain a /usr/opt/ prefix on workstations, or application servers.

Keep stuff separate from ports/packages that I've built by hand, or customized in any way.

well, freebsd and macos hier(7) don't mention either /opt or /usr/opt. solaris filesystem(7) says /opt is for 'unbundled application packages', which would seem to cover macports, but of course macos isn't solaris, so...

unixwitch, I don't think that would forbid you from using it.

homebrew went heavy for /opt in response to macos increasing its data protection on /usr

HELP!!!

HELP ME!!11!!!

experemental: what's wrong?

experemental: You would need to ask your actual question first.

THEYRS COMING!!!11

HELP!!!

I never found a Mac Makefile, not sure how to build a library on MacOS. Wrong channel, I guess

drugs are bad, mkay?

LINUX 6.0 IS COMING!!!!!HELP!!!HELP E!!

even debian is already on linux 6.1

unixwitch: I suspect MacPorts picked up /opt from Solaris from the time MacOS 10.5 went for officially counting as Unix.

is there a way to disallow pkg from updating a port?

audacious-plugins: 4.3.1_1 -> 4.3.1_3

ports should be handled by ports and pkg by packages

packages are just the result of building a port

pkg lock I guess

thanks nimaje