byakuren, just realized i was starting "cat truss.log | grep $(basename $PWD) | cut -d '"' -f 2 | sort | uniq" with cat but you started it by grepping open. i guess you did that because that limits to files whose contents are read (rather than just a stat or smth) right off the bat, right?

kerneldove: yeah exactly, i included the open query in that initial grep - plus cat-ing to grep just uses another process and pipeline in your os, so you might as well do it directly

are there any other syscalls that a proc can use to read the contents of a file than *open* ?

mmap and friends, maybe?

well you've got open* (open, openat...etc), and i guess mmap in case for some reason the process memory maps the file then reads the memory - all other ones i beleive read meta data of the file, like *stat (fstat, lstat), acccess and so on

hmm. so i guess to be safe i should cat (include all dirs) OR grep open + mmap?

grep -E ': open|: mmap' truss.log | ...etc

though a lot of the mmap output seems less than useful, as it's not file names per say - so i think you'll be fine not including it

(looking at the cargo build truss log specifically that is)

interesting. grep -E ': open|: mmap' truss.log | grep $(basename $PWD) | cut -d '"' -f 2 | sort | uniq eliminated the group of relative urls?

oh so did grep open ...

those relative paths were a __readlpathat call

rather than open

ahhh!

s/__readlpathat/__realpathat

i'm trying to track down an issue with the command and it seems grep \.rs matches some lines that don't have ".rs", am i doing it wrong?

try grep '\.rs' quoting it

the . is likely being escaped by your shell, and sent to grep as a normal dot, which is then treated as a singular wildcard

quotting it in a string like that should ensure grep gets the ecaped dot

sigh that worked lol ty

:p np

ya i think there's something wrong with the grep -E ': open|: mmap' truss.log part, it doesn't display as many .rs files as when i start it with cat truss.log ...

i double checked truss.log, and plenty of them *are* openat lines, not just fstat or something irrelevant

You can always test what the shell is doing with quoting by using echo to echo print out the result. Try "echo \." and we see only a single dot. "echo '\.'" shows that then it is "\.".

oh that's really cool ty

And for deep inspection I always use od as in "echo \. | od -tx1 -c" and then one can play with the quoting and see exactly what is being done there.

You can also use "wc -l" to check your impression that there's more occurences of one than the other.

my truss.log has an openat with a path including interface.rs, as i can see with grep -E ': open|: mmap' truss.log | grep interface, but when i run grep -E ': open|: mmap' truss.log | grep $(basename $PWD) | cut -d '"' -f 2 | sort | uniq | grep interface, there's no results ??

Probably lost it in the cut step at a quick guess in the dark. Start with the first command, pipe it to |less to see what it produces. Then add in the next stage in the pipeline. And then the next. Along the way to the final form you will see the problem.

may be the grep $(basename $PWD) is filtering that line out - as that contains the name of your directory, and if its a relative file that may not exist?

it's a relative path ya

Also you can avoid a process with "sort | uniq" by using "sort -u".

that's a shout! always forgot about sort -u - thanks for the reminder for me as well c:

ok ya i took out the basename part and it's there now, but so is a lot of other noisy build dirs hmm

maybe i can run the truss cargo build from my home dir so there's no relative dir that i can't filter on

nope even if i build from ~, the dirs are still relative damn (must be relative to repo's Cargo.toml)

man this is weird, i added --release to the truss cargo build command and it kept failing midway through that it couldn't find thread lol

there's a lot of noise to wade through going this truss method. what if i enabled atime in zfs so i could use the other method? find . -type f -atime -5m

yeah that could also work

how much more will it work my nvme mirror?

It would also be possible to create a zfs dataset with atime enabled and use it for whatever you are doing.

zfs create -o mountpoint=/var/mydata -o atime=on zroot/var/mydata

could i create the dataset within a dir in my ~ or would that be weird?

that's fine - you can just remove it when ur done with it no worries

i made an 'atimebuild' dir under /var/ just to be safe. now i'm waiting for some time to pass then i'll test!

You can make datasets anywhere you feel they are useful. It's also reasonable to give yourself permissions to create, modify, delete datasets in your home directory and many people do that too. Then you can do those actions as yourself without needing root.

can you make datasets within other datasets? because my home/dir is already a dataset of its own

Yes. That's typical. If you "zfs list" you will see that by default there are already nested datasets.

zfs datasets are light weight. People create them for any reason. The only thing to know is that they are different file systems and therefore _moving_ files to and from datasets is a data copy since it must copy data from one file system to another file system.

i wonder if the mv vs copy thing is why when i moved the repo dir i was working on from within my home into /var/atimebuild/ there were build errors. i rm'd the dir, cloned the repo fresh, then i could build

and btw the find command dumped into a file.txt and now i can see exactly the files touched during build within the repo!!!

tyvm byakuren and rwp

Hooray \o/

Glad its working, and you got your data c:

ty

to find the changes between versions i guess i do this process for both versions of the repo, then run diff using the list of changed files?

kerneldove, To find what files have changed ask the zfs file system with "zfs diff". Take a snapshot before the build. Then build. Then zfs diff against the snapshot and it will tell you exactly what files changed. It's fast because the file system knows what files changed.

but i'd still need atime enabled for that right?

rwp

kerneldove, zfs knows what files changed regardless of atime. atime will tell you which files were accessed even if they were not changed. But you said "list of changed files" and I thought of zfs-diff.

oh... i'm trying to identify which files in a repo are used in the building of a bin in the repo

that doesn't actually change any of them... except for using atime

tried using truss earlier and it probably worked fine but the log was really noisy (lots of files outside of the repo were included)

So... After you know what files are used during a build... What are you going to do with that information?

limit the diff between 2 versions of the repo, so i can know what's changed from the /relevant/ files, aka the files pertinent to building the bin

because the repo has files (code) for other bins that are irrelevant

rwp

In OpenBSD I can put `pkg_add -ui && syspatch` in a crontab, and it'll automatically update packages/ports and system files. But in FreeBSD I feel like that's not possible since you need to confirm with a "y", but could it be possible to do something like `pkg update && pkg upgrade && freebsd-update fetch` in a crontab, and have the server output the results in my mailbox?

Hi guys, I'm having a lot of trouble with my re0 ethernet device on FreeBSD 14.3. No matter what I do, I cannot establish a connection. I've double checked that it workes on other OS's, so this has to be a FreeBSD issue. Has anyone else seen re0 driver issues..?

is there any way to add more swap space to a system with zfs?

remiliascarlet: read the man pages, freebsd-update has the cron subcommand that is meant for that. Also, setup periodic to check for package updates available, and mail those to you. Preferable in my opinion to running unattended upgrade, but itlf that is what you really want, you can do that to.

I have a lot of BSD servers to manage, so preferrably instead of manually logging into each one of them multiple times a week just to update, I would rather do this automatically, and then read the output with the `mail` command for supervision.

pkg upgrade -y, is what you're after then

freebsd-update cron, will download the update if available and send you mail

I see.

I'd still use the notify approach if I were you, and write an ansible playbook to run the upgrades in parallel

kerneldove: you can, but you need unpartitioned free space on a drive to add or grow an existing swap partition

you can ask dvl, how to swap on zroot

If it was ufs, you could get away with a swap file, but that is discouraged on zfs. Don't know the exact reason

All my FreeBSD servers are ZFS.

It's overkill on anything other than the NAS, but I just like ZFS.

I think the advantages are worth the penalty, I use it even on a tiny vps. overkill or not, it's just so convenient

kerneldove: found it. Swapping to a file on zfs is simply unsupported. What is possible is swap on a zfs volume, but it could lead to deadlocks: openzfs/zfs #7734

Dates back to 2018 ^

dansimon: I've had issues in the past with Realtek NICs. If the "re" driver doesn't cut it for you, you might have better luck with net/realtek-re-kmod from the FreeBSD-kmods repo

swap best to ufs partition. 2nd best ufs swapfile (both ssd). 3ds best partition spinning rust. 4th best spilling rust ufs swapfile. zfs swap there be dragons

i have such a headache with if_re

divlamir: thanks, I'll give it a try :)

electron37 took 13h06m43s on 3.5ghz i7 parallel_jobs=1 allow_make_jobs=yes

probably faster with more than 1 job

f451: do you have ccache set up? that really helps with electron/node/llvm updates

not the first time obviously, but for updates

other problems happen though, like it'll hit swap or run out of resources elsewhere. this way allows me to keep using the machine this poudriere instance runs on. i'm not complaining about the time it took, not really. electron is rather large

ivy: yeah - ccache4 is the ccache

parallel_jobs=1 and allow_make_jobs means there are 4x llvm20 jobs running, load is btwn 4.00 & 5.00

but it completed and im happy about that :D

Can you potentially set a better nice value on it? I've done that to a bunch of cargo builds that would normally use 100% cpu and make my music skip, with a very high nice value it works much better and I don't get skips

byakuren: have to admit ive not looked or thought about re-nicing.

give it a shot!

:D

renice is for already-ruining commands

So just a nice -n 20 $COMMAND

i'm surprised there isn't a poudriere.conf option for that

(since you say it's very low priority - 20 is the lowest, maybe a 10 is a good middle ground though?)

i mean you can just run poudriere bulk under nice, but...

10 is the default priority

no, maybe the man page is hard to read

yeah - nice $CMD gives it a nice value of 10

it's a bit confusing that nice is separate from priority, but nice(1) calls the nice value the priority

"The lower the nice value of a process, the higher its scheduling priority."

it took as long as it did because i have it in my tmpfs blacklist (for wrkdir).

i think thats the most likely reason

nice 30 poudriere bulk blah .. yeah might try that

thanks for the suggestion

if you're running zfs - it'll also be using your ARC (zfs ram cache - up to 80% of free ram by default) quite a bit, which might be exausting (or making you think you're exhausting) your ram

20 is the most nice a procss can be c: - but it clamps the value within the acceptable range for ya

i have 2x poud instances here - the older one uses zfs. im using ufs on the other one exactly to eliminate arc pressire

20!

ok ill remember that

im really pleased rn its not hitting swap (thats on the ufs disk - the swap partition)

neat!

it seems like you have to create a port tree in poudriere to use as the overlay, which is less useful than i was hoping, i just want to testport from a temporary git worktree...

use -O to reference it iirc

right, but i was hoping -O would just take a path

if i have to create a ports tree in poudriere first i may as well just use that as -p, i guess overlays are intended for something else

poudriere bulk -j desktop -O sccache -f ports.txt

ah

no not that i kbow of - you need poudriere ports -c -M path -m null

that seems to work, thanks. i suppose i should write a script to do that

poudriere ports -c -p sccache -m null -M /usr/local/share/sccache/overlay

ivy: is yr poudriere the regular one or the -devel

f451: i always use -devel

yep

i dont think the regualr one supports overlays

yet. might be wrong there

I use /usr/ports/ports-mgmt/poudriere-dsh2dsh because in there the rebuild-rust-needlessly problem is fixed

ty divlamir

huh, i didnt know this: 1.2.p1 is *before* 1.2

f451: err, usually not. what software is versioned like that?

1.2-p1 is *after* 1.2

im looking at the porters handbook, example #5

section 5.2.2

pkg-version -t

that doesn't seem right at all, i wonder if there's a historical reason for that

i've never heard of .p1 meaning "prerelease"

yeah

.pre1, sure, that's quite common

then again - freebsd releng versions go 14.3-p3 not 14.3.p3

maybe it's the dot

in pkgbase the version is 14.3p3

which works because of rule (6)

yep

i do wonder why we don't just call it 14.3.3...

we used to have 3-part version numbers and they dropped around 3.x for reasons i'm not really clear on

i guess it might be confusing if you had 14.3.3 installed but uname -r said 14.3

(but people are already confused about why uname -r doesn't say 14.3p3, so...)

yeah i had the cds from 2.1.3 - remember Walnut Creek i wonder if they're still going

What does the p in -p# stand for? Part?

Macer: patch

Ah ok.

Macer: because historically the updates were literally released as source code patches, although nowadays you can also get them as binary updates from freebsd-update or pkgbase

ivy: same. basically if you're managing systems exposed to the internet it's a balancing act btwn the latest patched software and breaking things

f451: i meant more the releng process, rather than using releng releases :-)

ahah

although currently all my systems run main... i think i might switch to releng/15.0 and just backport patches i want

no stable/15 ?

if i was going to run stable i'd just run main, the idea is to make regular maintainance less exciting every time

heh

have to say, pkgbase is scary. i can think of a couple of situations it would be useful to get a new install running or tailor an embedded install, avoiding buildworld etc. On the other hand, single point of failure (pkg)

then again, some embedded - it'd be worthwhile to build on the embedded system for other near-identical ones to get the best out of the os and just dd the image to other instances

thats what i did for a few armv6 and v7 systems a few years ago

github.com/freebsd/pkg/blob/main/libpkg/pkg_version.c#L151 - this is the file and code for the -p1 type suffixes btw

so you can get an idea from the commit history about the reasons behind it all c:

f451: i really think we should import pkg to base, but there is some resistance to this right now due to the rate of development

yeah

i'm not sure it makes sense that the officially recommended way of building freebsd requires a tool we don't provide

LOL

also, that would let us do pkgbase cross-builds from Linux and macOS which currently doesn't work

(surprisingly, a lot of people rely on this for both development and production use)

i wish there was a gitlite in base as there used to be a svnlite years ago

we will probably import got into base eventually

I find it bizzare that FreeBSD allows this, but whatevs?: bpa.st/PJVQ

ive used got before, it's not a drop-in replacement though. i think by default it makes a shallow clone

why would it not? utf8 is utf8

CrtxReavr: the hostname is just a sequence of bytes. is there an advantage to adding code to disallow that?

f451: it is a drop-in replacement in the sense that its on-disk format is compatible with git, so you can clone with got and switch to git if you need something fancy. the UI might be a bit different, i'm not sure, never actually used it

Well. . . I can say that BIND doesn't allow it as a hostname. . .

Like I said, "whatevs."

f451: but got would be perfect for cloning src, ports, etc. after installing the base system, which you can't do right now, and we can't put git in base

yep

CrtxReavr: the system hostname is not required to be a DNS hostname, the system might not even be connected to an IP network at all

CrtxReavr: also you can use that as a hostname in BIND, but you'll have to convert it to punycode format first

Also: bpa.st/G6YA

xn--whatever would be a silly hostname

rtprio: right. but it should work to set a Unicode hostname with a punycode DNS record... i have no idea if our resolver even supports IDN though, but like in principle, it should work

(i don't think anyone should actually do this, but...)

It's supposedly in the works to allow unicode TLDs and domains.

CrtxReavr: in the works? it's been live for years

CrtxReavr: that's what i mean about converting the hostname to punycode, that's the way DNS encodes these non-ASCII hostnames

How show us a domain name example.

Now

Hi

o.O

This what yoiu mean by "punycode?": bpa.st/2TLQ

no, punycode names begin with 'xn--'

I'm ok with browser parsering those (and showing punycode), but I'm surprised that my terminal is parsering them correctly as URLs

lts: did that actually paste correctly? it's missing some characters here, i don't think my font has the full CJK set

but it does work in Firefox

Yeah, alacritty detects 人民网.中国 perfectly

(I don't think "parsering" is a word.)

"Parsing" sounds & looks correct.

xfce terminal doesn't. It stops at "www"

V_PauAmma_V, that could be an issue with the fonts you have installed.

And your terminal configured to use.

it should look like this: people.freebsd.org/~ivy/tmp/idn.png

"Parsing" indeed would be correct. Sorry, English as definitely-not-first language here

No, it displays properly, just isn't recognized as a URL.

ivy, am I miss-remembering, you don't you normally have a capital letter in your nick somewhere?

CrtxReavr: i never capitalise my IRC nick, if that's what you mean

huh

Indeed, a quick grep of my logs confirms that. . . I miss-remembered.

i actually never capitalise anything, except commit messages because people complain if you don't do that :-)

also sometimes work emails if it's to a client

ivy saves their pinkies for nose picking.

mintty doesn't seem to have an issue with that URL with it URL detection.

WezTerm doesn't like it, but i think that's because of the missing characters, i wonder why it doesn't take them from an alternate font... perhaps i don't have a proper monospace CJK font installed

What kinda TLD is that anyway?

it's the IDN version of .cn

en.wikipedia.org/wiki/.cn (para 2)

also icannwiki.org/.%E4%B8%AD%E5%9B%BD which is where i found that URL

actually there are two IDN versions of .cn apparently, one for traditional and one for simplified... i find that surprising since i thought PRC hated traditional script

I took a Chinese history class in college. . . prof was a middle-aged, white, American male. . .

He lectured that he thought China saw its best and longest period of peace and prosperity under the CCP.

I always wondered if he actually believd that, or if he was just trying to keep in the CCP's good graces so he could continue to travel there for research.

This is a very #freebsd-social discussion.

LxGHTNxNG: is it though? the channel is quiet, the discussion started here, there's no immediate need to move it because of the topic police

What ivy said.

Maybe LxGHTNxNG is lonely in #freebsd-social?

I am a god, and I rely on belief for my continued existence /s

hhe... It's been a long time since anyone invited to #social

there, i did one to CrtxReavr for old time's sake

i don't expect they'll join

#freebsd-social is pretty dead recently, i'm sure it used to be (a little) more popular

maybe everyone's on holiday

I've been on IRC since '91. From my perspective, the first rule of IRC, is that it's supposed to be fun.

Not a fan of topic-zealots, 'specially when the chat isn't interferring with people getting help or causing excessive scroll in the channel.

+1 to above, if there is no other active topic then redirecting discussion to a small channel is a net negative effect

Any good movies or tv shows to watch?

:))

lol

That’s a good one to watch.

zi, no lurking. 8-P

i didnt want to upset someone by saying something fun!

Make sure you don't have fun, either.

i try not to

If someone has fun, we can always redirect them to #freebsd-social

'fun' sounds like it should be a freebsd.org username, but it is not

lts: is that like telling someone to sit in a corner? :)

TommyC: youtube.com/watch?v=hMeO4qAG8WY

ivy, how are you enumerating the usernames?

grep fun /etc/passwd

zi: #freebsd-social is definitely a corner

CrtxReavr: i logged into freefall and ran 'finger fun' and it said: finger: fun: no such user

finger fun!

ivy, you have a .plan file?

CrtxReavr: as if i'd be that organised, have you met me?

freefall doesn't allow remote finger anyway, which i think is a shame

echo Wheeeeee! > ~/.plan

ivy has No Plan

zi: accurate

fun fact, you'lk get the same error if you finger me

is it .alias or .aliases ?

yamada: for mail forward? you want .forward

nah for alias commands

ivy, I think more systems would allow it if there was an option for not displaying a user's last login-from IP.

i miss it in the /etc/profile

which shell are you talking about? i don't think /bin/sh supports any specific file for that, just put them in .profile

ok

or you could create .aliases then source it from .profile with the . command

or modify /etc/profile to source $HOME/.aliases, or whatever, but i don't think there's any out-of-the-box support for that

ha. . . I forgot that I had a .plan on here.

A gleekzorp without a tornpee is like

a quop without a fertsneet (sort of).

oh man, just think of how much fun Mr HAPPY has

i guess not since $HOME/.login.conf already does that

It says "Secure site not available", not sure if it parsered it corectly LOL

sry, must have been in the backlog :-/

is it possible to get hardware support in freebsd for 1 million usd cash ^^

Depends on hardware, but probably yes

yamada: sure. get in touch with a contracter like Klara and they'll happily write drivers in exchange for $$$

Prolly cheaper to give a device drive dev a device and a programmers manual for it.

device driveR dev

hm

i cannot donate my hardware ^^

i need it myself

At a former employer, I had access to a lot of programmers manuals for a lot of devices, which I quite under-handedly gave to any FreeBSD driver dev that I could.

i bet in 2 3 years it will be supported anyway

what is the hardware? if it'd widely available, they can just buy their own copy of it (of course that would be included in the price)

Most hardware vendors are not so forthcoming with their programming manuals.

check the gigabyte.com site and pick somethign between 100-200 bucks

Sadly.

yamada: $200 is like an hour of developer time, so buying that would not be an issue :-)

i think freebsd 15 will support the mainboard chipset

:(

i wanted zfs

now i have to stick on kali linux or tumbleweed or cachyos

i'm not sure if you're serious about paying someone to add support, but if not, please report your non-working hardware somewhere (mailing lists, bugzilla) so people are at least aware

i didnt know that exists a list for non working hardware

Not exactly a list of non-working hardware, but there is a list of working hardware.

There's also drivers that are not in the default kernels, that either need to be loaded as a module, or you have to build a custom kernel for.

oh ok

What device and arch are you having issue iwth?

arch amd64 hardware a gigabyte.com/Motherboard/Z890-EAGLE-WIFI7

yamada: what hardware specifically is not supported?

wifi? network card?

Realtek LAN + Realtek Wifi 7 o-0

They don't even give the name of the chip in the specs those buggers

yamada, pastebin the results of: pciconf -lv

CrtxReavr, i can do that next weekend

Well, that's anti-climactic.

15.0-RELEASE spoiler people.freebsd.org/~ivy/tmp/bsdinstall-pkgbase.png

ivy: that is pretty sexy

ivy, meaning the OS can be installed/upgraded with pkg(8)?

CrtxReavr: that's been possible since at least 13.x

the new change here is you can select more components during install

well, i guess bsdinstall support for pkgbase is new in 15.0, but you could still install manually using pkg(8), for jails etc

tl;dr says: RTL8125 2.5GbE

yamada: freshports.org/net/realtek-re-kmod

yamada: wireless is more complicated, see: forums.freebsd.org/threads/mediatek-mt7921-wireless-driver.97165

aic: regarding if_re, YMMV see bugs.freebsd.org/bugzilla/show_bug.cgi?id=289168