<pie_>
i want a tool i can make some kind of machine dashboard out of
<pie_>
with toggle switches for vpn containers and crap
<pie_>
maybe clicky TUI style
<pie_>
so it doesnt die when my x session does
<infinisil>
pie_: Got no systemd service for vpn?
<pie_>
idek
<pie_>
i havent touched my system config in months because im still tryign to free up disk space and ram lol
<pie_>
(and had to spend time learning non *nix things)
supersandro2000 has quit [Disconnected by services]
supersandro2000 has joined #nixos-chat
<aleph->
joepie91: Stupid question what's the idiomatic way to sleep in a loop in node? Tried awaiting setTimeout in a loop but that didn't work, however I do need a way to wait, retry and block till I get one of several values. https://paste.rs/cYk
tilpner_ has joined #nixos-chat
tilpner has quit [Ping timeout: 265 seconds]
tilpner_ is now known as tilpner
rajivr has joined #nixos-chat
c4rc4s has quit [Remote host closed the connection]
<__red__>
Soo. I google for nixexpr cheatsheet
<__red__>
and much to my surprise - I didn't find one.
<__red__>
I'd, you know... offer to write one and all but the face that I couldn't find one gives you an indication as to how thorough my knowledge of nixexpr is ;-)
<__red__>
I'm going through nixcloud.io/tour and making notes so I guess I could make a start on it
<__red__>
if you want to do the pip thing for stuff that hasn't been packaged
<energizer>
"It always "just works" for however the package author is packaging it themselves" my pandas example shows that isnt true on nixos
<__red__>
I mean - if you really want to
<__red__>
Give me an example of something that isn't packaged
<infinisil>
energizer: The packagers aren't packaging it for NixOS
<__red__>
I've never actually used python so this will be a learning exercise for me
<energizer>
every time i want to add a python package dependency it's just `poetry add foo` on ubuntu; on nixos, it's often a few hours or more
<energizer>
(but it's getting faster as more things make it into the poetry2nix overrides.nix)
<__red__>
if you're not intending to make something packageable - you can just use an imperative dev environment
<energizer>
no because there are dependencies which won't be in the right place on nixos
ky0ko has quit [Remote host closed the connection]
<__red__>
I'm going to need an example I think, since every example I give you you disagree with
<__red__>
show me one that doesn't
<energizer>
install pandas on nixos without using nix
<__red__>
how would you install it on a different linux?
<__red__>
using pip?
<__red__>
and when you say install - you mean available in other directories ruight?
<__red__>
which directly contradicts when you said above that it didn't matter if it was installed or not
<__red__>
I mean - my default system doesn't even have python
<__red__>
or a compiler
<__red__>
so I have to start with something :-P
* __red__
waves a magnet over the spinning rust
<energizer>
by installed i mean `python -c 'import pandas'` exits 0
<energizer>
by installed i mean `python3 -c 'import pandas'` exits 0
<__red__>
okay
<__red__>
how would you install it on a different linux?
<__red__>
using pip?
<energizer>
i would use venv as i showed above
<__red__>
okay
ky0ko has joined #nixos-chat
<__red__>
one more moment caller...
<__red__>
Okay - so the problem is that pip downloads binary executables
<__red__>
which make assumptions regarding what is globally available on systems
<__red__>
So - I know wha the problem is
<__red__>
and I know a "hack" to fix it
<__red__>
but I know someone on this channel will be able to better fix it
<__red__>
So - here's the base problem
<__red__>
The package manager downloads binary .so files
<__red__>
when I ldd the .so file, it finds the correct locations for linux-vdso.so, libm.so.6, libgcc, libpthread, libc etc etc etc...
<__red__>
what it doesn't find is libstdc++.so.1
<__red__>
now I can make it work by just adding that path to LD_LIBRARY_PATH
<__red__>
but really - it should be able to find it itself
<__red__>
so thhe question is - what package should I add to nix-shell -p in order to get that library's path into scope?
<__red__>
(or is that thhe wrong question to ask)
kalbasit has joined #nixos-chat
endformationage has quit [Quit: WeeChat 2.9]
waleee-cl has quit [Quit: Connection closed for inactivity]
<__red__>
So you can use patchelf and modify the rpath and then it works
<__red__>
that's because its a library outside of glibc
<ldlework>
What's that meme with the really complicated machine name and the sales engineer like talking about all it's features in opqaue madeup jargon? Turbofrobnicator or something like that.
<cole-h>
(FWIW, that's the first result of the Google search "Turbofrobnicator" :P)
<__red__>
that sounds entertaining
<ldlework>
cole-h: hah
ky0ko has quit [Read error: Connection reset by peer]
ky0ko has joined #nixos-chat
insep_ has joined #nixos-chat
ky0ko has quit [Remote host closed the connection]
<insep_>
samueldr: hey i've heard you hate dtc_overlay as much as i do, have you ever found an alternative for it?
<samueldr>
huh?
<samueldr>
I don't think I hate it?
<samueldr>
and no, not really
<samueldr>
(if I hate something, it's how the kernel pushes the loading of FDTs into bootloaders rather than handling it itself)
<samueldr>
(the loading of their own "true" FDTs)
ky0ko has joined #nixos-chat
<insep_>
18:38 <samueldr> I wish they'd get mediatek to give the source that they are legally obligated to for the dtc_overlay binary in the kernel source tree
<insep_>
18:39 <samueldr> because of this, AFAIUI it can only be built using cross-compilation, or using qemu binfmt x86_64 emulation for that binary
<samueldr>
ah, that's a customized variant
<samueldr>
insep_: is it for a mediatek vendor kernel?
<insep_>
yes
<samueldr>
if so, no I don't know of an alternative
<samueldr>
I'm wondering even _if_ there are changes
<insep_>
:(
* samueldr
thinks
<samueldr>
maybe I thought `dtc_overlay` was something else
<insep_>
because someone is trying to port pmos to some device this kernel
<insep_>
but it obviously doesn't run
<samueldr>
it should on x86_64, but I guess you mean with native compilation
<samueldr>
qemu-user to the rescue?
<insep_>
no, because it's linked against glibc
<samueldr>
ooh, right
<samueldr>
forgot about alpine not being glibc
<samueldr>
HAH, searching for `dtc_overlay` on a popular search engine links to the logs I guess you just found
<insep_>
yes
<insep_>
that's how i found this channel
<samueldr>
:)
<samueldr>
the next best thing I have in mind is to shove this into something like ghidra
<samueldr>
the one from xiaomi I checked seemed to have debug information
<samueldr>
and since it's based on well-known trivially identifiable open source code
<insep_>
or ask someone with access to csdn to find sources for that binary :D
<samueldr>
that's another option, for sure
<samueldr>
I did my due diligence and asked xiaomi, and well... I did ask
<samueldr>
I SWEAR... it's as if I'm trolling myself
<samueldr>
I just tested, and with that I just booted on the xiaomi-begonia
<samueldr>
dtc is the bog standard dtc package
<samueldr>
MichaelRaskin: ^ dtc_overlay might be something that can be totally side-stepped using `dtc`
<insep_>
fun
<samueldr>
obviously, extremely YMMV, if the build fails to boot this is one of the variables to check against
<samueldr>
I could only verify against xiaomi-begonia, with that particular diff on top of Mobile NixOS, so I'm pretty confident that for at least that particular device it works
<samueldr>
btw, if anyone tries to do anything non-standard with xiaomi-begonia, be sure to read the device page to know the BIG caveats; REAL brick risks with that particular device!!
<samueldr>
non-standard meaning anything else than running stock
<insep_>
glory to mediatek, where invalid dts crashes bl
<samueldr>
hm?
<samueldr>
(I don't know what this refers to)
<insep_>
bl - bootloader
<samueldr>
yeah, but the particular crash you're talking about :)
<samueldr>
with xiaomi-begonia, it's xiaomi's ARB (anti-rollback) that is buggy
<samueldr>
at least the ARB being the fault is what I was informed of, it could also be a dumb(er) crash lol
<samueldr>
though it's been observed by android-people that it will do the same pretty-hard bricking with things that would cause ARB to fail
<samueldr>
(though could be a crash too!)
<samueldr>
I was mostly wondering if it was a reference to another similar problem on another device
cole-h has quit [Ping timeout: 264 seconds]
<joepie91>
aleph-: setTimeout does not return a Promise, and so cannot be awaited - the Promise-y equivalent is `Promise.delay`... if you use Bluebird then that is readily available, if you're using ES Promises then you need the `p-delay` module... if you avoid await entirely (as I do), then there's https://www.npmjs.com/package/promise-while-loop + the approach described in the docs
<eyJhb>
infinisil: !! I want that tool!
<eyJhb>
I do that as well all the time using pavucontrol....
<eyJhb>
samueldr: Is there any place, to look for the official kernel source/patches/drivers/whatever for most phones? Wondering if I could got my OnePlus One on a newer kernel
<eyJhb>
insep_: Now I just want to run PostmarketOs ...
<insep_>
we also don't have systemd
<insep_>
and glibc
<insep_>
so that's 3 things already over nixos
<eyJhb>
WEEEEEEEEEEEEEEEEEELL... I run Android, so :p
<eyJhb>
And why is glibc bad?
<insep_>
musl libc makes you look more edgy and gives elitism points
<insep_>
also there are almost 0 proprietary stuff compiled for musl
<insep_>
also there are some additional security features
<insep_>
also it's smaller
<insep_>
also it tries to not extend standard in dumb ways
<eyJhb>
If you go about 3 reasons, then you must give 7
FRidh has joined #nixos-chat
<eyJhb>
insep_: If I used glibc more, or was more into it, then it might be a bigger difference. And so far, systemd is OK :p
<eyJhb>
joepie91: ping
<joepie91>
pong
<eyJhb>
Do you have a minimal CSS library, as in, it does not have a too large footprint, but can make a decent website. Anything that you might have as go-to yourself
<eyJhb>
Currently I can see purecss and milligram
<eyJhb>
I have never really done much styling, but I guess I have to now joepie91 :( It is a sad day
<joepie91>
eyJhb: I've used purecss in the past yeah
<eyJhb>
Seems to be quite popular with many stars on Github as well
<eyJhb>
Very few commits however
<philipp[m]>
Ewww! Github hid the ssh url of the repo in favor of their own gh tool.
<insep_>
that's a good reason to switch to gitlab
<leons>
s/gitlab/!= github
<eyJhb>
philipp[m]: Hid how? I see all three tabs
<philipp[m]>
I only have two now.
<eyJhb>
I like Gitlab, but if one ever wants to run it yourself, it is HUGE
<philipp[m]>
Running gitlab for my job... It's not exactly fun.
<eyJhb>
^^^ Exactly
<philipp[m]>
Aaaah, alright. I'm sorry github.
<insep_>
<philipp[m] "Running gitlab for my job... It'"> at least you can run it :D
<eyJhb>
philipp[m]: Gogs/Gitea? :D
<philipp[m]>
Unfortunately we have a lot of time invested in gitlab-runners already, so switching now isn't really feasible.
<philipp[m]>
I'd go for gitolite+gitweb if I had the chance.
<MichaelRaskin>
First time I restarted our Gitlab-running VM, I decided something fail just because I did not give it enough time to start up
<philipp[m]>
MichaelRaskin: Oh, yes and don't dare to think that this vm can live with only one core because it's just running one dumb webapp...
<joepie91>
philipp[m]: I used to run gitolite
<joepie91>
Gogs/Gitea was a significant upgrade IMO
<philipp[m]>
In what way? I used to run gogs for my private stuff and went back to gitolite because there was less stuff that could break and I didn't have to worry about an insecure webui hanging around somewhere.
<joepie91>
gitolite is kinda rickety all over the place, and if you do want some sort of web interface for it (which I do), then Gogs/Gitea is going to be a much much better choice than some ancient fossil that's written in C, barely reviewed by anyone, and very likely full of memory vulns
<joepie91>
which AFAIK is pretty much the extent of your options when using Gitolite
<joepie91>
(and even then that only gives you a read-only view)
<joepie91>
gitolite is interesting as a sort of minimum-viable git hosting thing, but it really isn't a replacement for something like github IMO
<eyJhb>
Gitolite is Perl
<eyJhb>
ALso, did the migration go without any problems joepie91 ?
<joepie91>
I know that Gitolite is Perl, I'm talking about the additional software you will be running for UI :P
<joepie91>
eyJhb: currently doing it, for real this time
<joepie91>
repos are transferring now
<eyJhb>
Ahh, okay :p
<eyJhb>
How many you got?
<joepie91>
too many
<eyJhb>
Still amazed with the trust you have in yourself joepie91 ;)
<joepie91>
how's that? :P
<eyJhb>
I would not trust myself to take backups of it all :p
<eyJhb>
That's why Github/Gitlab is goto. I can push and not worry!
<MichaelRaskin>
philipp[m]: the core count was set before me; and now I am hoping to migrate to a larger local installation managed by people _only_ responsible for system administration.
<MichaelRaskin>
Apparently our VM is too small to do project-export on one of the repos…
<joepie91>
eyJhb: I have bad news for you...
<joepie91>
it does not work that way :P
<eyJhb>
Do not burst my bubble of happiness joepie91 :(
<joepie91>
lol
<eyJhb>
But I still trust them more than me :p
<joepie91>
dunno, I trust my provider a lot more than github
<eyJhb>
But it is mostly because, I have no money to actually setup a system with proper drives. I think my server has a bunch of drives with SMART failures
<eyJhb>
Don't you have remote backups as well?
<joepie91>
eyJhb: a VPS is cheap
<eyJhb>
Yeah, but storage is not :p
<joepie91>
yes, but I'm equating the two situations
<joepie91>
trust github + have backups vs. trust provider + have backups
<joepie91>
eyJhb: storage is also
<tilpner>
joepie91: I want to point out that gitolite has a very powerful access control mechanism, beyond anything I've seen in web-configured options
<eyJhb>
joepie91: which provider?
<f0x>
git doesn't take much storage either
<eyJhb>
f0x: Have you seem my repos? ;) :D
<eyJhb>
s/seem/seen/
<joepie91>
eyJhb: eg. Massive KVM here: https://ramnode.com/#pricing -- 325GB VPS for $5/month... or BuyVM, which charges $5/month for block storage
<joepie91>
err
<joepie91>
for 1TB block storage*
<joepie91>
or well more accurately $1.25/mo per 250GB
<joepie91>
I think max storage volume size is 4TB or 8TB or something, and you can have multiple
<f0x>
eyJhb: how bad is it?
<tilpner>
Even the simplest sort of AC, where you might want to allow someone to push commits which only contain changes to a subtree of a repository, is not possible on Github
<joepie91>
I am full of shit, it's max 8 storage blocks per VM of max 10TB each
<eyJhb>
joepie91: Seems doable.
<eyJhb>
f0x: I am not even sure anymore, but I know I have some that are 1+ GB each :p
<eyJhb>
Mostly because, CTF can take up a lot of space
<joepie91>
tilpner: mm, didn't have that when I last used it...
<tilpner>
When did you use it?
<joepie91>
years back
<joepie91>
I've been running Gogs for some time now
<joepie91>
I was using Gitolite before Gogs existed :P
<joepie91>
must've been 2011-2012 or thereabouts
<joepie91>
that I started using it
<f0x>
eyJhb: oof, i think my larger repos are my KiCad Projects, mostly because it also keeps zipped project snapshots (which actually helped a lot with some weird nightlies issues so glad to keep them)
<f0x>
but that's like <100Mb still
<joepie91>
well, initial repo sync completed
<joepie91>
time to shut down Gogs and start the downtime-y part of the migration!
<joepie91>
or well, almost :P
<eyJhb>
f0x: They can get fairly large quite quick... But then again, I don't use Github as a main backup for all stuff, I have 300-400 GB of "family data"
<eyJhb>
That I have just distributed to my brothers/father
__monty__ has joined #nixos-chat
<patagonicus>
I have a git-annex repo where the pure git part is a gigabyte or so. Git's not really designed to handle that, unfortunately. It works, but it's pretty slow even on beefy machines.
<patagonicus>
I learned that even with git-annex you shouldn't throw in hundred thousands of small files because git just doesn't like it. :D
<eyJhb>
No one likes that :p
<patagonicus>
The filesystem didn't have a problem with that many files.
<__monty__>
Hmm, too bad.
<__monty__>
Git-annex seems interesting but every time I get interested there's one reason or another keeping me from trying it.
<eyJhb>
patagonicus: You just need MORE of them :p But depends on what you do with them
<eyJhb>
Generally it is just no fun moving them, etc.
<patagonicus>
Yeah, I should have just started off with half a dozen repos instead of one big one, but it lured me in with a promise of simplicity. And honestly it was still usable.
cjpbirkbeck has joined #nixos-chat
<patagonicus>
__monty__: I'd still recommend git-annex. It's a really nice solution for checksums + location tracking, which enables you to provide redundancy guarantees with non-uniform storage (that doesn't even have to be online at the same time).
<patagonicus>
One (big) backup away from finally installing NixOS on my laptop. I tried when I got it a year ago, but the installer didn't even want to boot and I didn't have time back then to fiddle with that so I ended up using the preinstalled Ubuntu (with Nix on top, of course).
<joepie91>
how on earth does one get sendmail working on NixOS
<philipp[m]>
Are you talking about sendmail the ancient mailserver or a sendmail util that most mail servers ship?
<tilpner>
joepie91: Why do you have so many empty repositories and 282 pages of users?
<joepie91>
philipp[m]: well, let me phrase it differently. I have my Git server. it needs to send e-mails. I don't want to pay a SaaS e-mail sending company, and send the e-mails from my own server instead. Gitea expects a sendmail-compatible. what now?
<joepie91>
tilpner: the users are largely spammers, still need to do something about those, but my hands were kinda tied with my ancient Gogs
<joepie91>
empty repos are barely-started projects which have not really seen any commits yet
<tilpner>
Ahh, open registration and bots?
<joepie91>
yeah
<tilpner>
Does gitea improve on that somehow?
<joepie91>
yes, it has an API afaik :P
<joepie91>
so I can automate my spammer nuking
<tilpner>
Ahh, but nothing preconfigured
<joepie91>
nop. Gitea's spam-fighting mechanisms are pretty limited, and I refuse to use recaptcha
<joepie91>
but also 282 pages of users is basically nothing considering it's been running for years with a dubiously-challenging image captcha and no spammer removal
<joepie91>
so now that I have an API I can probably nuke all of it with a few well-placed regexes
<joepie91>
in under an hour
<tilpner>
What's the motivation for the spammers here? SEO by spamming backlinks?
<joepie91>
yeah
<joepie91>
the auto-generated names largely just don't seem to have figured out a way to insert spam and the effective spammers seem to be manually-created accounts, sometimes complete with logo of the company doing the spamming
<joepie91>
so overall I'm not super concerned :P
<joepie91>
I'm more concerned about the lack of e-mail heh
<joepie91>
the NixOS wiki seems surprisingly low on "how2email"
<joepie91>
has everyone just collectively given up on trying to do e-mail outside of Google?
<tilpner>
Huh, I wouldn't have expected people to manually register spam accounts on random gogs instances
<patagonicus>
joepie91: I've signed up for Migadu a long time ago and just never bothered worrying about email anymore. Unfortunately their offer isn't as great as it used to be - used to be unlimited everything except outgoing mail, but even the cheapest plan was way more than you'd use for a few people.
<joepie91>
patagonicus: I explicitly do not want to involve any external services
<joepie91>
tilpner: that's for a full-blown mailserver, no? I just want to send issue notifications etc. :)
<sphalerite>
joepie91: I'm using simple-nixos-mailserver to selfhost my mail, and at work we're running postfix
<patagonicus>
I saw that, just wanted to say that there's options other than Google. :P
<joepie91>
ah, in that sense, right
<tilpner>
joepie91: Yes, it may be overkill. I tried it once, but have decided I don't want to host my own mail since
<joepie91>
yeah I don't host my own e-mail for personal mail
<joepie91>
due to the deliverability nightmare
<MichaelRaskin>
This nightmare actually means you should send via whatever but receive on your own server
<eyJhb>
I tried setting up a mailserver in 2008, and I still have nightmares about it
<patagonicus>
Sigh. Using filesystems works a lot better if you actually mount them.
<MichaelRaskin>
I set up a mailserver that I do not tend much to, but that actually just keeps working…
<eyJhb>
patagonicus: I feel your pain, it happens
<eyJhb>
MichaelRaskin: Is it at the point, where you are afraid to touch it?
<MichaelRaskin>
No, too lazy to
FRidh has quit [Ping timeout: 240 seconds]
<insep_>
samueldr: so it worked
<insep_>
\o/
<joepie91>
Nov 15 14:56:53 machine-konjassiem-02.cryto.net systemd-coredump[21574]: Process 21572 (sendmail) of user 997 dumped core.
<joepie91>
I think I'm reconsidering doing computering
cjpbirkbeck has quit [Quit: Goodbye, take care]
tilpner has quit [Remote host closed the connection]
tilpner has joined #nixos-chat
<patagonicus>
joepie91: Something I say pretty much at least once a week at work is that computers in general were a mistake.
<philipp[m]>
joepie91: I usually just do authenticated smtp for those cases.
<MichaelRaskin>
Computers in general were not a mistake. It's the other story that oligopolisation and rapid expansion of market by advertising lead to what they normally lead
<joepie91>
yeah..
<joepie91>
also that, capitalism spoiled it
<MichaelRaskin>
So, it's obviously all downhill from Engelbart's The Demo or something around that
<joepie91>
(in a variety of ways)
<MichaelRaskin>
Well, capitalism of age where global oligopolies have logistics available to arise, and do not have break-first-ask-later antitrust to stop them
<MichaelRaskin>
Highly fragmented markets have different problems
waleee-cl has joined #nixos-chat
FRidh has joined #nixos-chat
FRidh has quit [Ping timeout: 272 seconds]
FRidh has joined #nixos-chat
<joepie91>
srhb: hi, do you happen to know why setgid was added to the restrictions list for Gitea here? https://github.com/NixOS/nixpkgs/pull/63840 -- it's not in the original recommendations, but it *is* in the PR contents, and it just broken sendmail/opensmtpd for me :)
<{^_^}>
#63840 (by dasJ, 1 year ago, merged): nixos/gitea: Sandbox the systemd service
<joepie91>
broke*
<joepie91>
but maybe there's a reason it's in there
<joepie91>
for the Googlers: if your sendmail/opensmtpd is breaking in Gitea with a SIGSYS ("bad system call"), then you can work around that with: systemd.services.gitea.serviceConfig.SystemCallFilter = lib.mkForce "~@clock @cpu-emulation @debug @keyring @memlock @module @mount @obsolete @raw-io @reboot @resources @swap";
<srhb>
joepie91: No, no idea. :)
<joepie91>
srhb: okay, in that case I'm going to file a PR :)
<joepie91>
thanks
<srhb>
joepie91: Thanks! It certainly could use some more attention than just emergency gum-and-glue when something breaks.
<joepie91>
hah
<joepie91>
it seems like a fairly well-built-out service though
<srhb>
The settings revamp helped a lot I think.
FRidh has quit [Ping timeout: 272 seconds]
FRidh has joined #nixos-chat
<das_j>
joepie91: Well, looks like I didn't test opensmtpd :/ sorry
<joepie91>
> Note: ssmtp is unmaintained. Consider using something like msmtp or OpenSMTPD instead.
<{^_^}>
error: syntax error, unexpected ')', expecting ID or OR_KW or DOLLAR_CURLY or '"', at (string):400:1
<das_j>
hm so maybe it was this way around
<das_j>
because yeah, my source was also the arch wiki ;)
<joepie91>
:P
<das_j>
joepie91: About your PR: You can also switch to a positive list instead of a negative list. `@system-service @setuid` should be enough probably
<joepie91>
das_j: I went for the minimum viable change :P
<das_j>
fair enough
<joepie91>
as I don't know the backstory behind the apparent coordinated effort to sandbox stuff by default
<joepie91>
so I don't really want to touch more without understanding that first
cole-h has joined #nixos-chat
<das_j>
I don't think it's coordinated at all ;) I sandbox stuff in my config repo and at random intervals I push out the changes to nixpkgs
<infinisil>
Is there some federation protocol that ensures there aren't any most-used servers?
<infinisil>
E.g. most matrix users are on the matrix.org server
<infinisil>
Which is not ideal for decentralization
<infinisil>
Is there a protocol where if you sign up, it searches e.g. for a server with a low amount of sign-ups?
<infinisil>
Hm probably won't work well, because you wouldn't trust a random instance to not just trash your data or whatever
<ashkitten>
infinisil: matrix might have that at some point, it's been mentioned many times that there might be a way for servers to load-balance by shifting users around between them (depends on portable identity ofc)
<__monty__>
Sounds like not-quite P2P to me.
<__monty__>
Which you could see as load-balancing with every request.
<__monty__>
Also, unless the client did the load-balancing, wouldn't instances be able to lie to keep you connected?
<cirno-999>
federations will always result in centralization
<cirno-999>
true p2p or bust :P
<__monty__>
Not with forced load balancing.
<__monty__>
I just suspect it means you have to artificially increase load on the network and I'm not sure that's worth the balance. Your client would have to include the federation node it's connecting to in every message and use the information in messages from others to estimate load on the nodes. (I'm assuming you can't trust nodes.)
<__monty__>
So you'd be intentionally load-balancing *away* from the people you actually talk to.
<joepie91>
infinisil: in federation, server choice involves a trust relationship
<joepie91>
honestly a major factor in the centralization, I think, that Synapse is not an obvious thing to run yourself
<__monty__>
Which means conversations need to be synchronized to more nodes *and* messages take more hops to reach participants.
<joepie91>
it's not like Gitea, which you dump somewhere in a corner and let run unattended
<joepie91>
Synapse uses way more resources, requires way more maintenance, etc.
<__monty__>
Even with easy to run nodes you still run into scaling issues though.
cjpbirkbeck has joined #nixos-chat
<__monty__>
If one party manages to engineer a super node with a more efficient internal protocol it'll attract more users and start benefitting from the network effect, because communication among those users is even more efficient.
<__monty__>
At least that's the argument I always see coming up for the failure of XMPP.
<joepie91>
the failure of XMPP was not really technical in nature
<joepie91>
XMPP failed because of a) a spec model that resulted in a swiss cheese feature matrix, and b) not enough powerful parties in the ecosystem to prevent a single megacorp (Google) from taking over
<joepie91>
(sub-problem of B being that there were no clients which could realistically compete with GTalk on UX)
<__monty__>
It's not the only criticism of XMPP. But it certainly is one of them.
<__monty__>
infinisil: Let me know if you find some research on this or something.
<cirno-999>
__monty__ | Not with forced load balancing.
<cirno-999>
that is effectively p2p then
<cirno-999>
i would consider a server an extension of the client then
<__monty__>
Not quite. It's a weaker scheme imo.
<infinisil>
Don't plan to look into this very much, but if I do I'll let you know :)
<lovesegfault>
If I have something running on port X, but I want it available on port Y _as well_ what do I do?
<lovesegfault>
can I "symlink" ports?
<cirno-999>
forward.
<cirno-999>
a stupid/simple workaround to be run it in docker and forward ports there
<cirno-999>
more intelligent would be to modify iptables
<cirno-999>
afaik.
__monty__ has quit [Quit: leaving]
<MichaelRaskin>
For silly approaches, there is socat