00:10
drakonis_ has quit [Ping timeout: 272 seconds]
00:25
<
qyliss >
Hydra's Darwin builders do that!
00:26
<
clever >
darwin, in qemu, on nixos, on apple hardware
00:26
drakonis_ has joined #nixos-dev
00:27
<
ivan >
on Linux you can also pass through an NVIDIA card to macOS and have working 3D
00:27
<
ivan >
on 10.13, anyway
00:27
<
clever >
ivan: oh, also
00:28
<
clever >
there are signs that 10.14 is getting better qemu support
00:28
<
clever >
it has support for the emulated gpu in qemu, along with virtio and 9planfs
00:28
<
clever >
it sounds like they havent really told anybody
00:28
<
samueldr >
now if only they dropped that clause in the EULA and sold licenses :/
00:29
<
clever >
and the author of the blog only noticed, because the resolution of his guest changed
00:29
<
samueldr >
between "minor" updates too, IIRC
00:29
<
samueldr >
>> After one of the minor updates
00:29
<
qyliss >
I wonder why they're doing that
00:29
<
samueldr >
literally the first words 🤦
00:29
drakonis has quit [Ping timeout: 252 seconds]
00:29
<
infinisil >
clever: Why does it need to run on mac hardware though?
00:29
<
samueldr >
it's likely useful for their own internal use
00:30
<
infinisil >
From that github project I'm getting that you don't need a mac
00:30
<
clever >
infinisil: legal reasons
00:30
<
ivan >
infinisil: mac machines have the secret string
00:30
<
samueldr >
there is a clause in the EULA, and a few copyrighted words needed that are probably technically meaningless
00:30
<
samueldr >
(just like the nintendo logo in roms)
00:31
<
clever >
please dont steal, lol
00:31
<
infinisil >
I wonder what the performance loss is due to it running through qemu instead of doing it directly
00:31
<
infinisil >
on the mac hardware
00:31
<
clever >
infinisil: kvm is still in effect
00:31
<
samueldr >
probably way less than anticipated, when it's CPU-bound
00:31
<
infinisil >
kvm is almost zero cost?
00:31
<
samueldr >
the cost is the other hardware, mainly storage
00:32
<
samueldr >
which can be helped along with spooky options that sync less or not at all to disk
00:32
<
clever >
the nixos stuff, is using zvols on a zfs pool
00:32
<
clever >
also, `drive id=MacHDD,cache=unsafe` this part tells qemu to just ignore when the guest asks to sync things to disk
00:33
<
clever >
so data can be in the kernels write cache, and at risk of loss during improper shutdowns, and the guest will think its all saved
00:33
<
clever >
which will give a small boost to performance
00:33
<
samueldr >
doing graphical stuff is... not always nice, since it runs without gpu accel (except maybe with iGVT)
00:33
<
clever >
and dont worry about data loss, it does `zfs rollback ${snapshot}
00:33
<
clever >
on bootup, so it always has data loss :P
00:38
Drakonis__ has joined #nixos-dev
00:49
justanotheruser has quit [Ping timeout: 248 seconds]
00:50
Drakonis__ has quit [Ping timeout: 245 seconds]
01:00
Synthetica has quit [Quit: Connection closed for inactivity]
01:03
ajs124 has quit [Quit: Gateway shutdown]
01:08
ajs124 has joined #nixos-dev
01:25
drakonis has joined #nixos-dev
01:27
drakonis1 has joined #nixos-dev
01:28
drakonis_ has quit [Ping timeout: 250 seconds]
01:31
<
drakonis1 >
ivan: changing PA's latency does no good
01:35
justanotheruser has joined #nixos-dev
01:44
<
ivan >
drakonis1: did you wineserver -k and start something from the shell you exported in
01:44
<
ivan >
if yes then I dunno, sorry
02:12
justanotheruser has quit [Quit: WeeChat 2.4]
02:15
<
drakonis1 >
i'm confused now
02:42
justanotheruser has joined #nixos-dev
02:47
<
edef >
so i'm working on making OpenSSH host-based authentication work nicely on NixOS
02:48
<
edef >
and among other things this needs ssh-keysign(1), which does come with openssh, and lives at ${openssh}/libexec/ssh-keysign
02:48
<
edef >
but it needs to be suid-root, which is fine, but OpenSSH expects to invoke it at that particular path
02:48
<
edef >
do we have policy around embedding /run/wrappers paths in packages?
03:35
Guanin has quit [Remote host closed the connection]
03:40
drakonis1 has quit [Quit: WeeChat 2.5]
03:46
drakonis1 has joined #nixos-dev
04:50
orivej has joined #nixos-dev
04:57
<
Shados >
edef: ...maybe just make a sh wrapper that execs the real suid version? Or would it pick up on the actual location?
04:57
orivej has quit [Ping timeout: 268 seconds]
04:58
<
Shados >
Or, similarly, a derivation that is just a symlink tree of the underlying openssh derivation + a symlink to the /run/wrappers path?
05:00
bgamari_ has quit [Ping timeout: 252 seconds]
06:01
bgamari has joined #nixos-dev
06:13
MichaelRaskin has quit [Ping timeout: 258 seconds]
06:20
drakonis1 has quit [Quit: WeeChat 2.5]
06:27
alp has joined #nixos-dev
07:05
johanot has joined #nixos-dev
07:12
Jackneill has joined #nixos-dev
07:13
MichaelRaskin has joined #nixos-dev
08:06
pie__ has quit [Ping timeout: 258 seconds]
08:41
Synthetica has joined #nixos-dev
09:20
pie__ has joined #nixos-dev
09:20
pie__ has quit [Remote host closed the connection]
09:28
johanot has quit [Quit: WeeChat 2.4]
09:29
johanot has joined #nixos-dev
09:52
cjpbirkbeck has quit [Quit: Quitting now.]
11:05
orivej has joined #nixos-dev
11:28
johanot has quit [Quit: WeeChat 2.4]
11:32
justanotheruser has quit [Ping timeout: 248 seconds]
11:42
justanotheruser has joined #nixos-dev
11:54
orivej has quit [Ping timeout: 245 seconds]
12:28
<
edef >
gchristensen: hmm, okay
12:48
bgamari has quit [Ping timeout: 244 seconds]
12:50
bgamari has joined #nixos-dev
12:58
bgamari has joined #nixos-dev
13:31
andi- has quit [Quit: WeeChat 2.5]
13:36
andi- has joined #nixos-dev
13:39
<
worldofpeace >
hey a hydra jobset for #63493 is really needed if we'd like to land that
13:48
alp has quit [Ping timeout: 258 seconds]
14:21
psyanticy has joined #nixos-dev
14:35
<
samueldr >
niksnut, domenkozar[m], gchristensen ^ I don't know who else has hydra privileged hands
15:15
alp has joined #nixos-dev
15:57
lopsided98_ has joined #nixos-dev
15:58
ma27_ has joined #nixos-dev
15:59
orivej has joined #nixos-dev
16:03
{^_^} has quit [*.net *.split]
16:03
Cale has quit [*.net *.split]
16:03
lopsided98 has quit [*.net *.split]
16:03
ma27 has quit [*.net *.split]
16:06
c00w has joined #nixos-dev
16:07
v0|d has joined #nixos-dev
16:10
Cale has joined #nixos-dev
16:35
orivej has quit [Ping timeout: 245 seconds]
17:11
cjpbirkbeck has joined #nixos-dev
17:11
FRidh has joined #nixos-dev
17:14
Jackneill has quit [Remote host closed the connection]
17:20
<
edef >
pondering whether i should split this PR or not
17:36
<
worldofpeace >
thanks a lot gchristensen ✨
17:43
<
qyliss >
edef: if in doubt, yes
18:36
aminechikhaoui has joined #nixos-dev
18:44
aminechikhaoui has joined #nixos-dev
18:49
johanot has joined #nixos-dev
18:50
aminechikhaoui has joined #nixos-dev
19:09
<
Guest4937 >
mainly due to x86_64-darwin builder woes, but perhaps we need a way to force an immediate channel update?
19:09
<
Guest4937 >
otherwise, we'll have to wait a few more days before we get the patched firefox
19:11
<
samueldr >
Guest4937: I don't think so :/
19:11
<
samueldr >
linux users can (and maybe should) instead prefer nixos-unstable rather than nixpkgs-unstable
19:12
<
samueldr >
(though it does have its issues getting eval'd lately, Too many heap sections: Increase MAXHINCR or MAX_HEAP_SECTS again)
19:18
<
edef >
what
*are* our Darwin builder woes
19:19
<
samueldr >
(and maybe in addition, being underprovisioned)
19:22
<
edef >
the latter i might be interested in improving
19:22
<
edef >
wait, are these Packet-hosted VMs?
19:23
<
samueldr >
edef: get in touch with gchristensen, I think graham's the one to talk to for improving all that
19:27
<
edef >
i guess i should retarget for staging
19:28
<
samueldr >
if you rebase to merge-base between staging and master before changing the branch in the PR it shouldn't notify anyone it doesn't have to (if even possible)
19:28
<
qyliss >
that's useful to know
19:29
<
samueldr >
no worries, I think we're all used to this happening sometimes :)
19:30
<
qyliss >
it's annoying, but I consider it GitHub's fault rather than any user's
19:45
<
edef >
at this point i'm just staring into the distance realising that my one-line change is rebuilding >1k packages
19:55
psyanticy has quit [Quit: Connection closed for inactivity]
20:09
cjpbirkbeck has quit [Quit: Quitting now.]
20:12
FRidh has quit [Quit: Konversation terminated!]
21:08
orivej has joined #nixos-dev
21:15
johanot has quit [Quit: WeeChat 2.4]
21:56
layus has quit [Ping timeout: 252 seconds]
21:57
layus has joined #nixos-dev
22:06
Guanin has joined #nixos-dev
22:06
drakonis1 has joined #nixos-dev
22:06
<
gchristensen >
okay so fwiw I've been having a heck of a time since I've been quite busy this week -- I'm sorry :(
22:07
<
gchristensen >
but I've built fresh images anda m about to roll out new macs
22:12
{^_^} has joined #nixos-dev
22:25
<
gchristensen >
I just rolled over most of the macs
22:25
<
gchristensen >
they take a while to boot though, and I'll probably go to sleep before they're done. ...
22:26
alp has quit [Ping timeout: 258 seconds]
22:41
<
Guest4937 >
sounds good
22:42
<
gchristensen >
ok Guest4937 we have macs again
22:42
<
gchristensen >
maybe I should make them implode after 1 week and recreatethemselves
22:42
<
samueldr >
that's a fun name there, I wonder if matrix masks the fac to Guest4937
22:42
<
samueldr >
the fact*
22:43
<
gchristensen >
hehe
22:43
<
samueldr >
con: implosion would spuriously make jobs fail, pro: fewer jobs failed than everything failing
22:43
<
gchristensen >
right
22:44
<
gchristensen >
hydra would detect it asa temproary failure I think
22:44
<
samueldr >
oh, then if it retries it's just lost time, at most 10h per machine I guess
22:52
<
gchristensen >
ideally they wouldn't implode of course
23:01
<
ekleog >
how many macs do we have? making them implode so that one implodes every week should lose relatively few jobs, while being enough to ensure the auto-respawning works properly
23:58
orivej has quit [Ping timeout: 244 seconds]
23:58
orivej has joined #nixos-dev