<gchristensen>
so ofborg has a very generous amount of linux capacity. I'm assuming what you're thinking is w.r.t. bors merging PRs?
<zimbatm>
yes
<zimbatm>
if ofborg can emulate nixpkgs close enough, then we can use bors to drive the builds
<zimbatm>
this would help to keep master green
<gchristensen>
right. so a concern that has stopped me from implementing things is what does it mean if things to fail on darwin or aarch64 but not master? and: how to suss out which problems are acceptable and which aren't
<zimbatm>
I guess the underlying concern is that people won't have the energy to debug foreign OS problems?
<zimbatm>
Darwin is constantly fighting to catch-up because they have less manpower than Linux people
<zimbatm>
that would help level the field
<zimbatm>
depending on what platforms we declare "official"
<zimbatm>
IMO, all the tests should pass on all the platforms all the time
<gchristensen>
I agree
<zimbatm>
then if we need to tweak something is should be on the test-level
<zimbatm>
that keeps the logic simple
<zimbatm>
we are already doing this with `doCheck = !(stdenv.isDarwin)`
<gchristensen>
a few challenges
<gchristensen>
right now I think ofborg works because it doesn't very much, just a few packages. I'm not against changing this, but it'd be a departure from how it works now in a lot of ways
<gchristensen>
ofborg doesn't do any build distribution currently, all PRs are built (per arch, of course) on a single system
<zimbatm>
I see
<gchristensen>
this is/was an important architectural feature because anyone can host a runner, and it would be risky to have people exchanging built store paths automatically from untrusted anyone else
<zimbatm>
yeah, I also don't want to risk detracting the project
<zimbatm>
alright, thanks for considering it still
<gchristensen>
well so there has been a recent development which opens up numerous opportunities for us, if we can get sufficient mac capacity
<gchristensen>
which is that there is now an AWS account setup to autoscale based on demand :)
<gchristensen>
and Packet.net provided another type 2a for community use, and it is going to soon run ofborg too
<zimbatm>
is the AWS account funded by AWS?
<zimbatm>
they have 100k grants for startups
<gchristensen>
no, a currently anonymous benefactor with sizable pockets
<zimbatm>
pockets are a good thing
<gchristensen>
(I'll leave it up to them to fess up to it, if they'd like)
<zimbatm>
that reduces the number of candidates to 87 :)
<gchristensen>
:D
<zimbatm>
during my coding retreat I spent some time to try and make macOS run under VirtualBox
<zimbatm>
I got the installer ISO booting but it had trouble finding the disks
<cransom>
not long ago, there used to be at least hackintosh images that woudl do that
<cransom>
and by not long ago, i mean, it could have been very long ago.
<gchristensen>
hah
<samueldr>
I have .box of macOS usable with vagrant and virtualbox, including provisioning through virtualbox scripts, though legally it is in the grey area
<gchristensen>
I don't want to know anything about legally gray boxes
<samueldr>
if there is interest, I can fish out the documents I used to bulid them
<gchristensen>
that said, if someone happens to have some macos systems lying around and would be happy too run an agent, I'll get you sete up
<simpson>
Has anybody pointed out yet the futility of trying to provide Free Software for a closed platform hostile to it?
<gchristensen>
yes
<simpson>
Okay, cool.
<gchristensen>
:)
<zimbatm>
I have a visit card of a macStadium guy I should try to find again
<simpson>
(There are *dozens* of us~)
<zimbatm>
samueldr: is it running the latest macOS release?
<sphalerite>
gchristensen: did you see the other day? I think I mentioned that I have a Mac that I'm not using actively
<sphalerite>
(Is this for ofborg? Or is there some context I missed out on?)
<samueldr>
I have been lazy and not made one for 10.13 yet
<samueldr>
whilst IANAL, the legality of running virtual machines of macOS is dependent on the host operating system
<sphalerite>
gchristensen: Since it's technically my dad's machine, I'll probably be taking it to Sweden next week, then set it up there if he's cool with that
<gchristensen>
sphalerite: we'd need regular access (via ssh at least) for updates, does that work?
<sphalerite>
yeah I'll have to ask him but I think it'll probably be fine
<sphalerite>
anyway, got to catch a bus tomorrow morning, so I should sleep
<sphalerite>
gnight folks!
<gchristensen>
see you
<gchristensen>
is it possible from an instantiation of nixpkgs to get the system it is for?
<gchristensen>
oh... yes ... `pkgs.system` :)
<clever>
stdenv should also have it
phreedom has quit [(Ping timeout: 240 seconds)]
<gchristensen>
hmm probably about time to figure out how to have ofborg send a single reply for multiple archs
pie_ has joined joined #nixos-dev
pie__ has quit [(Read error: Connection reset by peer)]
Sonarpulse has quit [(Ping timeout: 248 seconds)]
orivej_ has quit [(Ping timeout: 272 seconds)]
laggy_wifi is now known as gy_wifi
<gchristensen>
I have to go to bed, but if anyone has suggestions I'd love some tips. I'd like to have the netboot image work in ram, but without the unionfs. any pointers on how to do this? unionfs is taking lots of CPU during builds on https://github.com/nix-community/aarch64-build-box
<gchristensen>
please feel free to open an issue or PM me or send suggestions here
pie_ has quit [(Read error: Connection reset by peer)]
<makefu>
andi-: domenkozar: FRidh: openstack removal is at https://github.com/NixOS/nixpkgs/pull/32637 . however pytrainer seems to be broken as of right now (not due to the removal though)
<domenkozar>
peti: any chance we could have a jobset for multiple outputs branch?
<domenkozar>
I can set up one as well, but interested to hear if you have some feedback
__Sander__ has joined joined #nixos-dev
vcunat has joined joined #nixos-dev
<jtojnar>
how can pytrainer be broken but still fetched from binary cache
<jtojnar>
?
<makefu>
that is a damn good question
<domenkozar>
jtojnar: what happened?
<domenkozar>
we need to prevent it in the future
<jtojnar>
I thought I am on my repo and deleted release-17.09 branch
<domenkozar>
jtojnar: exactly how?
<domenkozar>
branch should be protected
<domenkozar>
from deletion and force-push
<jtojnar>
it looks like only master was
<domenkozar>
so using
<domenkozar>
git push origin :branch
<domenkozar>
?
phreedom has joined joined #nixos-dev
<domenkozar>
I've added protection back and extended it also for admins
<domenkozar>
jtojnar: all good, sorry for not protecting the branch :)
<domenkozar>
it's now setup correctly.
<LnL>
huh!
<domenkozar>
I think I might have removed protection when we played with status checks, but I honestly don't remember.
orivej has joined joined #nixos-dev
<LnL>
master/release branches should definitely be protected :)
orivej_ has joined joined #nixos-dev
orivej has quit [(Ping timeout: 260 seconds)]
<domenkozar>
LnL: now they are
<LnL>
:D
<LnL>
I've also done something like this before, git push origin $NOT_AN_EMPTY_STRING:master
<vcunat>
:-)
phreedom has quit [(Ping timeout: 272 seconds)]
orivej_ has quit [(Ping timeout: 255 seconds)]
orivej has joined joined #nixos-dev
<peti>
domenkozar: I think there is a jobset on Hydra still around that you could resurrect for testing purposes. We can just build the normal Nixpkgs set, IMHO, to test the code. That would include Darwin, too.
<domenkozar>
k
FRidh has quit [(Ping timeout: 240 seconds)]
FRidh has joined joined #nixos-dev
orivej has quit [(Ping timeout: 240 seconds)]
FRidh has quit [(Ping timeout: 256 seconds)]
FRidh has joined joined #nixos-dev
pie__ has quit [(Ping timeout: 264 seconds)]
ma27 has joined joined #nixos-dev
<gchristensen>
heads up: grahamcofborg now checks if the meta fields are valid in nixpkgs. let me know if you find a problem by opening an issue at https://github.com/grahamc/ofborg/issues please
<LnL>
o/
<aminechikhaoui>
gchristensen: did you ping me yesterday ? I wasn't online at that time
pie_ has joined joined #nixos-dev
<gchristensen>
hmm yes I did but I forget why :/
jtojnar has quit [(Ping timeout: 256 seconds)]
jtojnar has joined joined #nixos-dev
<aminechikhaoui>
np :)
<jtojnar>
could we maybe check top-level derivations for existence of maintainer and license keys?
<gchristensen>
we could
<jtojnar>
do you recall how was the big nixpkgs fork called? I think it was something like Tristan
<gchristensen>
Triton
<fadenb>
Unexpected mentioning of my firstname :P
<gchristensen>
but I'm not sure they're big, but it is continued to be maintained
<gchristensen>
why do you ask?
<jtojnar>
thanks
<jtojnar>
IIRC there was some divergence in GNOME packaging
<gchristensen>
ah, perhaps some inspiration?
<jtojnar>
I would like to check it out now that I am more familiar with nixpkgs
<gchristensen>
jtojnar: there are "lots" of errors if I look at mandator license or maintainer :)
<jtojnar>
yes, I have been adding them to GNOME related packages but it will be a lot of work
orivej has joined joined #nixos-dev
<gchristensen>
gsc.io/license-maintainer-errs
<gchristensen>
I don't actually know how to just check top level packages, so this probably checks ~everything
<gchristensen>
I added a small hack to exclude fetchurl drvs from the check
<gchristensen>
anyone know how these bootstraps came to be? pkgs/development/compilers/openjdk/bootstrap.nix its sketchy to be fetching weird URLs from dropbox
<Sonarpulse>
I don't know what to do about allowedReferences in bootstrapTools
<Dezgeg>
use nuke-refs more
<Sonarpulse>
oh ok
<Sonarpulse>
easy enough
<Dezgeg>
well, maybe worth investigating why they're occurring now and not before
<Sonarpulse>
Dezgeg: perhaps the allowed reqs is new?
<Sonarpulse>
it was in libgcc.a
<Sonarpulse>
it was in lib/<triple>/version
<Sonarpulse>
there was just a nuke-refs for libexec/*/*/* before
<Sonarpulse>
we could investigate, but doesn't look to concerning to me
<Sonarpulse>
static archive referencing headers is fine
<Dezgeg>
well make-bootstrap-tools.nix has had allowedReferences = []; since forever
<Dezgeg>
and I don't see why cross and non-cross should cause a noticeably different result (IIRC, the differences used to be only whether gcc has some libs like gmp statically linked or not)
<Sonarpulse>
Dezgeg: I am guessing the non-cross version would not have libs under the target platform name like that
<Sonarpulse>
nope, nevermind
jtojnar has joined joined #nixos-dev
<jtojnar>
and now I received a SUCCESS mail from NOVEMBER
<Sonarpulse>
merged with the PR that first broke mingw
<Sonarpulse>
gccCrossStageFinal fails with a GCC_NO_EXECUTABLE errors
<Sonarpulse>
which is weird because I saw that in a build != host == target gcc failure on arm
<Sonarpulse>
but never build == host != target failure
<Sonarpulse>
taktoa: so before what I did was "set; exit -1" preConfigure
<Sonarpulse>
or whatever
<Sonarpulse>
and compare the diff
<Sonarpulse>
its quite minimal
<Sonarpulse>
yet something is still going wrong
<Sonarpulse>
I suppose next step is build -j1 and comparing the log
<taktoa>
weird; I haven't run into that issue before. might be willing to look into it more deeply after my final exams are over (btw: I will be back in the NY metro area for winter break)
<Sonarpulse>
makes sense with finals
<Sonarpulse>
and cool!
<Sonarpulse>
yeah I am just gone a little bit around Christmas
<Sonarpulse>
so probably good overlap
<taktoa>
cool, maybe I could visit you guys at obsidian systems if you're interested. I'll also be hanging out and hacking on stuff with mckeankylej (who has done a lot of reflex and nix stuff) during winter break
<taktoa>
Sonarpulse: ^
<Sonarpulse>
taktoa: absolutely
<Sonarpulse>
yeah feel free to swing by! we can plan something
<taktoa>
alright, gonna get back to studying now
taktoa has quit [(Remote host closed the connection)]
<bgamari>
does anyone have a good sense for what is required to make nixos cross-compilable?
<bgamari>
I have a usable cross-compiled nixpkgs set
<bgamari>
it seems like lib/eval-config.nix can access pkgs as an argument
<bgamari>
s/access/accept/
<bgamari>
but I very quickly run into loops when I pass it my cross-compiled nixpkgs
<bgamari>
it seems _module.args.utils in attrsets.nix is how we enter the loop
<bgamari>
the warnings option seems to be in play as well
<Sonarpulse>
bgamari: I wish I did!
<Sonarpulse>
but know very little about internals of nixos
<Sonarpulse>
clever maybe, from not os?
<bgamari>
yes, that is my problem as well
* clever
waves
<Sonarpulse>
bgamari: I got the bootstrap tools building btw
<bgamari>
yay
<Sonarpulse>
tempted to kick of whole-world rebuild to see if my native-special casing prevents breakages, but might be better waiting for staging to hit mater
<bgamari>
is the `extend` function something I should know about?
<Sonarpulse>
bgamari: from the lib?
<Sonarpulse>
or a nixos one?
<Sonarpulse>
(/lib that is)
* bgamari
is trying to set boot.kernelPackages and encountering attribute 'extend' missing, at /home/ben/block-eng/nixpkgs/nixos/modules/system/boot/kernel.nix:26:31
<bgamari>
Sonarpulse, presumably the nixos one
<Sonarpulse>
bgamari: oh that is the lib extend
<Sonarpulse>
i think
<Sonarpulse>
that is weird
<bgamari>
the code in question is `kernelPackages: kernelPackages.extend ...`
<Sonarpulse>
bgamari: lib/fixed-points.nix
<bgamari>
hmm
<Sonarpulse>
inuxPackagesFor = kernel: lib.makeExtensible (self: with self; {
<Sonarpulse>
linuxPackagesFor = kernel: lib.makeExtensible (self: with self; {
<Sonarpulse>
in all-packages
<Sonarpulse>
so makeExtensible
<Sonarpulse>
no idea why cross would fuck it up
<bgamari>
Sonarpulse, it may be setting kernelPackages incorrectly