matthewbauer has quit [Remote host closed the connection]
<LnL>
whoops, did I miss anything?
<clever>
not much
__Sander__ has joined #nix-darwin
__Sander__ has quit [Ping timeout: 240 seconds]
<mitchty>
dhess: its dumb, but make -j1 got it further, and oddly the segfault appears to randomly crop up if you loop with make -jN, so you can get to the final stage but eventually you'll get a consistent segfault
<mitchty>
also HADDOCK_DOCS=NO helped but i'm not sure that matters much tbh
jtojnar_ has joined #nix-darwin
jtojnar has quit [Ping timeout: 256 seconds]
jtojnar_ is now known as jtojnar
<dhess>
mitchty: ugh, what a mess
<dhess>
given all that, it's odd that we both had a reproducible fault in the same place on different hardware
<mitchty>
dhess: yep :) i'm digging through the latest cores i got from trying to link with ld.bfd vs gold, its... confusing, feels like a stack overflow
<mitchty>
then again ghc executables stacks are just weird
<dhess>
hmm is there a way to increase the stack limit in ghc?
<dhess>
does it have one or does it just rely on the OS?
<mitchty>
i tried that too, musl libc has an 8m stack
<mitchty>
i bumped it just in case but no bueno
<dhess>
and only the one reply from upstream, that's disappointing
<dhess>
Surprised anyone is getting anywhere with ghc 8.2 on armv7. Are we the only ones using it? :)
<mitchty>
heh, well i can confirm the binutils fix there does seem to help
<mitchty>
and now i can link with bfd at least, i should compare the objects produced
<mitchty>
in either case, some of today is (re)learning more about ghc arm binaries
<dhess>
well at least you're learning something while working on this :)
<dhess>
I would like to dive into Core at some point as well but no time yet
<mitchty>
only problem is i feel like i'm yak shaving
<mitchty>
its cool and all but annoying when the goal is to use it more than debug
<mitchty>
also means I need to setup my "build arm and x86_64 stuff from git" build jobs
<dhess>
hopefully it will be worth it in the long run
<clever>
mitchty: qemu-user emulation or proper cross-compiler?
<mitchty>
clever: self hosted on hardware
<mitchty>
ghc 8.0.2 bulding 8.2.2, when it works takes about 29 hours
<dhess>
I wonder how much of my armv7l-linux Haskell stuff I could get away with cross-compiling
<dhess>
there's quite a bit of lens in there (therefore TH), but it's probably not the kind of TH that is system-dependent.
<mitchty>
i only cross compile the initial "build native compiler" stuff
<mitchty>
from then on its reuse things built natively
<dhess>
mitchty: well actually that's interesting.... what if you cross-compile ghc 8.2.2, i.e., x86_64-linux -> armv7l-linux
<mitchty>
i thought there was a fix for the cross th stuff
<dhess>
that would get around this self-building issue
<mitchty>
it might, dunno, won't help much since the packaging builders work the way i'm building it
<dhess>
mitchty: TH can do things like shell out to the system, apparently, so a 100% cross-compiling solution for TH will need some way to do that. I think angerman is working on that but as far as I know it's not fully baked
<mitchty>
dhess: ah yeah that would complicate everything
<dhess>
mitchty: yeah won't help you since you're the alpine maintainer :) But for my purposes it's fine. I should look into that.
<dhess>
My armv7l-linux Hydra remote builder spent 5+h last night just rebuilding ghc-8.0.2 last night against nixpkgs
<dhess>
would be nice to get that time back
<mitchty>
well i could try it, i gotta port ghc to x86 and aarch64 yet anyway
<mitchty>
and mips I suppose
<mitchty>
i think the issue is if your binutils isn't patched and using ld.bfd, cross compiling might just end up the same as native hosted
<dhess>
so it's a linker issue?
<mitchty>
well i was using ld.gold, but with the patch in the bugzilla linked i was finally able to use ld.bfd, but now i'm getting different segfaults
<mitchty>
but ones that make even less sense than the ones from gold
<mitchty>
aka that segfault in libc
<mitchty>
really early
<mitchty>
but i say try it, might be a musl libc thing
<dhess>
If I can figure out how to make nix-build use a cross-compiler I will give it a go
<dhess>
it has lots of other benefits for my case, namely that armv7l builders are rare and slow
<dhess>
If only the BeagleBoard people had an aarch64 version of the BeagleBone :(
<mitchty>
for that i geeked out on a tx2 dev board, 8GiB of ram :)
<dhess>
me too! I can't get NixOS running on it though :(
<dhess>
But even my TX1 isn't bad
<dhess>
and that runs NixOS just fine
<mitchty>
will try doing stuff in containers off the ubuntu os