<dhess>
looks like I can't upgrade to nixpkgs-unstable right now because qtwebkit is failing
<dhess>
so I will try with nixpkgs 310ad4345bbe42ae7360981243f6602a03fd232f
gchristensen has joined #nix-darwin
<dhess>
only anecdotal but I just built a couple of my own projects several times by cabal clean'ing and then cabal build'ing/test'ing in a loop, and didn't run into any issues.
<dhess>
also have been building a pretty decent number of Haskell packages (~100?) for macOS on my Hydra continuously for the past 3 days and haven't had any problems there, either. That's on the last 3 nixpkgs-unstable channel updates.
<dhess>
2 different Macs
johnw has joined #nix-darwin
<johnw>
dhess: back
<johnw>
dhess: indeed, a failed memory module
<dhess>
johnw: sorry to hear it
<johnw>
just a $75, several day inconvenience, that's all
<dhess>
hope it didn't corrupt anything persistent
<johnw>
yeah, that's the hope
<johnw>
I have ZFS snapshots going back a ways, so if I notice anything in the next month I should be able to recover it
<dhess>
apple blew it not putting data-level checksums in APFS
<dhess>
maybe in the future
<johnw>
I agree
<dhess>
johnw: ZFS snapshots of Time Machines or?
<johnw>
so, I rsync all of my data between machines using a utility I wrote called "pushme"
<johnw>
one of those destinations is a huge ZFS drive
<dhess>
ahh ok
<johnw>
with rolling daily, weekly, etc., snapshots
<dhess>
was it only manifesting on Setup's?
<johnw>
yes, only there
<dhess>
wow that's a weird one
<johnw>
so, GHC is strange in the way it allocates virtual memory
<johnw>
processes start by alocating a HUGE virtual address space, way more than other languages do
<johnw>
like 32G or something
hamishmack has joined #nix-darwin
<johnw>
I wonder if this changed which pages they were using
<johnw>
MemTest86 has found 3 errors so far. I just want to know exactly which module so I can avoid purchasing more than I need for this 4 year old machine
<dhess>
no way for it to control the mapping of virtual address to physical, though, and there was a particular physical address(es) that were causing the problem
<johnw>
that's true
<dhess>
so that's what's weird about it.
<johnw>
I just wonder if the usage pattern caused the OS to do something it doesn't normally do
<johnw>
oh, and very oddly, it *only* happened with the cabal ./Setup program
<johnw>
no other Haskell executables were affected either
<dhess>
ahh but Setup.hs is probably exactly the same in each project, compiles to the same image
<johnw>
likely
<dhess>
so maybe it was just being smart?
<johnw>
so a stray bit go into that executable?
<dhess>
is there de-duping in APFS?
<johnw>
i don't know, I don't use APFS yet
<dhess>
oh
<dhess>
well there's definitely no de-duping in HFS+ :)
<johnw>
in ZFS it costs too much memory, and I don't need the space
<dhess>
so that's not it. Anyway possibly related to the fact that Setup.hs compiled to the same image each time. But if so and that somehow got cached at the GHC+OS level there's some real magic going on there :)
<dhess>
yeah I don't use it in ZFS either
<johnw>
ok, now to pull 2 modules from this iMac so I can narrow down the faulty one
<johnw>
I had this happen to a NAS box once
<johnw>
was a really long time before I realized it was a memory module; thanks for reminding me tonight
<dhess>
of course
<johnw>
I have ECC in my NixOS ZFS box; I look forward to getting it in my next machine
<dhess>
I use it whenever practical. Mostly just a few embedded/HTPC-type devices now that don't.
<dhess>
and my laptop of course
<dhess>
it's a bit bizarre to me that it's not a standard feature from laptops up
<dhess>
laptops that cost what a MacBook does, anyway.
<johnw>
true
<johnw>
this shiny new touchbar macbook doesn't have it either
<dhess>
are you using any LSP-based emacs modes yet?
<dhess>
Yeah the iMac is basically a MacBook Pro with a slightly better GPU and a larger screen :)
<johnw>
i was interested in the iMac Pro, but I think I'm going to wait and see if the new Mac Pro materializes
<johnw>
I miss my old tower
<dhess>
those were beautiful.
<johnw>
I loved that machine
<johnw>
used it for 5 years, and then resold it as a decent price
<dhess>
Me too. I gave it to my dad, who's still using it as his main machine
<johnw>
I upgraded the GPU twice
<dhess>
me too!
<dhess>
I couldn't even see the boot screen. Crazy times.
<dhess>
and I use FileVault so that was interesting.
<johnw>
I use FV on my laptop noly
<johnw>
heh, found a way to provke an assertion violation in Z3
<johnw>
i'm cooking with gas tonight
<dhess>
on the iMac Pro FileVault is basically free because the encryption is offloaded to the T2 coprocessor.
<johnw>
i mean, it can't be *free* free, though
<johnw>
there's still a thermal cost
<dhess>
Eh compared to the CPU and GPU it's nothing really.
<johnw>
i worry about that closed in space
<dhess>
I'm sure the T2 makes no observable difference in the thermals
<dhess>
it's just a slightly better ARM than what's running in your Touch Bar on your MBP
<dhess>
and less powerful than any recent iPhone CPU
<dhess>
*and* it's already doing encryption by default, even if you don't have FileVault enabled. (A hardware-locked key so that you can't pull the SSD planes and use them in a different machine.)
<johnw>
ah
<dhess>
PC TPM-style.
<johnw>
like drives that do their own encryption
<dhess>
yes
<dhess>
The FileVault key is presumably just hashed into that key
<dhess>
I think you are right to be concerned about heat on the iMac Pro, but not because of anything the T2 is doing.
<dhess>
If you take a look at iFixit's teardown, that CPU+GPU heatsink is massive.
<johnw>
sometimes with my tower I would take the cover off and just blow a massive task fan at it
<dhess>
why, did it overheat?
<johnw>
the GPU did when I would overlock it
<dhess>
I never heard a peep from mine, even with newer GPUs
<dhess>
ohhh hehehe
<johnw>
I just liked have the OPTION of controlling temperature a bit
<johnw>
when I learn about things like them throttling the iPhone 6 due to battery, it feels so constraining
<johnw>
which happened to my mother-in-law
<johnw>
her iP6 feels like an iPhone 3
<johnw>
you know anything about mod_perl? or perlPackages?
<dhess>
perlPackages, just a tiny bit. mod_perl, no. I use nginx.
<johnw>
if I could get this CGI stuff working on nginx, I'd use that instead
<johnw>
but I spent hours last night and couldn't get it working
<johnw>
switched to mod_perl, and it worked right away
<dhess>
yeah this iPhone battery thing is bad optics, at the very least. They should have told people when they did that.
<johnw>
now I'm trying to properly nixify it
<dhess>
let's see, my perlPackages stuff is quite minor
<dhess>
oh
<dhess>
I do have a note in there, however. Hold on.
<dhess>
maybe relevant, not sure exactly what problem you're having.
<johnw>
thanks. I'm trying to package up Bugzilla to run under mod_perl in a container
<dhess>
I wouldn't think that was necessary for mod_perl but...
<johnw>
and I'm *this* close
<johnw>
dhess: here's a question
<dhess>
ok shoot, I will do my best
<johnw>
oh, never mind
<dhess>
oh come on, my best isn't *that* bad ;)
<johnw>
haha
<johnw>
dhess: ok, here's one for you: how can I temporarily run a mysql server that my build recipe can connect to?
<johnw>
because bugzilla won't create it's params.json file until it has initialized the database; but I only care about what will result, not the database, so I'm fine just standing one up for the duration of the build
<dhess>
hold on in the middle of fixing a PR
<dhess>
this needs to be created every time you build?
<johnw>
when I build the package yes
<johnw>
I'm wondering if this means I should make bugzilla service, because then I could depend on the mysql service, and have this step happen if the file is not present at startup time
<johnw>
having it happen at build time would be nice, but it's not *technically* right
<dhess>
that's what I would do I think. Anyway I can't think of any place I've done something like start a service during a build and then kill it.
<dhess>
if you want to get really funky could you patch in a test?
<dhess>
and add mysql as a nativeBuildInput?
<dhess>
and grab the output from the check phase?
zzamboni has quit [Quit: Leaving.]
<johnw>
hmm
<johnw>
that sounds horrific
<dhess>
yes it does
<dhess>
you could also do it in postBuild I guess.
<dhess>
can it connect to sqlite?
<johnw>
what is the difference between buildInputs and nativeBuildInputs?
<clever>
johnw: when doing a cross compile from x86 to arm, it will give you the arm version of the buildInputs and the x86 version of the nativeBuildInputs?
<dhess>
I believe nativeBuildInputs applies to both cross-compiles and native compiles
<dhess>
oh that is a better definition
<clever>
when doing a native x86 build, it just uses the x86 versions of both
zzamboni has joined #nix-darwin
<johnw>
ok, what I'm doing now is so evil, I expect members of the Hague to appear at my door tomorrow demanding restitution
zzamboni has quit [Ping timeout: 255 seconds]
zzamboni has joined #nix-darwin
__Sander__ has joined #nix-darwin
hamishmack has joined #nix-darwin
zzamboni has quit [Quit: Leaving.]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 240 seconds]
zzamboni has joined #nix-darwin
zzamboni has quit [Client Quit]
zzamboni has joined #nix-darwin
zzamboni has quit [Client Quit]
zzamboni has joined #nix-darwin
zzamboni has quit [Client Quit]
zzamboni has joined #nix-darwin
zzamboni has quit [Client Quit]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 268 seconds]
zzamboni has joined #nix-darwin
zzamboni has quit [Client Quit]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 240 seconds]
zzamboni has joined #nix-darwin
periklis has joined #nix-darwin
zzamboni has quit [Ping timeout: 265 seconds]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 265 seconds]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 248 seconds]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 265 seconds]
zzamboni has joined #nix-darwin
zzamboni has quit [Ping timeout: 252 seconds]
zzamboni has joined #nix-darwin
pjan_ has joined #nix-darwin
<pjan_>
LnL: I'm experiencing an interesting issue, that I can't seem to figure out solving. I have `nix.nixPath` set, have enabled bash.programs.enabled. It sources the /etc/bashrc when starting a new terminal, yet the `NIX_PATH` environment variable doesn't contain the path set (but /nix/var/nix/profiles/per-user/root/channels)
<LnL>
grep nix-daemon.sh /etc/profile
<pjan_>
yep, present
<LnL>
you don't want that since it runs after /etc/bashrc and will override stuff
<pjan_>
that's automatically set when installing nix, isn't it?
<johnw>
so, all those GHC segfaults yesterday were a bad RAM module btw
<LnL>
oh, that was causing the intermittent issues with your build?
<johnw>
yep
<LnL>
bummer, but it's easy to fix
<gchristensen>
oh, haha
<mitchty>
wish my ghc segfaults on my arm box were ram related
<dhess>
mitchy: ghc version?
<mitchty>
8.2.2
<dhess>
During the stage 2 compile of ghc itself?
<mitchty>
something in the stage2 build, not fully debugged it yet, takes ages to compile
<johnw>
hmm.. i can't seem to login to a mysql database with user/pass
<dhess>
mitchty: boom. Can reproduce.
<johnw>
but i can login if I'm that user (which of course apache isn't running as)
<mitchty>
dhess: oh? so its not just my port of ghc?
<dhess>
mitchty: is this with nixpkgs and the latest GHC stuff? Or a different distro?
<mitchty>
dhess: this is on alpine linux with musl libc
<mitchty>
around here make[1]: *** [utils/haddock/ghc.mk:21: utils/haddock/dist/build/ResponseFile.dyn_o] Segmentation fault (core dumped)
<dhess>
mitchty: It's not just you. If you grab nixpkgs from master or even one of the -unstable channels and try to build haskell.compiler on armv7l-linux, you will get the same thing
<dhess>
mitchty: yep, exactly. Reproducible every time.
<mitchty>
dhess: well thats good news, but now i'm curious whats wrong
<dhess>
mitchty: that's interesting, I figured it was something in the Nix recipe
<dhess>
mitchty: it's also odd that nobody has reported it as far as I can tell. Are we the only ones using ghc 8.2.2 on arm? :)
<mitchty>
looking to me like a ghc bug/reversion at this point
<mitchty>
dhess: probably :)
<dhess>
mitchty: 8.0.2 on Nixpkgs can build all of my apps just fine
<dhess>
ok well man am I glad I was idling here when you said that
<mitchty>
8.0.2 is fine as well, i tried bisecting things but it takes ages to build to bisect when it takes 29 hours to rebuild ghc
<mitchty>
heh, i'm glad too, been poking at it with gdb, looks kinda like its relating to some primitive integer operation
<mitchty>
my ability to debug ghc is pretty crap though
<mitchty>
gotta do a standup, bbiab
<mitchty>
dhess: so i'm guessing if we both are hitting the same bug its worthy of a new ghc bug, i'll fire one off tonight with what I've found so far
hamishmack has quit [Quit: hamishmack]
<johnw>
any mariadb masters here?
<copumpkin>
LnL: nope, never tried. I didn't even know it was there, and don't know why it is
<LnL>
hey! :D
<LnL>
just wanted to check, maybe I'll take a stab at getting rid of it then
<dhess>
mitchty: excellent, thanks!
szicari has joined #nix-darwin
hamishmack has joined #nix-darwin
<johnw>
LnL: I wonder if nix-darwin could be made to control what appears in the Sharing Preferences