<gchristensen>
niksnut: it looks like we're going to have to issue another release :( so sorry. I'm confused about the issue though, I've run the install 3x this morning now without issue
<gchristensen>
obviously it is a problem, but the scope may not be quite so big
<shlevy>
Anyone around with a fresh darwin machine/VM they can test on?
<LnL>
the sudo call won't have the local variables
<domenkozar>
sudo -E will
<gchristensen>
shlevy: I do, and already did
<shlevy>
Is it possible this is a OS X version issue?
<gchristensen>
no
<shlevy>
OK
<domenkozar>
but probably best to pass explicitly
<gchristensen>
I tried on 2 versions this morning
<gchristensen>
I do , domenkozar
<gchristensen>
_sudo "to update the default channel in the default profile" \HOME="$ROOT_HOME" NIX_SSL_CERT_FILE="$NIX_SSL_CERT_FILE" "$NIX_INSTALLED_NIX/bin/nix-channel" --update nixpkgs
<LnL>
well the log doesn't say it's using -E
<gchristensen>
the problem is here: if [ -z "$NIX_SSL_CERT_FILE" ] || ! [ -f "$NIX_SSL_CERT_FILE" ]; then
<gchristensen>
not sudo
<LnL>
yes that fails with set -u
<shlevy>
Surprised shellcheck didn't catch that
<LnL>
but I'm guessing sudo is why it's not defined
<gchristensen>
me too, shlevy
<shlevy>
LnL: You mean running the installer itself under sudo?
<LnL>
doesn't the installer set that somewhere
<gchristensen>
they can't run the installer with sudo
<gchristensen>
I'm testing my fx
<shlevy>
domenkozar: Should we just set NIX_SSL_CERT_FILE= in travis while we wait for the Nix bump to go through?
<gchristensen>
yes
<shlevy>
Alternatively we can ask niksnut to roll back to 2.0.2
<shlevy>
The install link, I mean
<domenkozar>
shlevy: as a user workaround, deploying travis fix will take longer than shipping Nix :)
<shlevy>
OK :)
<gchristensen>
wow
<LnL>
lol
<gchristensen>
ok I've replicated the problem, running `set -u` prior to running the installer itself
<shlevy>
gchristensen: do you have merge rights to Nix?
<gchristensen>
I do
<shlevy>
OK
<gchristensen>
the shame :(
<shlevy>
Who has the ability to make nixos-homepage changes take effect? niksnut , anyone else?
<gchristensen>
rob?
<shlevy>
ikwildrpepper: Around? :)
<gchristensen>
merged
<shlevy>
Nice
<gchristensen>
I was just fixing other issues with `set -u` in the profile when all thi started... jynxed it I guess. I'm sorry for letting this through. it is time to think more thoroughly about how to verify installer patches
<shlevy>
I think there's something unexpected going on here... Possibly to do with piping into /bin/sh?
<shlevy>
Why isn't set -u mattering?
<gchristensen>
it is mattering :$
<gchristensen>
I must have been sloppy about my testing
<LnL>
I test both install and uninstall on travis for nix-darwin
<shlevy>
Oh, even if you didn't set it yourself
<shlevy>
?
<shlevy>
gchristensen: Maybe we should take some time to list out all of the different "configurations" possible and get exhaustive coverage,it's not that much... OS, single vs multi, a few env switches (or maybe just NIX_SSL_CERT_FILE?)
<gchristensen>
yes I agree
<gchristensen>
I've made a hacky install matrix test before, across different distros
<shlevy>
Refusing to install over a preexisting insall definitely helps
<gchristensen>
I'm glad you see that :P I've gotten a lot of requests for making it automatically "fix" an existing install
<shlevy>
Even if that should exists, it's easily an independent product :P
Synthetica has joined #nixos-dev
<shlevy>
Uninstall and reinstall, making a dump of paths if you really want to first
<shlevy>
Definitely shouldn't be part of the core installer
<gchristensen>
yeah
<gchristensen>
I'm working on a build of the installer, a bit slow since I've blown away my store recently.
<gchristensen>
... I'm compiling cmake and git on the nixpkgs-unstable branch? O.o
<gchristensen>
ok I'll be able to test the Nix installer in approximately 4 days, after I finish compiling Parsec and shellcheck etc.
<shlevy>
gchristensen: Why do you need to test against unstable?
<gchristensen>
its just the channel that was setup
<shlevy>
Ah
<gchristensen>
I guess I should switch to darwin's 18.03 :)
<gchristensen>
the darwin 18.03 channel also has me building cmake and git (?)
<gchristensen>
shlevy: the reason it is building against unstable is it fetchGit's nixpkgs directly
<cransom>
one of my favorite hacks to avoid a chunk of the main due to garbage collection is `nix copy --all --to file://somewhere` and run a webserver on that directory. all the old derivations end up in a cold-ish storage, compressed. if i gc, i don't end up building for days again.
<dtz>
woooo more mem fixes \o/
<gchristensen>
LnL: I think my macos machine is about to give up the ghost :(
<clever>
cransom: nix copy can also just --from the same file://somewhere
<cransom>
clever: sure, but over time, i don't need to keep every package i've built in the store. and the machine that holds all of this does binary cache duties for the laptop as well.
Sonarpulse has joined #nixos-dev
<clever>
cransom: oh, also, nix copy --to local?root=/somewhere will create a full /somewhere/nix/store, uncompressed, and it expects a chroot when in-use
<clever>
it also has its own /nix/var/nix/profiles
Sonarpulse has quit [Read error: Connection reset by peer]
Sonarpulse has joined #nixos-dev
Sonarpulse has quit [Remote host closed the connection]
ixxie has joined #nixos-dev
FRidh has joined #nixos-dev
<gchristensen>
niksnut: I like what you're doing with the install script :)
<niksnut>
yeah, makes a lot more sense to include it in each release directory and symlink from the homepage
<gchristensen>
I have, just moments ago, updated an internal install script to have a fetched-and-modified copy of the 2.0.3 install script ... I completely agree :D
<cransom>
interesting. reproducing that hydra link locally fails for me as the default TMPDIR has an extra / in it.
<srhb>
gchristensen: That's not the nixpkgs one though is it
<srhb>
gchristensen: I think 2.0.2 was fine in the nix jobset as well
<srhb>
Er, project
<ikwildrpepper>
shlevy: pong
<gchristensen>
oh I seee
<ikwildrpepper>
the website should update itself every 15-20 minutes, I believe
<ikwildrpepper>
has it not updated yet?
<gchristensen>
we wanted you to merge a PR :) its alldone
<ikwildrpepper>
yay
<ikwildrpepper>
sorry, wasn't in the office today due to airconditioning not working :(
<ikwildrpepper>
too hot to work (for Dutch people)
<domenkozar>
gchristensen: thanks for swift action :)
<copumpkin>
niksnut you have anything planned for my "content-addressable" binary cache issue or is it esoteric enough that I should just fix it on my side and we assume that nobody else will encounter it
<gchristensen>
domenkozar: my deepest appologies for it being necessary :o
<domenkozar>
gchristensen: the only thing unnecessary is status quo
contrapumpkin has quit [Read error: Connection reset by peer]
<aristid>
LnL: i'm not sure if i summarised the results of my rustc experiments: on my server (16 cores, 32 gb ram), i consistently get segfaults in various stages of the build, it seems it's usually in the vendored llvm part. on my laptop, it seems to succeed (i only tried once because it takes ages there), 4 cores & 8 gb ram
<aristid>
LnL: enableParallelBuilding = false on the server does not help, but it seems that some part of the expression is built with -j16 even if i do that
<LnL>
aristid: 1.26?
<aristid>
LnL: 1.25
<LnL>
the vendored llvm build used to ignore enableParallelBuilding but I thought somebody fixed that
<LnL>
this does sound kind of similar to the weird errors I ran into when trying to separate llvm into a separate drv
contrapumpkin has joined #nixos-dev
<aristid>
LnL: not really sure what to do from here
<LnL>
can you reproduce this before the staging merge?
<dtz>
new download.cc stuff has problems if contentEncoding is "br", ends up attempting to decompress the result.data() when the contents were sent to request.dataCallback ... i think
<dtz>
pretty sure I've looked at this, found that bug, and reported it previously. Probably not with an ideal fix though.... but can't find any of that inmy logs/revs for shame
<gchristensen>
is it possible to kill a buildstep that isn't its own job?
<niksnut>
"lucifer" is the old hydra
<niksnut>
contrapumpkin: yeah, might be best to fix your .narinfo files
<LnL>
can we cancel staging builds until master is back in a somewhat reasonable state?
<dtz>
this code's a doozy :D
<gchristensen>
LnL: sure
<gchristensen>
LnL: can you link to the specific jobset you want canceled?
<LnL>
I can do it, just want to check if everybody think it's ok with that
<dtz>
curl can decompress things it's built w/if you tell it to, I think...
<dtz>
or after we receive header dynamically add decompression sink?! lol
<samueldr>
the `types.enuml` should be `types.enum l`, xmlformat reduces spaces when there is nothing else than sub elements
<gchristensen>
ooh
<samueldr>
(let me copy the paragraph in their doc)
<gchristensen>
I see it
<gchristensen>
semantically those terms could use work
<samueldr>
at a glance, it looks like there's no way to treat `varname` and `replaceable` as space-separated inside a `term` with xmlformat
<samueldr>
and my docbook experience is severely limited
<gchristensen>
I'll experiment and find something
<samueldr>
oh wow, my head is spinning (looking at the term page in the docbook TDG)
<gchristensen>
yeah I don't think we want term
<samueldr>
it's varlistentry that requires term OR listitem, so either varlistentry isn't right (I don't know) or instead of term, listitem?
Cale has joined #nixos-dev
<gchristensen>
varlistitem should be fine, I think listitem would be better
<samueldr>
hm, at a glance it seems that term is right
<gchristensen>
yeah could be
<samueldr>
> Each varlistentry contains one or more terms and their description or definition
<{^_^}>
undefined variable 'Each' at (string):41:1
<gchristensen>
well it isn't exactly a term
<gchristensen>
this seems more like a Term to me: <term>Strings</term>
<gchristensen>
let me talk to someone and get some advice
<gchristensen>
great catch, samueldr
<samueldr>
I dislike how xml handles spaces, or more to the point, how it doesn't really and they can but do not necessarily have meaning for the software using it
<gchristensen>
indeed
<samueldr>
and then it's converting to another thing (not really xml) which has its own set of rules for spaces
<gchristensen>
"when this occurs, any whitespace in text nodes adjacent to non-reformatted nodes is discarded."