<samueldr>
error happened inside ofborg while borging on the PR
<samueldr>
mainly it'S for ofborg handlers to look for errors
<siraben>
Ah
<gchristensen>
I'd like to expand the team of handlers, btw
<gchristensen>
cole-\h has been doing really great, and I'd love for there to be others
ajs124 has joined #nixos-dev
Scriptkiddi has joined #nixos-dev
das_j has joined #nixos-dev
orivej has joined #nixos-dev
cole-h has joined #nixos-dev
<cole-h>
Just FYI: starting the process to redploy ofborg right now. I expect no interruptions, but, well, not everything goes as planned :)
<cole-h>
Deploy succeeded, nothing out of the ordinary to report.
srk has quit [Ping timeout: 268 seconds]
srk has joined #nixos-dev
<siraben>
yay, 99.82% reproducible!
cole-h has quit [Ping timeout: 240 seconds]
* energizer
glances warily at linux-5.4.83.drv
<energizer>
(if you measure by mass instead of count it's probably a lot lower)
<samueldr>
energizer: refresh harder
<samueldr>
no linux kernel on r13y
<samueldr>
only 3 unreproduced paths
pinpox has joined #nixos-dev
<energizer>
i refreshed as hard as i know how to
<energizer>
it's still there
<samueldr>
:/ things were iffy earlier yesterday too
<samueldr>
cache is hard
<energizer>
i was gonna fiddle with one of the nonrepro packages but i couldn't figure out how to build it on my computer
<energizer>
is there a way to ask nix to give me the shell whose builder builds $path, and then tell it to go ahead and build it
<symphorien[m]>
genericBuild is the command
<symphorien[m]>
It will fail at installPhase
<energizer>
i tried that but it didn't produce the path. i can show you
<energizer>
b7m9y5mxvi54bxhl8j75rxhmfrnj2m07-python3.8-pytest-6.1.2 is one of the paths on r13y. nix-shell $(nix-store -q --deriver b7m9y5mxvi54bxhl8j75rxhmfrnj2m07-python3.8-pytest-6.1.2) does not exist and cannot be created
<energizer>
er
<energizer>
b7m9y5mxvi54bxhl8j75rxhmfrnj2m07-python3.8-pytest-6.1.2 is one of the paths on r13y. nix-shell $(nix-store -q --deriver /nix/store/b7m9y5mxvi54bxhl8j75rxhmfrnj2m07-python3.8-pytest-6.1.2) does not exist and cannot be created
<energizer>
so i tried pulling master and running nix-shell -A python3Packages.pytest but that doesn't produce those files either
<energizer>
(running genericBuild)
<energizer>
i tried with --pure too
<energizer>
on the other hand, the whole point is that it's not reproducible so maybe i shouldn't be surprised :)
<energizer>
or maybe now it is reproducible but the website is out of date
<aminechikhaoui>
If (defined $limit) ? I don’t know perl, just from looking at the log function a bit down in that file
<gchristensen>
hmm it looks like I could also do my $limit = trim $c->request->params->{"limit"} || 500;
<siraben>
worldofpeace: edited the release notes in the PR
<gchristensen>
if anyone familiar with perl is around: https://github.com/NixOS/hydra/pull/853 <- this is fixing an issue causing hydra to be visibly (but not totally) down
<{^_^}>
hydra#853 (by grahamc, 2 minutes ago, open): search: limit results to 50, default to 10
<supersandro2000>
I just realized why my nixpkgs-review hangs
<supersandro2000>
it asks hydra about the packages....
<gchristensen>
hopefully this PR doesn't damage it in aty way
<supersandro2000>
siraben: I also saw someone who might do python = python3 removals. Maybe something for you in the future if it isn't done in a few weeks
abathur has quit [Quit: abathur]
<gchristensen>
most web queries 200ms, evaluator 150ms, queue runner 800ms, search page 500,000ms, notify 150ms someone who is good at reviewing pull requests please help me mereg this. my family is dying
<andi->
gchristensen: sure, looking
<gchristensen>
thank you :D
<gchristensen>
hopefully this handles it for today, I have a lot of chores to do ;_
<andi->
gchristensen: you have a comment I forgot to make that a new review but please add the LOCAL there unless I read the postgresql docs wrong
<gchristensen>
andi-: may I get a final check and possibly thumb?
<andi->
gchristensen: one more nit: the semicolon after the statement is likely not needed as the statement is already closed. Usually you only need that on the sql console or for multiple statements. Does this ORM treat it differently?
<gchristensen>
andi-: the examples I found in the repo used semis
<gchristensen>
not all do though
<andi->
ok, i'll set that slip :D
<gchristensen>
I dropped the ; and fixed lukegb's comment too
<gchristensen>
I also tried a $c->model('DB')->schema->storage->dbh->do("SELECT pg_sleep(21)"); to double check it worked :)
<lukegb>
Nice, thanks 😀
<gchristensen>
another round of thumbs? thank you both, this has been meaningfully useful and I'm glad for it
<andi->
Thumbs-as-a-Service
<clever>
and i just filled a pr to fix thumb mode in arm-embedded, lol
<andi->
You can book the IRC2Thumbs extra for just $9.99
<gchristensen>
lol
<andi->
gchristensen: wait, since when doesn't hydra CI run the ldap test etc?
<gchristensen>
I don't know :(
<andi->
ahh, it requires KVM and SH Actions doesn't support that.
<gchristensen>
woohoo, those terrible queries are killed
<gchristensen>
thank you both, andi- + lukegb !
<andi->
thank you for doing something about it
<gchristensen>
gotta keep the community unblocked :)
<andi->
Would be interesting to know what people are commonly searching for
<andi->
once we know that we can just remove the derviation or jobset or outputname or whatever search feature
<andi->
maybe nobody is using the search due to the performance -> we can do anything!!! :D
<gchristensen>
just pkgs.chromium afaict, back in an hour or so we can look at more logs together :)
<lukegb>
I think it takes the (uncompressed) initial initrd and then adds the squashfs image to it, then recompresses the whole thing
<pie_>
that double compression sounds funky
<pie_>
(what Im trying to do is make the squashfs step faster because its really bogging down my iteration speed)
<lukegb>
didn't gchristensen do something for making iteration faster there
<pie_>
i managed to change the squashfs compression options but it still seems too slow so im trying to turn compression off alltogether, it's only 2 gigs.
<__monty__>
makeWrapper's use of `exec -a` is failing me (on macOS). Would changing it to use a C wrapper maybe be an option (to get access to execle and friends)?
<gchristensen>
makeWrapper should use nixpkgs's own bash, not sure how it could be a problem?
<gchristensen>
aszlig: can you take a look in to fixing the link? it is causing quite a lot of load, even with the mitigations
<__monty__>
gchristensen: I'm not sure how it could be a problem either but I'm still running into it : )
<gchristensen>
__monty__: is the thing you're exec'ing a script?
<__monty__>
Yes.
<samueldr>
gchristensen: one of the problems is that it's using the bash from the "nativeBuildInputs", but maybe not __monty__'s
<__monty__>
Ah, so exec -a is setting bash's $0?
<gchristensen>
__monty__: is the thing that is being exec'd have a shebang which is pointing to a script?
<aszlig>
gchristensen: a jobset however would
<aszlig>
... have the issue that it's systemd-dependant
<__monty__>
The shebang is "env bash" but replaced by the bash in the store of course.
<samueldr>
with makeWrapper it's actuall @shell@ replaced by runtimeShell, no?
<__monty__>
I don't think the wrapped script's shebang poses a problem though. Is my understanding correct that the wrapper really invokes bash when it tries to execute the script and so `exec -a` is setting bash's $0?
<samueldr>
plausible
<samueldr>
if it's not running the exectuable, and/or being wrapped multiple times
<samueldr>
(note that you didn't say what "failing me" meant recently)
<samueldr>
if it's about not renaming the child process, that sounds about right
<samueldr>
if you have e.g. foo -> .foo-wrapped -> your-own-wrapper -> actual-exec
<samueldr>
or uh, even .foo-wrapped being your-own-wrapper
<aszlig>
gchristensen: yeah, but only because it hits the limit
<pie_>
Wow there is a lot of copying in this proces. Theres at least like, 3-4 round trips through the disk.
<pie_>
And its not even like I can put it in a ramdisk because nix store?
<tilpner>
If the entire closure fits into your ramdisk, you can change the entire store location
<gchristensen>
I assumed you wanted the query to return results :/
<pie_>
tilpner: maybe.
<tilpner>
I used lz4 compression for the squashfs, when I needed quicker loops
<pie_>
Yeah I just switched to doing that but its taking forever to build the initrd too.
<pie_>
Im not sure I didn't make it slower somehow.
teto has quit [Ping timeout: 260 seconds]
<tilpner>
Ahh, my initrd had two files
<pie_>
With this the initrd is 3.2 gigs and takes 15 minutes https://bpa.st/6VSXU (i dont even know if this will run yet)
<tilpner>
Why is your initrd that large?
<pie_>
its the netboot installer or something
<pie_>
hm still. good question
<pie_>
Yeah, how big is netboot-minimal?
justanotheruser has quit [Ping timeout: 265 seconds]
aminechikhaoui has joined #nixos-dev
<pie_>
Im timing the normal process right now.
teto has joined #nixos-dev
AlwaysLivid has joined #nixos-dev
<sphalerite>
energizer: re what you said in #nixos-chat: nice that you're taking a look! bz2 can take streamed input, so the file may very well be created and then gradually filled — even if it's compressed.
<__monty__>
samueldr: No, it's just a bash script with a shebang and a single wrapper. The bash script uses $0 in help output for example and that's rendered as ".nb-wrapped", which is confusing as the user never called ".nb-wrapped".
<energizer>
sphalerite: ah, nice
<samueldr>
__monty__: would it be fair to ask "does bash respect argv0 being replaced in scripts?"
<samueldr>
and is it really macOS specific?
<samueldr>
(not trying to dismiss your issue, I see it)
<samueldr>
can you check the output of `ps` or similar if argv0 is replaced when running that bash script from the wrapper? maybe add sleep 100 or something to the help command
<samueldr>
I'm curious if it's really something like bash doesn't care about argv[0] and always uses some other value for $0
justanotheruser has joined #nixos-dev
<pie_>
tilpner: with the default config it takes 22 minutes...
<pie_>
so that only shaved off 7 minutes or so
<pie_>
Hm. I guess I could try making this part smaller and setting a remote builder.
<__monty__>
samueldr: What I think is happening is that bash's $0 is changed but bash has no reason to pass that on to the script, it sets $0 to the name or path it was passed.
<samueldr>
just looked on my Linux, a pure bash repro
<samueldr>
and yeah, $0 with -a is not set to the arbitrary value I passed
<samueldr>
exec -a my-utility bash -c 'echo I AM $0' # is fine though
<samueldr>
I'm pretty sure a C wrapper wouldn't change the behaviour, since it seems to be on the callee that it happesn
<samueldr>
happens*
<samueldr>
(I could be wrong)
<__monty__>
Yes, I agree. I was misunderstanding the problem. I figured somehow I was using my system's bash, which is too old for the -a flag I think. Hence reaching for a C wrapper.
<__monty__>
Thanks for working through this with me. I think I'll PR adding a note in the docs to mention that makeWrapper has this interaction with scripts.
<__monty__>
And I'll just patch the script to hardcode the proper name.
<samueldr>
n/p, it's also nice to have an answer about it, since that's the kind of things I see all over, Usage: .whatever-wrapped
<__monty__>
Ah, I was thinking "Just my luck that no one's run into this before >.<" : D
<__monty__>
I'll give this another think though, maybe there *is* a nice, general solution.
<samueldr>
ran into this as a user, didn't look further :)
<samueldr>
gchristensen: yeah, I was almost certain _that_ index type wouldn't help enough
<samueldr>
but I never had heard of it before taking a stroll through the whole irc logger code base
<gchristensen>
there are only so many index types :P
<__monty__>
Btw, could the urls on the package search page be made to stand out a little more? I had a hard time realizing they were even there and even now I really need to look for them. I'm talking about "source" et al. The hydra jobset urls stand out a lot more.
<samueldr>
I think the `gin` function is part of the magic
<samueldr>
(which AFAIUI is really a thing that works only with "dictionaries")
<samueldr>
(like words)
<__monty__>
Not sure this is the appropriate channel for this. Feel free to point me at a more appropriate one : )
<samueldr>
appropriate enough
<samueldr>
if you can give details about how you'd fix it, an issue on the nixos-search repo would be how to progress with that
<gchristensen>
let's see what happens: hydra=> CREATE INDEX expIndexBuildOutputsPathGin ON BuildOutputs USING GIN(to_tsvector('english', regexp_replace(path, '[\./-]', ' ', 'g')));
<samueldr>
gchristensen: note that I don't know (yet) what gin is
<samueldr>
(I haven't been following the latest developments of nixos-search really)
<__monty__>
The OP already mentions styling. I don't think more than an underline and a different color is necessary. Maybe homepage and source could stand out more than license for example but that's minor and probably subjective.