<Phillemann>
What's the "has: clean-up" label for?
init_6 has joined #nixos-dev
hedning_ has joined #nixos-dev
hedning_ is now known as hedning
init_6 has quit [Ping timeout: 240 seconds]
init_6 has joined #nixos-dev
avn has quit [Remote host closed the connection]
init_6 has quit []
apaul1729 has joined #nixos-dev
<samueldr>
anyone actually knowledgeable with the kernel, or has leads or contacts, could take a look at #54509? not sure what can and should be done, and it probably needs to be reported to upstream if it hasn't already by aszlig
<aszlig>
samueldr: sorry, i didn't get back to posting this on the mailing list
<aszlig>
samueldr: i still have a draft email with the details, but i was on the wrong track there and the actual problem was a different one
<aszlig>
samueldr: do you have a non-NixOS system available right now?
<gchristensen>
what do you need one for
<gchristensen>
?
<aszlig>
i vaguely remember that i tested it on a non-nixos system, but i don't remember the result
<samueldr>
no, only managing the release
<gchristensen>
(I can get you one)
<aszlig>
gchristensen: for writing the bug report to the unionfs mailinglist
<samueldr>
(which here I mean making sure the problem isn't overlooked)
<gchristensen>
aszlig: what distro would you like?
<aszlig>
or well, let me rephrase it:
<samueldr>
I think aszlig needs a non-nixos repro of the issue?
<aszlig>
does the problem also occur on non-nixos hosts when running NixOS VM tests on kernel >= 4.19?
<gchristensen>
I can't do that, but I can provide a Linux machine
<samueldr>
the "Operation not permitted" isn't limited to the VM tests, a build-vm vm also exhibited the same issue, but I think they do share the same way to share the store
<gchristensen>
aszlig: do you have IRC logs between you and me from 2018-12-08?
<aszlig>
gchristensen: oh, i think so, let me check
<gchristensen>
"yeah, still works fine" is something I said, if that helps you finda key part of the convo :)
<aszlig>
gchristensen: found it... but it was on nixos not non-nixos distro
<gchristensen>
ah...
<aszlig>
i wanted to narrow it down whether it changes anything if the host kernel is 4.19
<aszlig>
and the test you did was on 4.19
<aszlig>
but still nixos
<samueldr>
4.18.20 here on nixos host if it matters
<aszlig>
samueldr: we need a non-nixos host :-D
<samueldr>
who would do such a thing?
<aszlig>
because i wanted to give the kernel maintainers a way te reproduce it
<gchristensen>
step one: install nixos ;)
<aszlig>
so i have to make sure that it works with nix on non-nixos systems
<aszlig>
gchristensen: well... the original maintainer i wrote the original bug report already asked me whether i could make a test case for non-nixos systems
<aszlig>
and i tried very hard to reproduce it on something like fedora
<aszlig>
(and it was actually quite painful)
<gchristensen>
aszlig: well, if I get you a, say, ubuntu box, would that be useful?
<aszlig>
but couldn't reproduce it (with fedora as *guest* of course)
<aszlig>
gchristensen: well, don't we already have users who just use nix on a non-nixos system?
<aszlig>
... and also non-macos system of course :-D
<gchristensen>
all the ones I know of saw the light
<aszlig>
gchristensen: yeah, same here, even the machines non-technical relatives run nixos X-D
apaul1729 has quit [Quit: Ping timeout (120 seconds)]
<infinisil>
I feel like the checkmarks in the PR template are really useless
<infinisil>
Nobody actually reads the contribution guidelines, but tickes the box of course (similar to accepting terms and conditions)
<infinisil>
Sometimes people don't even make sure the build succeeds, but tick the box for having built it still
<infinisil>
And now I just saw a PR that checked the box for having executed all binary files in the package. There are no binaries in the package though..
<infinisil>
And that's really annoying when we have such a high rate of incoming PR's, but only such little reviewers. It makes me not trust these checkmarks, even though I really want to trust them
<qyliss>
I don’t really find them at all useful
<qyliss>
I fill them out every time, but feels kind of pointless
<infinisil>
Yeah same
<samueldr>
I was toying with the idea of providing a default template for PRs, for manual PRs, the usual like now, but with fewer checkboxes (maybe only: I have read the contributing guide etc.)
<samueldr>
but also have something like nix-review which could output a pre-configured PR template
<tilpner>
I open the contribution guidelines every time, in case they changed
<infinisil>
I want a checkmark for "I have formatted my commit messages according to the guidelines <link>"
<tilpner>
But I agree that the box is mostly useless
<samueldr>
so it could list all the binaries, or do nix path-info automatically
<tilpner>
If the author didn't read them, it'll be evident from their commits
<tilpner>
(And if it's not evident, it didn't matter)
<samueldr>
I'm kinda thinking a good chunk of the users in this channels both aren't the ones mis-filling them, and whom it's the most important to fill correctly
<tilpner>
And the closure size box seems like something borg should do
<samueldr>
right
<samueldr>
but yeah, main idea was to have a tooling to generate the PR bodies, for those that want to use it, maybe streamlining some contributions
<infinisil>
How would that work? Is there some GitHub functionality for that?
<infinisil>
Or just with a manual tool and copy pasting over?
<samueldr>
no functionality AFAICT, which was my main blocker
<samueldr>
yeah
<samueldr>
or with github api token
<infinisil>
I think I'd prefer having more automated checks
<samueldr>
but then, ofborg does a part of what could be really useful already
<{^_^}>
#54392 (by ivan, 1 week ago, open): ffmpeg, mpv: enable hardware-accelerated decoding with CUDA
<Profpatsch>
Oh, somebody bounded the width of the manual! Tasty.
<Profpatsch>
Though it kinda side-scrolls now.
<ivan>
oh it's got conflicts I need to resolve
<gchristensen>
ivan: fix those up, and Ill trigger ofborg
<ivan>
gchristensen: fixed
<gchristensen>
I hear tell ofborg should learn a new impacted-packages range, what should it be?
<ivan>
is that a question for me? I am confused
<gchristensen>
for the channel
<gchristensen>
500-1000? 500-10,000?
<infinisil>
Alternatively I just made the suggestion to use percentage (of all derivations) ranges instead (in #nixos-borg), so we could have labels for <1%, 1-20% and >20%
<gchristensen>
change 100-500 to 100-1000, and do a 1000-10000, and a 10000+?
<gchristensen>
( infinisil's proposal on %s has merit, but isn't within scope of "oh, I could do that tonight")
<infinisil>
( Fair enough )
<Profpatsch>
zimbatm: lol, I managed to copy the wrong hash.
<Profpatsch>
Didn’t we design the new error message together?
<Profpatsch>
gchristensen: why not an order of magnitude?
<Profpatsch>
Or why not: 2^n? :)
<gchristensen>
like 100-1000, 1000-10000, and a 10000+?
<Profpatsch>
Hm, it gets hard to read
<gchristensen>
we can add ,s
<gchristensen>
2^ns are pretty nice divisions
<Profpatsch>
or 100–1k, 1k–10k, 10k+
<gchristensen>
even better
<Profpatsch>
There was this way of estimating orders of magnitude
<gchristensen>
ping lnl on this chat
<gchristensen>
probably need an rfc on this
<Profpatsch>
It went by logarithmic scale, so 3–20 is 10^1, 30–200 is 10^2, 300–2000 is 10^3 and so on.
<Profpatsch>
“ To round a number to its nearest order of magnitude, one rounds its logarithm to the nearest integer.”
<gchristensen>
is 1k changed packages ok for master?
<Profpatsch>
So you can do n=round(log10(pkgs)) and display 10^n
<gchristensen>
let's get some Hot Takes
<gchristensen>
is 1k changed packages ok for master?
<gchristensen>
infinisil, LnL, samueldr
<samueldr>
depends on the cost of building said packages!
<andi->
gchristensen: that is an interesting topic... lately I have heard <5k would be fine within a day.. I am hacing the same discussion about ~5kish packages for the firefox bump as well..
<LnL>
not sure the buckets are that important for those large numbers, but something between <1k and >10k can be useful
<infinisil>
gchristensen: I think 1000 is alright for master
<LnL>
depending on what type of packages it is it can be ok
<infinisil>
In general
<gchristensen>
is 5k changed packages ok for master?
<infinisil>
I'd like to know how many packages we have in total (that get built by hydra, not counting arch duplicates)