<niksnut>
probably you lose privacy-invading location services, so no big loss there ;-)
<gchristensen>
sgtm
<aszlig>
dito
<niksnut>
aszlig: I can't imagine what caused me to write that line :-|
<niksnut>
btw most API calls are to the safe browsing API (~589K in the last month)
<niksnut>
also 45K to the chrome spelling API
<aszlig>
mhm, the same thing that's used by firefox as well
<niksnut>
and 1.4K to the speech API
<gchristensen>
:/ safe-browsing API is good :/
<niksnut>
I didn't disable that, just the maps API
<gchristensen>
ahh ok
<gchristensen>
(* and not privacy-invading)
<niksnut>
well...
<niksnut>
maybe it sends a lot of URLs to google
<gchristensen>
nah
<gchristensen>
the browser downloads a list of hashes and compares client-side
<gchristensen>
*well, it CAN send URLs to the API, but the browser uses the bulk API of fetching the hash list)
<aszlig>
gchristensen: well, even if it's hashed it still means that google can have a mapping between hashes and some well-known urls...
<gchristensen>
yes but you download the entire list of hashes, and the diffs to the hash list over time
<aszlig>
and they can answer things such as "has the user accessed site x?"
<gchristensen>
you never send hashes off to google
<aszlig>
iirc chromium was using the lookup api the last time i was involved in chromium
<gchristensen>
:o
<aszlig>
(the lookup api with hashing of course)
<aszlig>
niksnut: so can i remove that line?
pie___ has quit [Ping timeout: 240 seconds]
<gchristensen>
niksnut: seems we should opt to use more space and less RAM, it is much cheaper to use more space than to get more RAM in pretty much every environemnt
<gchristensen>
I just had to upgrade my machine from 4gb to 8gb to import a package, so I'm a bit fresh from that perspective
<dtz>
the certificate thing? page itself seems nice :3
<gchristensen>
yeah
page has joined #nixos-dev
Drakonis[m] has quit [Changing host]
Drakonis[m] has joined #nixos-dev
Drakonis[m] has joined #nixos-dev
<infinisil>
Is there a way to prevent hydra from even evaluating a top-level nixpkgs attribute?
<gchristensen>
why for?
* gchristensen
doesn't like where this is going ...
<infinisil>
Because it would be a builtins.fetchGit
<infinisil>
I'm expirementing with getting modules to work with NUR
<infinisil>
Nicely*
<gchristensen>
probably better to just have users put imports = [ <nur/modules.nix> ]; or something
<infinisil>
Although, that wouldn't work now that I think about it
<gchristensen>
of course, the module system is not POL so it is dangerous
<infinisil>
POL?
<gchristensen>
POLA*
<infinisil>
POLA?
<gchristensen>
principle of least authority
<gchristensen>
any module can change any thing
<LnL>
either it's a drv in which case meta.hydraPlatforms or it's a set and then avoid recurseIntoAttrs
<infinisil>
(I'm gonna scratch that idea), but it's a set, but to even evaluate it you'd need builtins.fetchGit, which is what I would've wanted to prevent
<infinisil>
But yeah gchristensen, the NIX_PATH thing is probably the best way
<infinisil>
If it works..
<aszlig>
or maybe even imports = [ <nur/foo/bar> ]; instead of a modules.nix that contains *all* modules
<aszlig>
so if people add repositories with modules that define options unconditionally you don't end up with a messy system
<aszlig>
infinisil: but i'm not quite sure why you want hydra to avoid evaluating the top-level attribute
<infinisil>
Eh doesn't matter
<aszlig>
i mean if the top-level attribute is builtins.fetchGit, why not make it an input?
<infinisil>
an input to what?
<aszlig>
jobset input
<infinisil>
Not sure what you're getting at, the idea of putting it in nixpkgs is so not gonna work anyways
<aszlig>
ah, so you wanted to get it in nixpkgs?
<infinisil>
Yeah, just the url or a way to fetch it though, without a pinned hash
<aszlig>
ah, okay, then never mind... i thought it was about building NUR itself with hydra
<dtz>
im so glad I now know that borg prints set of changed paths
<dtz>
was rather tired/sad recomputing locally--always "hoping" it was same results used elsewhere, etc. lol
<dtz>
and surprisingly expensive to compute
<LnL>
yeah...
<dtz>
anyway \o/ that it's included :D:D was my point
<LnL>
indeed, I often pick out a bunch of attributes for testing a library
<infinisil>
(but then you can't get a reproducible system because can't pin it)
<aszlig>
... and the contents for nur.nixos.org would be built from a git repos
<infinisil>
Ah I see, neat idea!
<aszlig>
that way it can also be statically served
<infinisil>
How do you mean that?
<aszlig>
the docroot for nur.nixos.org
<aszlig>
ideally it should only contain a bunch of json files or something similar with meta-info about the target repository
<aszlig>
it could even be a 1:1 clone of the git repos
<infinisil>
Or just redirect to github.com/<user>/<repo>/archive/<rev>.tar.gz?
<aszlig>
infinisil: you mean for the target repository?
<infinisil>
Won't work with non-github repos though
<infinisil>
yeah
<aszlig>
yeah, but the metadata could also point to a git repository, so something like "import nur:infinisil" should use fetchGit
<infinisil>
Yeah that would be neat
<aszlig>
but implementing this in nix would be more involved because import then needs to do an implicit call of fetchGit, depending on the metadata
<aszlig>
lemme do a first PoC
<infinisil>
Since this requires changes in Nix it would take a while though, and let's see first if NUR becomes something even. Until then people will have to live with a manual import