00:01 <jaeger> Admittedly I don't check the wayland stuff since nvidia and wayland hate each other

00:02 <Romster> farkuhar, their is a stance of not creating feature creep to the point it's impossible to maintain and test all features.

00:02 <Romster> wayland and nvidia support is about to get a lot better.

00:03 <Romster> i even observed a so.1 symlink for nvidia-settings aftger i've ran that but i didn't bother to add symlinkes to all the so files to so.1

00:04 <farkuhar> ivandi: actually no, 'depinst' and 'depends' still only find the minimal set of ports needed to build the targets specified on the command line, regardless of whether --softdeps is passed or not. The new flag is meant to ensure that update/sysup operations are carried out in the right order, but first-time installation of a port (and selection of its optional dependencies) is still under the user's control.

00:06 <jaeger> interesting... I wonder how that isinst check worked in my local tests

00:06 <jaeger> It shouldn't... but it did

00:06 <Romster> yeah it works if you don't subshell it with if test....

00:06 <Romster> as [ is a test bash built in command

00:06 <Romster> bash match is done in [[ ]]

00:06 <jaeger> Yeah, what I'm saying is that it worked as it is

00:07 <jaeger> But I don't know how

00:07 <Romster> but i couldn't get it to work with prt-get isinst ... as it's more than one word even with quotation marks around it so i went with keeping the subshell and echo the error code out at the end.

00:07 <jaeger> Where does the /usr/share/vulkan/icd.d thing come from? That's in direct conflict to nvidia's documentation

00:08 <Romster> also the logic of ne 0 not equal to zero is the reverse of is installed to install them files if wayland is found.

00:08 <jaeger> Is that configured via vulkan and nvidia is just behind the times or something else?

00:08 <Romster> i found it in archlinux

00:09 <jaeger> looks like mesa uses it, too

00:09 <jaeger> but nothing about it in the vulkan footprints

00:09 <Romster> it gets found and used correctly from that path and mesa uses that path too.

00:09 <Romster> vulkan does read it from that path

00:09 <Romster> and it saves rejemrge run

00:09 <ivandi> farkuhar: thanks for the clarification, i'll still need my install.lst file.

00:10 <jaeger> Which app specifically uses those files?

00:10 <Romster> granted one could exclude /etc/vilkan stuff from pkgadd.conf

00:10 <farkuhar> ivandi: although if you give both the port and its desired optional dependencies as arguments to the 'depinst' command, then the --softdeps flag will affect the ordering, so in that sense it can still prove useful.

00:10 <jaeger> Fighting with some proxmox bullshit right now but I'll test it when I can

00:10 <Romster> vulkaninfo for one

00:10 <jaeger> ok

00:11 <Romster> i haven't hit any issues from months of testing in my use case.

00:11 <Romster> it could use more testers

00:11 <Romster> i know i've been idle for months

00:12 <Romster> then i come up with this.

00:12 <Romster> even had beer-man go off at me for having crux core member in my github profile so i removed that to make him happy.

00:13 <jaeger> All good, I appreciate the input, just always want to verify it myself, too

00:13 <Romster> i know, i've spent time reasoning every change. in that.

00:14 <Romster> been gaming more since it's the only thing that calms my adhd down from work, and not as much mental capacity to deal with much else.

00:14 <Romster> that's why i've been quiet for so long.

00:15 <jaeger> Mental health is important, of course, do what you need to

00:15 <Romster> currently working on dmarc reports for work stuff.

00:16 <farkuhar> Romster: glad to see you back and working on crux again! As dlcusa so memorably put it, "maintainers are owed, not owing."

00:16 <jaeger> The firmware I skipped because I wasn't installing NGX at all, for what that's worth

00:17 <Romster> farkuhar, also on the note of feature creep, i had the intention of extending prt-get or some other tool to use the optional: line to give the option to display and select extra optional deps for a depinst.

00:18 <Romster> farkuhar, i also changed all the "# Nice to have:" lines to "# Optional:" but then i seen at some point a "# Nice to have:" line added again recently.

00:21 <Romster> farkuhar, things like harfbuzz-icu we had in the past which is now only built in harfbuzz itself breaks depinst when some port needs harfbuzz with icu but depinst ordering makes icu install after harfbuzz. so i had the idea of a port that lists icu would check and find that harfbuzz has a optional icu, then it would either get icu installed before harfbuzz or warn the user that harfbuzz needs a rebuild to support icu.

00:21 <Romster> i didn't get that far.

00:22 <Romster> jaeger, i wondered why i wasn't using my gpu very heavily then i found in my dmesg that firmware wasn't found, did a search and ah...

00:23 <jaeger> Out of curiosity, what are you using NGX for?

00:23 <Romster> jaeger, little suprising since you got wine-staging and you got nvidia, but perhaps you don't use your gpu that heavily.

00:23 <farkuhar> Romster: with my latest commits to the forked repo, both headers ('Optional' and 'Nice to have') are recognized as synonymous. But anyway, I think it's more in keeping with jw's vision to have prt-get remain noninteractive, so a *menu* for selecting optional dependencies is not how I would proceed.

00:23 <jaeger> The GPU works fine for me in wine-staging, but I don't use any AI features

00:24 <Romster> jaeger, i was playing with cuda a bit, not done anything serious with it but i noticed it was missing stuff to work

00:24 <jaeger> ah

00:24 <Romster> so for completeness i thought i'd install those.

00:25 <Romster> i'm also using https://github.com/GloriousEggroll/proton-ge-custom instead of steams proton for some games.

00:26 <Romster> i got cuda cores on my rtx3070 i got curious what i can use it for.

00:26 <jaeger> Maybe some stable diffusion or neural network training, etc.

00:27 <Romster> also opencl uses it too that i was messing with

00:27 <jaeger> A couple years back I benchmarked my laptop's crappy MX150 GPU against the Threadripper 2950X CPU for some neural network training and the MX150 absolutely smoked the CPU

00:27 <jaeger> Kinda crazy how big the gap is

00:28 <Romster> that's amazing

00:30 <Romster> i have had ideas of using cuda cores for audio DSP

00:33 <Romster> farkuhar, yeah prt-get needs to be non-interactive, i hate that it now breaks on wine libreoffice and stuff that depends on harfbuzz built with icu. where the old way of building harfbuzz-icu and extracting the icu libs worked flawlessly.

00:36 <jaeger> A minor feature I have on my list for pkgutils/prt-get replacements is a quiet option for isinst for checking in scriptys

00:36 <jaeger> scripts

00:38 <Romster> there's been a couple of installs over the years that i forgot to pass the pre/post scripts, and it broke a build. it might be an idea for prt-get to warn that a possible port may not work correctly if a pre-install is found in the deptree

00:38 <Romster> but this is crux for experienced users so i just enable it or go manually and run what i missed.

00:39 <Romster> it's difficult how user friendly you make crux vs "well you f&^ked up, easy to resolve and move on"

00:48 <farkuhar> Romster: I've noticed some variety in how the pre/post scripts create new entries in /etc/passwd or /etc/group. Some maintainers delay the useradd and chown commands until post-install (leaving everything in .footprint as root/root), while other maintainers prefer to set up the new accounts in pre-install, and do the chown command inside build(). The Wiki page on pre/post scripts offers no firm guidance one way or the other.

00:50 <Romster> anything that needs files owned as a set user needs to be done pre. if all files are owned as root and only run later as some other user then most would do post

00:51 <Romster> i did in the past when i was forking pkgutils think of adding a user/group feature to pkgmk and Pkgfile

00:51 <Romster> seems most pre installs just need a user and/or a group added.

00:52 <Romster> i've seen gentoo added group/user packages/files for each of those so they can depend on what each ebuild needs.

00:52 <Romster> far more complicated how gentoo does it but it does make sense.

00:53 <jaeger> Hrmm... I don't see any change in vulkaninfo with the icd file move... any idea what else to use to verify it?

00:53 <Romster> jaeger, that's the thing it still reads it just fine at the /usr/share... path

00:54 <jaeger> It seems to work fine in both

00:54 <jaeger> At least as far as vulkaninfo is concerned

00:54 <Romster> if you keep the icd at etc and in share you'll see two entries in vulkaninfo iirc

00:55 <jaeger> So is there any point in moving it, other than to deduplicate and use the same path as mesa does?

00:56 <Romster> to deduplicate and use the same path as mesa does, and not having to run rejmerge on that file due to the version of vulkan used changing over different nvidia versions

00:56 <Romster> that's about all.

00:56 <jaeger> ok

00:56 <Romster> for that small part

00:56 <Romster> i did it for consistency with mesa

00:56 <Romster> even though the nvidia readme ways etc

00:56 <Romster> says*

00:57 <Romster> not like you ever edit that file anyways

01:25 <jaeger> Guess I did install the ngx lib after all, oops. I'd forgotten that.

03:34 <jaeger> Interesting... I DO see the glib-32 issue with includes on another system even though I didn't in the clean container

03:36 <jaeger> For some reason it wants to use /usr/local/include

13:23 <beerman> color me surprised..

13:25 <beerman> jaeger yeah same error i got on my desktop then.. pretty weird

22:39 <jaeger> It looks like /usr/lib/libnvidia-egl-wayland.so* could conflict between egl-wayland and nvidia - I'm guessing probably the one packaged by nvidia is the "correct" one but not sure. Currently there's a version difference, 1.1.10 vs 1.1.11, and the nvidia port doesn't create the .so.1 and .so symlinks yet...

22:40 <jaeger> Maybe egl-wayland isn't required anymore if the nvidia port is updated to install those?

22:42 <jaeger> Ah, I see egl-wayland gets its source from the nvidia github repo so I guess they're both "correct"