| 29 May 2024 |
 connor (he/him) | The alternative was Equinix, which was much more expensive | 19:34:14 |
 hexa | ugh yeah, eqx mtl is prohibitively expensive 😄 | 19:54:00 |
| 30 May 2024 |
| connor (he/him) | I didn't realize that after Equinix bought Packet they started charging egrees fees of 5 cents a GB lmao | 00:27:07 |
 Gaétan Lepage | In reply to @connorbaker:matrix.org
https://github.com/Mic92/nixpkgs-review/blob/14339add462bfb5f3181979899debe62d97325ce/nixpkgs_review/buildenv.py#L23-L46 Ok, that is very suprising for me, because in my experiene, it doesn't attempt at building them.
I am running the version currently shipped in nixpkgs | 06:59:18 |
| hexa | In reply to @connorbaker:matrix.org
I didn't realize that after Equinix bought Packet they started charging egrees fees of 5 cents a GB lmao which is funny, because equinix is a really big player owning all kinds of datacenters around the world, and packet was comparably small | 11:12:45 |
| 31 May 2024 |
|  @maxwell325:matrix.org joined the room. | 12:15:35 |
 zopieux | is the Cachix occasionally pruned of older derivations? I pinned nixpkgs to github:NixOS/nixpkgs/2748d22b45a99fb2deafa5f11c7531c212b2cefa a few weeks ago, so that it would build to cuda-maintainers cached outputs, but launching this today it builds from source :( | 20:52:48 |
 aidalgol | I think the free plan provides fairly limited space, so as new derivations get pushed, the oldest ones get purged. | 23:32:00 |
| connor (he/him) | IIRC all of the plans use something like LRU eviction
Ours is sponsored by Domen so it’s not the entry-level cache, but I don’t have visibility into the entries | 23:34:39 |
| 1 Jun 2024 |
| connor (he/him) | I’m noticing that the CMake config file generated by OpenCV and used by downstream dependencies requires they use the same version of CUDA. Is that expected?
I’m also noticing some package build failures when they depend on OpenCV because that same OpenCV CMake config file requires CUDA_TOOLKIT_ROOT_DIR be specified, which isn’t set by our hooks. I understand it’s the legacy variable and that the FindCUDA module is deprecated, but how bad would it be to set that variable for the redist packages? We do it currently for the legacy runfile installer. https://github.com/NixOS/nixpkgs/blob/50b636da4a71664532583c800209e7a9e88c9e36/pkgs/development/cuda-modules/cudatoolkit/default.nix#L336 | 20:34:31 |
| connor (he/him) | I’m not by a computer atm but if I can remember I’ll try to post some examples of failures later | 20:34:53 |
| connor (he/him) | My current thought was to set it equal to CUDAToolkit_ROOT in our setup hook: https://github.com/NixOS/nixpkgs/blob/6a697d3418f733c222ac1f95e395cd26bdc2463e/pkgs/development/cuda-modules/setup-hooks/setup-cuda-hook.sh#L57 | 20:35:51 |
 SomeoneSerge (matrix works sometimes) | In reply to @connorbaker:matrix.org
I’m noticing that the CMake config file generated by OpenCV and used by downstream dependencies requires they use the same version of CUDA. Is that expected?
I’m also noticing some package build failures when they depend on OpenCV because that same OpenCV CMake config file requires CUDA_TOOLKIT_ROOT_DIR be specified, which isn’t set by our hooks. I understand it’s the legacy variable and that the FindCUDA module is deprecated, but how bad would it be to set that variable for the redist packages? We do it currently for the legacy runfile installer. https://github.com/NixOS/nixpkgs/blob/50b636da4a71664532583c800209e7a9e88c9e36/pkgs/development/cuda-modules/cudatoolkit/default.nix#L336 I don't remember it checking versions, but if it does it should be satisfied with the cuda we pass in $cxxdev/nix-support/propagated-build-inputs | 21:37:49 |
| SomeoneSerge (matrix works sometimes) |
because that same OpenCV CMake config file requires CUDA_TOOLKIT_ROOT_DIR be specified, which isn’t set by our hooks.
I didn't know it still depends on FindCUDA; what are examples of reverse dependencies failing without the variable? | 21:40:48 |
| 2 Jun 2024 |
| connor (he/him) | SomeoneSerge (UTC+3): actiona
$ nix build --impure -L nixpkgs/master#actiona^*
warning: Nix search path entry '/nix/var/nix/profiles/per-user/root/channels' does not exist, ignoring
trace: warning: cudaPackages.autoAddDriverRunpath is deprecated, use pkgs.autoAddDriverRunpath instead
actiona> Running phase: qtPreHook
actiona> Running phase: unpackPhase
actiona> unpacking source archive /nix/store/0zxvw6cx77nr2chamx8a6kwgaqf8s0aj-source
actiona> source root is source
actiona> Running phase: patchPhase
actiona> applying patch /nix/store/0ihg2n2gfxmn8wsmj85ib4a0zgjnhpz2-fix-paths.patch
actiona> patching file executer/src/executer.cpp
actiona> patching file gui/src/mainwindow.cpp
actiona> patching file tools/src/languages.cpp
actiona> applying patch /nix/store/kgc09dw780kjyccxrrj53i7hsm7papy8-disable-tts.patch
actiona> patching file actions/system/CMakeLists.txt
actiona> Hunk #2 succeeded at 136 (offset 7 lines).
actiona> Hunk #3 succeeded at 157 (offset 7 lines).
actiona> patching file actions/system/src/actionpacksystem.hpp
actiona> Running phase: updateAutotoolsGnuConfigScriptsPhase
actiona> Updating Autotools / GNU config script to a newer upstream version: ./gui/src/3rdparty/breakpad/autotools/config.sub
actiona> Updating Autotools / GNU config script to a newer upstream version: ./gui/src/3rdparty/breakpad/autotools/config.guess
actiona> Running phase: configurePhase
actiona> fixing cmake files...
actiona> cmake flags: -DCMAKE_FIND_USE_SYSTEM_PACKAGE_REGISTRY=OFF -DCMAKE_FIND_USE_PACKAGE_REGISTRY=OFF -DCMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=OFF -DCMAKE_INSTALL_LOCALEDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/share/locale -DCMAKE_INSTALL_LIBEXECDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/libexec -DCMAKE_INSTALL_LIBDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/lib -DCMAKE_INSTALL_DOCDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/share/doc/actiona -DCMAKE_INSTALL_INFODIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/share/info -DCMAKE_INSTALL_MANDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/share/man -DCMAKE_INSTALL_OLDINCLUDEDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/include -DCMAKE_INSTALL_INCLUDEDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/include -DCMAKE_INSTALL_SBINDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/sbin -DCMAKE_INSTALL_BINDIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/bin -DCMAKE_INSTALL_NAME_DIR=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2/lib -DCMAKE_POLICY_DEFAULT_CMP0025=NEW -DCMAKE_OSX_SYSROOT= -DCMAKE_FIND_FRAMEWORK=LAST -DCMAKE_STRIP=/nix/store/9bv7dcvmfcjnmg5mnqwqlq2wxfn8d7yi-gcc-wrapper-13.2.0/bin/strip -DCMAKE_RANLIB=/nix/store/9bv7dcvmfcjnmg5mnqwqlq2wxfn8d7yi-gcc-wrapper-13.2.0/bin/ranlib -DCMAKE_AR=/nix/store/9bv7dcvmfcjnmg5mnqwqlq2wxfn8d7yi-gcc-wrapper-13.2.0/bin/ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_INSTALL_PREFIX=/nix/store/8n9g8fd1jhr5nw6n9s2dg8bxv3yc0ha1-actiona-3.10.2 -DCMAKE_SKIP_BUILD_RPATH:BOOL=TRUE
actiona> -- The CXX compiler identification is GNU 13.2.0
actiona> -- Detecting CXX compiler ABI info
actiona> -- Detecting CXX compiler ABI info - done
actiona> -- Check for working CXX compiler: /nix/store/9bv7dcvmfcjnmg5mnqwqlq2wxfn8d7yi-gcc-wrapper-13.2.0/bin/g++ - skipped
actiona> -- Detecting CXX compile features
actiona> -- Detecting CXX compile features - done
actiona> -- Found PkgConfig: /nix/store/zdvrzlvzbn9ymb0z8na50w995j8np16z-pkg-config-wrapper-0.29.2/bin/pkg-config (found version "0.29.2")
actiona> -- Checking for one of the modules 'xtst'
actiona> CMake Warning (dev) at /nix/store/r008mflixfchlfscby4h0mjgqvz059pw-opencv-4.9.0/lib/cmake/opencv4/OpenCVConfig.cmake:86 (find_package):
actiona> Policy CMP0146 is not set: The FindCUDA module is removed. Run "cmake
actiona> --help-policy CMP0146" for policy details. Use the cmake_policy command to
actiona> set the policy and suppress this warning.
actiona> Call Stack (most recent call first):
actiona> /nix/store/r008mflixfchlfscby4h0mjgqvz059pw-opencv-4.9.0/lib/cmake/opencv4/OpenCVConfig.cmake:108 (find_host_package)
actiona> actiontools/CMakeLists.txt:289 (find_package)
actiona> This warning is for project developers. Use -Wno-dev to suppress it.
actiona>
actiona> CMake Error at /nix/store/q1nssraba326p2kp6627hldd2bhg254c-cmake-3.29.2/share/cmake-3.29/Modules/FindCUDA.cmake:883 (message):
actiona> Specify CUDA_TOOLKIT_ROOT_DIR
actiona> Call Stack (most recent call first):
actiona> /nix/store/r008mflixfchlfscby4h0mjgqvz059pw-opencv-4.9.0/lib/cmake/opencv4/OpenCVConfig.cmake:86 (find_package)
actiona> /nix/store/r008mflixfchlfscby4h0mjgqvz059pw-opencv-4.9.0/lib/cmake/opencv4/OpenCVConfig.cmake:108 (find_host_package)
actiona> actiontools/CMakeLists.txt:289 (find_package)
actiona>
actiona> -- Configuring incomplete, errors occurred!
| 20:15:24 |
| connor (he/him) | https://github.com/NixOS/nixpkgs/blob/095f1acb70302bd74cd5f3ab02a64bdfac36daa8/pkgs/applications/misc/actiona/default.nix#L13 | 20:18:22 |
| aidalgol | SomeoneSerge (UTC+3): I see you put a 👍️ on my comment about libXNVCtrl being tied to the driver version. I'm not 100% certain of my reading of the upstream code. Do you think the "NV-CONTROL Protocol" is just the interface between libXNVCtrl and the client, or does it rely on changes in the nvidia driver as well? | 21:40:27 |
| SomeoneSerge (matrix works sometimes) | In reply to @connorbaker:matrix.org
https://github.com/NixOS/nixpkgs/blob/095f1acb70302bd74cd5f3ab02a64bdfac36daa8/pkgs/applications/misc/actiona/default.nix#L13 My workstation went down, can't access it rn. Can you try (getOutput "cxxdev" opencv) instead of opencv? | 21:41:52 |
| SomeoneSerge (matrix works sometimes) | In reply to @aidalgol:matrix.org
SomeoneSerge (UTC+3): I see you put a 👍️ on my comment about libXNVCtrl being tied to the driver version. I'm not 100% certain of my reading of the upstream code. Do you think the "NV-CONTROL Protocol" is just the interface between libXNVCtrl and the client, or does it rely on changes in the nvidia driver as well? No clue whatsoever. I'd just try putting libxnvctrls from different packages into LD_LIBRARY_PATH and running this mangohud (iirc?) | 21:43:04 |
| aidalgol | Oh, you know what? The "NV-CONTROL Protocol Version History" comment hasn't changed since at least 340, so we're fine. 🙃 | 21:47:04 |
| aidalgol | I should really look at the change history before shooting my mouth off in issue comments. | 21:47:28 |
| 3 Jun 2024 |
| connor (he/him) | In reply to @ss:someonex.net
My workstation went down, can't access it rn. Can you try (getOutput "cxxdev" opencv) instead of opencv? that works! | 01:20:38 |
| connor (he/him) | I mean, I did also have to edit the CUDA setup hook to set the legacy cuda toolkit root dir CMake variable, but it does work | 01:21:14 |
|  lockshaw (UTC-7) joined the room. | 04:09:50 |
| aidalgol | SomeoneSerge (UTC+3): My attempt at moving libXNVCtrl: https://github.com/aidalgol/nixpkgs/commits/libxnvctrl-pure/ Not sure how to define it so that we can reference it sensibly from nvidia-settings to override the src. | 06:09:11 |
|  shekhinah changed their display name from shekhinah to Cat, Nyan - Postdoctoral researcher in Social and Behavioral Sciences. | 09:38:38 |
| connor (he/him) | hexa (UTC+1): the deprecation warnings you were getting for aliases? I think I found one possible reason:
https://github.com/NixOS/nixpkgs/blob/57610d2f8f0937f39dbd72251e9614b1561942d8/pkgs/development/python-modules/torch/default.nix#L435-L439
even though autoAddDriverRunpath is defined in the global scope, it's overridden there inside the with expression
| 13:30:57 |
| hexa | thanks for following up on that | 13:31:37 |
| connor (he/him) | tbh I'm just going through my PR right now trying to fix all the build failures and it caught my eye | 14:26:27 |
| SomeoneSerge (matrix works sometimes) | In reply to @connorbaker:matrix.org
tbh I'm just going through my PR right now trying to fix all the build failures and it caught my eye Oh I begin to remember we actually ran into a similar with; situation in one of the reviews 🤦 | 16:58:59 |
