You are not logged in.
Can someone try to install python-torch-geometric AUR package and try to reproduce this bug? https://github.com/acxz/pkgbuilds/issues/219
Is it possible that I am the only one using this package and/or the only one with this bug? I have a pretty standard arch linux installation and PyTorch works
Thank you
Offline
The bug is reproducible, but there's something weird going on .
https://aur.archlinux.org/packages/pyth … rch_sparse does state it's required by python-torch-geometric .
However the python-torch-geometric page doesn't mention python-pytorch_sparse at all.
No idea how that can happen .
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Online
The bug is reproducible, but there's something weird going on .
https://aur.archlinux.org/packages/pyth … rch_sparse does state it's required by python-torch-geometric .
However the python-torch-geometric page doesn't mention python-pytorch_sparse at all.No idea how that can happen .
Not only that: the current PKGBUILD for pytorch sparse is outdated and fails to install, error below.
ERROR test/test_matmul.py - RuntimeError:
ERROR test/test_metis.py - RuntimeError:
ERROR test/test_overload.py - RuntimeError:
ERROR test/test_permute.py - RuntimeError:
ERROR test/test_saint.py - RuntimeError:
ERROR test/test_spmm.py - RuntimeError:
ERROR test/test_spspmm.py - RuntimeError:
ERROR test/test_storage.py - RuntimeError:
ERROR test/test_tensor.py - RuntimeError:
ERROR test/test_transpose.py - RuntimeError:
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Interrupted: 15 errors during collection !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
============================================================================================ 15 errors in 2.10s =============================================================================================
==> ERROR: A failure occurred in check().
Aborting...
-> error making: python-pytorch_sparse
So I think I have to write to the Pytorch Geometric AUR package to update the dependencies and to try to update the Pytorch sparse PKGBUILD because it is clearly outdated and unmaintained. Any help will be appreciated, it is the first time i try to update a PKGBUILD.
Thank you
Offline
The python-pytorch_sparse looks like it uses an older, deprecated method to build. it may need a rewrite to comply with https://wiki.archlinux.org/title/Python … guidelines .upstream build instructions are all targeted at pip users, so won't be much help.
For now I suggest you increase pkgver to 0.6.16 and verify if that builds .
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Online
Offline
gcc is in the base-devel meta-package which is assumed to be present when building aur packages, so doesn't need to be listed.
I built in a clean chroot with extra-x86_64-build.
Sofar I've found python-pytorch is needed as dependency . The build error you got is because parallel-hashmap is not present .
Add that package and build succeeds .
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Online
Thank you very much. So parallel-hashmap should be added as a dependency. Pytorch should be a dependency of pytorch geometric
Offline
https://aur.archlinux.org/packages/python-torch-sparse fails to build if it can't find the torch module, this makes python-pytorch a hard dependency (at buildtime, no idea about runtime).
So both python-pytorch and parallel-hashmap need to be atleast makedepends.
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Online