Closed
Description
Since yesterday or more specifically since pytorch=1.14.0.dev20221017 the CMake workflows are failing on Windows. For example: https://app.circleci.com/pipelines/github/pytorch/vision/21251/workflows/cb54fe99-650c-4084-9717-639ee6b20032/jobs/1724374
The command that is failing is the installation of PyTorch core:
$ conda install -yq pytorch=1.14.0.dev20221018 pytorch-mutex=1.0=cpu numpy -c nvidia -c pytorch-nightly
[...]
UnsatisfiableError: The following specifications were found
to be incompatible with the existing python installation in your environment:
Specifications:
- pytorch=1.14.0.dev20221018 -> python[version='>=3.10,<3.11.0a0|>=3.8,<3.9.0a0|>=3.9,<3.10.0a0']
Your python: python=3.7
If I interpret this right, it seems that the nightly requires Python>=3.8 although we are running on 3.7.
The day before everything worked fine:
$ conda install -yq pytorch=1.14.0.dev20221016 pytorch-mutex=1.0=cpu numpy -c nvidia -c pytorch-nightly
[...]
## Package Plan ##
environment location: C:\tools\miniconda3
added / updated specs:
- numpy
- pytorch-mutex==1.0=cpu
- pytorch=1.14.0.dev20221016
The following packages will be downloaded:
package | build
---------------------------|-----------------
blas-1.0 | mkl 6 KB
intel-openmp-2021.4.0 | haa95532_3556 2.2 MB
mkl-2021.4.0 | haa95532_640 114.9 MB
mkl-service-2.4.0 | py37h2bbff1b_0 49 KB
mkl_fft-1.3.1 | py37h277e83a_0 135 KB
mkl_random-1.2.2 | py37hf11a4ad_0 216 KB
numpy-1.21.5 | py37h7a0a035_3 25 KB
numpy-base-1.21.5 | py37hca35cd5_3 4.4 MB
pytorch-1.14.0.dev20221016 | py3.7_cpu_0 139.1 MB pytorch-nightly
pytorch-mutex-1.0 | cpu 3 KB pytorch-nightly
typing_extensions-4.3.0 | py37haa95532_0 42 KB
------------------------------------------------------------
Total: 261.0 MB
Given that this only happens on Windows and couldn't find any commit that introduces the 3.8 requirement in core, this seems like a bug to me.
cc @seemethere