Switch to approach of removing it from the constraints.txt file compl…

…etely.

In testing, it appears that unnamed constraints don't really have an effect on
what pip installs anyway, so they only seem to cause problems if included in
the constraints.txt file.

constraints.txt

@@ -288,7 +288,6 @@ tf-estimator-nightly==2.8.0.dev2021122109
 threadpoolctl==3.1.0
 toml==0.10.1
 toolz==0.10.0
-git+https://github.com/NVIDIA/torch-harmonics.git@8826246cacf6c37b600cdd63fde210815ba238fd
 torch==1.12.1
 torchvision==0.13.1
 tornado==6.1

docker/dataflow/Dockerfile

@@ -3,12 +3,6 @@ FROM apache/beam_python3.8_sdk:${BEAM_VERSION}
 
 
 COPY constraints.txt /etc/
-# Remove all constraints from GitHub, since the version of pip in this image
-# does not support unnamed constraints.  See fv3net#2304 for more details.  This
-# fix works because we do not need to install any packages from GitHub in the
-# dataflow image, despite them being present in the constraints.txt file for
-# other images.
-RUN sed -i '/git+https/d' /etc/constraints.txt
 ENV PIP_CONSTRAINT=/etc/constraints.txt
 
 RUN apt-get update && apt-get install -y gfortran

pip-requirements.txt

@@ -58,7 +58,3 @@ bokeh>=2.4.2,<3 # needed for dask distributed dashboard
 
 # fv3fit testing tools
 hypothesis
-
-# Needed for xtorch_harmonics.  Pin to older version since latest version leads
-# to an import error.
-torch-harmonics@git+https://github.com/NVIDIA/torch-harmonics.git@8826246cacf6c37b600cdd63fde210815ba238fd