Open fgreg opened 3 years ago
Thanks for putting in the issue!
This is also demonstrated in CI builds (e.g. https://app.circleci.com/pipelines/github/nasa/cumulus-process-py/14/workflows/6aa01e9c-0de9-4c7b-87f9-eb1753af4431/jobs/949) .
I've added this to our internal tracker as CUMULUS-2151
With the new pip dependency resolver introducing the idea of 'backtracking' in order to resolve dependencies in an environment, we are noticing very long installation times for this python package because of the
boto3~=1.4
requirement.To reproduce what I'm talking about, create a new virtual environment (I'm using anaconda below), make sure pip is up to date, install the latest boto3 package, then try installing this package.
The behavior I see is pip will tell me:
And it ends up 'backtracking' all the way to version 1.4 of boto3 before resolving. I've seen a few examples where it never resolves and just gives an error.
I think this could be resolved by being more specific about which version of boto3 is required; I think ideally the project should be converted to use something like a pipfile and pipfile.lock to specify exact versions of libraries it is intended to be built against.