I'm confused about what the

--environment

flag for

run

does. I cannot find docs for the CLI and the help message isn't very helpful:

--environment [local|conda|pypi]
                                  Execution environment type  [default: local]

• Why do I need to specify this argument if I have a @pypi decorator around my step, why is that decorator not enough information? • Since specifying the environment creates a blank slate for all steps in the workflow, how am I supposed to specify my dependencies for all steps? Tragically, my

end

step fails with an import error because my

start

step assigns a dataframe to

self

and only the start step has the

pypi

decorator with pandas specified • How am I to mix different compute environments in a same workflow if this

--environment

arg is global? I wanted to use pypi for one step but a docker image for another.