Test python import

a.py

class A(object):
  def __init__():
    print 'A'

b.py

from a import A

class B(A):
  def __init__():
    print 'B'

c.py

from a import A

class C(A):
  def __init__():
    print 'C'

test.py

from a import A

print A.__subclasses__()
# []

from b import B

print A.__subclasses__()
# [<class 'b.B'>]

from c import C

print A.__subclasses__()
# [<class 'b.B'>, <class 'c.C'>]

If you just write "import b" instead of "from b import B" this will work as well. Likewise, if you import (say) c from b.py, there's no need to import it from test.py.

https://gist.github.com/robertwb/c171444f53a2dd55787a38186d4b1e45

Thanks @robertwb, one follow up questions for the beam plugin case

myfs.py installed via pip somewhere

class MyFS(FileSystem):
   def __init__():
       print 'This is my fs implementation'

my_pipeline.py

p = Pipeline() | TextIO("myfs://path/to/file")
p.run()

python -m my_pipeline --runner DataflowRunner how does this find the myfs implementation when it is not imported anywhere. Is the hope that the user is going to import this in the pipeline ?

The user must cause this to be imported to indicate that they want to use it. (Perhaps directly, perhaps some runners will import certain filesystems themselves.)

Thanks