Propagate self.env if it was set. (#916)

Breaking example, where we initialize `reqs` in the constructor, which gets set in the env, but is not propagated in `.to` because the env has no name. Pretty sure we can just send `self.env` and later we have a check if it has no name to take a different path.

See test for repro.

runhouse/resources/module.py

@@ -443,13 +443,8 @@ def to(
  system = (
  _get_cluster_from(system, dryrun=self.dryrun) if system else self.system
  )
- if not env:
- if (
- not self.env or (isinstance(self.env, Env) and not self.env.name)
- ) and system:
- env = system.default_env
- else:
- env = self.env
+
+ env = self.env if not env else env
 
  env = _get_env_from(env)
 
@@ -458,6 +453,9 @@ def to(
  if isinstance(env, Env):
  env = env.to(system, force_install=force_install)
 
+ if isinstance(env, Env) and not env.name:
+ env = system.default_env
+
  # We need to backup the system here so the __getstate__ method of the cluster
  # doesn't wipe the client of this function's cluster when deepcopy copies it.
  hw_backup = self.system

tests/test_resources/test_envs/test_env.py

@@ -30,6 +30,12 @@ def np_summer(a, b):
  return int(np.sum([a, b]))
 
 
+def torch_import():
+ import torch
+
+ return str(torch.__version__)
+
+
 @pytest.mark.envtest
 class TestEnv(tests.test_resources.test_resource.TestResource):
  MAP_FIXTURES = {"resource": "env"}
@@ -313,3 +319,8 @@ def test_env_to_with_provider_secret(self, cluster):
  os.environ["HF_TOKEN"] = "test_hf_token"
  env = rh.env(name="hf_env", secrets=["huggingface"])
  env.to(cluster)
+
+ @pytest.mark.level("local")
+ def test_env_in_function_factory(self, cluster):
+ remote_function = rh.function(torch_import, env=["torch"]).to(system=cluster)
+ assert remote_function() is not None