ENH: run: Abort if dataset is dirty

This prevents problematic cases where the user has uncommitted changes
to files and doesn't realize that the run will take the files from
HEAD.  It also prevents unnecessary work in scenarios like the one
described in gh-447 where the dirty check on the remote side will fail
due to subdataset modifications.

Closes #447.

reproman/support/jobs/orchestrators.py

@@ -484,6 +484,10 @@ def __init__(self, resource, submission_type, job_spec=None,
         if self._resurrection:
             self.head = self.job_spec.get("head")
         else:
+            if self.ds.repo.dirty:
+                raise OrchestratorError("Local dataset {} is dirty. "
+                                        "Save or discard uncommitted changes"
+                                        .format(self.ds.path))
             self._configure_repo()
             self.head = self.ds.repo.get_hexsha()
             _datalad_check_container(self.ds, self.job_spec)

reproman/support/jobs/tests/test_orchestrators.py

@@ -405,6 +405,13 @@ def get_orc():
             ssh, submission_type="local", job_spec=job_spec)
 
     with chpwd(dataset.path):
+        # We abort if the local dataset is dirty.
+        create_tree(dataset.path, {"local-dirt": ""})
+        with pytest.raises(OrchestratorError) as exc:
+            get_orc()
+        assert "dirty" in str(exc.value)
+        os.unlink("local-dirt")
+
         orc0 = get_orc()
         # Run one job so that we create the remote repository.
         orc0.prepare_remote()