Solely iterable datasets will now be able to use compression.

All other keywords are ignored, as they are either used by hickle itself or serve no purpose with hickle.
This fixes #140.

hickle/hickle.py

@@ -292,6 +292,10 @@ def create_hkl_dataset(py_obj, h_group, call_id=None, **kwargs):
     # Set the name of this dataset
     name = 'data%s' % ("_%i" % (call_id) if call_id is not None else '')
 
+    # If this obj is iterable, use compression if given
+    if hasattr(py_obj, '__iter__') and not isinstance(py_obj, (str, bytes)):
+        kwargs = {'compression': kwargs.pop('compression', None)}
+
     # Try to create the dataset
     try:
         h_subgroup = create_dataset(py_obj, h_group, name, **kwargs)
@@ -310,7 +314,7 @@ def create_hkl_dataset(py_obj, h_group, call_id=None, **kwargs):
             pass
 
         # Create the pickled dataset
-        h_subgroup = create_dataset(py_obj, h_group, name, error, **kwargs)
+        h_subgroup = create_dataset(py_obj, h_group, name, error)
 
     # Save base type of py_obj
     h_subgroup.attrs['base_type'] = base_type

hickle/loaders/load_builtins.py

@@ -72,9 +72,6 @@ def create_scalar_dataset(py_obj, h_group, name, **kwargs):
             iterable.
     """
 
-    # Make sure 'compression' is not in kwargs
-    kwargs.pop('compression', None)
-
     # If py_obj is an integer and cannot be stored in 64-bits, convert to str
     if isinstance(py_obj, int) and not (-2**63 <= py_obj < 2**64):
         py_obj = bytes(str(py_obj), 'ascii')