[Draft] Fix MR Reads

Start to fix MR reads on higher levels, where we have declared
but undefined regions and need to filter by valid boxes.

Author: ax3l

openpmd_viewer/openpmd_timeseries/data_reader/io_reader/utilities.py

@@ -11,6 +11,16 @@
 import numpy as np
 
 
+def chunk_to_slice(chunk):
+    """
+    Convert an openPMD_api.ChunkInfo to np.s_
+    """
+    #print("chunk={} {}".format(chunk, type(chunk)))
+    stops = [a + b for a, b in zip(chunk.offset, chunk.extent)]
+    indices_per_dim = zip(chunk.offset, stops)
+    index_tuple = map(lambda s: slice(s[0], s[1], None), indices_per_dim)
+    return tuple(index_tuple)
+
 def get_data(series, record_component, i_slice=None, pos_slice=None,
              output_type=np.float64):
     """
@@ -46,8 +56,21 @@ def get_data(series, record_component, i_slice=None, pos_slice=None,
     if i_slice is not None and not isinstance(i_slice, list):
         i_slice = [i_slice]
 
+    chunks = record_component.available_chunks()
+    full_indices = np.indices(record_component.shape)[0]
+    full_shape = full_indices.shape
+    print("full_shape.shape={}".format(full_shape))
+    print("full_shape={}".format(full_shape))
+
     if pos_slice is None:
-        data = record_component[()]
+        # mask invalid regions with zero
+        data = np.zeros_like(record_component)
+        for chunk in chunks:
+            s = chunk_to_slice(chunk)
+            # read only valid region
+            x = record_component[s]
+            series.flush()
+            data[s] = x
     else:
         # Get largest element of pos_slice
         max_pos = max(pos_slice)
@@ -60,8 +83,17 @@ def get_data(series, record_component, i_slice=None, pos_slice=None,
             list_index[dir_index] = i_slice[count]
         # Convert list_index into a tuple
         tuple_index = tuple(list_index)
-        # Slice dset according to tuple_index
-        data = record_component[tuple_index]
+        print("tuple_index={}".format(tuple_index))
+
+        # prepare sliced data according to tuple_index
+        #slice_indices = full_indices[tuple_index]
+        #slice_shape = slice_indices.shape
+        #data = np.zeros(slice_shape, dtype=output_type)
+        # write now in index space between intersection of slice_indices and chunk indices
+        #for chunk in chunks:
+        #    ...
+        #data = record_component[tuple_index] # todo
+        data = np.zeros_like(record_component)[tuple_index]  # just avoid invalid reads for now
 
     series.flush()
 

setup.py

@@ -38,12 +38,12 @@ def run_tests(self):
       tests_require=['pytest', 'jupyter'],
       install_requires=install_requires,
       extras_require = {
-        'all': ["ipympl", "ipywidgets", "matplotlib", "numba", "openpmd-api", "wget"],
+        'all': ["ipympl", "ipywidgets", "matplotlib", "numba", "openpmd-api~=0.13.3,~=0.14.0", "wget"],
         'GUI':  ["ipywidgets", "ipympl", "matplotlib"],
         'plot': ["matplotlib"],
         'tutorials': ["ipywidgets", "ipympl", "matplotlib", "wget"],
         'numba': ["numba"],
-        'openpmd-api': ["openpmd-api"]
+        'openpmd-api': ["openpmd-api~=0.13.3,~=0.14.0"]
         },
       cmdclass={'test': PyTest},
       platforms='any',