Add a TileDB Array Proxy (#22)

* Add a tiledb data proxy

* Usage testing changes

nctotdb/readers.py

@@ -33,6 +33,36 @@
         "_FillValue",
     ])
 
+
+# Inspired by https://github.com/SciTools/iris/blob/master/lib/iris/fileformats/netcdf.py#L418.
+class TileDBDataProxy:
+    """A proxy to the data of a single TileDB array attribute."""
+
+    __slots__ = ("shape", "dtype", "path", "var_name")
+
+    def __init__(self, shape, dtype, path, var_name):
+        self.shape = shape
+        self.dtype = dtype
+        self.path = path
+        self.var_name = var_name
+
+    @property
+    def ndim(self):
+        return len(self.shape)
+
+    def __getitem__(self, keys):
+        with tiledb.open(self.path, 'r') as A:
+            data = A[keys][self.var_name]
+        return data
+
+    def __getstate__(self):
+        return {attr: getattr(self, attr) for attr in self.__slots__}
+
+    def __setstate__(self, state):
+        for key, value in state.items():
+            setattr(self, key, value)
+
+
 class Reader(object):
     """
     Abstract reader class that defines the API.
@@ -223,16 +253,20 @@ def _extract(self, array_name):
 
         return named_array_path, dim_paths
 
-    def _array_shape(self, nonempty_domain):
+    def _array_shape(self, nonempty_domain, slices=False):
         """
         Use the TileDB array's nonempty domain (i.e. the domain that encapsulates
         all written cells) to set the shape of the data to be read out of the
         TileDB array.
 
         """
         # We need to include the stop index, not exclude it.
-        slices = [slice(start, stop+1, 1) for (start, stop) in nonempty_domain]
-        return tuple(slices)  # Can only index with tuple, not list.
+        if slices:
+            slices = [slice(start, stop+1, 1) for (start, stop) in nonempty_domain]
+            return tuple(slices)  # Can only index with tuple, not list.
+        else:
+            # TileDB describes the nonempty domain quite annoyingly!
+            return [filled[1]+1 for filled in nonempty_domain]
 
     def _handle_attributes(self, attrs, exclude_keys=None):
         """
@@ -256,17 +290,16 @@ def _from_tdb_array(self, array_path, naming_key, array_name=None, to_dask=False
             metadata = {k: v for k, v in A.meta.items()}
             if array_name is None:
                 array_name = metadata[naming_key]
-            array_inds = self._array_shape(A.nonempty_domain())
             # This may well not maintain lazy data...
             if to_dask:
-                # Borrowed from:
-                # https://github.com/dask/dask/blob/master/dask/array/tiledb_io.py#L4-L6
                 schema = A.schema
+                dtype = schema.attr(array_name).dtype
                 chunks = [schema.domain.dim(i).tile for i in range(schema.ndim)]
-                points = da.from_array(A[array_inds][array_name],
-                                       chunks,
-                                       name=naming_key)
+                array_shape = self._array_shape(A.nonempty_domain())
+                proxy = TileDBDataProxy(array_shape, dtype, array_path, array_name)
+                points = da.from_array(proxy, chunks, name=naming_key)
             else:
+                array_inds = self._array_shape(A.nonempty_domain(), slices=True)
                 points = A[array_inds][array_name]
         return metadata, points