Fix read of query returning only empty var-length strings for attribute

If a query range only returns empty strings for a var-length attribute, then
the data buffer will have zero-length and the offsets denote the empty
cells. Remove the zero-length check. See:

  https://github.com/TileDB-Inc/TileDB/blob/dev/test/src/unit-empty-var-length.cc

Author: ihnorton

tiledb/core.cc

@@ -863,8 +863,6 @@ class PyQuery {
 
     auto start = std::chrono::high_resolution_clock::now();
 
-    if (buf.size() < 1)
-      TPY_ERROR_LOC(std::string("Unexpected empty buffer array ('") + name + "')");
     if (off.size() < 1)
       TPY_ERROR_LOC(std::string("Unexpected empty offsets array ('") + name + "')");
 

tiledb/libtiledb.pyx

@@ -5402,12 +5402,11 @@ cdef class SparseArrayImpl(Array):
             if name == '__attr':
                 final_name = ''
             if self.schema._needs_var_buffer(name):
-                if len(results[name][1]) > 0:
+                if len(results[name][1]) > 0: # note: len(offsets) > 0
                     arr = q.unpack_buffer(name, results[name][0], results[name][1])
                 else:
                     arr = results[name][0]
                     final_dtype = self.schema.attr_or_dim_dtype(name)
-                    print((arr, len(arr), final_dtype))
                     if (len(arr) < 1 and
                             (np.issubdtype(final_dtype, np.bytes_) or
                              np.issubdtype(final_dtype, np.unicode_))):

tiledb/tests/test_libtiledb.py

@@ -1467,6 +1467,33 @@ def test_varlen_write_bytes(self):
 
             assert_array_equal(A, T.multi_index[1:len(A)][''])
 
+    def test_varlen_sparse_all_empty_strings(self):
+        # this test addresses a fix for specific need for reads on a
+        # large existing array, see
+        #   https://github.com/TileDB-Inc/TileDB-Py/pull/475
+        # we currently have to write a placeholder at the end to
+        # avoid zero-length cell error
+        # TODO: follow-up with improved testing for empty var-length/strings
+        A = np.array(['','','','','','\x00'], dtype=object)
+        dim_len = len(A)
+        uri = self.path("varlen_all_empty_strings")
+
+        ctx = tiledb.Ctx()
+
+        dom = tiledb.Domain(tiledb.Dim(ctx=ctx, domain=(1, dim_len), tile=dim_len), ctx=ctx)
+        att = tiledb.Attr(name="a1", dtype=np.str_, var=True, ctx=ctx)
+
+        schema = tiledb.ArraySchema(dom, (att,), sparse=True, ctx=ctx)
+
+        tiledb.Array.create(uri, schema)
+
+        with tiledb.open(uri, mode='w', ctx=ctx) as T:
+            T[np.arange(1,dim_len+1)] = {'a1' : A}
+
+        with tiledb.open(uri, mode='r', ctx=ctx) as T:
+            # check interior range
+            assert_array_equal(A[1:-1], T[2:-1]['a1'])
+            assert_array_equal(A[1:-1], T.multi_index[2:dim_len-1]['a1'])
 
     def test_varlen_write_unicode(self):
         A = np.array(['aa','bbb',