get_batch_iterator: allow max_positions=None (#673)

Summary:
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/673

This function breaks when leaving the argument `max_positions` with the default value `None`, which is presumably not the intended behavior.

Reviewed By: theweiho, myleott

Differential Revision: D15937221

fbshipit-source-id: 1f5dc1c27ad9b6a89501d2dc015de12181059349

fairseq/tasks/fairseq_task.py

@@ -132,9 +132,10 @@ class FairseqTask(object):
             indices = dataset.ordered_indices()
 
         # filter examples that are too large
-        indices = data_utils.filter_by_size(
-            indices, dataset.size, max_positions, raise_exception=(not ignore_invalid_inputs),
-        )
+        if max_positions is not None:
+            indices = data_utils.filter_by_size(
+                indices, dataset.size, max_positions, raise_exception=(not ignore_invalid_inputs),
+            )
 
         # create mini-batches with given size constraints
         batch_sampler = data_utils.batch_by_size(