Thanks. I've filed https://issues.apache.org/jira/browse/BEAM-11882 .
If you want to take a stab at fixing it, you could try replacing the arguemnt passed to merge_accumulators at https://github.com/apache/beam/blob/release-2.28.0/sdks/python/apache_beam/transforms/combiners.py#L963 with a new object whose __iter__ method returns iter(accumulators) and create a pull request. On Wed, Feb 24, 2021 at 2:45 PM Stephen Dewey <stepheneliotde...@gmail.com> wrote: > Oh, I forgot to mention that I am using SDK 2.27.0 and Python 3.8 > > On Wed, Feb 24, 2021 at 5:27 PM Stephen Dewey <stepheneliotde...@gmail.com> > wrote: > >> Hi, I am reporting a minor bug. >> >> Based on this answer by Pablo: >> https://stackoverflow.com/a/42283279/783314 >> >> It appears that you want to always have an _UnwindowedValues in >> DirectRunner whenever it exists in DataflowRunner, to provide consistency >> between the two. >> >> What I have noticed is that if you subclass beam.CombineFn in Python, >> the accumulators received by the merge_accumulators method (as its >> argument) will be _UnwindowedValues in DataflowRunner, but not in >> DirectRunner. This leads to an error if somebody passes that value to, say, >> len(). The error will be: TypeError: object of type '_UnwindowedValues' >> has no len() >> >> Hope this helps! >> Stephen >> >
