if length > 0:
                data["__streamdata__"] = stream. Read(length)
            elif length < 0:
                data["__streamdata__"] = read_until_regex(
                    stream, re.compile(b"endstream")
                )

...

        if "__streamdata__" in data:
            return StreamObject.initialize_from_dictionary(data)
        retval = DictionaryObject()
        retval.update(data)
        return retval

I think the problem lies in this piece of code. A StreamObject of length 0 will not have "__streamdata__" in data, and thus it will be parsed into a DictionaryObject instead of a StreamObject.

I've run pytest in my Windows,. The tests failed because tika-950337.pdf contains a length-0 content stream in page 2 (in test_compress_raised, test_workflows.py), which will be parsed to be a DictionaryObject. However compress_content_streams() need to call get_data() method of the content, so it failed.