Bug in type inference for Python 3.10

[damccorm]

After debugging, the issue arises here

beam/sdks/python/apache_beam/typehints/trivial_inference.py

Line 328 in 70cee1d

return Any

The method above is not able to infer outputs properly in Python 3.10 and results in an outputting wrong output type. 

# Python <= 3.9 

from apache_beam.typehints.trivial_inference import infer_return_type_func  

fn = lambda x: False
wrapper = lambda x, *args, **kwargs: [x] if fn(x, *args, **kwargs) else []
output_type = infer_return_type_func(wrapper, [int])
# output_type would be List[int]

# Python == 3.10

output_type = infer_return_type_func(wrapper, [int]) 
# fails with error beam/sdks/python/apache_beam/typehints/opcodes.py", line # 322, in dict_update base = state.stack[-arg]
# IndexError: list index out of range

# in python
3.9
from dis import dis
print(dis(wrapper))

'''
 0 LOAD_GLOBAL              0 (fn)
 2 LOAD_FAST                0 (x)
 4 BUILD_LIST               1         
 6 LOAD_FAST                1 (args)
 8 LIST_EXTEND              1
10 LIST_TO_TUPLE
12 BUILD_MAP                0
14 LOAD_FAST               2 (kwargs)
16 DICT_MERGE               1
18 CALL_FUNCTION_EX         1  
20 POP_JUMP_IF_FALSE       28
22 LOAD_FAST                0 (x)          
24 BUILD_LIST               1
26 RETURN_VALUE
28 BUILD_LIST              0
30 RETURN_VALUE
'''

# in python 3.10 

print(dis(wrapper))
'''
 0 LOAD_GLOBAL              0 (fn)
 2 LOAD_FAST                0 (x)
 4 BUILD_LIST               1
 6 LOAD_FAST                1 (args)
 8 LIST_EXTEND            1
10 LIST_TO_TUPLE
12 BUILD_MAP                0          
14 LOAD_FAST                2 (kwargs)
16 DICT_MERGE               1
18 CALL_FUNCTION_EX         1
20 POP_JUMP_IF_FALSE       14 (to 28)
22 LOAD_FAST               0 (x)
24 BUILD_LIST               1
26 RETURN_VALUE     
28 BUILD_LIST               0
30 RETURN_VALUE
'''

Imported from Jira BEAM-14547. Original Jira may contain additional context.
Reported by: Anand Inguva.

[damccorm]

Unable to assign user @AnandInguva. If able, self-assign, otherwise tag @damccorm so that he can assign you. Because of GitHub's spam prevention system, your activity is required to enable assignment in this repo.

[AnandInguva]

.take-issue

[AnandInguva]

@yeandy this is one of the bug related to Python 3.10 type inference. Can you assign this to yourself?

[AnandInguva]

Failed tests for Python 3.10 on inference can be found at https://ci-beam.apache.org/job/beam_PreCommit_Python_Commit/24196/ or in github actions as well

[yeandy]

.take-issue

[yeandy]

Not sure if I can override the assignee by commenting .take-issue. Edit: It assigned to me now.