deepyaman commented on code in PR #23141:
URL: https://github.com/apache/flink/pull/23141#discussion_r1295230147


##########
flink-python/dev/dev-requirements.txt:
##########
@@ -15,20 +15,20 @@
 pip>=20.3
 setuptools>=18.0
 wheel
-apache-beam==2.43.0
-cython==0.29.24
+apache-beam>=2.43.0,<2.49.0
+cython>=0.29.24
 py4j==0.10.9.7
 python-dateutil>=2.8.0,<3
-cloudpickle==2.2.0
-avro-python3>=1.8.1,!=1.9.2,<1.10.0
-pandas>=1.3.0,<1.4.0
-pyarrow>=5.0.0,<9.0.0
+cloudpickle>=2.2.0

Review Comment:
   Hi @HuangXingBo! Thank you for reviewing.
   
   I tested it out, and, unfortunately, I can't resolve dependencies between 
Ibis and PyFlink (based on the loosened dependencies in this PR) if I pin 
`cloudpickle==2.2.0`. I didn't realize that there's only been one release 
since, but I confirmed that `cloudpickle==2.2.1` does indeed resolve.
   
   All I see in [the 
changelog](https://github.com/cloudpipe/cloudpickle/blob/master/CHANGES.md) for 
2.2.1 is fixing `NamedTuple` pickling; do you think it's possible to provide a 
more conservative range that would at least include 2.2.1? My guess is some 
other dependencies pinned `>=2.2.1` because they rely on `NamedTuple` pickling 
working (but I haven't verified).
   
   Perhaps something like:
   
   ```suggestion
   cloudpickle~=2.2.0
   ```
   
   or even
   
   ```suggestion
   cloudpickle~=2.2
   ```
   
   depending on what you level of future compatibility you're worried about.
   
   Personally, I think it would also be fine with leaving it uncapped, and 
users can always bound `cloudpickle` themselves if a breaking release comes 
out. Preferable to capping it too aggressively, where it requires another 
release for users to use newer versions. (I quite believe in the rationale in 
[a popular post on the 
topic](https://iscinumpy.dev/post/bound-version-constraints/), but in the end 
you know better how it applies to PyFlink.)



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to