THRIFT-5693 - fix bug in serialization of enum default values

[andrewcunnin]

This change fixes a bug where the default values set in enum fields were unable to be serialized. Previously default enum values were being stored as enum objects upon initialization, where Thrift generally expects I32 values. With the changes in this pull request, serialization and deserialization are fixed and the python 3.4 style enums continue to work as expected.

JIRA ticket here: https://issues.apache.org/jira/browse/THRIFT-5693

• Did you create an Apache Jira ticket? (Request account here, not required for trivial changes)
• If a ticket exists: Does your pull request title follow the pattern "THRIFT-NNNN: describe my issue"?
• Did you squash your changes to a single commit? (not required, but preferred)
• Did you do your best to avoid breaking changes? If one was needed, did you label the Jira ticket with "Breaking-Change"?
• If your change does not involve any code, include [skip ci] anywhere in the commit message to free up build resources.

[fishy]

we actually don't have python set up in github actions yet so this is not really tested (we only tested that it did not break other languages that's setup in github actions)

can you show the diff between old and new compiler generated python code?

[andrewcunnin]

Gotcha, here's a diff of the ThriftTest.thrift generated Python:
https://gist.github.com/andrewcunnin/802f4d8fa779c181e2272b5219b52a07

[fishy]

I see this diff from ttypes.py:

@@ -2931,7 +2931,7 @@ class OptionalEnum(object):
     )
 
 
-    def __init__(self, e=    Numberz.FIVE,):
+    def __init__(self, e=5,):
         self.e = e
 
     def read(self, iprot):

and also this diff:

@@ -7198,7 +7198,7 @@ OptionalBinary.thrift_spec = (
 all_structs.append(OptionalEnum)
 OptionalEnum.thrift_spec = (
     None,  # 0
-    (1, TType.I32, 'e', None,     Numberz.FIVE, ),  # 1
+    (1, TType.I32, 'e', None, 5, ),  # 1
 )
 fix_spec(all_structs)
 del all_structs

which I think is caused by this part of the change. from the diff I don't think this change is necessary? (it's probably still correct, but the old code generated seems to be better?)

[andrewcunnin]

Ah, good catch. That default value change was because I thought Thrift was generally expecting int values even representing Python enums, but after re-testing I've found that that was only an issue for serialization and deserialization. Removed that change and here's the updated diff: https://gist.github.com/andrewcunnin/ba688afe22f943c7b9f96e67adb3bc53

[fishy]

the code looks good to me from the diff generated, but I'm not familiar with the python code enough so I'd still prefer to have some tests to verify it, and I failed to run python3 tests locally similar to how we didn't get the CI setup work in #2787 (comment).

[fishy]

excessive blank lines. I think the general python style is 2 blank lines between classes/top level stuff and one blank line between functions inside the same class. here you are making it 3 blank lines between classes.

[fishy]

same here, you are making it 4 blank lines between the class above and suite.

[andrewcunnin]

For the testing, is running the cross test and running make -k check in test/py sufficient? Or were you referring to a different set of tests that I should be running separately?

[fishy]

For the testing, is running the cross test and running make -k check in test/py sufficient? Or were you referring to a different set of tests that I should be running separately?

we have tests under both test/py and lib/py/test, and then we also have cross tests between different language libraries.