Github user zentol commented on the issue:

    https://github.com/apache/flink/pull/3838
  
    Here's a preliminary changelog:
    
    General:
    - rebase branch to current master
    - incremented version to 1.5-SNAPSHOT
    - fixed kafka-connector dependency declaration
        - set to provided
        - scala version set to scala.binary.version 
        - flink version set to project.version
    - applied checkstyle
        - disabled method/parameter name rules for API classes
    - assigned flink-python-streaming to 'libraries' travis profile
    
    API:
    - PDS#map()/flat_map() now return PythonSingleOutputStreamOperator
    - renamed PDS#print() to PDS#output()
        - print is a keyword in python and thus not usable in native python APIs
    - added PythonSingleOutputStreamOperator#name()
    - removed env#execute methods that accepted local execution argument as 
they are redundant due to environment factory methods
    
    Moved/Renamed:
    - made SerializerMap top-level class and renamed it to AdapterMap
    - Moved UtilityFunctions#adapt to AdapterMap class
    - renamed UtilityFunctions to InterpreterUtils
    - moved PythonobjectInputStream2 to SerializationUtils
    - renamed PythonObjectInputStream2 to 
SerialVersionOverridingPythonObjectInputStream
    
    Functions:
    - Introduced AbstractPythonUDF class for sharing 
RichRunction#open()/close() implementations
    - PythonOutputSelector now throws FlinkRuntimeException when failing during 
initialization
    - added generic return type to Serializationutils#deserializeObject
    - added new serializers for PyBoolean/-Float/-Integer/-Long/-String
    - PyObjectSerializer not properly fails when an exceptioin occurs
    - improved error printing
    
    - PythonCollector now typed to Object and properly converts non-PyObjects
    - jython functions that use a collector now have Object has output type
        - otherwise you would get ClassCastException if jython returns 
something that isn't a PyObject
    
    PythonStreamBinder
    - adjusted to follow PythonPlanBinder structure
    - client-like main() exception handling
    - replaced Random usage with UUID.randomUIID()
    - now loads GlobalConfiguration
    - local/distributed tmp dir now configurable
        - introduced PythonOptions
    - no longer generate plan.py but instead import it directly via the 
PythonInterpreter
    
    Environment:
    - Reworked static environment factory methods from 
PythonStreamExecutionEnvironment into a PythonEnvironmentFactory
    - program main() method now accepts a PythonEnvironmentFactory
    - directories are now passed properly to the environment instead of using 
static fields
    - removed PythonEnvironmentConfig
    
    Tests:
    - removed 'if __name__ == '__main__':' blocks from tests since the 
condition is never fulfilled
    - removed python TestBase class
    - removed print statements from tests
    - standardized test job names
    - cleaned up PythonStreamBinderTest / made it more consistent with 
PythonPlanBinderTest
    - run_all_tests improvements
        - stop after first failure
        - print stacktrace on failure
        - no longer relies on dirname() to get cwd but uses the module file 
location instead


---

Reply via email to