[ 
https://issues.apache.org/jira/browse/BEAM-1677?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Tibor Kiss updated BEAM-1677:
-----------------------------
    Description: 
*Proposal:*
Leverage PyCharm's static code analysis capabilities to detect possible errors 
if a Python-SDK related PR is submitted.

*Problem statement:*
The Python SDK test goal has linting phase, which utilizes {{pep8}} and 
{{pylint}} to validate PEP8 compliance and do static code analysis. These steps 
are helpful but {{pylint}}'s capabilities are somewhat limited compared to 
modern IDE's (like PyCharm). 

In particular: Unresolved attributes, unbound local variables and invalid calls 
(due to argument number mismatch) type of errors  are not detected by pylint, 
but they do cause runtime error. See BEAM-1649 for problems not detected by 
pylint but reported by PyCharm.

*Technical details:*
PyCharm provides command line interface [1] to trigger code inspection. The 
results are created as entries in XML documents, containing their location, 
severity, category and textual description of the problem.

Since the inspector reports back typos, warnings and errors it is required that 
the result XML files are properly parsed and filtered.
A good signal-to-noise ratio is crucial for this improvement to be useful and 
accepted, therefore it is required to make good selection of events to minimize 
false positives.
      
The filtered results shall be posted back to GitHub through Commit Status API 
[2] to notify the contributor and reviewer of the quality of the change.
      
The static code analysis could be done on either Apache Build Jenkins servers 
or in Travis CI.
      
*Alternatives considered:*
 - Eclipse's PyDev plugin [3] has similar code inspection capabilities as 
PyCharm. The lack of command line interface makes it difficult to integrate it 
with the current build system.
 - SonarQUBE's Python plugin [4] was briefly investigated. The number of rules 
seems very limited compared to PyCharm's code analysis.
      
*Required changes:*
- PyCharm to be installed in Jenkins Slave or Travis-CI's container
- PyCharm project & inspection profile created for Python SDK
- Result parser & poster application/script created
- Trigger to run the result parser & poster set up
      
*Example Code Analysis* run on OS X (given a PyCharm project is available in 
beam/sdks/python/.idea directory):
{code}
beam/sdks/python $ /Applications/PyCharm.app/Contents/MacOS/pycharm inspect 
$(pwd) $(pwd)/.idea/inspectionProfiles/Project_Default.xml 
$(pwd)/inspection-results -v2 -d $(pwd)/apache_beam
{code}
      
*Example result:*
{code:title=PyUnresolvedReferencesInspection.xml}
<problem>
<file>file://$PROJECT_DIR$/apache_beam/io/range_trackers.py</file>
<line>342</line>
<module>python</module>
<entry_point TYPE="file" 
FQNAME="file://$PROJECT_DIR$/apache_beam/io/range_trackers.py" />
<problem_class severity="WARNING" attribute_key="WARNING_ATTRIBUTES">Unresolved 
references</problem_class>
<description>Unresolved attribute reference '_end_position' for class 
'OrderedPositionRangeTracker'</description>
</problem>
{code}
      
*References:*
[1] PyCharm Command Line Interface: 
https://www.jetbrains.com/help/pycharm/2016.3/command-line-code-inspector.html
[2] GitHub Commit Status API: https://developer.github.com/v3/repos/statuses/
[3] Eclipse PyDev Plugin: http://www.pydev.org/manual_adv_code_analysis.html
[4] SonarQUBE Python plugin: 
https://www.sonarsource.com/why-us/products/codeanalyzers/sonarpython/rules.html
      
CC: [~altay] / [~jbonofre] / [~jasonkuster]

  was:
*Proposal:*
Leverage PyCharm's static code analysis capabilities to detect possible errors 
if a Python-SDK related PR is submitted.

*Problem statement:*
The Python SDK test goal has linting phase, which utilizes {{pep8}} and 
{{pylint}} to validate PEP8 compliance and do static code analysis. These steps 
are helpful but {{pylint}}'s capabilities are somewhat limited compared to 
modern IDE's (like PyCharm). 

In particular: Unresolved attributes, unbound local variables and invalid calls 
(due to argument number mismatch) type of errors  are not detected by pylint, 
but they do cause runtime error. See BEAM-1649 for problems not detected by 
pylint but reported by PyCharm.

*Technical details:*
PyCharm provides command line interface [1] to trigger code inspection. The 
results are created as entries in XML documents, containing their location, 
severity, category and textual description of the problem.

Since the inspector reports back typos, warnings and errors it is required that 
the result XML files are properly parsed and filtered.
A good signal-to-noise ratio is crucial for this improvement to be useful and 
accepted, therefore it is required to make good selection of events to minimize 
false positives.
      
The filtered results shall be posted back to GitHub through Commit Status API 
[2] to notify the contributor and reviewer of the quality of the change.
      
The static code analysis could be done on either Apache Build Jenkins servers 
or in Travis CI.
      
*Alternatives considered:*
 - Eclipse's PyDev plugin [3] has similar code inspection capabilities as 
PyCharm. The lack of command line interface makes it difficult to integrate it 
with the current build system.
 - SonarQUBE's Python plugin [4] was briefly investigated. The number of rules 
seems very limited compared to PyCharm's code analysis.
      
*Required changes:*
- PyCharm to be installed in Jenkins Slave or Travis-CI's container
- PyCharm project & inspection profile created for Python SDK
- Result parser & poster application/script created
- Trigger to run the result parser & poster set up
      
*Example Code Analysis* run on OS X (given a PyCharm project is available in 
beam/sdks/python/.idea directory):
{code}
beam/sdks/python $ /Applications/PyCharm.app/Contents/MacOS/pycharm inspect 
$(pwd) $(pwd)/.idea/inspectionProfiles/Project_Default.xml 
$(pwd)/inspection-results -v2 -d $(pwd)/apache_beam
{code}
      
*Example result:*
{code title="PyUnresolvedReferencesInspection.xml"}
<problem>
<file>file://$PROJECT_DIR$/apache_beam/io/range_trackers.py</file>
<line>342</line>
<module>python</module>
<entry_point TYPE="file" 
FQNAME="file://$PROJECT_DIR$/apache_beam/io/range_trackers.py" />
<problem_class severity="WARNING" attribute_key="WARNING_ATTRIBUTES">Unresolved 
references</problem_class>
<description>Unresolved attribute reference '_end_position' for class 
'OrderedPositionRangeTracker'</description>
</problem>
{code}
      
*References:*
[1] PyCharm Command Line Interface: 
https://www.jetbrains.com/help/pycharm/2016.3/command-line-code-inspector.html
[2] GitHub Commit Status API: https://developer.github.com/v3/repos/statuses/
[3] Eclipse PyDev Plugin: http://www.pydev.org/manual_adv_code_analysis.html
[4] SonarQUBE Python plugin: 
https://www.sonarsource.com/why-us/products/codeanalyzers/sonarpython/rules.html
      
CC: [~altay] / [~jbonofre] / [~jasonkuster]


> Extend automated PR check with PyCharm's code inspector for Python-SDK
> ----------------------------------------------------------------------
>
>                 Key: BEAM-1677
>                 URL: https://issues.apache.org/jira/browse/BEAM-1677
>             Project: Beam
>          Issue Type: Improvement
>          Components: build-system
>            Reporter: Tibor Kiss
>
> *Proposal:*
> Leverage PyCharm's static code analysis capabilities to detect possible 
> errors if a Python-SDK related PR is submitted.
> *Problem statement:*
> The Python SDK test goal has linting phase, which utilizes {{pep8}} and 
> {{pylint}} to validate PEP8 compliance and do static code analysis. These 
> steps are helpful but {{pylint}}'s capabilities are somewhat limited compared 
> to modern IDE's (like PyCharm). 
> In particular: Unresolved attributes, unbound local variables and invalid 
> calls (due to argument number mismatch) type of errors  are not detected by 
> pylint, but they do cause runtime error. See BEAM-1649 for problems not 
> detected by pylint but reported by PyCharm.
> *Technical details:*
> PyCharm provides command line interface [1] to trigger code inspection. The 
> results are created as entries in XML documents, containing their location, 
> severity, category and textual description of the problem.
> Since the inspector reports back typos, warnings and errors it is required 
> that the result XML files are properly parsed and filtered.
> A good signal-to-noise ratio is crucial for this improvement to be useful and 
> accepted, therefore it is required to make good selection of events to 
> minimize false positives.
>       
> The filtered results shall be posted back to GitHub through Commit Status API 
> [2] to notify the contributor and reviewer of the quality of the change.
>       
> The static code analysis could be done on either Apache Build Jenkins servers 
> or in Travis CI.
>       
> *Alternatives considered:*
>  - Eclipse's PyDev plugin [3] has similar code inspection capabilities as 
> PyCharm. The lack of command line interface makes it difficult to integrate 
> it with the current build system.
>  - SonarQUBE's Python plugin [4] was briefly investigated. The number of 
> rules seems very limited compared to PyCharm's code analysis.
>       
> *Required changes:*
> - PyCharm to be installed in Jenkins Slave or Travis-CI's container
> - PyCharm project & inspection profile created for Python SDK
> - Result parser & poster application/script created
> - Trigger to run the result parser & poster set up
>       
> *Example Code Analysis* run on OS X (given a PyCharm project is available in 
> beam/sdks/python/.idea directory):
> {code}
> beam/sdks/python $ /Applications/PyCharm.app/Contents/MacOS/pycharm inspect 
> $(pwd) $(pwd)/.idea/inspectionProfiles/Project_Default.xml 
> $(pwd)/inspection-results -v2 -d $(pwd)/apache_beam
> {code}
>       
> *Example result:*
> {code:title=PyUnresolvedReferencesInspection.xml}
> <problem>
> <file>file://$PROJECT_DIR$/apache_beam/io/range_trackers.py</file>
> <line>342</line>
> <module>python</module>
> <entry_point TYPE="file" 
> FQNAME="file://$PROJECT_DIR$/apache_beam/io/range_trackers.py" />
> <problem_class severity="WARNING" 
> attribute_key="WARNING_ATTRIBUTES">Unresolved references</problem_class>
> <description>Unresolved attribute reference '_end_position' for class 
> 'OrderedPositionRangeTracker'</description>
> </problem>
> {code}
>       
> *References:*
> [1] PyCharm Command Line Interface: 
> https://www.jetbrains.com/help/pycharm/2016.3/command-line-code-inspector.html
> [2] GitHub Commit Status API: https://developer.github.com/v3/repos/statuses/
> [3] Eclipse PyDev Plugin: http://www.pydev.org/manual_adv_code_analysis.html
> [4] SonarQUBE Python plugin: 
> https://www.sonarsource.com/why-us/products/codeanalyzers/sonarpython/rules.html
>       
> CC: [~altay] / [~jbonofre] / [~jasonkuster]



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Reply via email to