Re: [Codenarc-user] Running rules which require compilerPhase > Phases.CONVERSION using Gradle's CodeNarc plugin

[Chris M. <cm...@gm...>]

Marcin,

Yes, your proposal seems reasonable, and I am okay putting out a CodeNarc
release afterward.

Thanks for your in-depth analysis.

Chris

On Fri, Mar 10, 2017 at 9:53 AM, Marcin Erdmann <mar...@pr...>
wrote:

> So I had a look and this is my current thinking:
>
> 1. Add a new *classpath* nested element and *classpathref* property to
> CodeNarc's ant task. They would be analogous to the ones with the same
> names defined for groovyc ant task (http://groovy-lang.org/
> groovyc.html#_ant_task) and thus use a way of defining classpaths ant
> users are familiar with (by being backed with org.apache.tools.ant.types.
> Path).
> 2. Use the newly added classpath definition to create a new
> *UrlClassLoader* with current context classloader as it's parent in
> *CodeNarcTask.execute()*. Set the newly created classloader as the
> context classloader around the call to *CodeNarcRunner.execute()*.
> 3. Get a new version of CodeNarc released so that the new functionality
> can be integrated with Gradle's CodeNarc plugin.
> 4. Add a new *FileCollection compilationClasspath* property to Gradle's
> CodeNarc task that will be used to configure *classpath* nested element
> of *CodeNarcTask* in *CodeNarcInvoker*. If the *compilationClasspath* property
> is not set then *classpath* nested element is not configured allowing for
> backwards compatibility with older CodeNarc versions.
>
> Another option would be to change which classloader is used when CodeNarc
> compiles analysed code to the one that was used to load its classes, but
> I'd rather not do that because:
> - it's a potentially breaking change
> - you should not be forced to use the same classpath for both the tool
> (Codenarc) and the classes you compile
>
> Chris, I would highly appreciate if you could confirm that you would be ok
> with such change and would be happy to do a release after it's merged in to
> unlock integrating the change with Gradle. Upon receiving a go ahead from
> you I will send a message to Gradle dev list asking if they are ok with my
> proposed change in point 4.
>
> Cheers,
> Marcin
>
> On Wed, Mar 8, 2017 at 7:56 AM, Marcin Erdmann <mar...@pr...>
> wrote:
>
>> Thanks for your response, Chris.
>>
>> It's good to understand that using the context classloader when compiling
>> analysed code is not a conscious design decision but just the way it works
>> at the moment.
>>
>> I will do some more digging around on Friday which is my OSS day at work.
>> I will check what the context classloader is set to when the extended rules
>> are tested within CodeNarc's test suite and investigate if specifying a
>> classpath to be used for the context classloader as part of CodeNarc's ant
>> task is an option and how we would deal with that backwards incompatible
>> change in Gradle's CodeNarc plugin.
>>
>> Cheers,
>> Marcin
>>
>> On Wed, 8 Mar 2017 at 02:32, Chris Mair <cm...@gm...> wrote:
>>
>>> Ugh. Sorry for the delayed response. I wasn't sure how to respond, and I
>>> was hoping that maybe someone else might have some useful input.
>>>
>>> I am not very familiar with the internals of the Gradle CodeNarc plugin,
>>> so I'm not sure what to suggest from that side.
>>>
>>> I am not philosophically opposed to changing the CodeNarc classloader,
>>> but I am nervous about unintended consequences. Support for the "enhanced"
>>> rules was a later addition, and presumably less used, though I'm still
>>> surprised no one else has complained about this.
>>>
>>> On Fri, Mar 3, 2017 at 8:39 AM, Marcin Erdmann <
>>> mar...@pr...> wrote:
>>>
>>> Hi,
>>>
>>> I wrote a custom rule which requires SEMANTIC_ANALYSIS compiler phase
>>> and I'm trying to run it using Gradle. Unfortunately I'm getting
>>> compilation errors upon analysis telling me that compiler cannot load any
>>> of the classes imported by the analysed code, even though I checked and
>>> they are available in the codenarc configuration of my project which is
>>> added by Gradle's CodeNarc plugin and used to load CodeNarc classes, see:
>>>
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> code-quality/src/main/groovy/org/gradle/api/plugins/
>>> quality/CodeNarcPlugin.java#L72
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> code-quality/src/main/groovy/org/gradle/api/plugins/
>>> quality/CodeNarcPlugin.java#L89
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> code-quality/src/main/groovy/org/gradle/api/plugins/
>>> quality/CodeNarc.java#L104
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> code-quality/src/main/groovy/org/gradle/api/plugins/
>>> quality/internal/CodeNarcInvoker.groovy#L45
>>>
>>> I then dug around and realised that when CodeNarc compiles analysed code
>>> it uses the context class loader, see:
>>> - https://github.com/CodeNarc/CodeNarc/blob/v0.26.0/src/main/
>>> groovy/org/codenarc/source/AbstractSourceCode.groovy#L88
>>> - https://github.com/groovy/groovy-core/blob/GROOVY_2_1_8/sr
>>> c/main/org/codehaus/groovy/control/CompilationUnit.java#L94
>>> https://github.com/groovy/groovy-core/blob/GROOVY_2_1_8/src/
>>> main/org/codehaus/groovy/control/CompilationUnit.java#L136
>>> - https://github.com/groovy/groovy-core/blob/GROOVY_2_1_8/sr
>>> c/main/org/codehaus/groovy/control/ProcessingUnit.java#L64
>>> - https://github.com/groovy/groovy-core/blob/GROOVY_2_1_8/sr
>>> c/main/org/codehaus/groovy/control/ProcessingUnit.java#L103
>>>
>>> And I finally realised that Gradle's BasicAntBuilder replaces the
>>> context class loader to only contain Ant classes which means that not even
>>> Groovy classes are available on the compilation classpath when CodeNarc
>>> compiles the analysed code:
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> code-quality/src/main/groovy/org/gradle/api/plugins/
>>> quality/internal/CodeNarcInvoker.groovy#L45
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> core/src/main/java/org/gradle/api/internal/project/
>>> antbuilder/DefaultIsolatedAntBuilder.java#L151
>>> - https://github.com/gradle/gradle/blob/v3.3.0/subprojects/
>>> core/src/main/java/org/gradle/api/internal/project/ant/
>>> BasicAntBuilder.java#L76
>>>
>>> Given the above, here are my questions:
>>>
>>> 1. Has anybody ever managed to get their projects set up correctly and
>>> get violations for one of the built-in CodeNarc Enhanced Classpath Rules,
>>> e.g. CloneWithoutCloneable? If that's the case can they please explain how
>>> they achieved it? If one is to believe comments in Gradle's codebase (e.g.
>>> https://github.com/gradle/gradle/blob/v3.3.0/subprojec
>>> ts/core/src/main/java/org/gradle/api/internal/project/
>>> antbuilder/DefaultIsolatedAntBuilder.java#L140) then it looks like a
>>> similar problem to the one occurring when using Gradle would occur when
>>> using Ant.
>>>
>>> 2. Is there a reason for CodeNarc to use the context class loader
>>> instead of the class loader that loaded CodeNarc when compiling the
>>> analysed code? Given that it is a breaking change, would a PR changing that
>>> class loader to the one which loaded CodeNarc be a valid option? It is a
>>> breaking change but I don't see a reason why somebody would depend on
>>> current behaviour as it actually prevents usage of Enhanced Classpath
>>> Rules. Or should I look into changing Gradle's CodeNarc plugin to set the
>>> context class loader as per CodeNarc's expectations?
>>>
>>> Cheers,
>>> Marcin
>>>
>>>
>>>
>>> ------------------------------------------------------------
>>> ------------------
>>> Check out the vibrant tech community on one of the world's most
>>> engaging tech sites, SlashDot.org! http://sdm.link/slashdot
>>> _______________________________________________
>>> Codenarc-user mailing list
>>> Cod...@li...
>>> https://lists.sourceforge.net/lists/listinfo/codenarc-user
>>>
>>>
>>>
>