Litho: NoSuchFieldError: savedParameterNames

[cushon]

There is a reference to com.sun.tools.javac.code.Symbol.MethodSymbol.savedParameterNames in MethodExtractorUtils:

https://github.com/facebook/litho/blob/ce87d198c8f5ac0cee2c9302cb48076b06253ef6/litho-processor/src/main/java/com/facebook/litho/specmodels/processor/MethodExtractorUtils.java#L99

That field is an internal implementation detail of javac and is not part of the supported API. It is being removed in JDK 10. See this change: http://hg.openjdk.java.net/jdk/jdk/rev/89f6aa26fd6c, which fixes JDK-8177486 and JDK-8007720.

The removal of the field causes DelegateMethodExtractor to crash when running on a JDK 10 javac:

Caused by: java.lang.NoSuchFieldError: savedParameterNames
    at com.facebook.litho.specmodels.processor.MethodExtractorUtils.getSavedParameterNames(MethodExtractorUtils.java:87)
    at com.facebook.litho.specmodels.processor.MethodExtractorUtils.getMethodParams(MethodExtractorUtils.java:45)
    at com.facebook.litho.specmodels.processor.DelegateMethodExtractor.getDelegateMethods(DelegateMethodExtractor.java:63)
    at com.facebook.litho.specmodels.processor.LayoutSpecModelFactory.create(LayoutSpecModelFactory.java:79)
    at com.facebook.litho.specmodels.processor.LayoutSpecModelFactory.create(LayoutSpecModelFactory.java:33)
    at com.facebook.litho.specmodels.processor.AbstractComponentsProcessor.process(AbstractComponentsProcessor.java:66)
    at jdk.compiler/com.sun.tools.javac.processing.JavacProcessingEnvironment.callProcessor(JavacProcessingEnvironment.java:969)

[pavlospt]

@cushon are we sure that this is something that works? Can you please share some info on how you tested it? Seems to me that at least a previous existing functionality looks broken on a quick look at the code diff, especially at the line http://hg.openjdk.java.net/jdk/jdk/rev/89f6aa26fd6c#l1.36 (the i is not being incremented properly like it was before). Also is this something final that will end up on JDK 10?

[cushon]

@pavlospt thanks for catching that. As I mentioned in the compiler-dev thread I think the patch is OK, and the index that isn't incremented is effectively dead code.

For JDK 10 it should be sufficient to compile with -parameters and then use the names on the symbols returned by getParameters(). With JDK 8 and 9 some of the litho tests were running into JDK-8007720 - normally savedParameterNames should be null by the time annotation processing happens.

[pavlospt]

@cushon thanks for the follow up :)

[passy]

Hey @cushon! Thanks for reporting this. I'm kinda jealous of you for tracking the JDK so closely. We have to maintain compatibility with JDK 7, which means we make use of -parameters yet.

Because of the limitation stated in the patch description itself, I had to implement a workaround for this anyway. We are now persisting parameter names in META-INF as resources during codegen, which should make this entire fallback unnecessary. The only place where this might still be used is in tests.

I'll see if I can remove it altogether and if not, I will wrap this in a try/catch to make sure this doesn't break with JDK 10.

[passy]

I went with a try/catch to maintain backwards compatibility. I hope this works for you!

[cushon]

Thanks - that achieves binary compatibility with JDK 10, but not source compatibility. WDYT about Symbol.MethodSymbol.class.getField("savedParameterNames").get(methodSymbol)?

[passy]

Sounds good to me. Thanks, I'll make that change!

[vinc3m1]

friendly bump :) will help clean up my import script to have source compatibility here before the next release

[passy]

Thanks for the reminder. I'll make that change tomorrow!

On Wed, 10 Jan 2018, 21:00 Vince Mi, [e-mail] wrote:

friendly bump :) will help clean up my import script to have source
compatibility here before the next release

—
You are receiving this because you modified the open/close state.

Reply to this email directly, view it on GitHub
https://github.com/facebook/litho/issues/295#issuecomment-356734997, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAAmshatLeWBUptiK8_DI-rnO3id0Zntks5tJSTUgaJpZM4RN5sq
.

>

timezone: eu/london | utc+1
@passy