condy-folding bug; static field initializer is discarded

[Brian Goetz]

Yes, I noted the incorrect dead-code elimination of the putstatic in my 
previous.  I think that's the same issue -- that we're not generating a 
putstatic here, so the field appears null?



On 8/10/2018 2:49 PM, Jorn Vernee wrote:
> Hello,
>
> I think I found another bug in the condy-folding branch. I noticed 
> when trying to do a static get of a constant in another compilation 
> unit that the value was always `null`. Here is the code:
>
> ```
> import java.lang.constant.*;
> import java.lang.constant.DirectMethodHandleDesc.Kind;
> import static java.lang.constant.ConstantDescs.*;
>
> public class MyConstants {
>
>     // Not ICE from different compilation unit
>     public static final DynamicConstantDesc<String> INV_myMethod
>         = DynamicConstantDesc.<String>of(BSM_INVOKE).withArgs(
>             MethodHandleDesc.of(
>                 Kind.STATIC,
>                 ClassDesc.of("MyConstants"),
>                 "myMethod",
>                 CR_String
>             )
>         );
>
>     public static String myMethod() {
>         return "Hello: ";
>     }
> }
> ```
>
> If I then do a `System.out.println(MyConstants.INV_myMethod);` from a 
> different compilation unit, the value is `null`. Looking at javap 
> output, there is no <clinit> generated for the initializer. If I do 
> that manually like so:
>
> ```
>     public static final DynamicConstantDesc<String> INV_myMethod;
>
>     static {
>         INV_myMethod
>             = DynamicConstantDesc.<String>of(BSM_INVOKE).withArgs(
>                 MethodHandleDesc.of(
>                     Kind.STATIC,
>                     ClassDesc.of("MyConstants"),
>                     "myMethod",
>                     CR_String
>                 )
>             );
>     }
> ```
>
> I get a value as expected when printing:
>
> ```
> DynamicConstantDesc[ConstantBootstraps::invoke(MethodHandleDesc[STATIC/MyConstants::myMethod()String])Object] 
>
> ```
>
> This looks like a bug?
>
> Jorn