RFR: 8269738: AssertionError when combining pattern matching and function closure

[Vicente Romero]

On Wed, 7 Jul 2021 18:51:52 GMT, Vicente Romero <vromero at openjdk.org> wrote:

>> Hi all,
>> 
>> The method `TransPatterns.BasicBindingContext::bindingDeclared` uses `varSymbol.owner` to construct a new `VarSymbol`. 
>> 
>> 
>>      res = new VarSymbol(varSymbol.flags(), varSymbol.name, varSymbol.type, varSymbol.owner);
>> 
>> 
>> But the `varSymbol.owner` may be also a `VarSymbol` when it is at the `init` clause of the static variables or instance variables.
>> You can see the following code. The `owner` of the `Interger i` is `staticNum1` or `instanceNum1` which is `VarSymbol`.
>> 
>> 
>>     static Number num1 = 1;
>>     static Number staticNum1 = (num1 instanceof Integer i) ? ((Supplier<Integer>) () -> i).get() : null;
>>     Number instanceNum1 = (num1 instanceof Integer i) ? ((Supplier<Integer>) () -> i).get() : null;
>> 
>> 
>> Then, the class `LambdaToMethod` can't capture the expected variables `Integer i` and the following phases such as `Gen` may fail or throw exceptions.
>> 
>> This patch uses the `currentMethodSym` which means a class or instance field initializer to replace `varSymbol.owner` and adds the corresponding test.
>> 
>> ---
>> 
>> And I have a problem:
>> 
>>> The `owner` of the `Interger i` is `staticNum1` or `instanceNum1` which is `VarSymbol`.
>> 
>> The statement above means the `env.info.scope.owner` in the method `Attr::visitBindingPattern` is a `VarSymbol`. Is it a expected result? Or the original `env.info.scope.owner` should be modified to a class or instance field initializer which the following phases need.
>> 
>> 
>>     // Attr.java
>>     public void visitBindingPattern(JCBindingPattern tree) {
>>         ......
>>         BindingSymbol v = new BindingSymbol(tree.var.mods.flags, tree.var.name, tree.var.vartype.type, env.info.scope.owner);
>>         ......
>>     }
>> 
>> 
>> ---
>> 
>> Thanks for taking the time to review.
>> 
>> Best Regards,
>> -- Guoxiong
>
> lgtm

> @vicente-romero-oracle Thanks for your review. Should we raise this bug to `P3` and fix it in JDK 17?

yep, it makes sense to do it

-------------

PR: https://git.openjdk.java.net/jdk/pull/4678