lambda method naming changes in jdk 24

[Egor Ushakov]

Thanks for the details, we'll try to fix it on IDEA side.

Egor

On 7/14/2025 11:32 PM, Maurizio Cimadamore wrote:
>
> Hi Egor,
> I confirm that this change is deliberate. There were several changes 
> required to make JEP 513 (Flexible Constructor Bodies) work. Some of 
> these changes resulted in significant rewrites of the compiler backend 
> (to move lambda translation ahead of inner class translation). The 
> bulk of the work (and some history) is here:
>
> https://github.com/openjdk/jdk/pull/19836
>
> This change created some compatibility issues for serializable 
> lambdas, which were later addressed here:
>
> https://github.com/openjdk/jdk/pull/20349
>
> This latter change dropped a lot of accidental complexity from the 
> javac LambdaToInner pass, and that meant making the logic for 
> computing the names serializable and non-serializable lambdas more robust.
>
> This meant that, unfortunately, the (unspecified) name of some 
> non-serializable lambda changed. This compatibility issue is described 
> in the associated CSR:
>
> https://bugs.openjdk.org/browse/JDK-8337558
>
> The difference you see is caused by when the name of the translated 
> lambda method is generated -- it used to be computed _after_ the end 
> of the lambda body, now it is computed _before_.  This means that as 
> soon as we see the outermost lambda for `r3` we give it the first 
> number -- then the nested lambda for `r4` gets the second number.
>
> This allows us to avoid having to "patch" the names of the translated 
> symbols, which minimizes the amount of mutation in our code, making it 
> less brittle. So, unless there's a big reason as to why these names 
> should go back the way they were, we'd like to keep the code the way 
> it is :-)
>
> Cheers
> Maurizio
>
> On 11/07/2025 16:08, Egor Ushakov wrote:
>> Hi everyone!
>>
>> it looks like in jdk 24 compiler started naming lambda methods 
>> slightly differently,
>> consider the code:
>> public class Lambdas1 {
>>      public static void main(String[]args) {
>>          Runnable r3 = () -> {
>>              Runnable r4 = () ->System.out.println(new Exception().getStackTrace()[0]);
>>              r4.run();
>>              System.out.println(new Exception().getStackTrace()[0]);
>>          };
>>          r3.run();
>>      }
>> }
>>
>> with jdk 23 the output is:
>> Lambdas1.lambda$main$*0*(Lambdas1.java:7)
>> Lambdas1.lambda$main$*1*(Lambdas1.java:9)
>>
>> with jdk 24:
>> Lambdas1.lambda$main$*1*(Lambdas1.java:4)
>> Lambdas1.lambda$main$*0*(Lambdas1.java:6)
>>
>> I know that this is not specified anywhere, but Intellij IDEA relied 
>> on this and it caused issues with debugger features:
>> https://youtrack.jetbrains.com/issue/IDEA-375811
>>
>> Was this change really intended? Could anyone please point me to the 
>> fix that caused this?
>>
>> Thanks!
>> Egor
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/compiler-dev/attachments/20250731/fb8d675b/attachment.htm>