On Fri, 25 Apr 2025 07:54:03 GMT, Johannes Döbler <d...@openjdk.org> wrote:
>>> @AlanBateman I have modified it to use interface + record. Is this what you
>>> want?
>>
>> I don't object to changing it to interface + record but it feels more like
>> needless code churn. I really disliked the next version that used
>> printStackTrace0(Object printer) as it immediately invites another re-write.
>
> The record has fewer source lines but doesn't the generated record class have
> a getter, toString(), hashCode() and equals(Object) which are not needed here?
> What about using a lock parameter and a method handle to pass in the required
> features of PrintStream and PrintWriter:
>
> public void printStackTrace(PrintWriter pw) {
> printStackTrace(pw, pw::println);
> }
>
> public void printStackTrace(PrintStream ps) {
> printStackTrace(ps, ps::println);
> }
>
> private void printStackTrace(Object lock, Consumer<Object> println) {
> synchronized(lock) {
> ...
> println.accept(this);
> ...
> }
> }
FYI lambda expressions depend on java.lang.invoke, and if anything initialized
before that throws an exception and needs to print stack trace, the VM will
fail with a traceless `StackOverflowError` I think. Consumer class is fine for
early loading, but to implement that, we fall back to records again.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/24795#discussion_r2060345723
