RFR: 8373570: Javac stack overflow on method-local class with nested record referring to enclosing type

[Vicente Romero]

On Mon, 15 Dec 2025 11:02:36 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

> The fix for https://git.openjdk.org/jdk/pull/21410 introduced a regression: javac will no longer reject local classes created from a context that is not an inner class of the class enclosing the local class.
> 
> In other words, javac is failing to apply this restriction in the JLS  (15.9.2):
> 
>> If U is not an inner class of O or O itself, then a compile-time error occurs
> 
> The issue is causes by the new method `Resolve::findLocalClassOwner`. This method is meant to start at the point where the local class is instantiated, and walk up the enclosing scopes, until it finds the place where the local class is defined. If any static context has been traversed doing this search, an error is issued.
> 
> Unfortunately, neither static local classes, nor records/interfaces count as "static contexts" in the javac implementation -- that is, they do not increase the "static" counter in the `env.info` variable.
> 
> Instead, most code in `Resolve` does this (see e.g. `Resolve::findFun`):
> 
> 
> while (env1.outer != null) {
>     if (isStatic(env1)) staticOnly = true;
>     ...
>     if ((env1.enclClass.sym.flags() & STATIC) != 0) staticOnly = true;
>     env1 = env1.outer;
> }
> 
> 
> The new method acts a bit differently, and it's missing the logic to check the staticness of the enclosing class, hence the regression.

lgtm

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

Marked as reviewed by vromero (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/28821#pullrequestreview-3580197472