[lworld] RFR: 8367698: [lworld] New lint category for code that would not be allowed in the prologue

[Maurizio Cimadamore]

On Wed, 17 Sep 2025 00:57:34 GMT, Vicente Romero <vromero at openjdk.org> wrote:

> This PR introduces a new Xlint category: `initialization`. When enabled it will indicate if some code in an identity class couldn't be placed in the prologue phase. This applies to field initializers and constructors with no explicit `super` invocation. So for example for code like:
> 
> 
> class Test {
>     int i = 0;
> 
>     Test() {
>         this.i = 1;
>     }
> }
> 
> 
> a warning will be issued as if there were a super invocation at the end of the constructor, the compiler would issue an error.
> 
> PS. The new lint warning had to be disabled during the build as in other case it wouldn't finish

src/jdk.compiler/share/classes/com/sun/tools/javac/code/Lint.java line 285:

> 283:          * construction due to a this dependency.
> 284:          */
> 285:         INITIALIZATION("initialization"),

IMHO this should contain the word `strict` somewhere.

src/jdk.compiler/share/classes/com/sun/tools/javac/comp/Attr.java line 1257:

> 1255:                     }
> 1256:                     if (!prologueCode.isEmpty()) {
> 1257:                         CtorPrologueVisitor ctorPrologueVisitor = new CtorPrologueVisitor(localEnv, addedSuperInIdentityClass && allowValueClasses);

I suppose I'd rather not run the visitor if the lint warning is disabled?

src/jdk.compiler/share/classes/com/sun/tools/javac/comp/Attr.java line 1601:

> 1599:                             v.type = chk.checkLocalVarType(tree, tree.init.type, tree.name);
> 1600:                         }
> 1601:                         if (allowValueClasses && v.owner.kind == TYP && !v.isStatic()) {

checking again whether value classes are enabled here seems redundant?

test/langtools/tools/javac/SuperInit/InitializationWarningTest.java line 9:

> 7:  */
> 8: 
> 9: class InitializationWarningTest implements Iterable<Object> {

Since the code base is shared -- perhaps a better testing methodology would be to run _all_ the existing early init negative tests in this new mode, to make sure they generate a warning when there was an error before?

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

PR Review Comment: https://git.openjdk.org/valhalla/pull/1595#discussion_r2355572298
PR Review Comment: https://git.openjdk.org/valhalla/pull/1595#discussion_r2355557583
PR Review Comment: https://git.openjdk.org/valhalla/pull/1595#discussion_r2355570464
PR Review Comment: https://git.openjdk.org/valhalla/pull/1595#discussion_r2355577796