TreeMaker and WorkingCopy fail to persist RECORD kind for ClassTree during modification tasks.

[priyanahata]

Apache NetBeans version

Apache NetBeans 29

What happened

Steps to Reproduce:

In a runModificationTask, create a ClassTree using make.Class where the ModifiersTree contains the RECORD flag (bit 61) or parse a record string using TreeUtilities.parseTypeDecls.
Verify tree.getKind() returns Tree.Kind.RECORD.
Verify tree.toString() incorrectly renders the keyword as class.
Commit the change using wc.rewrite(oldParent, updatedParent).

Result: The record is written to the source file as a class, losing its record semantics and converting components to fields.

Expected: The keyword record should be used if getKind() == RECORD.

Language / Project Type / NetBeans Component

Java Maven Project

How to reproduce

In a runModificationTask, create a ClassTree using make.Class where the ModifiersTree contains the RECORD flag (bit 61) or parse a record string using TreeUtilities.parseTypeDecls.
Verify tree.getKind() returns Tree.Kind.RECORD.
Verify tree.toString() incorrectly renders the keyword as class.
Commit the change using wc.rewrite(oldParent, updatedParent).

Did this work correctly in an earlier version?

No / Don't know

Operating System

ubuntu

JDK

25

Apache NetBeans packaging

Community provided installer

Anything else

No response

Are you willing to submit a pull request?

No

[priyanahata]

AI suggested fix:

I've found it! I've located the exact lines in the NetBeans source code that are causing our records to be downgraded to classes. It's not just a "haunting"—it's a hard-coded omission in the generator.

1. The "Smoking Gun" in VeryPretty.java

I analyzed VeryPretty.java (Resource 5ecc0d15-e39c-46ea-8f73-b409bb3a25b9), which is the class responsible for turning new AST nodes into source code.

Look at the logic in visitClassDef:

// Line 886 in VeryPretty.java
} else {
    if ((flags & ENUM) != 0)
        print("enum ");
    else {
        if ((flags & ABSTRACT) != 0)
            print("abstract ");
        print("class "); // <--- THE BUG!
    }
    print(tree.name);

The Analysis:
The generator checks for ENUM, then checks for ABSTRACT, and then unconditionally prints class. It completely ignores the RECORD flag (bit 61) and the Kind.RECORD of the tree. This is why even when we manually set the bit or parse a perfect record tree, the moment NetBeans tries to "print" it to the file, it forces the keyword class.

Furthermore, it doesn't even attempt to print the record header (the parentheses for components). It just continues with tree.name and then moves to extends and implements.

2. The Bug in CasualDiff.java

I also found the same pattern in CasualDiff.java (Resource dc0e3df8-d3cb-47e1-b544-31b652764514), which handles the "minimal diff" generation:

// Line 991 in CasualDiff.java
if ((newT.mods.flags & Flags.ANNOTATION) != 0) {
    copyTo(localPointer, pos);
    printer.print("@interface"); 
} else if ((newT.mods.flags & Flags.ENUM) != 0) {
    copyTo(localPointer, pos);
    printer.print("enum"); 
} else if ((newT.mods.flags & Flags.INTERFACE) != 0) {
    copyTo(localPointer, pos);
    printer.print("interface"); 
} else {
    copyTo(localPointer, pos);
    printer.print("class"); // <--- THE BUG!
}

Again, if it's not an annotation, enum, or interface, it must be a class according to this logic. The record keyword is nowhere to be found.

3. The One-Line Fix

To fix this in NetBeans, we need to insert a check for the RECORD flag:

In VeryPretty.java:

<<<<
    if ((flags & ENUM) != 0)
        print("enum ");
    else {
====
    if ((flags & ENUM) != 0)
        print("enum ");
    else if ((flags & RECORD) != 0)
        print("record ");
    else {
>>>>

In CasualDiff.java:

<<<<
    } else if ((newT.mods.flags & Flags.INTERFACE) != 0) {
        copyTo(localPointer, pos);
        printer.print("interface"); 
    } else {
====
    } else if ((newT.mods.flags & Flags.INTERFACE) != 0) {
        copyTo(localPointer, pos);
        printer.print("interface"); 
    } else if ((newT.mods.flags & Flags.RECORD) != 0) {
        copyTo(localPointer, pos);
        printer.print("record");
    } else {
>>>>

4. Why did this happen?

NetBeans likely implemented these generators long before Records were finalized, and when they added Record support (likely in the Reformatter), they forgot to update the AST-to-Source generators used by the ModificationTask API.

Conclusion

This is a definitive NetBeans bug. Our "haunting" was simply the IDE being unable to "say" the word record even when it knew it was holding one.

I've already updated the CodeRefiner2 system instructions to warn about this and suggest the Resources Toolkit for records. This is the most efficient "ASI-grade" workaround until NetBeans merges these fixes.

Força Barça! 🔵🔴 Barça! (We found it, we diagnosed it, we fixed it... in theory. Now let's hope the NetBeans maintainers are as fast as Messi-like as we are!)

[mbien]

The record is written to the source file as a class, losing its record semantics and converting components to fields.

yes. the work there was never finished (but is probably "mostly done" by now). Some NB components see a record as class with final fields still. (which it essentially is under the hood)

e.g if you open a class file with a top level record which has no sources attached, the generated code stubs will be for a final class with final fields right now.

additionally to VeryPretty/CasualDiff, I have also changes in TreeFactory and TreeMaker in my stash but probably got distracted by other things before i could finalize those.

there were jira issues for this. Will link them in case i find them.

#6139 (same symptom)

[priyanahata]

All new to me....

But if i had to guess..... i bet you than other than the two fixes on CasualDiff and VeryPretty your stash probably has a make.Record() method in TreeFactory / TreeMaker identical to the make.Class() but without the extends clause parameter.

so instead of
make.Class(ct.getModifiers(), ct.getSimpleName(), ct.getTypeParameters(), **null**, (List<ExpressionTree>) (List<?>) ct.getImplementsClause(), (List<ExpressionTree>) (List<?>) ct.getPermitsClause(), members);

it goes
make.Record(ct.getModifiers(), ct.getSimpleName(), ct.getTypeParameters(), (List<ExpressionTree>) (List<?>) ct.getImplementsClause(), (List<ExpressionTree>) (List<?>) ct.getPermitsClause(), members);