[core] empty synthetic class kept after method inlining

[obus-globus]

Issue details

A synthetic class whose only method gets inlined into its caller is left in the output as an empty shell:

public final /* synthetic */ class c {
}

jadx already removes empty synthetic classes (ClassModifier.isEmptySyntheticClass), but this one survives because the check runs before the inline pipeline and looks at the raw method list.

Minimal smali repro.

Trampoline.smali:

.class public final synthetic Lhusk/Trampoline;
.super Ljava/lang/Object;
.method public static synthetic bridge a(Ljava/lang/StringBuilder;)I
    .registers 1
    invoke-virtual {p0}, Ljava/lang/StringBuilder;->length()I
    move-result v0
    return v0
.end method

Caller.smali:

.class public Lhusk/Caller;
.super Ljava/lang/Object;
.method public static check(Ljava/lang/StringBuilder;)Z
    .registers 2
    invoke-static {p0}, Lhusk/Trampoline;->a(Ljava/lang/StringBuilder;)I
    move-result v0
    const/4 v1, 0x5
    if-lt v0, v1, :cond_0
    const/4 v0, 0x1
    return v0
    :cond_0
    const/4 v0, 0x0
    return v0
.end method

Caller inlines to return sb.length() >= 5;, but husk/Trampoline.java is emitted as an empty public final /* synthetic */ class Trampoline {}.

Jadx version

master (8c28a853)

[obus-globus]

Some root-cause context for whoever picks this up.

The empty class only appears after inlining, and the emptiness is not detectable until codegen.
ClassModifier (which already drops empty synthetic classes via isEmptySyntheticClass) runs before MarkMethodsForInline (runAfter = ClassModifier), so the trampoline method is still present when ClassModifier sees the class. The decision to actually drop the inlined method is made at codegen in ClassGen.skipMethod, because it depends on the callee useIn reaching empty, which is only known once every caller class has been processed; there is no global post-process barrier (processing is lazy per class-tree). JadxDecompiler.getClasses() also caches the top-level list filtered on DONT_GENERATE before per-class codegen, so a flag set during codegen is too late for the tree.

So a ClassGen-only fix leaves the class in getClasses() (an empty node in the GUI) and in JSON output (CodeGen.generate routes JSON to JsonCodeGen, which never calls ClassGen.makeClass). A complete fix looks like it needs the emptiness re-checked after inline decisions settle, somewhere that runs before getClasses() caches. Happy to help if you can point at the right hook.