JVM IR: Don't generate bridges for default argument stubs (KT-46389)

compiler/testData/codegen/box/bridges/kt46389.kt

@@ -0,0 +1,23 @@
+// IGNORE_BACKEND: JS, JS_IR
+
+// The code in this test should be prohibited in the frontend, see KT-36188.
+
+// This test fails with an IllegalStateException on the JS(-IR) backend,
+// but the behavior would probably not match the JVM backend even if
+// the test passed. Compare with kt46389_jvmDefault.
+
+interface I {
+    fun h(x: String = "O"): Any
+}
+
+interface I2 : I
+
+open class A {
+    inline fun h(x: String = "K") = x
+}
+
+class B : A(), I2
+
+fun box(): String {
+    return "${(B() as I).h()}${B().h()}"
+}

compiler/testData/codegen/box/bridges/kt46389_jvmDefault.kt

@@ -0,0 +1,30 @@
+// IGNORE_BACKEND: JS, JS_IR
+// !JVM_DEFAULT_MODE: all
+// JVM_TARGET: 1.8
+
+// The code in this test should be prohibited in the frontend, see KT-36188.
+
+// Before the fix to kt46389, the IR backend would have added a bridge
+// for `h$default` in `B`, which would have lead both calls to use "K"
+// as the default value. This would arguably be less surprising, but is
+// impossible to match without jvm-default on the JVM(-IR)backend.
+//
+// Meanwhile, the two calls using "K" as the default value is the current
+// behavior on the JS backend, which is why the test is muted for the JS
+// backend.
+
+interface I {
+    fun h(x: String = "O"): Any
+}
+
+interface I2 : I
+
+open class A {
+    fun h(x: String = "K") = x
+}
+
+class B : A(), I2
+
+fun box(): String {
+    return "${(B() as I).h()}${B().h()}"
+}