Attempting to upgrade to the new version (?), I've noticed some oddities with SetOpaque:

• Since Prevent JS_SetOpaque from overriding internal class state #658 it can fail. But it returns JS_FALSE (0) on success and JS_TRUE (1) on failure, which I find somewhat counter-intuitive. Maybe it's not too late to change it?
• Previously, I could set an opaque on the global object, but now I can't because it's a JS_CLASS_OBJECT which is < JS_CLASS_INIT_COUNT.

The latter one is a bit problematic, because my type conversion functions expect the global object to have an opaque. I would change its class ID if I could, but at least when I was writing my wrapper for qjs I couldn't, and I don't see any changes in -ng that would enable this.

I don't know what the preferred fix is for this one; this patch works for me:

diff --git a/quickjs.c b/quickjs.c
index 9cac6de..e268ce3 100644
--- a/quickjs.c
+++ b/quickjs.c
@@ -10102,7 +10102,8 @@ JS_BOOL JS_SetOpaque(JSValue obj, void *opaque)
     if (JS_VALUE_GET_TAG(obj) == JS_TAG_OBJECT) {
         p = JS_VALUE_GET_OBJ(obj);
         // User code can't set the opaque of internal objects.
-        if (p->class_id >= JS_CLASS_INIT_COUNT) {
+        if (p->class_id >= JS_CLASS_INIT_COUNT ||
+            p->class_id == JS_CLASS_OBJECT) {
             p->u.opaque = opaque;
             return 0;
         }