Similar to https://bugs.webkit.org/show_bug.cgi?id=97375 IDBAny is used for attributes of type "any", and these should be replaceable with ScriptValue: IDBCursor.idl: readonly attribute IDBAny source; IDBRequest.idl: readonly attribute IDBAny source; ^^^ Another IDB type (IDBObjectStore, IDBIndex, etc) - would just need a neutral way to call ScriptValue(to[V8|JS](object)) IDBDatabase.idl: readonly attribute IDBAny version; ^^^ Number or String - would need a way to initialize ScriptValue from a number/string IDBIndex.idl: readonly attribute IDBAny keyPath; IDBObjectStore.idl: readonly attribute IDBAny keyPath; ^^^ Null, String or DOMStringList IDBRequest.idl: readonly attribute IDBAny result ^^^ All of the above, or arbitrary script value.
IDBVersionChangeEvent.idl: readonly attribute IDBAny oldVersion; IDBVersionChangeEvent.idl: readonly attribute IDBAny newVersion; ^^^ Number or String ... I've been thinking about doing this in three parts: (1) Remove IntegerType/StringType/NullType/KeyPathType in favor of just ScriptValueType + the various IDB*Types (2) Reduce the IDBAny type set down to ScriptValue, and have a binding helper that does (RefPtr<T>) -> ScriptValue via toV8/toJS(). (3) Eliminate IDBAny itself IMHO #1 is valuable on its own as it removes the need to create scratch SerializedScriptValues which then are deserialized into ScriptValues. I've started #1 then abandoned this change about 5 times so far. The crux of it is that we need a ScriptExecutionContext at the point where the ScriptValue is created to get an isolate. Right now this easily done in IDBRequest but requires "dirtying" some of the other call sites - adding [CallWith=ScriptExecutionContext] IDL annotations here and there. I've toyed with several API variations, e.g.: * Require a ScriptExecutionContext to be passed to IDBAny constructors - callers need to are otherwise unchanged. * Add IDBBindingUtilities functions for stringToScriptValue(isolate, str), numberToStringValue(isolate, double), etc; IDBAny creators pass in the ScriptValue * Add constructors to ScriptValue: ScriptValue(isolate, double), ScriptValue(isolate, string), ScriptValue::null(isolate), Opinions?
I vote for constructors on ScriptValue, the same way SerializedScriptValue already has intValue() and such.
(In reply to comment #2) > I vote for constructors on ScriptValue, the same way SerializedScriptValue already has intValue() and such. Agreed.