Steps to reproduce: 1. Set pref devtools.chrome.enabled to true 2. Visit http://www.squarefree.com/shell/shell.html 3. Open scratchpad; select menu Environment -> Browser. 4. Paste this code into Scratchpad and run it (Ctrl+R or Cmd+R). var dbg = new Debugger(gBrowser.selectedBrowser.contentWindow); dbg.onDebuggerStatement = function (frame) { alert("hello world"); } 5. In the squarefree tab, type "debugger" and hit enter. Expected: an alert that says "hello world" Observed: step 4 fails with: /* Exception: Debugger is not defined @Scratchpad:1 */

Comment on attachment 618771 [details] [diff] [review] v1 r?bholley for the C++/IDL. r?dcamp for everything else.

Comment on attachment 618771 [details] [diff] [review] v1 >diff --git a/js/ductwork/debugger/IJSDebugger.idl b/js/ductwork/debugger/IJSDebugger.idl >--- a/js/ductwork/debugger/IJSDebugger.idl >+++ b/js/ductwork/debugger/IJSDebugger.idl > [scriptable, uuid(2fc14cc6-4ed0-4bbf-a7dd-e535bf088eb5)] > interface IJSDebugger : nsISupports > { > /** >- * Define the global Debugger constructor. >+ * Define the global Debugger constructor on a given global. > */ > [implicit_jscontext] >- void addClass(); >+ void addClass(in jsval global); > }; Needs an iid rev. >diff --git a/js/ductwork/debugger/JSDebugger.cpp b/js/ductwork/debugger/JSDebugger.cpp >+ JSObject* obj = &global.toObject(); >+ obj = JS_UnwrapObject(obj); Just pass , /* stopAtOuter = */ false here, skip the subsequent innerization, and throw if the result doesn't have the global flag. >+ if (!obj) { >+ return NS_ERROR_FAILURE; > } > >- if (!JS_DefineDebuggerObject(cx, global)) { >- return NS_ERROR_FAILURE; >+ { >+ JSAutoEnterCompartment aec; >+ if (!aec.enter(cx, obj)) { >+ return NS_ERROR_FAILURE; >+ } >+ >+ obj = JS_ObjectToInnerObject(cx, obj); >+ if (!obj) { >+ return NS_ERROR_FAILURE; >+ } >+ } >+ >+ { >+ JSAutoEnterCompartment aec; I'd prefer ac2. But the entire first AutoEnterCompartment is unnecessary given the above. So let's just unwrap, enter the compartment, check that it's the global, and define the object. r=bholley with that.

(In reply to Bobby Holley (:bholley) from comment #3) > Needs an iid rev. Oops. Thanks. > >+ JSObject* obj = &global.toObject(); > >+ obj = JS_UnwrapObject(obj); > > Just pass , /* stopAtOuter = */ false here, skip the subsequent > innerization, and throw if the result doesn't have the global flag. JS_UnwrapObject doesn't have an optional second param. Actually I think it's extern "C" and used from C code (barf). So I added another function, JS_UnwrapObjectAndInnerize. Ready to land when the tree reopens.

(In reply to Jason Orendorff [:jorendorff] from comment #4) > JS_UnwrapObject doesn't have an optional second param. Actually I think it's > extern "C" and used from C code (barf). So I added another function, > JS_UnwrapObjectAndInnerize. Erm, any reason not to just kill the extern C stuff and use js::UnwrapObject from jsfriendapi? Or maybe move it to jsapi if need be? Seems silly to have both.

this was landed directly on central? Bobby, do we need a follow-up to address your question in Comment 5 or can this stick as-landed? This also looks like something that could use some documentation.

(In reply to Rob Campbell [:rc] (:robcee) from comment #7) > this was landed directly on central? > > Bobby, do we need a follow-up to address your question in Comment 5 or can > this stick as-landed? These shims are silly, but there's no duplicated code, so I don't really mind.