287846 - the checkin for bug 287107 broke Forumzilla

Status: RESOLVED FIXED

(Core :: XPConnect, defect, P1)

Description

[Myk Melez [:myk] [@mykmelez]]

Since the fix for bug 287107 was checked in, the Forumzilla extension for
Thunderbird dies with the following error:

JavaScript error: , line 0: uncaught exception: [Exception... "Component
returned failure code: 0x80570016 (NS_ERROR_XPC_GS_RETURNED_FAILURE)
[nsIJSCID.getService]"  nsresult: "0x80570016
(NS_ERROR_XPC_GS_RETURNED_FAILURE)"  location: "JS frame ::
chrome://forumzilla/content/forumzilla.js :: onLoad :: line 35"  data: no]

Lines 32-35 of forumzilla.js:

    feedService = 
      Components
        .classes["@forumzilla.mozdev.org/feed/feedservice;1"]
          .getService(Components.interfaces.nsIFeedService);

The feed service is a JavaScript XPCOM component that comes with Forumzilla.

Reinstalling Forumzilla doesn't solve the problem, nor does installing into a
fresh profile.  The fix for bug 287490 doesn't have any effect either, but
everything works fine if I back out the fix for bug 287107.

Comment 1

[Brendan Eich [:brendan]]

I never saw bugmail from being cc'd on this bug.  Hrm.  Anyone see mail log
traces saying why?

Should this bug be WFM now, or FIXED?

/be

Comment 2

[Asa Dotzler [:asa]]

This is windows as well as Linux.

I don't think this is fixed at all -- unless something changed in the last
couple of hours and bonsai doesn't show anything.

Comment 3

[Myk Melez [:myk] [@mykmelez]]

> I never saw bugmail from being cc'd on this bug.  Hrm.  Anyone see mail log
> traces saying why?

All mozilla.org mail was bouncing for two hours today around the time I filed
this bug after dbaron accidentally horked the aliases database.  That's probably
why you didn't receive mail about it (I didn't either).

> Should this bug be WFM now, or FIXED?

Neither unless something has changed since this afternoon, when I pulled and
built again to confirm the bug on the tip.  Note that bug 287490, the other
regression from bug 287107 (which this bug may just be a dupe of), has since
been reopened, as two users report that the patch on that bug didn't fix the
problem for them.

Comment 4

[Brendan Eich [:brendan]]

Myk, can you breakpoint
http://lxr.mozilla.org/mozilla/source/js/src/xpconnect/src/xpcjsid.cpp#900 and
see why that's being hit?

/be

Comment 5

[Myk Melez [:myk] [@mykmelez]]

> Myk, can you breakpoint
> http://lxr.mozilla.org/mozilla/source/js/src/xpconnect/src/xpcjsid.cpp#900 and
> see why that's being hit?

rv=0x80004002 and srvc is a null pointer at line 900 when the app tries to load
the feed service.

Comment 6

[Brendan Eich [:brendan]]

Attachment: fix?

Comment 7

[Brendan Eich [:brendan]]

Attachment: fix?

Comment 8

[Brendan Eich [:brendan]]

Working with myk over IRC, finding that one thing that might be adding noise is
a lack of nsISupports handling in JS-scripted QueryInterface impls.  Myk fixed
his to handle nsISupports and that moved the symptom.  Could download code have
a similar lack in one of its scripted QI impls?

/be

Comment 9

[Brendan Eich [:brendan]]

Fooey.  I backed out.  Something's not right in a bunch of places in our JS QI
impls, which return null after setting Components.returnCode ineffectually (due
to bug 287107).  My fix for bug 287107 makes these QIs truly fail with
NS_NOINTERFACE yet not throw, and that leaves callers unhappy.

Keeping bug open to investigate more.

/be

Comment 10

[Brendan Eich [:brendan]]

Attachment: private patch for myk, me, and anyone else willing to help debug

<brendan> k, here's what happened
<brendan> rev 1.73 of xpcwrappedjsclass.cpp regressed things so a JS method
couldn't set Components.returnCode, and succeed (not throw or error out), yet
have the returnCode propagate back to the XPConnected caller
<brendan> that means these QueryInterface impls in JS, such as yours, are
wasting their time setting Components.returnCode without also throwing
(failing)
<brendan> cuz it's never fetched via GetPendingResult in any
xpconnect/src/*.cpp (grep yourself) on a success path
<brendan> my fix for 287107 made these returnCode statements do what they used
to
<brendan> and that showed up bugs
<brendan> your QI didn't handle nsISupports
<brendan> and maybe download mgr bugs of that kind
<brendan> i'll patch privately and debug more, later
<myk> i'll do the same and see if i can fix any bugs in my components
<brendan> i'll put the private patch up and then crash -- it's tricky

Comment 11

[Brendan Eich [:brendan]]

Technically fixed, again.  Hope not to reopen if the private patch shakes out
all the missing QI cases.

/be