75607 - Can't create factories for several components in embedding case

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[edburns]

For some reason, on unix only (not on win32), my embedding app can't
work because the following assertions occurr:

+++ nsGenericModule JS component loader: unable to create factory for
{6bd13476-1dd2-11b2-bbef-f0ccb5fa64b6}
###!!! ASSERTION: Factory creation failed: 'NS_SUCCEEDED(rv)', file
nsNativeComponentLoader.cpp, line 157
###!!! Break: at file nsNativeComponentLoader.cpp, line 157
+++ nsGenericModule nsStringBundleModule: unable to create factory for
{d85a17c1-aa7c-11d2-9b8c-00805f8a16d9}
###!!! ASSERTION: Factory creation failed: 'NS_SUCCEEDED(rv)', file
nsNativeComponentLoader.cpp, line 157
###!!! Break: at file nsNativeComponentLoader.cpp, line 157
+++ nsGenericModule nsChromeModule: unable to create factory for
{d8c7d8a2-e84c-11d2-bf87-00105a1b0627}
###!!! ASSERTION: Factory creation failed: 'NS_SUCCEEDED(rv)', file
nsNativeComponentLoader.cpp, line 157
###!!! Break: at file nsNativeComponentLoader.cpp, line 157
+++ nsGenericModule appshell: unable to create factory for
{e34783f5-ac08-11d2-8d19-00805fc2500c}
###!!! ASSERTION: Factory creation failed: 'NS_SUCCEEDED(rv)', file
nsNativeComponentLoader.cpp, line 157
###!!! Break: at file nsNativeComponentLoader.cpp, line 157
+++ nsGenericModule nsProfileModule: unable to create factory for
{02b0625b-e7f3-11d2-9f5a-006008a6efe9}
###!!! ASSERTION: Factory creation failed: 'NS_SUCCEEDED(rv)', file
nsNativeComponentLoader.cpp, line 157
###!!! Break: at file nsNativeComponentLoader.cpp, line 157
+++ nsGenericModule nsPrefModule: unable to create factory for
{dc26e0e0-ca94-11d1-a9a4-00805f8a7ac4}
###!!! ASSERTION: Factory creation failed: 'NS_SUCCEEDED(rv)', file
nsNativeComponentLoader.cpp, line 157
###!!! Break: at file nsNativeComponentLoader.cpp, line 157

I've investigated this quite thoroughly and here's what I have come up
with.

After debugging this further, I find that in 

NS_IMETHODIMP
nsGenericModule::GetClassObject(nsIComponentManager *aCompMgr,
                                const nsCID& aClass,
                                const nsIID& aIID,
                                void** r_classObj)

during the for loop when it's looking for a component whose CID equals
the requested cid, it doesn't find one.  Does this mean that
autoregistration is failing somehow?

#0  nsGenericModule::GetClassObject (this=0x81d3b18, aCompMgr=0x81c9ba8, 
    aClass=@0x809f098, aIID=@0x49137ad0, r_classObj=0xbe5ff3d4)
    at nsGenericFactory.cpp:259
#1  0x490c71e8 in nsNativeComponentLoader::GetFactoryFromModule (
    this=0x81cc350, aDll=0x80f5b90, aCID=@0x809f098, aFactory=0xbe5ff3d4)
    at nsNativeComponentLoader.cpp:1184
#2  0x490c3c5d in nsNativeComponentLoader::GetFactory (this=0x81cc350, 
    aCID=@0x809f098, aLocation=0x809f0f8 "rel:libstrres.so", 
    aType=0x809f0d0 "application/x-mozilla-native", _retval=0xbe5ff3d4)
    at nsNativeComponentLoader.cpp:136
#3  0x49126232 in nsFactoryEntry::GetFactory (this=0x809f098, 
    aFactory=0xbe5ff3d4, mgr=0x81c9ba8) at nsComponentManager.h:217
#4  0x490bf195 in nsComponentManagerImpl::FindFactory (this=0x81c9ba8, 
    aClass=@0xbe5ff4d4, aFactory=0xbe5ff3d4) at nsComponentManager.cpp:1028

I find that the nsGenericModule ctor for "JS component loader" is called
once in the embedding case, with mComponents is:

debug: edburns: printing argument components for JS component loader
        @mozilla.org/widgets/window/gtk;1
        @mozilla.org/widgets/child_window/gtk;1
        @mozilla.org/widgets/button/gtk;1
        @mozilla.org/widgets/checkbutton/gtk;1
        @mozilla.org/widgets/filewidget/gtk;1
        @mozilla.org/widgets/horizscroll/gtk;1
        @mozilla.org/widgets/vertscroll/gtk;1
        @mozilla.org/widgets/textwidget/gtk;1
        @mozilla.org/widget/appshell/gtk;1
        @mozilla.org/widget/toolkit/gtk;1
        @mozilla.org/widget/lookandfeel/gtk;1
        @mozilla.org/widget/label/gtk;1
        @mozilla.org/sound;1
        @mozilla.org/widget/transferable;1
        @mozilla.org/widget/clipboard;1
        @mozilla.org/widget/htmlformatconverter/gtk;1
        @mozilla.org/widget/fontretrieverservice/gtk;1
        @mozilla.org/widget/dragservice;1
        @mozilla.org/filespecwithui;1

But it's called twice in the mozilla case, with the mComponents the
first time being:

debug: edburns: printing argument components for JS component loader
        @mozilla.org/moz/jsloader;1
        @mozilla.org/moz/jssubscript-loader;1

and the second time being:

debug: edburns: printing argument components for JS component loader
        @mozilla.org/widgets/window/gtk;1
        @mozilla.org/widgets/child_window/gtk;1
        @mozilla.org/widgets/button/gtk;1
        @mozilla.org/widgets/checkbutton/gtk;1
        @mozilla.org/widgets/filewidget/gtk;1
        @mozilla.org/widgets/horizscroll/gtk;1
        @mozilla.org/widgets/vertscroll/gtk;1
        @mozilla.org/widgets/textwidget/gtk;1
        @mozilla.org/widget/appshell/gtk;1
        @mozilla.org/widget/toolkit/gtk;1
        @mozilla.org/widget/lookandfeel/gtk;1
        @mozilla.org/widget/label/gtk;1
        @mozilla.org/sound;1
        @mozilla.org/widget/transferable;1
        @mozilla.org/widget/clipboard;1
        @mozilla.org/widget/htmlformatconverter/gtk;1
        @mozilla.org/widget/fontretrieverservice/gtk;1
        @mozilla.org/widget/dragservice;1
        @mozilla.org/filespecwithui;1

This first time invocation is key to the GetClassObject working.

Comment 1

[edburns]

I have posted this to n.p.m.embedding, and n.p.m.xpcom, and received no 
response.  Thus I'm filing a bug.  

This severely blocks webclient progress.

I've tried to craft a workaround, by calling RegisterComponentLib on all the 
offending components, but have had no luck.

Ray, please help

Comment 2

[edburns]

The inimitable Igor Kushnirsky helped me out with this bug.  It turns out the 
reason for the factory not being found is that webclient links with
 -lwidget_gtk on Unix.  Once I prevented my app from linking with 
-lwidget_gtk I was able to run.  However, I was linking with 
-lwidget_gtk for a reason, so I can call gdk_event_handler_set
(handle_gdk_event, NULL, NULL).  Now the problem remains that I have to ensure 
that I call this function after initing XPCOM.

Comment 3

[Ashu Kulkarni]

We use the gdk_event_handler_set method to specify that all callbacks should be 
handled by the handle_gdk_event function. However, we dont override this 
function anywhere. What happens if we dont make this gdk_event_handler_set 
function call. Maybe the gtk window uses this handle_gdk_event function as 
default for all callbacks. It's worth a try. 

_Ashu

Comment 4

[edburns]

What happens, and I don't know if this was caused by commenting out the 
gdk_set_event_handler call, is that links cannot be traversed.  It's possible 
that this is caused by the lack of a gdk_set_event_handler.

Comment 5

[Igor Kushnirskiy]

> Now the problem remains that I have to ensure 
> that I call this function after initing XPCOM.

From my prospective you can call the function anytime.
Restriction is that no components might be linked with your library.

dlopen and dlsym may be used for getting access to the function of the library.

Comment 6

[Ashu Kulkarni]

Why does linking with lwidget_gtk cause problems in loading the factory? Is 
this a known bug?

Comment 7

[Ray Whitmer]

Reassigning to default xpcom owner

Comment 8

[Doug Turner (:dougt)]

reassigning kandrot bugs.

Comment 9

[Doug Turner (:dougt)]

ed, please reopen if this is still a problem.