Test failure in embeddedobj on some Windows systems

[Stephan Bergmann]

On 30/11/2022 11:45, Mike Kaganski wrote:
> On 30.11.2022 13:18, Stephan Bergmann wrote:
>> If anybody has an (explicitly, or implicitly via --enable-debug or 
>> --enable-dbgutil) --enable-sal-log Windows build for which 
>> CppunitTest_embeddedobj_msole succeeds and fully executes that above 
>> testSaveOnThread because it is run with 96 DPI (so that that above "96 
>> DPI" check succeeds):  I would be interested in the resulting 
>> workdir/CppunitTest/embeddedobj_msole.test.log

So (after a bit more offline back and forth:) both of us get into the

>             if (hr == RPC_E_WRONG_THREAD)
>             {
>                 // Assume that the OLE object was loaded on the main thread.
>                 vcl::solarthread::syncExecute([this, &hr, &pDataObject, &aFormat, &aMedium]() {
>                     // Make sure that the current state is embed::EmbedStates::RUNNING.
>                     RunObject();
>                     // Now try again on the correct thread.
>                     hr = pDataObject->GetData(&aFormat, &aMedium);
>                 });
>             }

block in OleComponent::GetExtent 
(embeddedobj/source/msole/olecomponent.cxx) and into its call (done on a 
different thread, by that syncExecute) of

> void OleComponent::RunObject()
> {
>     OSL_ENSURE( m_pNativeImpl->m_pOleObject, "The pointer can not be set to NULL here!" );
>     if ( !m_pNativeImpl->m_pOleObject )
>         throw embed::WrongStateException(); // TODO: the object is in wrong state
> 
>     if ( !OleIsRunning( m_pNativeImpl->m_pOleObject ) )
>     {
>         HRESULT hr = OleRun( m_pNativeImpl->m_pObj );
> 
>         if ( FAILED( hr ) )
>         {
>             if ( hr == REGDB_E_CLASSNOTREG )
>                 throw embed::UnreachableStateException(); // the object server is not installed
>             else
>                 throw io::IOException();
>         }
>     }
> }

(also in embeddedobj/source/msole/olecomponent.cxx).  Where things start 
to differ is that while Mike's run apparently succeeds without throwing 
an exception (so either OleIsRunning(...), or not FAILED(hr)), my run 
goes into the hr == REGDB_E_CLASSNOTREG branch throwing an 
UnreachableStateException (which that syncExecucte then rethrows on the 
original thread).