Any JNI/UNO bridge experts around? Some assumptions about field ids are apparently not valid for Dalvik

[Noel Grandin]

Well, that assumption would hold for any of the Sun/IBM derived VM's, 
but I can certainly see how Dalvik might be doing something different in 
that case.

It might be safer and simpler to add a getContext() method to 
com.sun.star.uno.Exception and then look that up and access it.
That has a higher chance of using the same methodID for both classes.

On 2012-03-23 14:44, Tor Lillqvist wrote:
> It seems that the code assumes the field id for the "Context" field in
> com.sun.star.uno.Exception is also valid for that field in
> com.sun.star.uno.RuntimeException. This apparently holds for "normal"
> JVMs, but not Dalvik? I am still digging into what is actually going
> on, need to write some minimal test programs, but just wanted to ask
> about this already now in case somebody immediately can think of
> something like "yeah, I remember back in 2001 when we discussed
> whether that is safe" ;)
>
> I added some debugging output to the JNI_compound_type_info
> constructor in bridges/sources/jni_uno/jni_info.cxx, where it calls
> GetFieldId for the Exception and RuntimeException types to see what
> the returned field ids are:
>
> diff --git a/bridges/source/jni_uno/jni_info.cxx
> b/bridges/source/jni_uno/jni_info.cxx
> index a5ed28a..8e15979 100644
> --- a/bridges/source/jni_uno/jni_info.cxx
> +++ b/bridges/source/jni_uno/jni_info.cxx
> @@ -302,2 +302,3 @@ JNI_compound_type_info::JNI_compound_type_info(
>                   (jclass) jo_class.get(), "Context", "Ljava/lang/Object;" );
> +            fprintf(stderr, "%s:%d %s: m_fields=[%p,%p]\n",
> __FUNCTION__, __LINE__, OUStringToOString(uno_name,
> RTL_TEXTENCODING_ASCII_US).getStr(), m_fields[0], m_fields[1]);
>               jni.ensure_no_exception();
> @@ -311,2 +312,3 @@ JNI_compound_type_info::JNI_compound_type_info(
>
> +            fprintf(stderr, "%s:%d %s: %d fields\n", __FUNCTION__,
> __LINE__, OUStringToOString(uno_name,
> RTL_TEXTENCODING_ASCII_US).getStr(), (int)nMembers);
>               for ( sal_Int32 nPos = 0; nPos<  nMembers; ++nPos )
> @@ -337,2 +339,3 @@ JNI_compound_type_info::JNI_compound_type_info(
>                   jni.ensure_no_exception();
> +                fprintf(stderr, "  %s: %p\n", member_name.getStr(),
> m_fields[nPos]);
>                   OSL_ASSERT( 0 != m_fields[ nPos ] );
>
> and get this output:
>
> JNI_compound_type_info:303 com.sun.star.uno.Exception: m_fields=[0x0,0x44e29a68]
> JNI_compound_type_info:303 com.sun.star.uno.RuntimeException:
> m_fields=[0x0,0x44e26c68]
>
> Then shortly after that I get a crash:
>
> JNI WARNING: instance fieldID 0x44e29a68 not valid for class
> Lcom/sun/star/uno/RuntimeException;
>
> Note that the fieldID here is the one printed as being for the
> Exception type, while the warning message says it is used for
> RuntimeExcpetion.
>
> (Yeah, despite claiming that is a "warning", Dalvik then throw its
> hands up and quits:
> http://androidxref.com/source/xref/dalvik/vm/CheckJni.cpp#364 )
>
> Bells ringing, anyone?
>
> Firstlty, I guess I need to quickly check what that debugging output
> says with a "normal" JVM, whether the field ids are
>
> --tml
> _______________________________________________
> LibreOffice mailing list
> LibreOffice at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/libreoffice
>

Disclaimer: http://www.peralex.com/disclaimer.html