Re: [Quickfix-developers] fatal error in java api.

[Joerg T. <Joe...@ma...>]

Hi Andrew,

> Hello.. This bit of code where I set the null value...
> 
> 	quickfix.fix41.OrderCancelRequest message =
>  	 new quickfix.fix41.OrderCancelRequest
>  	 (new OrigClOrdID("1"),
>  	 new ClOrdID("2"),
>  	 new Symbol("MSFT"),
>  	 new Side(Side.BUY));
> 
> 	message.setString(100, null);
> 
> causes this..
> 
> Unexpected Signal : EXCEPTION_ACCESS_VIOLATION (0xc0000005) occurred at 

Actually, the JNI layer code should check for NULL values and throw a 
java.lang.NullPointerException. This a bit more polite than crashing 
with a Segmentation Violation, but anyway: you should not set a field to 
"null." What result would you expect?

Looking at src/java/Conversions.h:

     136 inline void setString( FIX::FieldMap& map, jint field, jstring 
value )
     137 {
     138   const char* uvalue = ENV::get()->GetStringUTFChars( value, 0 );
     139   map.setField( field, uvalue );
     140   ENV::get()->ReleaseStringUTFChars( value, uvalue );
     141 }

Between lines 138 and 139 a check whether uvalue is NULL should be inserted:

    138a   if ( NULL == uvalue ) {
    138b        return;
    138c   }

In this case, GetStringUTFChars() has already thrown an exception, so it 
is not necessary to throw a new one. This is the default idiom in JNI. I 
still have to check whether this also cares for java null values, i.e. 
whether GetStringsUTF() also checks for a jstring null value.

I submitted a bug report on the QuickFIX bugtracker:

http://www.quickfixengine.org/bugtracker/bug.php?op=show&bugid=27

Cheers, Jörg

-- 
Joerg Thoennes
                                http://macd.com
Tel.: +49 (0)241 44597-24      Macdonald Associates GmbH
Fax : +49 (0)241 44597-10      Lothringer Str. 52, D-52070 Aachen