Re: [PATCH] [JAVA] Double.parseDouble(null) throw NullPointerException

[Andrew Haley <aph@redhat.com>]

On 08/07/2011 12:00 PM, Jie Liu wrote:

> When I use gcj on an RTOS(RTEMS), Double.parseDouble(null) throw
> NumberFormatException, but it should throw NullPointerException. So I
> add the patch below:
> 
> Index: natVMDouble.cc
> ===================================================================
> --- natVMDouble.cc	(revision 172224)
> +++ natVMDouble.cc	(working copy)
> @@ -19,6 +19,7 @@
>  #include <java/lang/VMDouble.h>
>  #include <java/lang/Character.h>
>  #include <java/lang/NumberFormatException.h>
> +#include <java/lang/NullPointerException.h>
>  #include <jvm.h>
> 
>  #include <stdio.h>
> @@ -162,6 +163,9 @@
>  jdouble
>  java::lang::VMDouble::parseDouble(jstring str)
>  {
> +  if(str == NULL)
> +      throw new NullPointerException();
> +
>    int length = str->length();
> 
>    while (length > 0
> 
> The testsuite/Throw_2.java has been PASS after this patch. what do you
> think about this patch?

This patch is not OK, and must not go in.

Look at the comment at the top of Throw_2:

// Check that NullPointerExceptions thrown from library code are
// caught.  This detects a number of failures that can be caused by
// libgcj being built incorrectly.  In particular, we ensure that a
// SEGV in native (i.e. C++) code in libgcj is handled correctly.

// Regrettably, we cannot guarantee that Double.parseDouble() will
// always be native code, or that it will never be inlined.  It could
// be argued that we should add a method to libgcj that will be
// guaranteed forever to be native, but I'm reluctant to add to the
// library for the sole purpose of performing this test.

Andrew.