gnu.gcj.RawData vs. long

gnu.gcj.RawData vs. long

[Craig Vanderborgh]

Hello all:

I am working to make some of our code more portable.  We "gcj-ified"
our source several years back, including making liberal use of
gnu.gcj.RawData * class members to store pointers for native (CNI)
code within object instances.

The problem is that gnu.gcj.RawData * is not portable.  The question
is: can I use "long" for the type instead without garbage collection
problems?  Can my native code store the pointer value in a Java "long"
in the CNI with impunity, or will Boehm GC think these pointers needs
to be considered for collection?

Please advise.

Thanks in advance,
Craig Vanderborgh
Voxware Incorporated

Re: gnu.gcj.RawData vs. long

[Bryce McKinlay]

On Mon, Jul 18, 2011 at 5:08 PM, Craig Vanderborgh
<craigvanderborgh@gmail.com> wrote:

> The problem is that gnu.gcj.RawData * is not portable.  The question
> is: can I use "long" for the type instead without garbage collection
> problems?  Can my native code store the pointer value in a Java "long"
> in the CNI with impunity, or will Boehm GC think these pointers needs
> to be considered for collection?

Hi Craig.

You must use gnu.gcj.RawData* if the reference being stored points to
something allocated by the GC. The GC will not mark references held in
non-pointer types.

If it's something that was allocated outside the GC (ie, with malloc),
it should be fine to use a long.

Bryce