Yeah, I guess so. We'd have to add @SuppressWarnings("serial") to some
non-public classes but so what.
Éamonn
On 25/8/11 6:01 PM, Joe Darcy wrote:
Eamonn McManus wrote:
Could we perhaps have the JDK build use an annotation processor that
requires every public class in java.* javax.* etc that is
serializable to either declare a serialVersionUID or have
@SuppressWarnings("serial")? The amount of grief we have had over the
years with this would easily justify the effort.
We could achieve that effect using the "-Xlint:serial -Werror" option
to javac during the JDK build.
-Joe
Éamonn
On 25/8/11 9:00 AM, Joe Darcy wrote:
Alan Bateman wrote:
Joe Darcy wrote:
Hi Alan.
I did check for that point in my review :-)
The VirtualMachineError class is abstract so as long as all its
subclasses declare a serialVersionUID, like InternalError does, I
think we're fine.
We need to check the serialization protocol but I'm pretty the
object stream class of the supertype goes into the stream too. Also
I quickly checked an an up-to-date build of jdk8/tl and its
currently unable to deserialize streams contain any of the virtual
machine errors like OutOfMemoryError, StackOverflowError, etc.
*sigh*
Thanks for checking on this Alan; I'll prepare a changeset with the
explicit serialVersionUID on InternalError.
-Joe
