David, Jason,
Thank you for your comments and suggestions. They all were taken in
account and as a result - the new webrev:
http://cr.openjdk.java.net/~dmeetry/8009581/webrev.2/
On 05/24/2013 08:26 AM, David Holmes wrote:
On 22/05/2013 2:51 AM, Jason Mehrens wrote:
Aleksej,
Actually, the readObject calls the super.initCause, because there is no
initCause in XPathException.
I would think that subclasses of XPE will see calls to this.initCause
from readObject. That wouldn't have happened prior to this change.
I think this is why super.initCause() (or super.getCause()) must be
called here. If you invoke this.xxx() then you may well be invoking a
subclass specialization and you don't know what it will do.
Agree with that, we want to call the specific 'initCause', not the
subclass specialization.
About 'super.getCause() == null' check: yes it can be done in such way.
In current version I caught the IllegalStateException to correctly
process the situation when the cause was already initialized.
I think you'll always have to catch ISE. If super.getCause is not
null you know initCause will fail. I would think it would be cheaper
to null check than to fillStackTrace. But, I haven't tested that.
I think readObject only needs to account for deserializing an older
version of the exception which will have a non-null local cause, but
Throwable.cause is null. That should be rare. The common case would be
deserializing the new form, in which case initCause would fail. So I
agree with Jason that checking super.getCause() is more efficient than
always calling initCause.
Also agree, the ' super.getCause() == null' check added to readObject.
David
Jason
Aleksej
