Hello,
While you discuss TWR hazards, be aware that closing a sink (OutputStream)
as part of a "finally" branch which hides exceptions is dangerous in some
I/O scenarios because hardware write errors could be delayed until the
close. NFS for example has this tendency, but also some virtual
filesystems do a lot of commit work on close. So typically you add a close
inside the try as well.
Greetings
Bernd
Am 10.10.2013, 04:36 Uhr, schrieb Joseph Darcy <joe.da...@oracle.com>:
It is a hazard (I thought I had published a blog entry on this very
tropic, but apparently not). The most robust pattern is
try(OriginalResource r1 = new OriginalResource;
WrappingResource r2 = new WrappingResource(r1);
AnnotherWrappingResource r3 = new WrappingResource(r2)) { ...}
One thing to watch out for in this pattern is a non-idempotent close.
Calling close on r3 will presumably propagate a close call to r2, and
then r2 to r1. So give the desguaring the try-with-resource and the
expected behavior of the wrapping in a normal termination situation,
close on r3 gets called once, close on r2 gets called twice (first from
the close on r3, second close from try-with-resources), and close on r1
gets called three times.
HTH,
-Joe
--
http://bernd.eckenfels.net
