Neil Richards wrote:
Hi,
I've noticed that the fix introduced to address bug 6735255 [1] [2] had
an unfortunate side-effect.
That fix stores references to the InputStream objects returned from
ZipFile.getInputStream(ZipEntry) in a HashSet (within ZipFile), so that
the close() method may be called upon those objects when ZipFile.close()
is called.
It does this to conform to the existing API description, and to avoid a
native memory leak which would occur if the InputStream is GC'd without
its close() being called.
However, by holding these InputStreams in a set within their ZipFile
object, their lifecycle is now prolonged until the ZipFile object either
has its close() method called, or is finalized (prior to GC).
I've created 7031076 to track this and generated a webrev from the
change-set that you attached:
http://cr.openjdk.java.net/~alanb/7031076/webrev/
Sherman mentioned off-list that he plans to look at this.
-Alan.
