The created mr jar file is being tested in #139 with versionedStream(). Without 
the fix there will be a npe.

On Oct 6, 2018, at 12:17 AM, Alan Bateman <[email protected]> wrote:

> On 05/10/2018 17:48, Xueming Shen wrote:
> Hi
> 
> Please help review the change for JDK-8211728
> 
> issue: https://bugs.openjdk.java.net/browse/JDK-8211728
> webrev: http://cr.openjdk.java.net/~sherman/8211728/webrev
> 
> The "intention" of JEP 238 is that the resources under META-INF 
> can't/shouldn't
> be versioned. But it appears this is not documented & specified 
> clearly/explicitly.
> And the JarFile implementation has loophole dealing with this case. The spec
> clarification is being traced/addressed separately via JDK-8211758. The change
> proposed here is to fix the JarFile implementation.
> 
> Note: the direct trigger of the bug is that versionedStream() does not expect 
> the
> possibility VersinedEntry() might return null when the entry is actually under
> META-INF/versions/{n}/META-INF, in which it has a "basename" of META-INF/xxx
> but doesn't have a real entry under /META-INF (in which the 
> getVersionEntry(...)
> is invoked with (name="META-INF/xxx", je=null).
The update to versionedStream looks okay but the test seems to be checking 
JarFile::stream rather than versionedStream. For the bug we need the stream to 
end without NPE and we also need to check that there isn't an element for 
META-INF/Foo.class (because this is not in the base section).

-Alan

Reply via email to