[classlib][archive] org.apache.harmony.archive.tests.java.util.jar.ManifestTest fails

• Key: HARMONY-6121
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Major
• Assignee: Tim Ellison
• Reporter: Tim Ellison
• Component/s: Classlib

Description

java.lang.NullPointerException
at java.io.InputStream.read(InputStream.java:148)
at java.io.InputStream.read(InputStream.java:121)
at org.apache.harmony.luni.util.InputStreamExposer.expose(InputStreamExposer.java:118)
at java.util.jar.Manifest.read(Manifest.java:208)
at org.apache.harmony.archive.tests.java.util.jar.ManifestTest.testRead(ManifestTest.java:420)
at java.lang.reflect.VMReflection.invokeMethod(VMReflection.java)

Comments

[Tim Ellison]

Proposed patch for review.

[Nathan Beyer]

Hmm the previous code is a little odd. This looks much better. +1

[Tim Ellison]

New patch (version 2) with more robust line detection.

[Alexei Fedotov]

I have a strong feeling that I have already implemented readFully() method for reading ("wrapping") manifests in HARMONY-4569. The stream was examined to be an instance of ByteArrayStream, and for those streams a fast reflection path worked. Otherwise the stream was read using slower methods.

[1] https://issues.apache.org/jira/secure/attachment/12378468/ByteBuffer.java

[Tim Ellison]

Yes, and that fast reflection path is still there in the InputStreamExposer. I modified it to throw an UnsupportedOperationException when the reflection 'trick' does not work rather than trying to read the entire stream, since the regression testcase passes in a stream that always returns a byte (an "infinite" stream).

The original code caught OutOfMemoryException and converted it to an IO exception, but that did not work on some machines. The new implementation, if it cannot use the fast reflection path, reads a chunk and looks for a CR/CR LF within a reasonable length of data, and throws the IO exception if it does not see one.

[Tim Ellison]

Fixed in ARCHIVE and LUNI modules are r757150.

[Alexei Fedotov]

Tim, thanks for explaining! I was just checking svn annotate to understand what happened. I understand and support you fix.