PATCH [6901992] : Possible InvalidJarIndexException due to bug in sun.misc.JarIndex.merge()

[Diego Belfer]

Hi Chris,

There is no way to generate a jar without directory entries using the jar
tool; there is no option for that. What do you think is the best way to
handle this ?
I don't want to re-implement the logic for creating a jar using the
JarOutputStream class.

Do you think it is possible to add a new option to the Jar tool Main class
to exclude directory entries? The option does not need to be exposed by the
command line tool to final users if this an issue, although I think it may
be useful for them too.

Best,
Diego


On Wed, Jun 13, 2012 at 7:12 PM, Diego Belfer <dbelfer at gmail.com> wrote:

> Chris,
>
> I was thinking of something similar. I will create the jars and their
> contents using Java code. There is no need of creating real class files,
> using a few resource files and some directories will be enough.
>
> Best,
> Diego
>
>
> On Wed, Jun 13, 2012 at 6:46 PM, chris hegarty <chris.hegarty at oracle.com>wrote:
>
>> Diego,
>>
>> I'm thinking that some of the trivial source files, to compile and built
>> into the jars, could be simply created and written by the test itself,
>> rather than checking them all in. If this makes it cleaner. I really don't
>> like all the file in test/sun/misc/JarIndex/**metaInfFilenames, but at
>> least it is quite understandable.
>>
>> -Chris.
>>
>>
>> On 13/06/2012 20:36, Diego Belfer wrote:
>>
>>> Hi Chris,
>>>
>>> That's right. The only non-cleanup change is the one in the merge.
>>>
>>> Regarding the test case,  I will re-write them in order to generate the
>>> jars on fly. I'd scanned the jdk/test folder and found a few jars,
>>> that's why I included them.  I have seen your test case, I will use it
>>> as a sample.
>>>
>>> I had not seen your comment in the bug report. Maybe there are other
>>> cases which trigger the InvalidJarIndexException, but, as far as I could
>>> see, the validIndex method checks that at least one entry of the jar
>>> matches the target path found in the index. If directory entries are not
>>> present in the jar, stripped paths generated during the merge and used
>>> by the index will return jars which may not contain entries for them,
>>> triggering the exception.
>>> When all directory entries are present, if a jar contains an entry for
>>> "xxx/yyy/resource.file", it  will contain entries for "xxx", "xxx/yyy"
>>> and "xxx/yyy/resource.file".
>>>
>>>
>>> Best,
>>> Diego
>>>
>>>
>>> On Wed, Jun 13, 2012 at 12:05 PM, Chris Hegarty
>>> <chris.hegarty at oracle.com <mailto:chris.hegarty at oracle.**com<chris.hegarty at oracle.com>>>
>>> wrote:
>>>
>>>    Hi Diego,
>>>
>>>    Thanks for picking up this bug.
>>>
>>>    I think your changes look fine. Mainly cleanup except for add ->
>>>    addExplicit/addMapping in merge, right? BTW the cleanup makes this
>>>    more readable.
>>>
>>>    Unfortunately, the tests you created require checking in a binary
>>>    jar file. This is a real no no for the OpenJDK, we really need to
>>>    create these jars on the fly. I did similar for
>>>    test/sun/misc/JarIndex/__**metaInfFilenames/, but I really wish I
>>>
>>>    generated the source files for these tests rather than checking in
>>>    so many pointless files.
>>>
>>>    I can look at helping with writing suitable tests for this.
>>>
>>>
>>>     > That's because I was using jars containing "directory entries"
>>>     > (I was unaware that jar files may not include them)
>>>
>>>    Strangely I added the comment "Remove directories from jar files
>>>    being indexed." to the workaround section of the bug. You seem to be
>>>    seeing the opposite, right?
>>>
>>>    -Chris.
>>>
>>>
>>>
>>>    On 13/06/2012 06:11, Diego Belfer wrote:
>>>
>>>        Hi,
>>>
>>>        I have finally reproduced the InvalidJarIndexException bug as
>>>        reported in
>>>        the ticket. I mentioned in a previous email, that the only way
>>>        I'd found
>>>        for getting the error was to use an invalid index file
>>>        (INDEX.LIST), which
>>>        did not have any sense. That's because I was using jars containing
>>>        "directory entries" (I was unaware that jar files may not
>>>        include them)
>>>
>>>        After reviewing the URLClasspath$JarLoader class and the
>>>        validIndex method,
>>>        I notice it is possible to get the exception for a Jar file
>>>        which does not
>>>        include directory entries. In order to trigger the issue, the
>>>        Jar must be
>>>        referenced by an intermediary INDEX.LIST and the intermediary
>>>        Jar index
>>>        should have been merged to its parent index. Although, jar tool
>>>        includes
>>>        directory entries in the generated jar files, Eclipse default
>>>        option for
>>>        exporting jars does not include them (AFAIK), so this might be
>>>        quite common.
>>>
>>>        I have created a new PATCH which includes an additional test
>>>        case which
>>>        uses the URLClassLoader to trigger the InvalidIndexException.
>>>
>>>        The patch is attached, please consider it for review.
>>>
>>>        Best,
>>>        Diego Belfer [muralx]
>>>
>>>
>>>
>>>        On Mon, Jun 11, 2012 at 4:47 PM, Diego Belfer<dbelfer at gmail.com
>>>        <mailto:dbelfer at gmail.com>>  wrote:
>>>
>>>
>>>            Hi,
>>>
>>>            Here is a patch that fixes the merge method of the JarIndex.
>>>            This bug was
>>>            reported as the cause of the bug 6901992. Although, I was
>>>            not able to
>>>            reproduce the BUG itself (InvalidJarIndexException), I did
>>>            verified that
>>>            the method had a bug, and resources/classes where not found
>>>            in a jarIndex
>>>            with merged contents.
>>>
>>>            If you think it is possible to commit this fix without
>>> actually
>>>            reproducing the original bug report, please consider this
>>>            patch for review.
>>>
>>>            Thanks,
>>>            Diego Belfer [muralx]
>>>
>>>
>>>
>>>
>