RFR JDK-8147539: (zipfs) ZipPath should throw ProviderMismatchException when invoking register()

[Alexandre (Shura) Iline]

> On May 25, 2016, at 11:31 AM, Xueming Shen <xueming.shen at oracle.com> wrote:
> 
> On 5/25/16, 11:06 AM, Alexandre (Shura) Iline wrote:
>> Sherman,
>> 
>> Not declaring any module dependencies in test is equivalent to declaring that the test only depends on java.base. In such case, the test will be picked up for execution by JTReg, no matter what modules are available. One way to test such behavior would be to supply -javaoptions:”-limitmods java.base” in the JTReg command line. The tests will be selected for execution, but would they work?
>> 
>> I have taken another look on the code now, it looks like with the absence of jdk.zipfs test/jdk/nio/zipfs/Basic.java test will fail. I think it would be good to skip this test altogether if there is no jdk.zipfs available.
> 
> Hi Shura,
> 
> Yes, the test will fail if the jdk.zipfs is absence. With the assumption that zipfs module should
> be available to the jdk images/runtime by default. If it's not there/available, something is wrong

Well, not necessarily. 

Since JTReg is taking on the burden to select tests for execution based on the available modules, be it -limitmods option or a custom built limited JDK image, the test suite will be used to execute tests for compact profiles, for example. We will be relying on JTReg to select proper tests. The plan is to get rid of the test groups for compact profiles when the test markup is complete.

Shura

> and we would want to know why. So the intention here is not to skip the zipfs test silently, if
> the zipfs is missing from the jdk image. That said, arguably these existing tests are the unit tests
> for the zipfs functionality, it might be desired to have separate test(s) for testing the availability
> of the zipfs...
> 
> sherman
> 
>> 
>> Shura
>> 
>>> On May 25, 2016, at 10:49 AM, Xueming Shen<xueming.shen at oracle.com>  wrote:
>>> 
>>> 
>>> Should it? My understanding is that those tests don't use zipfs directly
>>> from zipfs module, it access it via java.base's FileSystemProvider interface.
>>> It depends on the jdk runtime to pick up the zipfs "provider" to function. So
>>> if the jdk runtime fails to pick up the zipfs module for whatever reason,
>>> that's something we want to detect as well?
>>> 
>>> -Sherman
>>> 
>>> On 05/25/2016 10:18 AM, Alexandre (Shura) Iline wrote:
>>>> Hi.
>>>> 
>>>> Should the tests also declare “@modules jdk.zipfs”?
>>>> 
>>>> Shura
>>>> 
>>>>> On May 25, 2016, at 9:39 AM, Xueming Shen<xueming.shen at oracle.com>   wrote:
>>>>> 
>>>>> Hi,
>>>>> 
>>>>> Please help review the changes for the following zipfs related bug fixes
>>>>> 
>>>>> JDK-8147539: (zipfs) ZipPath should throw ProviderMismatchException when invoking register()
>>>>> JDK-8153248: (zipfs) ZipPath#getFileName( ) returns inconsistent result
>>>>> JDK-8146754: (zipfs) ZipPath.relativize() returns wrong result for path ending with slash /
>>>>> JDK-8139956: (zipfs) ZipPath does not produce correct empty path on relativize()
>>>>> 
>>>>> issue:
>>>>> https://bugs.openjdk.java.net/browse/JDK-8147539
>>>>> https://bugs.openjdk.java.net/browse/JDK-8153248
>>>>> https://bugs.openjdk.java.net/browse/JDK-8146754
>>>>> https://bugs.openjdk.java.net/browse/JDK-8139956
>>>>> 
>>>>> webrev:
>>>>> http://cr.openjdk.java.net/~sherman/8139956_8146754_8147539_8153248/webrev/
>>>>> 
>>>>> The changes are relative easy.
>>>>> (1) ZipPath.initOffsets() should deal with empty path specially, as the UnixPath impl does
>>>>> (2) ZipPath.normalize(byte[]) should take care of tailing "/" (it is taken care of at
>>>>>      ZipPath.normalize(byte[], int), but is missed in the "main" method)
>>>>> (3) ZipPath.register() should throw PME as suggested.
>>>>> 
>>>>> Thanks,
>>>>> Sherman
>