[OpenJDK 2D-Dev] CR 6802962 test fix for openjdk8 - requesting a sponsor

[valerii.potokov at oracle.com]

Hi Andrew,

I see your concern about shell launcher usage for this case.
I'll re-check how it works under cygwin and will send results how it went.

 > could you please provide some explanations for suggested changes?

Below is my understanding. I am really sorry if some points are well 
known, I just had to include them for a consistency reason.

The fix version is following a workaround that was implemented already 
for jdk5 as a simplest way to resolve a known issue. Otherwise it 
appears as a showstopper for a whole testing session invocation in 
particular for "jdk_awt" target under jtreg (it is not in the 
ProblemList.txt file at present).

The jtreg invokes the own main method and the codebase (not the codebase 
of the test class file) when the test case is executed under jtreg.jar 
(that includes the underlying javatest.jar libraries invocation also).

As a result, the system class loader is initialized at a very early 
point in the startup sequence;
it copies the classpath into its own data structures, and the classpath 
property is not read again.

A wrapper file is a solution that is used for jtreg based invocations 
for the tests with a "sensitive" functionality (the security manager in 
this case). It separates a "system/harness" level invocation from an 
actual test functions and implement a sort of "emulation" of a manual 
*class file invocation.

A convenience of such approach is - we have a shell wrapper template 
that serves such "separation" in a transparent way and it's a part of 
"The JDK Test Framework: Tag Language Specification". It also supports 
all jtreg/javatest tags in a unified way. We formally can use the java 
file wrapper script (instead of the shell ) for the tests like this one 
and avoid a potential necessity to adjust it for the cygwin 
particularity. A wrapper script is the only solution that is currently 
seen.

But wouldn't be it better if the "java file wrippers usage" be unified 
(on the level of The JDK Test Framework specification) in the same way 
as currently the shell script wrapper is. Otherwise the shell scripts 
wrappers are still used (when they are required) for most for the others 
test cases in WS anyway.

Please suggest if you see a necessity to replace the shell script by the 
java wrapper for the particular test case. Also if I have missed some 
points. I'll take care about this. As I mentioned earlier, I'll re-check 
how the current version works on windows under cygwin and let you know 
if any adjustments is required.

Thanks for looking in this,
Valerii


On 01/10/12 01:35 PM, Andrew Brygin wrote:
> Hello Valerii,
>
>  could you please provide some explanations for suggested changes?
>
>  The main concern regarding suggested shell launcher is that a list of
>  supported OS names is too limited. In particular, it does not include
>  Cygwin* which likely makes this test failed on windows under cygwin
>  environment (which is default for openjdk). Could you please make sure
>  that modified test can be executed under cygwin?
>
> Thanks,
> Andrew
>
> On 10.01.2012 15:36, valerii.potokov at oracle.com wrote:
>> Hi All,
>>
>> I've attached a patch for 2d-dev category test fix.
>> The fix is related to openjdk8 (also to openjdk7).
>>
>> test/java/awt/FontClass/CreateFont/fileaccess/FontFile.java
>>
>> The failure is a showstopper, it causes a hang of test execution when 
>> launched using "make jdk_awt".
>>
>> The reason of failure is known and has been fixed already in jdk6.
>>
>> It has a corresponding CR 6802962 in the Bugster.
>>
>> I am requesting a sponsor to review and to integrate the fix.
>>
>> Regards,
>> Valerii
>