[rfc][icedtea-web] Fix for PR1592 and jnlp applets/applications

[Jiri Vanek]

On 12/12/2013 10:34 PM, Andrew Azores wrote:
> On 12/11/2013 10:57 AM, Jiri Vanek wrote:
>> On 12/09/2013 05:25 PM, Andrew Azores wrote:
>>> Hi,
>>>
>>> The fix for PR1592 currently only affects plugin applets. The attached patch fixes this so that it
>>> also enables the same functionality when launching from JNLP, as well as the fix for the reproducer
>>> to expect this behaviour.
>>>
>>> Thanks,
>>>
>>
>>
>> I forgot to mention - the tests will need a bit of update:
>>
>> You should verify various states of requested permissions.
>> probably just:
>>
>> a)signed x unsigned x requesting permissions
>> b)signed x unsigned x not requesting permissions
>> a x signed doing something "forbidden" , unsigned doing something "forbidden"
>> b x signed doing something "forbidden" , unsigned doing something "forbidden"
>> a x signed doing something not "forbidden" , unsigned doing something not "forbidden"
>> b x signed doing something not "forbidden" , unsigned doing something not "forbidden"
>>
>> and the ways when call is via second jar, and calls via reflection.
>>
>> It should be just multiplication of your lready existing usecases.
>>
>> Thanx,
>> j.
>>
>>
>
> I'm not 100% on what you mean, but I've added new tests that I think might be doing somewhat what you want?
>
> Also, I guess I'm doing something wrong or just missing something somewhere. I tried to pass parameters to the JNLP applets through ServerAccess.executeJavawsHeadless(List<String>, String, String[]), but I kept getting runtime java.awt.GraphicsEnvironment.checkHeadless exceptions when I did this. I need to pass parameters to the applet to specify which specific test to run (which the plugin applet is doing), so unfortunately right now the JNLP tests are all specified by each having their own JNLP file, with only the <param> tag different :( is there a better way around this?

No, you are not doing anything wrong.  I run into this  issue recently too, although under different conditions.

My suspects are , that some exception is thrown somewhere, is catch, goes to logs *to early* and it cause some unpredicted behaviour.
Another  suspicion is wrong loading of config (maybe whole jnlp runtime). It really have to change to regular singleton :-/
Another candidate is my DirectoryValidator patch

You may notice that in  "unittests warning cleanup" I added small protection into centring of dialogues. Now they work even in headles system. The need was caused ( I guess) by your patch "create, center, show". (from previous "create show center" - here the center was done only if system was not headless)

I think I will push it today, so you may retry?
>
> Next, right now every test case is using reflection for the various method calls between the JARs. Since there are two different reproducers built separately, is there a way I can use normal method calls for these tests? Maybe some special make flag voodoo I don't know about to add one of the reproducers to the other's compile classpath? Or might I need to convert this whole test into a custom reproducer with its own Makefile to achieve this? Even if I use a custom reproducer, might I not still run into the same problem I had before, with the signing information being different across a single package?
>
> The "main" attached patch is the one that matters more. The "resources" one is just all the changed JNLP files.

review later :)
>
> Thanks,
>

Thanx on my side!
   J.