jtreg failing in awt tests

[Jonathan Gibbons]

At least in part, it's related to the underlying JavaTest harness.

It would be good to improve things in this area.

There are a bunch of options which are now "strange defaults", but I 
don't know how much problems we would cause to change. For example,

1. othervm is now the default, but I recommend agentvm
2. all tests is the default but I would recemmend most folk use -a (for 
automatic tests only)
3. report generation.

Maybe as jdk 10 gets underway, that would be a good time to reconsider 
some of these choices.

-- Jon


On 11/23/2016 06:44 PM, Martin Buchholz wrote:
> Thank you!  My own jtreg running infrastructure now uses -noreport as 
> the default.
>
> It's a little surprising that report generation is a global operation 
> while test running is "local", although it's understandable because 
> jtreg wants to report all tests NOT run as well.  I don't see this 
> explained anywhere in the docs. The 10 second tax for the jdk repo is 
> just small enough that everyone just lives with it, not knowing about 
> -noreport.
>
> I'm guessing most people would be happier if -noreport was the default.
>
> On Wed, Nov 23, 2016 at 1:42 PM, Jonathan Gibbons 
> <jonathan.gibbons at oracle.com <mailto:jonathan.gibbons at oracle.com>> wrote:
>
>     Actually, jtreg is (has always been) optimised to *run* tests
>     efficiently, with no time tax.
>     But yes, there is a time tax, which comes from writing the report
>     at the end of the test run, and that is actually where the clash
>     is (probably) being detected.
>
>     And, my guess is that a developer running "jtreg MyTest.java"
>     doesn't need/want/care about the report. If so, there are two ways
>     to disable the report:
>
>     1. Options:  -noreport  (or -nr for short)
>     2. System property: javatest.noReportRequired
>
>     -- Jon
>
>     On 11/23/2016 01:19 PM, Martin Buchholz wrote:
>>     Jonathan: Here's a small jtreg feature request:
>>     It's nice for release engineers to have jtreg check the whole
>>     test/ tree for correct test definitions.  But for developers who
>>     are just doing
>>
>>     jtreg MyTest.java
>>
>>     they're not interested in awt failures, and don't want to pay the
>>     10-second tax to check every test definition for every jtreg
>>     invocation.
>>
>>     On Wed, Nov 23, 2016 at 12:57 PM, Jonathan Gibbons
>>     <jonathan.gibbons at oracle.com
>>     <mailto:jonathan.gibbons at oracle.com>> wrote:
>>
>>
>>
>>         On 11/23/2016 12:47 PM, Martin Buchholz wrote:
>>
>>             Am I the only one seeing jtreg test failures in latest
>>             jdk9/dev, apparently due to
>>
>>             https://bugs.openjdk.java.net/browse/JDK-8160766
>>             <https://bugs.openjdk.java.net/browse/JDK-8160766>
>>
>>             Error: Test clashes with another test with a similar name:
>>             .../jdk/test/java/awt/Focus/DisposedWindow/DisposeDialogNotActivateOwnerTest/DisposeDialogNotActivateOwnerTest.java
>>              .../jdk/test/java/awt/Focus/DisposedWindow/DisposeDialogNotActivateOwnerTest/DisposeDialogNotActivateOwnerTest.html
>>
>>             (even though I'm not actually running any of the awt tests?)
>>
>>
>>
>>         Martin,
>>
>>         It looks like this is a test bug, introduced in this changeset.
>>
>>         changeset:   16112:88faebbdbf9b
>>         user:        arapte
>>         date:        Fri Nov 04 21:55:19 2016 +0530
>>         summary:     8160766: [TEST_BUG] java/awt/Focus/DisposedWindow
>>
>>
>>         The problem edit looks like this:
>>
>>
>>          /*
>>         -  test
>>         -  @bug       6386592
>>         -  @summary   Tests that disposing a dialog doesn't activate
>>         its invisible owner.
>>         -  @author anton.tarasov at sun.com
>>         <mailto:anton.tarasov at sun.com>: area=awt.focus
>>         -  @run       applet DisposeDialogNotActivateOwnerTest.html
>>         +  @test
>>         +  @key headful
>>         +  @bug 6386592 8160766
>>         +  @summary Tests that disposing a dialog doesn't activate
>>         its invisible owner.
>>          */
>>
>>         Note that plain "test" was edited to "@test", meaning that
>>         the file previously was not a standalone jtreg test, and it
>>         was changed to be one.  This causes a clash with an HTML test
>>         of the same base name.
>>
>>         jtreg correctly reports the clash, because both tests would
>>         lead to the same .jtr file.  i.e. you can't have two tests
>>         that only differ in their filename extension.
>>
>>         -- Jon
>>
>>
>
>