RFR(S): 8049071: Add jtreg jobs to JPRT for Hotspot

[Mikael Vidstedt]

Thanks Igor, appreciate it!

Cheers,
Mikael

> On Jul 12, 2014, at 11:41, Igor Veresov <igor.veresov at oracle.com> wrote:
> 
> Looks fine..
> 
> igor
> 
>> On Jul 10, 2014, at 1:06 PM, Mikael Vidstedt <mikael.vidstedt at oracle.com> wrote:
>> 
>> 
>> Anybody? Pleeeease?
>> 
>> Cheers,
>> Mikael
>> 
>>> On 2014-07-07 16:12, Mikael Vidstedt wrote:
>>> 
>>> Fixed the comment, removed the loop (the loop logic is btw taken directly from jdk/test/Makefile, but I'll follow up on a fix for that separately).
>>> 
>>> Anybody else want to have a look?
>>> 
>>> top: http://cr.openjdk.java.net/~mikael/webrevs/8049071/webrev.01/top/webrev/
>>> hotspot: http://cr.openjdk.java.net/~mikael/webrevs/8049071/webrev.01/hotspot/webrev/
>>> 
>>> Thanks,
>>> Mikael
>>> 
>>>> On 2014-07-03 22:29, David Holmes wrote:
>>>>> On 4/07/2014 2:46 PM, David Holmes wrote:
>>>>> Hi Mikael,
>>>>> 
>>>>> Generally looks okay - took me a minute to remember that jtreg groups
>>>>> combine as set unions :)
>>>>> 
>>>>> A couple of things:
>>>>> 
>>>>> 226 # Unless explicitly defined below, hotspot_<x> is interpreted as the
>>>>> jtreg test group <x>
>>>>> 
>>>>> The jtreg group is actually called hotspot_<x>
>>>>> 
>>>>> 227 hotspot_%:
>>>>> 228         $(ECHO) "Running tests: $@"
>>>>> 229         for each in $@; do \
>>>>> 230                 $(MAKE) -j 1 TEST_SELECTION=":$$each"
>>>>> UNIQUE_DIR=$$each jtreg_tests; \
>>>>> 231         done
>>>>> 
>>>>> While hotspot_% can match multiple targets each target will be distinct
>>>>> - ie $@ will only every have a single value and the for loop will only
>>>>> execute once - and hence is unnecessary. This seems borne out with a
>>>>> simple test:
>>>>> 
>>>>>> cat Makefile
>>>>> hotspot_%:
>>>>>        @echo "Running tests: $@"
>>>>>        @for each in $@; do \
>>>>>          echo $$each ;\
>>>>>        done
>>>>> 
>>>>>> make hotspot_a hotspot_b
>>>>> Running tests: hotspot_a
>>>>> hotspot_a
>>>>> Running tests: hotspot_b
>>>>> hotspot_b
>>>> 
>>>> Though if you have a quoting issue with the invocation:
>>>> 
>>>>> make "hotspot_a hotspot_b"
>>>> Running tests: hotspot_a hotspot_b
>>>> hotspot_a
>>>> hotspot_b
>>>> 
>>>> things turn out different.
>>>> 
>>>> David
>>>> 
>>>> 
>>>>> Cheers,
>>>>> David
>>>>> 
>>>>>> On 3/07/2014 11:08 AM, Mikael Vidstedt wrote:
>>>>>> 
>>>>>> Please review this enhancement which adds the scaffolding needed to run
>>>>>> the hotspot jtreg tests in JPRT.
>>>>>> 
>>>>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8049071
>>>>>> Webrev (/):
>>>>>> http://cr.openjdk.java.net/~mikael/webrevs/8049071/webrev.00/top/webrev/ 
>>>>>> Webrev (hotspot/):
>>>>>> http://cr.openjdk.java.net/~mikael/webrevs/8049071/webrev.00/hotspot/webrev/ 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Summary:
>>>>>> 
>>>>>> We want to run the hotspot regression tests on every hotspot push. This
>>>>>> change enables this and adds four new test groups to the set of tests
>>>>>> being run on hotspot pushes. The new test sets still need to be
>>>>>> populated.
>>>>>> 
>>>>>> Narrative:
>>>>>> 
>>>>>> The majority of the changes are in the hotspot/test/Makefile. The
>>>>>> changes are almost entirely stolen from jdk/test/Makefile but have been
>>>>>> massaged to support (at least) three different use cases, two of which
>>>>>> were supported earlier:
>>>>>> 
>>>>>> 1. Running the non-jtreg tests (servertest, clienttest and
>>>>>> internalvmtests), also supporting the use of the "hotspot_" for when the
>>>>>> tests are invoked from the JDK top level
>>>>>> 2. Running jtreg tests by selecting test to run using the TESTDIRS
>>>>>> variable
>>>>>> 3. Running jtreg tests by selecting the test group to run (NEW)
>>>>>> 
>>>>>> The third/new use case is implemented by making any target named
>>>>>> hotspot_% *except* the ones listed in 1. lead to the corresponding jtreg
>>>>>> test group in TEST.groups being run. For example, running "make
>>>>>> hotspot_gc" leads to all the tests in the hotspot_gc test group in
>>>>>> TEST.groups to be run and so on.
>>>>>> 
>>>>>> I also removed the packtest targets, because as far as I can tell
>>>>>> they're not used anyway.
>>>>>> 
>>>>>> Note that the new component test groups in TEST.group -
>>>>>> hotspot_compiler, hotspot_gc, hotspot_runtime and hotspot_serviceability
>>>>>> - are currently empty, or more precisely they only run a single test
>>>>>> each. The intention is that these should be populated by the respective
>>>>>> teams to include stable and relatively fast tests. Tests added to the
>>>>>> groups will be run on hotspot push jobs, and therefore will be blocking
>>>>>> pushes in case they fail.
>>>>>> 
>>>>>> Cheers,
>>>>>> Mikael
>