Hi Oliver,
But is j.u.c actually required to be in the runtime under test? I was
thinking
that j.u.c was only required for the VM actually running the harness,
and all
that gets run on the VM under test is the actual test method. If this
was true,
then we could run TestNG with the RI (which has j.u.c) and use the
jvm option to specify the Harmony VM (which would not need j.u.c).
I afraid we cannot do like that. At least I was not successful last
time I tried to run tests using the jvm="<harmony>" option. As far as
I understand TestNG requires j.u.c for running every single test
method because parallel running can be specified on the method level.
I mean in TestNG you can write something like this:
@Test(threadPoolSize = 7, invocationCount = 29)
This means that this method should be invoked from different threads.
And it seems that TestNG needs j.u.c to implement multithreading.
Yes agreed, it is good to make group membership explicit as it facilitates
inclusion/exclusion of groups, and makes it obvious which group tests
belong to. Perhaps the same should be done for api tests, so we have a
type.api group?
So you suggest to add @Test (groups={os.any, type.api}) to every api
test (that runs on every platform) without any defaults at all?
I thought I had sent a mail out on this in the original thread, but I guess
I never did (unless Thunderbird is hiding mail from me again!).
Just checked - there is no such mail in my gmail box, at least in the
"[classlib] Testing conventions - a proposal" thread.
So, for example, if we were on a Windows x86 32bit machine, the Ant
scripts would run test groups "os.shared", "os.windows", "os.windows.x86"
and (if we ever get any 32/64bit specific tests) "os.windows.x86.32bit",
or something similar.
Well, I like it in general. The only thing I still feel uncomfortable
with is "os.shared". When some code is shared among different
platforms it makes sense. But here we have a test shared by several
OSes. Does this sound natural? But I don't feel strongly about that
and will not object if everybody likes this.
With Best Regards,
2006/8/14, Oliver Deakin <[EMAIL PROTECTED]>:
Alexei Zakharov wrote:
> Hi Oliver,
>
>> So perhaps the build system should be changed temporarily so that
>> we dont self host the test harness? i.e. until we get
>> java.util.concurrent,
>> run Ant and the subsequent TestNG process with RI or other non-Harmony
>> VM, and launch the tests with Harmony VM using the jvm option.
>
> The bad news is that TestNG requires j.u.c stuff even for single test
> execution (i.e. in any case if jvm="<path to harmony>"). :( So if we
> want to run annotated tests with TestNG (even from the command line)
> we need j.u.c.
But is j.u.c actually required to be in the runtime under test? I was
thinking
that j.u.c was only required for the VM actually running the harness,
and all
that gets run on the VM under test is the actual test method. If this
was true,
then we could run TestNG with the RI (which has j.u.c) and use the
jvm option to specify the Harmony VM (which would not need j.u.c).
Please correct me if Im wrong - I'm just trying to explore our options in
the hope that we can start to move to TestNG soon.
>
>> That's odd - Thread.class in luni-kernel (VME v4) definitely contains a
>> getId() method.
>
> May be I did something wrong - I will check tomorrow.
>
>> I do like them - in fact I was going to link his mail after that point
>> but couldn't find it.
>
> Here is the link:
>
http://mail-archives.apache.org/mod_mbox/incubator-harmony-dev/200607.mbox/[EMAIL
PROTECTED]
>
Thanks - it was good to reread.
>
> As far as I remember there was additions to the George's list like
> using @Test (groups={"os.any"} ) rather than simple @Test for API
> tests that run on any platform.
>
Yes agreed, it is good to make group membership explicit as it facilitates
inclusion/exclusion of groups, and makes it obvious which group tests
belong to. Perhaps the same should be done for api tests, so we have a
type.api group?
I thought I had sent a mail out on this in the original thread, but I guess
I never did (unless Thunderbird is hiding mail from me again!).
We have had discussions on the list about platform specific native
code organisation, and I think organisation of platform specific tests
can be handled in a similar way.
So we would have an "os.shared" (or os.any, Im just going with shared
as that is what we use for the native code) group that would run on all
platforms. If there are platform specific tests, they would be grouped
by os, then architecture (and then possibly word size).
So, for example, if we were on a Windows x86 32bit machine, the Ant
scripts would run test groups "os.shared", "os.windows", "os.windows.x86"
and (if we ever get any 32/64bit specific tests) "os.windows.x86.32bit",
or something similar.
I think this tallies with what George was suggesting, and makes sense to
me. Are there any objections to this approach?
Regards,
Oliver
>> I really mean that we should make sure that everyone is happy with
>> a certain set of group names before going ahead and applying them.
>> Once they are decided upon, they should be added to the testing
>> conventions webpage.
>
> Yes, agree.
>
>
> With Best Regards,
>
> 2006/8/10, Oliver Deakin <[EMAIL PROTECTED]>:
>> Alexei Zakharov wrote:
>> >> > We now have this, so let the TestNG debate continue :)
>> >> Unfortunately, we still need java.util.concurrent :-(
>> >
>> > Yeah, TestNG 5.0 still throws java.lang.NoClassDefFoundError :
>> > java.util.concurrent.LinkedBlockingQueue on Harmony+j9v4.
>>
>> So perhaps the build system should be changed temporarily so that
>> we dont self host the test harness? i.e. until we get
>> java.util.concurrent,
>> run Ant and the subsequent TestNG process with RI or other non-Harmony
>> VM, and launch the tests with Harmony VM using the jvm option.
>> At least it will allow us to move forward with TestNG (if that's what we
>> decide) without having to wait for java.util.concurrent. Then when
>> we have j.u.c, start self-hosting again.
>> Comments?
>>
>> >
>> > I've also got an error while trying to compile TestNG 5.0 tests with
>> > Harmony+j9v4+ecj: The method getId() is undefined for the type Thread
>>
>> That's odd - Thread.class in luni-kernel (VME v4) definitely contains a
>> getId() method. I don't know anything about the TestNG tests - how are
>> they run? is luni-kernel.jar definitely at the front of the
>> bootclasspath?
>>
>> >
>> >> > - If we go ahead with TestNG, we need to select a set of group
>> names
>> >> > to use to indicate exclusion, platform specificness etc.
>> >
>> > Don't you like the names suggested by George?
>>
>> I do like them - in fact I was going to link his mail after that point
>> but couldn't
>> find it. I really mean that we should make sure that everyone is
>> happy with
>> a certain set of group names before going ahead and applying them.
>> Once they are decided upon, they should be added to the testing
>> conventions webpage.
>>
>> >
>> >> > - Decide whether some physical separation of tests on disk is
>> >> necessary, for instance to separate classpath and bootclasspath
>> tests
>> >
>> > IMHO it is ok to separate classpath and bootclasspath tests because it
>> > will be easer to implement such solution technically.
>>
>> I agree, although I don't feel strongly about it.
>>
>> Regards,
>> Oliver
>>
>> >
>> > Regards,
>> >
>> > 2006/8/10, Richard Liang <[EMAIL PROTECTED]>:
>> >>
>> >>
>> >> Oliver Deakin wrote:
>> >> > Richard Liang wrote:
>> >> >>
>> >> >>
>> >> >> Alexei Zakharov wrote:
>> >> >>> Hi Richard,
>> >> >>>
>> >> >>>> Not sure if we really want to involve another migration: TestNG
>> >> >>>> javadoc
>> >> >>>> -> TestNG annotation. Any comments?
>> >> >>>
>> >> >>> Well, IMHO this depends on time constraints - when do we plan to
>> >> have
>> >> >>> the support for anotations? If the answer is about a couple of
>> >> weeks -
>> >> >>> no problem, we can wait. But if this is several months...
>> >> >>> About the "migration" - I don't think this will be a real
>> painfull
>> >> >>> migration, all infrastructure will remain the same. We will only
>> >> need
>> >> >>> to convert javadocs to annotations (one-one correspondence)
>> and this
>> >> >>> task can be easily automated.
>> >> >> Sounds reasonable. :-) Maybe drlvm guys or Oliver could tell
>> us when
>> >> >> we will have a VM with annotation support?
>> >> >
>> >> > We now have this, so let the TestNG debate continue :)
>> >> >
>> >>
>> >> Unfortunately, we still need java.util.concurrent :-(
>> >>
>> >>
>> >> > I guess we need to decide a few things before we go ahead with
>> this:
>> >> > - Whether TestNG is generally accepted by the Harmony community
>> >> > as our test harness of choice for unit testing. I think this will
>> >> > probably
>> >> > require a vote of some kind before we could make the move.
>> >> > - If we go ahead with TestNG, we need to select a set of group
>> names
>> >> > to use
>> >> > to indicate exclusion, platform specificness etc.
>> >> > - Decide whether some physical separation of tests on disk is
>> >> necessary,
>> >> > for instance to separate classpath and bootclasspath tests.
>> >> >
>> >> > Comments/additions?
>> >>
>> >> Agree. And we could provide proposals for these questions case by
>> case,
>> >> and let community make decision.
>> >>
>> >> Best regards,
>> >> Richard
>> >> >
>> >> > Regards,
>> >> > Oliver
>> >> >
>> >> >
>> >> >>
>> >> >>>
>> >> >>> Thanks,
>> >> >>>
>> >> >>> 2006/8/1, Richard Liang <[EMAIL PROTECTED]>:
>> >> >>>>
>> >> >>>>
>> >> >>>> Alexei Zakharov wrote:
>> >> >>>> > Hi,
>> >> >>>> >
>> >> >>>> > I have created this new thread as a single place for
>> discussions
>> >> >>>> > started in "Re: [testing] Peace" and "[classlib] Testing
>> >> >>>> conventions –
>> >> >>>> > a proposal" threads.
>> >> >>>> >
>> >> >>>> > What did we have in the previous threads?
>> >> >>>> > * Test classification proposed by Vladimir
>> >> >>>> > * Test classification and group names proposed by George
>> >> >>>> > * Solution for Ant and TestNG scripting issues (still being
>> >> >>>> discussed)
>> >> >>>> >
>> >> >>>> > Since a lot of people are asking about TestNG and wanting
>> >> TestNG I
>> >> >>>> > decide to put some effort and take a closer look at this
>> piece of
>> >> >>>> > software. Thus during the last few days I was playing with
>> TestNG
>> >> >>>> - I
>> >> >>>> > tried to run different kind of tests with it, to perform
>> various
>> >> >>>> > workloads, generate reports in different ways and etc. The
>> >> >>>> purpose of
>> >> >>>> > all this activity was to try TestNG in a real work,
>> understand is
>> >> >>>> > TestNG really worth our credits and how expensive can be
>> >> moving to
>> >> >>>> > TestNG from our currently implemented testing infrastructure.
>> >> Now I
>> >> >>>> > have some thoughts and facts I'd like to share with the
>> >> community.
>> >> >>>> > I've put it in the form of list for convenience.
>> >> >>>> >
>> >> >>>> > * TestNG works ok in normal conditions, no visible bugs
>> >> >>>> > * It is possible to define and use various (possibly
>> >> intersecting)
>> >> >>>> > test groups with TestNG
>> >> >>>> > * TestNG-style metadata is more convenient than JUnit test
>> suites
>> >> >>>> (now
>> >> >>>> > I agree with this statement). IMHO this is the main TestNG
>> >> benefit.
>> >> >>>> > * It is possible to run TestNG from command line
>> >> >>>> > * There is also the special ant task for running TestNG
>> >> >>>> > * Not everything can be configured with the ant task or
>> >> command-line
>> >> >>>> > params, sometimes extra test suite definition file
>> >> "testng.xml" is
>> >> >>>> > needed
>> >> >>>> > * It is possible to run jUnit tests with TestNG
>> ("testng.xml" is
>> >> >>>> > needed in this case)
>> >> >>>> > * It is possible to run junit tests we currently have in
>> Harmony
>> >> >>>> with
>> >> >>>> > TestNG without any problems and modifications of the source
>> code.
>> >> >>>> > However, we probably should write some number of TestNG test
>> >> suite
>> >> >>>> > definition files "testng.xml" to be able to run all our junit
>> >> >>>> tests (I
>> >> >>>> > have tried tests for bean module and some tests for luni)
>> >> >>>> > * We can mix jUnit tests and TestNG tests in the single test
>> >> suite
>> >> >>>> > configuration – i.e. single testng.xml file. We can add TestNG
>> >> >>>> > metadata to some test classes and leave other test classes
>> >> untouched
>> >> >>>> > * TestNG generates HTML reports in its own style, not a very
>> >> >>>> > convenient one IMHO
>> >> >>>> > * It is also possible to generate JUnitReports from the output
>> >> >>>> > generated by TestNG
>> >> >>>> > * Such reports will have a little bit different structure
>> since
>> >> >>>> TestNG
>> >> >>>> > doesn't provider any information about enclosing type for test
>> >> >>>> > methods. Names for tests (replacement for JUnit "test
>> >> classes") and
>> >> >>>> > test suites should be externally configured in "testng.xml"
>> >> >>>> > * TestNG for Java 5 doesn't work on Harmony because some
>> >> necessary
>> >> >>>> > classes from java.util.concurrent package are missing and it
>> >> seems
>> >> >>>> > that jsr14 target (we are currently using) doesn't support
>> >> >>>> annotations
>> >> >>>> > * TestNG for Java 1.4 (javadoc version) currently works on
>> >> Harmony
>> >> >>>> > * I have half-way done script that converts TestNG 1.4
>> metadata
>> >> >>>> > (javadoc) tests to TestNG 1.5 (5.0 annotations) tests.
>> >> >>>> >
>> >> >>>>
>> >> >>>> Excellent summary! Thanks a lot
>> >> >>>>
>> >> >>>> > The question I'd like to raise now is – aren't we ready for
>> >> TestNG
>> >> >>>> > right now?
>> >> >>>> I suppose we will use Java 5.0 annotations of TestNG, so it
>> seems
>> >> >>>> now we
>> >> >>>> are not ready for TestNG. But we can continue our feasibility
>> >> study,
>> >> >>>> just like what you have done, to know if TestNG really meets our
>> >> >>>> requirements or if there are any potential problems. Maybe we
>> could
>> >> >>>> list
>> >> >>>> a prerequisite list. e.g,
>> >> >>>> 1) Harmony can fully self-host TestNG with Java5 annotations
>> >> >>>> 2) Test groups are well-defined and agreed in community
>> >> >>>> 3) Guidelines to write TestNG testcases
>> >> >>>> 4) Take one module to run a pilot case
>> >> >>>> ....
>> >> >>>>
>> >> >>>> Please correct me if I'm wrong
>> >> >>>>
>> >> >>>> > For example, we could replace our harness from jUnit to
>> >> >>>> > TestNG and lazily start converting of some failing and
>> platform
>> >> >>>> > dependent tests to javadoc version of TestNG. The rest of the
>> >> tests
>> >> >>>> > will remain jUnit in fact. And when our VM will be ready to
>> >> handle
>> >> >>>> > annotations we can convert all our TestNG 1.4 tests to TestNG
>> >> 1.5. I
>> >> >>>> > understand that this idea may seem to be too early. But
>> anyway we
>> >> >>>> will
>> >> >>>> > need to change things some day since many people are
>> unhappy with
>> >> >>>> the
>> >> >>>> > current testing infrastructure (me for example).
>> >> >>>> Not sure if we really want to involve another migration: TestNG
>> >> >>>> javadoc
>> >> >>>> -> TestNG annotation. Any comments?
>> >> >>>>
>> >> >>>> >
>> >> >>>> > Thought? Suggestions? Opposite opinions?
>> >> >>>> >
>> >> >>>> > With Best Regards,
>> >> >>>> >
>> >> >>>> >
>> >> >>>>
>> >> >>>> --
>> >> >>>> Richard Liang
>> >> >>>> China Software Development Lab, IBM
>> >
>> >> --
>> >> Richard Liang
>> >> China Software Development Lab, IBM
>> >
>>
>> --
>> Oliver Deakin
>> IBM United Kingdom Limited
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: [EMAIL PROTECTED]
>> For additional commands, e-mail: [EMAIL PROTECTED]
>>
>>
>
>
--
Oliver Deakin
IBM United Kingdom Limited
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
--
Alexei Zakharov,
Intel Middleware Product Division
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
