Fwd: Re: adding testng after removig of junit

[Jiri Vanek]

On 04/14/2011 01:58 PM, Dr Andrew John Hughes wrote:
> On 19:59 Tue 12 Apr     , Jiri Vanek wrote:
>
> snip...
>

This is still the same arround:(
>>>
>>> In brief:
>>>
>>> * With the current solution, we can directly produce the JTreg-like
>>> output we want via our subclass and already do.  From what I
>>> understand of the TestNG patch, it would produce XML and then we'd
>>> have to convert it.  Exactly why is that better?
>>
>> Because we DO NOT want jtreg like result.
>
> Who is 'we'?  If you've read the previous discussion, you'll know that
> JTreg output was stated as a requirement so that the test suite worked
> well with the build bots.

We == everybody who will USE this testsuite. JTreg output is Minimal 
requirement. But for JUnit it probably also Maximal. You didn't get it. 
TestNG output is execlent. Why to avoid all those benefits or to write 
(maybe) complicated class and invent wheel again?

>
>> Pavel is doing magic to create
>> reports as are now from such a bad output. From xml - we have all
>> informations. Pavel will use it to produce reports containing stack
>> traces, will be able to see directly which ones should fail and which
>> should pass. Will be able to separate groups,Will be able to create very
>> solid statistics, and whatever we can imagine. It is pitty to lost all
>> the information. jtreg-like is just poor summary - and without ant it is
>> probably impossible to reach something better.
>
> You make a lot of unsubstantiated claims here.  If you want different output,
> you are welcome to submit a different outputter along the lines of the one
> Omair wrote for jtreg output and provide a way of using that instead.

This luncher and listener classes  are disgusting.  (Nothing against 
Omair's code, but against concept). It is unclear for everybody who will 
be looking inside project. And I'm afraid it  cause troubles.

>
>>
>> Another advance is extreme configurability (see bottom).
>>
>
> What is 'extreme configurability' and why would we want it?

Why to ecape solution which have it? You can easily group tests, run 
just some subset of test, exlude tests, specific order of tests... And 
it is clearly to be seen in XML - in comparsion with make file find 
*.class mess. Which I consider really unsuitable, and can't believe it 
pass review.


>
>>
>>>
>>> * Switching to TestNG also adds a mass of dependencies which aren't
>>> needed with JUnit, making it harder for people to run the test suite.
>>>
>>
>> testng are three jars. Junit is one jar. I do not see difference whether
>> wee need to make special steps to add one jar to classpath or three. Al
>> four of them are easily to be found and downloaded. And all are packaged
>> in fedora (an in most of distros).
>
> Under what licenses?  This discussion is the first I've heard of TestNG
> and the support for it in various distros seems to be immature.  In comparison,
> JUnit is well-established and without extraneous dependencies.
>

It is under appache licence which is much more suitable for FSOS/any 
project then Common Public License.
Even when some distro have missing some of those three libraries (which 
I do not believe as they are widely used libraries), then it is just 
benefit for them to add it - or just downlood jars  and use them on 
classpath.
>>
>>> Unless you can produce a clear example of why switching to TestNG
>>> for IcedTea-Web is essential and which overrides the need for all
>>> those additional dependencies, I see no reason to drop a solution that
>>> works and has already been extensively reviewed.
>>>
>>
>> What I see essential is output of test framework and configureability.
>> junit is not about to be run from commadline.
>
> Of course it can be run from the command line.  We're doing it right now.
>

Not exactly. We had to create class luncher for it, and makefile 
execution is disaster.

>> It is ant tool.
>
> No it isn't.
>
>> And we do
>> not want ant for sure (we have make).
>> You will probably never use or run this test. Me, Pavel and Omair will
>> do it most of time.
>
> This is a FOSS project.  You can't make assumptions about who will run the
> tests.  Distros, for example, will want to run these tests to verify that
> IcedTea-Web has built and installed correctly.  Having less dependencies
> makes this easier.
>

One jar or three?
> Clearly you can't give a clear example of why TestNG is so much better and
> have to resort to cheap personal attacks instead.
>
Please don't tkae it personaly. I suspect you from seeing it from 
make/distros  perspective. I see it from .. "java developer: point of view.

>   And for myself i really wont to configure what i'm
>> running. This Luncher class and makefile "send all classes to classpath"
>> is dreadfull hack.
>>
>> Btw.. I have already posted all of those dis/advantages;)
>>>> J.
>>>
>>
>>
>
>
>

Regards J.