harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Geir Magnusson Jr." <ge...@apache.org>
Subject Re: Backward compatibility
Date Fri, 13 May 2005 00:12:10 GMT
I didn't cc Gerry - I assume he's subscribed.

On May 11, 2005, at 6:22 PM, Gerry Steele wrote:

> I'm a big fan of the Apache foundation but this is one product I'm  
> not too
> sure is such a good idea as of yet for reasons several:
>
>
>>> Deprecated or non deprecated, we want Harmony to pass the TCK, so
>>>
>>>> whatever the TCK wants us to do, we'll do it.
>>>>
>
>
> I hope you understand what sticking to the TCK entails. When it  
> comes to
> implementing GUI stuff for instance, your platform will have to  
> fully copy
> the official JVM's Swing/AWT widgets and all other details in order  
> for the
> automation and robot driven tests to pass. The JCK testbase for  
> tiger is
> immense. To get it setup and run is a skill on its own. To get it  
> to pass
> all tests takes a serious am mount of tweaking and a noteable  
> knowledge of
> the javatest harness. It will require implementations on things as  
> extensive
> as CORBA and RMI. We would need passive agents, tname servers etc.

Indeed.  We face the same thing with the J2EE TCK.  It's a lot of  
work, but we've proven that it can be done in the Apache environment.

>
> Also, when running the TCK bear in mind that you'll have to run the  
> harness
> with the Sun VM.
>
> I'm not sure about the particular extent of the testsuite provided  
> with the
> TCK you guys are talking about (if there is interest can find out  
> more), but
> the JCK, which is basically a TCK for the entire J2SE jre and jdk  
> will be
> going on impossible to pass for an alternative implmentation as  
> everything
> is written with the Sun JDK/JRE in mind and test cases are adapted  
> in ways
> that will create an infinite unpredictable series of problems when  
> trying to
> adapt your code.

I believe that you may be mistaken, or it's just a problem of  
phrasing :)   I'm sure that any tests that are for specific extra- 
specification Sun-specific behavior will be added to the exception  
list and therefore not have to pass.  We've run into a few things  
like this w/ the TCK for J2EE.  For me, they are just TCK  
implementation mistakes, rather than any requirement to have Sun  
implementations as part of the tested code.

>
> Another reason is that I'm not quite sure I see the point. It will  
> take 4-5
> years or more to even come close to a product like tiger. Sun are  
> already
> working heavily on mustang and dolphin (to a lesser degree on the  
> latter).
> As well as this, sun research have many projects looking at the  
> future of
> the Java VM such as the Barcelona project which will drastically  
> change the
> implementation of the JVM. For instance to make it more network  
> orientated
> or to improve resource sharing.

It's clear that Sun does put a lot of effort into this.  But don't  
underestimate what an open source community can do.

>
> The latter things (which are yet to see real sun implementation)  
> might be
> something you guys might then want to take advantage of in order to  
> leverage
> a selling point of Harmony. Without something like that it's just  
> another
> attempt at a VM that will be playing catch-up forever.

We'll see :)

>
> Also, don't forget about quality. Sun put a serious amount of money  
> and
> manpower into ensuring the quality and compatibility of the JVM. A  
> lot of
> corporations depend on this. They have a regular update release  
> cycle. For
> instance we are currently working on 1.3.1_16, 1.4.2_09, 5.0_04 &  
> 5.0_05.

Yes - quality is a major factor here.  If we can't have the same  
quality, people won't care, and the code won't be used.  That's been  
something we've been aware of since day 1.

>
> In a project of this size some of the the test suites take several  
> days to
> run. Some take many many hours of man power. For excessive  
> thoroughness
> there also manual JCK and regression test suites. Which, trust me,  
> will not
> be performed by someone who isn't being paid for it. Things like  
> this don't
> fit well with the community model.

I'm not so sure.  I can easily see that we can find people that would  
want to be paid to do it, and people that might pay them to do it.  
There's nothing wrong with commercial involvement in projects at  
apache, as long as it's a standard Apache-style, transparent  
meritocracy.  Throughout the ASF, you'll find individuals  
participating in projects that are paid by someone to do so.

>
> Another worry I have is that the effort here might be better  
> redirect to
> some other project. We already have Java. Even if harmony does make  
> it to a
> useable release people will still prefer to use the Sun VM. It will  
> be the
> platform people build on and it will be the one they trust.
>
> I'll be very interested in how this turns out.

As will we all.  Please consider helping us out :)

geir

-- 
Geir Magnusson Jr                                  +1-203-665-6437
geirm@apache.org



Mime
View raw message