You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by "Geir Magnusson Jr." <ge...@apache.org> on 2005/12/01 06:39:02 UTC

Where to place the core classlib code?

Now that the code from IBMs contrib (HARMONY-14) is in the sandbox  
(well - it's uploading now...), we'll need to shift things around a bit.

do we do something as obvious as

enhanced/trunk/classlib/

?

Or do we get ahead of things and do :

enhanced/trunk  /* for general stuff, including project build scripts  
and such */

enhanced/classlib/trunk  /* for HARMONY-14 and then the security code  
layered on *?

?

geir

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



Re: Where to place the core classlib code?

Posted by "Geir Magnusson Jr." <ge...@apache.org>.
On Dec 1, 2005, at 11:38 AM, Tim Ellison wrote:

> Geir Magnusson Jr. wrote:
>> Ok - so we should just move it then to
>>
>> enhanced/classlib/trunk
>>
>> and it goes from there.
>
> Agreed.
>
>> Now, I'm really worried about getting some build sanity from the   
>> start,
>> because other than a corked wine, there are few things more
>> disappointing than "the build is broken"
>
> Unless/until someone shows up with a comprehensive test suite we  
> have to
> start by writing tests for everything that gets patched.  A successful
> build is a clean compile and testsuite pass.  I suggest that we make a
> pact to self-host from the start, so the build system becomes the  
> first
> real test case that we have :-) .

Well, as soon as possible. :)

I'd actually like to do both.  Test with ourselves, and test w/ other  
VMs, as I'd hate to see us hide our own bug :)

>
>> I think we need that each subproject will need freedom for approaches
>> for building as we're going to have a diverse set of  
>> technologies.   But
>> if we can federate so that
>> what's in enhanced/build is the master that can kick off a subproject
>> build, we have a bit os sanity.
>
> Absolutely.
>
>> We don't have to do it right now,  but
>> we should keep this in mind as enh/classlib evolves in the very   
>> near term.
>
> I think we _should_ do it right now, to avoid the refactoring pain  
> later.

I said "not right now" because right now there are no other things  
peer to classlib/ other than build/ so I think the effort might be  
better spent ensuring that building classlib/ (resolving all the open  
ideas and threads there) works well...

But, yes.  Agreed.

geir

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



Re: Testsuite (Was: Where to place the core classlib code?)

Posted by Tim Ellison <t....@gmail.com>.
Indeed, Mauve is a great resource.

Regards,
Tim

Mark Wielaard wrote:
> On Thu, 2005-12-01 at 16:38 +0000, Tim Ellison wrote:
> 
>>Unless/until someone shows up with a comprehensive test suite we have
>>to start by writing tests for everything that gets patched.
> 
> 
> Take a look at http://www.sourceware.org/mauve/
> 
> Mauve is a collaboration of several companies and organizations started
> by Cygnus, Transvirtual, Hewlett-Packard and the GNU project and is used
> by GNU Classpath, gcj, kaffe, ikvm, cacao, etc. Mauve has several
> modules and tens of thousands of tests for testing different parts of
> the compilers, runtimes and core library implementations. The main
> module has tests for the core library, there is a module jacks for
> compiler tests, there is a module verifier for byte code verifier tests,
> there is a visual tester for testing AWT implementations, a module for
> rmic tests and a serialization testsuite.
> 
> Cheers,
> 
> Mark
> 

-- 

Tim Ellison (t.p.ellison@gmail.com)
IBM Java technology centre, UK.

Re: Testsuite (Was: Where to place the core classlib code?)

Posted by "Geir Magnusson Jr." <ge...@apache.org>.
Well, First we should get mauve pointed at what we have and start  
using it.  Then I think it would be very cool to help switch it to  
JUnit (or TestNG...)

geir

On Dec 5, 2005, at 5:07 PM, Anthony Green wrote:

> On Mon, 2005-12-05 at 20:34 +0100, Mark Wielaard wrote:
>> On Thu, 2005-12-01 at 16:38 +0000, Tim Ellison wrote:
>>> Unless/until someone shows up with a comprehensive test suite we  
>>> have
>>> to start by writing tests for everything that gets patched.
>>
>> Take a look at http://www.sourceware.org/mauve/
>
> Speaking of Mauve (a project born of class library licensing
> frustrations!), there's some real interest amongst the Mauve community
> to migrate the test cases to JUnit.  The Mauve infrastructure predates
> JUnit.  This would be a universally useful task for some enthusiastic
> contributor.
>
> AG
>
>

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



Re: Testsuite (Was: Where to place the core classlib code?)

Posted by Anthony Green <gr...@redhat.com>.
On Mon, 2005-12-05 at 20:34 +0100, Mark Wielaard wrote:
> On Thu, 2005-12-01 at 16:38 +0000, Tim Ellison wrote:
> > Unless/until someone shows up with a comprehensive test suite we have
> > to start by writing tests for everything that gets patched.
> 
> Take a look at http://www.sourceware.org/mauve/

Speaking of Mauve (a project born of class library licensing
frustrations!), there's some real interest amongst the Mauve community
to migrate the test cases to JUnit.  The Mauve infrastructure predates
JUnit.  This would be a universally useful task for some enthusiastic
contributor.

AG



Testsuite (Was: Where to place the core classlib code?)

Posted by Mark Wielaard <ma...@klomp.org>.
On Thu, 2005-12-01 at 16:38 +0000, Tim Ellison wrote:
> Unless/until someone shows up with a comprehensive test suite we have
> to start by writing tests for everything that gets patched.

Take a look at http://www.sourceware.org/mauve/

Mauve is a collaboration of several companies and organizations started
by Cygnus, Transvirtual, Hewlett-Packard and the GNU project and is used
by GNU Classpath, gcj, kaffe, ikvm, cacao, etc. Mauve has several
modules and tens of thousands of tests for testing different parts of
the compilers, runtimes and core library implementations. The main
module has tests for the core library, there is a module jacks for
compiler tests, there is a module verifier for byte code verifier tests,
there is a visual tester for testing AWT implementations, a module for
rmic tests and a serialization testsuite.

Cheers,

Mark

-- 
Escape the Java Trap with GNU Classpath!
http://www.gnu.org/philosophy/java-trap.html

Join the community at http://planet.classpath.org/

Re: Where to place the core classlib code?

Posted by Tim Ellison <t....@gmail.com>.
Geir Magnusson Jr. wrote:
> Ok - so we should just move it then to
> 
> enhanced/classlib/trunk
> 
> and it goes from there.

Agreed.

> Now, I'm really worried about getting some build sanity from the  start,
> because other than a corked wine, there are few things more 
> disappointing than "the build is broken"

Unless/until someone shows up with a comprehensive test suite we have to
start by writing tests for everything that gets patched.  A successful
build is a clean compile and testsuite pass.  I suggest that we make a
pact to self-host from the start, so the build system becomes the first
real test case that we have :-) .

> I think we need that each subproject will need freedom for approaches 
> for building as we're going to have a diverse set of technologies.   But
> if we can federate so that
> what's in enhanced/build is the master that can kick off a subproject 
> build, we have a bit os sanity.

Absolutely.

> We don't have to do it right now,  but
> we should keep this in mind as enh/classlib evolves in the very  near term.

I think we _should_ do it right now, to avoid the refactoring pain later.

-- 

Tim Ellison (t.p.ellison@gmail.com)
IBM Java technology centre, UK.

Re: Where to place the core classlib code?

Posted by "Geir Magnusson Jr." <ge...@apache.org>.
On Dec 1, 2005, at 6:59 AM, Tim Ellison wrote:

> Geir Magnusson Jr. wrote:
>>
>>
>>
>> Yes - we've talked about this before, and I just wanted to focus on
>> classlib...
>>
>> The problem I have with this is that we won't be able to branch the
>> whole thing easily like we would if we had :
>>
>> enhanced/
>>      trunk/
>>          classlib/
>>          vm/
>>          tools/
>>      branches/
>>          harmony-v0.5-branch/
>>                  ...
>>      tags/
>>          harmony-v0.5-rc1/
>>                  ...
>>          harmony-v0.5-rc2/
>>                  ...
>>
>> Thinking out load -  it's not obvious to me that we'd want to do that
>> anyway, given the eventual mass of this thing.  The deliverable  
>> goal  is
>> a full stack and tools, would we really want to be assembling from
>> truly independent sub-parts?   Yah, I guess we do...  that allows the
>> sub-components to be independently developed and tested.
>
> Right, in a system this size we have to break it down into subprojects
> (what you called 'subdivisions' below).  The *repository* branches  
> will
> be most useful within a subproject, and if we want different
> combinations of subprojects then they are likely build-time targets.
>
>> It might be that the structure of each subdivision (not using the   
>> word
>> "component"
>
> are you comfortable with "subproject"?
>
>> needs to vary independently also because of  requirement
>> specific to each (for example, a modular class library  would have  
>> ore
>> than one build artifact...
>
> exactly.  We have said that the classlibs are going to be modular too,
> so we will have flexibility within the classlib subproject/ 
> subdivision too.
>

Ok - so we should just move it then to

enhanced/classlib/trunk

and it goes from there.

Now, I'm really worried about getting some build sanity from the  
start, because other than a corked wine, there are few things more  
disappointing than "the build is broken"

I think we need that each subproject will need freedom for approaches  
for building as we're going to have a diverse set of technologies.   
But if we can federate so that
what's in enhanced/build is the master that can kick off a subproject  
build, we have a bit os sanity.  We don't have to do it right now,  
but we should keep this in mind as enh/classlib evolves in the very  
near term.

geir

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



Re: Where to place the core classlib code?

Posted by Tim Ellison <t....@gmail.com>.
Geir Magnusson Jr. wrote:
> 
> On Dec 1, 2005, at 6:04 AM, Tim Ellison wrote:
> 
>> I thought the idea was that each contribution started working directly
>> in the sandbox?  Isn't that what Dan and Archie are doing?
> 
> 
> We threw it in the sandbox because we had no better ideas and was a 
> good starting point.  Since I'm really hoping that we don't have 
> multiple classlibrary efforts here at Harmony (although that's  welcome
> too, I suppose, if someone really wants to do it...), I  figure we can
> start by moving this code out now, and then if one or  more of our VMs
> (Archie or Dan's) start to support it, we pull it out  too and put in
> the appropriate place - whatever that is - so we can  start having a
> comprehensive build.

Ok, sounds good.

>>
>> Once things are pulled out from the sandbox I propose that the
>> repository is laid out to represent a few (non-trivial) sub-projects
>> that have well-defined interaction, to allow some development  autonomy,
>> but build together into a single system, something like:
>> enhanced
>>   /classlib
>>     /trunk
>>     /branches
>>     /tags
>>   /vm
>>     /trunk
>>     /branches
>>     /tags
>>   /jit
>>     /trunk
>>     /branches
>>     /tags
>>   /build
>>     /trunk
>>     /branches
>>     /tags
>>
>> and so on, with further subdivision within the sub-projects are  required
>> (e.g. ACL areas within the classlib as defined by the ACQ).
>>
>> Is this the sort of thing you mean?
> 
> 
> Yes - we've talked about this before, and I just wanted to focus on 
> classlib...
> 
> The problem I have with this is that we won't be able to branch the 
> whole thing easily like we would if we had :
> 
> enhanced/
>      trunk/
>          classlib/
>          vm/
>          tools/
>      branches/
>          harmony-v0.5-branch/
>                  ...
>      tags/
>          harmony-v0.5-rc1/
>                  ...
>          harmony-v0.5-rc2/
>                  ...
> 
> Thinking out load -  it's not obvious to me that we'd want to do that 
> anyway, given the eventual mass of this thing.  The deliverable goal  is
> a full stack and tools, would we really want to be assembling from 
> truly independent sub-parts?   Yah, I guess we do...  that allows the 
> sub-components to be independently developed and tested.

Right, in a system this size we have to break it down into subprojects
(what you called 'subdivisions' below).  The *repository* branches will
be most useful within a subproject, and if we want different
combinations of subprojects then they are likely build-time targets.

> It might be that the structure of each subdivision (not using the  word
> "component" 

are you comfortable with "subproject"?

> needs to vary independently also because of  requirement
> specific to each (for example, a modular class library  would have ore
> than one build artifact...

exactly.  We have said that the classlibs are going to be modular too,
so we will have flexibility within the classlib subproject/subdivision too.


Regards,
Tim

> 
> geir
> 
> 
>>
>> Regards,
>> Tim
>>
>>
>> Geir Magnusson Jr. wrote:
>>
>>> Now that the code from IBMs contrib (HARMONY-14) is in the sandbox
>>> (well - it's uploading now...), we'll need to shift things around  a
>>> bit.
>>>
>>> do we do something as obvious as
>>>
>>> enhanced/trunk/classlib/
>>>
>>> ?
>>>
>>> Or do we get ahead of things and do :
>>>
>>> enhanced/trunk  /* for general stuff, including project build scripts
>>> and such */
>>>
>>> enhanced/classlib/trunk  /* for HARMONY-14 and then the security code
>>> layered on *?
>>>
>>> ?
>>>
>>> geir
>>>
>>
>> -- 
>>
>> Tim Ellison (t.p.ellison@gmail.com)
>> IBM Java technology centre, UK.
> 
> 

-- 

Tim Ellison (t.p.ellison@gmail.com)
IBM Java technology centre, UK.

Re: Where to place the core classlib code?

Posted by "Geir Magnusson Jr." <ge...@apache.org>.
On Dec 1, 2005, at 6:04 AM, Tim Ellison wrote:

> I thought the idea was that each contribution started working directly
> in the sandbox?  Isn't that what Dan and Archie are doing?

We threw it in the sandbox because we had no better ideas and was a  
good starting point.  Since I'm really hoping that we don't have  
multiple classlibrary efforts here at Harmony (although that's  
welcome too, I suppose, if someone really wants to do it...), I  
figure we can start by moving this code out now, and then if one or  
more of our VMs (Archie or Dan's) start to support it, we pull it out  
too and put in the appropriate place - whatever that is - so we can  
start having a comprehensive build.

>
> Once things are pulled out from the sandbox I propose that the
> repository is laid out to represent a few (non-trivial) sub-projects
> that have well-defined interaction, to allow some development  
> autonomy,
> but build together into a single system, something like:
> enhanced
>   /classlib
>     /trunk
>     /branches
>     /tags
>   /vm
>     /trunk
>     /branches
>     /tags
>   /jit
>     /trunk
>     /branches
>     /tags
>   /build
>     /trunk
>     /branches
>     /tags
>
> and so on, with further subdivision within the sub-projects are  
> required
> (e.g. ACL areas within the classlib as defined by the ACQ).
>
> Is this the sort of thing you mean?

Yes - we've talked about this before, and I just wanted to focus on  
classlib...

The problem I have with this is that we won't be able to branch the  
whole thing easily like we would if we had :

enhanced/
      trunk/
          classlib/
          vm/
          tools/
      branches/
          harmony-v0.5-branch/
                  ...
      tags/
          harmony-v0.5-rc1/
                  ...
          harmony-v0.5-rc2/
                  ...

Thinking out load -  it's not obvious to me that we'd want to do that  
anyway, given the eventual mass of this thing.  The deliverable goal  
is a full stack and tools, would we really want to be assembling from  
truly independent sub-parts?   Yah, I guess we do...  that allows the  
sub-components to be independently developed and tested.

It might be that the structure of each subdivision (not using the  
word "component" needs to vary independently also because of  
requirement specific to each (for example, a modular class library  
would have ore than one build artifact...

geir


>
> Regards,
> Tim
>
>
> Geir Magnusson Jr. wrote:
>> Now that the code from IBMs contrib (HARMONY-14) is in the sandbox
>> (well - it's uploading now...), we'll need to shift things around  
>> a bit.
>>
>> do we do something as obvious as
>>
>> enhanced/trunk/classlib/
>>
>> ?
>>
>> Or do we get ahead of things and do :
>>
>> enhanced/trunk  /* for general stuff, including project build scripts
>> and such */
>>
>> enhanced/classlib/trunk  /* for HARMONY-14 and then the security code
>> layered on *?
>>
>> ?
>>
>> geir
>>
>
> -- 
>
> Tim Ellison (t.p.ellison@gmail.com)
> IBM Java technology centre, UK.

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



Re: Where to place the core classlib code?

Posted by Tim Ellison <t....@gmail.com>.
I thought the idea was that each contribution started working directly
in the sandbox?  Isn't that what Dan and Archie are doing?

Once things are pulled out from the sandbox I propose that the
repository is laid out to represent a few (non-trivial) sub-projects
that have well-defined interaction, to allow some development autonomy,
but build together into a single system, something like:
enhanced
  /classlib
    /trunk
    /branches
    /tags
  /vm
    /trunk
    /branches
    /tags
  /jit
    /trunk
    /branches
    /tags
  /build
    /trunk
    /branches
    /tags

and so on, with further subdivision within the sub-projects are required
(e.g. ACL areas within the classlib as defined by the ACQ).

Is this the sort of thing you mean?

Regards,
Tim


Geir Magnusson Jr. wrote:
> Now that the code from IBMs contrib (HARMONY-14) is in the sandbox 
> (well - it's uploading now...), we'll need to shift things around a bit.
> 
> do we do something as obvious as
> 
> enhanced/trunk/classlib/
> 
> ?
> 
> Or do we get ahead of things and do :
> 
> enhanced/trunk  /* for general stuff, including project build scripts 
> and such */
> 
> enhanced/classlib/trunk  /* for HARMONY-14 and then the security code 
> layered on *?
> 
> ?
> 
> geir
> 

-- 

Tim Ellison (t.p.ellison@gmail.com)
IBM Java technology centre, UK.