You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@cloudstack.apache.org by Ewan Mellor <Ew...@eu.citrix.com> on 2012/08/09 00:30:40 UTC

Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

We need to push this discussion on build systems to a conclusion.  Let me summarize so far.

* We need to be able to build CloudStack in various configurations (support for the range of hypervisors turned on or off, etc).

* We would like the build tool to be able to get the necessary dependencies automatically, to avoid the user having to download a dozen libraries from all sorts of different places.

* Many of our dependencies are version-dependent, so we would want to get a specific version or at the very least have greater-than constraints on the version that is downloaded.

* We want to be able to package CloudStack in RPMs and .debs that correctly depend on packages available on the target platform.


Gradle has been proposed: http://www.gradle.org/.  No-one seems to know much about it, other than the fact that Hibernate and Spring have switched to using it, so it can't be too bad.

Maven has been proposed: http://maven.apache.org.  We have some expertise offered from Brett Porter (a committer there).  It's an Apache project, and it's very widely used.  We don't know much about it though, outside of Brett.  I think that there's a legitimate criticism that it makes it difficult for packagers, because it hides the system dependencies.  That might not be reason enough not to use it though.  I don't know if Gradle suffers from the same problem.

Scripting it all ourselves with Ant is a possibility.  David started this effort on the deps-ctrl branch.  This puts us in control of our own destiny, and means that we don't have to learn a new tool (or fit in with the way that the tool works).  It does sound like we're just going to end up re-inventing half of Maven that way though.


For me, it comes down to developer resources.  We need people who care about this topic to volunteer some time to get the build working smoothly.  If those people want to use Maven or Gradle, I don't think anyone else is going to object too strongly, because no-one seems to know very much about those tools.  At the least, we get the opportunity to try one of them, and see if we like it.  Nothing is ever set in stone, and if we decide to rewrite the build system in time for the 4.1 release, then we can do that.

So that's the big question: is someone going to volunteer to get the build system tamed for the 4.0 release?  It's a dirty job, but someone's got to do it, and you'll earn significant kudos if you do!

Thanks,

Ewan.


Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Robert Schweikert <rj...@suse.com>.
On 08/10/2012 03:52 PM, Ewan Mellor wrote:
>> -----Original Message-----
>> From: Robert Schweikert [mailto:rjschwei@suse.com]
>> Sent: 10 August 2012 05:35
>> To: cloudstack-dev@incubator.apache.org
>> Subject: Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source
>> tree/source releases.]
>>
>> On 08/09/2012 09:06 PM, David Nalley wrote:
>>> On Thu, Aug 9, 2012 at 8:47 PM, Ewan Mellor <Ew...@eu.citrix.com>
>> wrote:
>>>>> -----Original Message-----
>>>>> From: Robert Schweikert [mailto:rjschwei@suse.com]
>>>>>
>>>>> [Snip]
>>>>>
>>>>>> * We want to be able to package CloudStack in RPMs and .debs that
>>>>> correctly depend on packages available on the target platform.
>>>>>
>>>>> This is IMHO not a "job" of the project. This is up to the packagers
>>>>> and package maintainers for the various distros. I will maintain
>>>>> openSUSE and SLE builds in OBS and would very much prefer that the
>>>>> build system know nothing about how to build an rpm. We should
>>>>> maintain a "reference spec file" in the source tree and I will
>>>>> contribute to that, but having the build system crank out a .deb or
>>>>> .rpm package is just not the best approach.
>>>>
>>>> David, could you reply to this?  You generally have opinions on how
>> packages should be built.  I happen to disagree with the comment above, but
>> I'm not a packager so there's a strong chance that I haven't got a clue what
>> I'm talking about, and so I will do whatever people want in this regard.
>>>
>>>
>>> So I think this is a terminology issue, perhaps.
>>> My perspective (with packager hat on)
>>> So assuming we pick $tool to replace ant. I tend to agree - $tool
>>> should not build RPMs or .debs.
>>> There are tools to build RPMs and .debs - (rpmbuild and dpkg) and they
>>> should build the packages. They will call $tool to actually build the
>>> software, but we ought not get into the loop of $tool calling
>>> rpmbuild/dpkg calling $tool.
>>> Also - $tool should have an option for turning dependency resolution
>>> off. (Maven certainly has this, and I am sure other tools do as well.)
>>> Packagers don't want dependency resolution - the guidelines they
>>> operate under don't permit bundled or automagically downloaded
>>> dependencies. Instead those need to be packaged independently and
>>> listed as a requirement (or build requirement) for the package. Any
>>> sane $tool will handle this.
>>>
>>> This doesn't mean that we wouldn't/couldn't automate RPM/deb builds in
>>> something like jenkins. We just don't want another debacle like waf.
>>> Those packages typically will not be as good as distro-maintained
>>> packages in my experience. The distro also won't be kicking out RPMs
>>> for every iteration either, so it's a mixed bag.
>>>
>>> Robert, does this adequately reflect the concerns?
>>
>> Yup,
>
> OK, so concerns reflected.  What are we going to do?  It sounds like we should:
Nope, not quite

1. Have a script that packages the source into a tarball

> 1. Have the build system produce a tarball of build artifacts.

We want to package the source as a tarball, not the artifacts (see my 1. 
above). $tool, from David's message, compiles the code into .jar, .war, 
.whatever. In this step the binaries are placed within the source tree 
or in a created directory of our choosing relative to the source tree. 
$tool has an install option that honors $DESTDIR and than places the 
binaries into the appropriate directories under $DESTDIR. Any 
directories missing in $DESTDIR get created by $tool

This approach works for both packagers and "install from source" users. 
For "install from source" users $DESTDIR is set to /

> 2. Have a separate script that takes the tarball and a spec file and runs rpmbuild to produce an RPM.
> 3. Ditto for .deb.
> 4. Have Jenkins run those scripts, so that we have RPMs and .debs for every build.
> 5. Expect that the distros are going to ignore our packages and make their own.
>
> Is that the summary of it?

Yes, with the exception on point 1 as noted above.

Robert



-- 
Robert Schweikert                           MAY THE SOURCE BE WITH YOU
SUSE-IBM Software Integration Center                   LINUX
Tech Lead
rjschwei@suse.com
rschweik@ca.ibm.com
781-464-8147

Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Wido den Hollander <wi...@widodh.nl>.

On 08/10/2012 09:52 PM, Ewan Mellor wrote:
>>> So I think this is a terminology issue, perhaps.
>>> My perspective (with packager hat on)
>>> So assuming we pick $tool to replace ant. I tend to agree - $tool
>>> should not build RPMs or .debs.
>>> There are tools to build RPMs and .debs - (rpmbuild and dpkg) and they
>>> should build the packages. They will call $tool to actually build the
>>> software, but we ought not get into the loop of $tool calling
>>> rpmbuild/dpkg calling $tool.
>>> Also - $tool should have an option for turning dependency resolution
>>> off. (Maven certainly has this, and I am sure other tools do as well.)
>>> Packagers don't want dependency resolution - the guidelines they
>>> operate under don't permit bundled or automagically downloaded
>>> dependencies. Instead those need to be packaged independently and
>>> listed as a requirement (or build requirement) for the package. Any
>>> sane $tool will handle this.
>>>
>>> This doesn't mean that we wouldn't/couldn't automate RPM/deb builds in
>>> something like jenkins. We just don't want another debacle like waf.
>>> Those packages typically will not be as good as distro-maintained
>>> packages in my experience. The distro also won't be kicking out RPMs
>>> for every iteration either, so it's a mixed bag.
>>>
>>> Robert, does this adequately reflect the concerns?
>>
>> Yup,
>
> OK, so concerns reflected.  What are we going to do?  It sounds like we should:
> 1. Have the build system produce a tarball of build artifacts.
> 2. Have a separate script that takes the tarball and a spec file and runs rpmbuild to produce an RPM.
> 3. Ditto for .deb.

No need for deb, that's where the debian/rules file is for.

You should be able to just run:

$ dpkg-buildpackage

That will call debian/rules which can then execute the scripts which 
compiles and generates the artifacts.

> 4. Have Jenkins run those scripts, so that we have RPMs and .debs for every build.
> 5. Expect that the distros are going to ignore our packages and make their own.
>
> Is that the summary of it?
>
> Ewan.
>

RE: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Ewan Mellor <Ew...@eu.citrix.com>.
> -----Original Message-----
> From: Robert Schweikert [mailto:rjschwei@suse.com]
> Sent: 10 August 2012 05:35
> To: cloudstack-dev@incubator.apache.org
> Subject: Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source
> tree/source releases.]
> 
> On 08/09/2012 09:06 PM, David Nalley wrote:
> > On Thu, Aug 9, 2012 at 8:47 PM, Ewan Mellor <Ew...@eu.citrix.com>
> wrote:
> >>> -----Original Message-----
> >>> From: Robert Schweikert [mailto:rjschwei@suse.com]
> >>>
> >>> [Snip]
> >>>
> >>>> * We want to be able to package CloudStack in RPMs and .debs that
> >>> correctly depend on packages available on the target platform.
> >>>
> >>> This is IMHO not a "job" of the project. This is up to the packagers
> >>> and package maintainers for the various distros. I will maintain
> >>> openSUSE and SLE builds in OBS and would very much prefer that the
> >>> build system know nothing about how to build an rpm. We should
> >>> maintain a "reference spec file" in the source tree and I will
> >>> contribute to that, but having the build system crank out a .deb or
> >>> .rpm package is just not the best approach.
> >>
> >> David, could you reply to this?  You generally have opinions on how
> packages should be built.  I happen to disagree with the comment above, but
> I'm not a packager so there's a strong chance that I haven't got a clue what
> I'm talking about, and so I will do whatever people want in this regard.
> >
> >
> > So I think this is a terminology issue, perhaps.
> > My perspective (with packager hat on)
> > So assuming we pick $tool to replace ant. I tend to agree - $tool
> > should not build RPMs or .debs.
> > There are tools to build RPMs and .debs - (rpmbuild and dpkg) and they
> > should build the packages. They will call $tool to actually build the
> > software, but we ought not get into the loop of $tool calling
> > rpmbuild/dpkg calling $tool.
> > Also - $tool should have an option for turning dependency resolution
> > off. (Maven certainly has this, and I am sure other tools do as well.)
> > Packagers don't want dependency resolution - the guidelines they
> > operate under don't permit bundled or automagically downloaded
> > dependencies. Instead those need to be packaged independently and
> > listed as a requirement (or build requirement) for the package. Any
> > sane $tool will handle this.
> >
> > This doesn't mean that we wouldn't/couldn't automate RPM/deb builds in
> > something like jenkins. We just don't want another debacle like waf.
> > Those packages typically will not be as good as distro-maintained
> > packages in my experience. The distro also won't be kicking out RPMs
> > for every iteration either, so it's a mixed bag.
> >
> > Robert, does this adequately reflect the concerns?
> 
> Yup,

OK, so concerns reflected.  What are we going to do?  It sounds like we should:
1. Have the build system produce a tarball of build artifacts.
2. Have a separate script that takes the tarball and a spec file and runs rpmbuild to produce an RPM.
3. Ditto for .deb.
4. Have Jenkins run those scripts, so that we have RPMs and .debs for every build.
5. Expect that the distros are going to ignore our packages and make their own.

Is that the summary of it?

Ewan.


Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Robert Schweikert <rj...@suse.com>.
On 08/09/2012 09:06 PM, David Nalley wrote:
> On Thu, Aug 9, 2012 at 8:47 PM, Ewan Mellor <Ew...@eu.citrix.com> wrote:
>>> -----Original Message-----
>>> From: Robert Schweikert [mailto:rjschwei@suse.com]
>>>
>>> [Snip]
>>>
>>>> * We want to be able to package CloudStack in RPMs and .debs that
>>> correctly depend on packages available on the target platform.
>>>
>>> This is IMHO not a "job" of the project. This is up to the packagers
>>> and
>>> package maintainers for the various distros. I will maintain openSUSE
>>> and SLE builds in OBS and would very much prefer that the build system
>>> know nothing about how to build an rpm. We should maintain a "reference
>>> spec file" in the source tree and I will contribute to that, but having
>>> the build system crank out a .deb or .rpm package is just not the best
>>> approach.
>>
>> David, could you reply to this?  You generally have opinions on how packages should be built.  I happen to disagree with the comment above, but I'm not a packager so there's a strong chance that I haven't got a clue what I'm talking about, and so I will do whatever people want in this regard.
>
>
> So I think this is a terminology issue, perhaps.
> My perspective (with packager hat on)
> So assuming we pick $tool to replace ant. I tend to agree - $tool
> should not build RPMs or .debs.
> There are tools to build RPMs and .debs - (rpmbuild and dpkg) and they
> should build the packages. They will call $tool to actually build the
> software, but we ought not get into the loop of $tool calling
> rpmbuild/dpkg calling $tool.
> Also - $tool should have an option for turning dependency resolution
> off. (Maven certainly has this, and I am sure other tools do as well.)
> Packagers don't want dependency resolution - the guidelines they
> operate under don't permit bundled or automagically downloaded
> dependencies. Instead those need to be packaged independently and
> listed as a requirement (or build requirement) for the package. Any
> sane $tool will handle this.
>
> This doesn't mean that we wouldn't/couldn't automate RPM/deb builds in
> something like jenkins. We just don't want another debacle like waf.
> Those packages typically will not be as good as distro-maintained
> packages in my experience. The distro also won't be kicking out RPMs
> for every iteration either, so it's a mixed bag.
>
> Robert, does this adequately reflect the concerns?

Yup,

Thanks,
Robert



-- 
Robert Schweikert                           MAY THE SOURCE BE WITH YOU
SUSE-IBM Software Integration Center                   LINUX
Tech Lead
rjschwei@suse.com
rschweik@ca.ibm.com
781-464-8147

Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by David Nalley <da...@gnsa.us>.
On Thu, Aug 9, 2012 at 8:47 PM, Ewan Mellor <Ew...@eu.citrix.com> wrote:
>> -----Original Message-----
>> From: Robert Schweikert [mailto:rjschwei@suse.com]
>>
>> [Snip]
>>
>> > * We want to be able to package CloudStack in RPMs and .debs that
>> correctly depend on packages available on the target platform.
>>
>> This is IMHO not a "job" of the project. This is up to the packagers
>> and
>> package maintainers for the various distros. I will maintain openSUSE
>> and SLE builds in OBS and would very much prefer that the build system
>> know nothing about how to build an rpm. We should maintain a "reference
>> spec file" in the source tree and I will contribute to that, but having
>> the build system crank out a .deb or .rpm package is just not the best
>> approach.
>
> David, could you reply to this?  You generally have opinions on how packages should be built.  I happen to disagree with the comment above, but I'm not a packager so there's a strong chance that I haven't got a clue what I'm talking about, and so I will do whatever people want in this regard.


So I think this is a terminology issue, perhaps.
My perspective (with packager hat on)
So assuming we pick $tool to replace ant. I tend to agree - $tool
should not build RPMs or .debs.
There are tools to build RPMs and .debs - (rpmbuild and dpkg) and they
should build the packages. They will call $tool to actually build the
software, but we ought not get into the loop of $tool calling
rpmbuild/dpkg calling $tool.
Also - $tool should have an option for turning dependency resolution
off. (Maven certainly has this, and I am sure other tools do as well.)
Packagers don't want dependency resolution - the guidelines they
operate under don't permit bundled or automagically downloaded
dependencies. Instead those need to be packaged independently and
listed as a requirement (or build requirement) for the package. Any
sane $tool will handle this.

This doesn't mean that we wouldn't/couldn't automate RPM/deb builds in
something like jenkins. We just don't want another debacle like waf.
Those packages typically will not be as good as distro-maintained
packages in my experience. The distro also won't be kicking out RPMs
for every iteration either, so it's a mixed bag.

Robert, does this adequately reflect the concerns?

--David

RE: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Ewan Mellor <Ew...@eu.citrix.com>.
> -----Original Message-----
> From: Robert Schweikert [mailto:rjschwei@suse.com]
>
> [Snip]
>
> > * We want to be able to package CloudStack in RPMs and .debs that
> correctly depend on packages available on the target platform.
> 
> This is IMHO not a "job" of the project. This is up to the packagers
> and
> package maintainers for the various distros. I will maintain openSUSE
> and SLE builds in OBS and would very much prefer that the build system
> know nothing about how to build an rpm. We should maintain a "reference
> spec file" in the source tree and I will contribute to that, but having
> the build system crank out a .deb or .rpm package is just not the best
> approach.

David, could you reply to this?  You generally have opinions on how packages should be built.  I happen to disagree with the comment above, but I'm not a packager so there's a strong chance that I haven't got a clue what I'm talking about, and so I will do whatever people want in this regard.

Thanks,

Ewan.


Re: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Robert Schweikert <rj...@suse.com>.
On 08/08/2012 06:30 PM, Ewan Mellor wrote:
> We need to push this discussion on build systems to a conclusion.  Let me summarize so far.
>
> * We need to be able to build CloudStack in various configurations (support for the range of hypervisors turned on or off, etc).

Yes

>
> * We would like the build tool to be able to get the necessary dependencies automatically, to avoid the user having to download a dozen libraries from all sorts of different places.

Not necessarily. This cannot be a "hard coded" "feature" of the build 
tool. For package build systems used by distributions network access is 
often not available, thus automatic download of dependencies will fail. 
Whatever tool is chosen must be able to handle the "do not download 
anything" case.

>
> * Many of our dependencies are version-dependent, so we would want to get a specific version or at the very least have greater-than constraints on the version that is downloaded.
>
> * We want to be able to package CloudStack in RPMs and .debs that correctly depend on packages available on the target platform.

This is IMHO not a "job" of the project. This is up to the packagers and 
package maintainers for the various distros. I will maintain openSUSE 
and SLE builds in OBS and would very much prefer that the build system 
know nothing about how to build an rpm. We should maintain a "reference 
spec file" in the source tree and I will contribute to that, but having 
the build system crank out a .deb or .rpm package is just not the best 
approach.

My $0.02
Robert


-- 
Robert Schweikert                           MAY THE SOURCE BE WITH YOU
SUSE-IBM Software Integration Center                   LINUX
Tech Lead
rjschwei@suse.com
rschweik@ca.ibm.com
781-464-8147

RE: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Ewan Mellor <Ew...@eu.citrix.com>.
> -----Original Message-----
> From: Alex Huang [mailto:Alex.Huang@citrix.com]
> Sent: Wednesday, August 08, 2012 4:19 PM
> To: cloudstack-dev@incubator.apache.org
> Subject: RE: Build systems [was RE: [DISCUSS] Binaries (jars) in our
> source tree/source releases.]
> 
> 
> 
> > -----Original Message-----
> > From: Ewan Mellor [mailto:Ewan.Mellor@eu.citrix.com]
> > Sent: Wednesday, August 08, 2012 3:31 PM
> > To: cloudstack-dev@incubator.apache.org
> > Subject: Build systems [was RE: [DISCUSS] Binaries (jars) in our
> source
> > tree/source releases.]
> >
> > We need to push this discussion on build systems to a conclusion.
> Let me
> > summarize so far.
> >
> > * We need to be able to build CloudStack in various configurations
> (support
> > for the range of hypervisors turned on or off, etc).
> >
> > * We would like the build tool to be able to get the necessary
> dependencies
> > automatically, to avoid the user having to download a dozen libraries
> from all
> > sorts of different places.
> >
> > * Many of our dependencies are version-dependent, so we would want to
> > get a specific version or at the very least have greater-than
> constraints on the
> > version that is downloaded.
> >
> > * We want to be able to package CloudStack in RPMs and .debs that
> correctly
> > depend on packages available on the target platform.
> >
> >
> > Gradle has been proposed: http://www.gradle.org/.  No-one seems to
> know
> > much about it, other than the fact that Hibernate and Spring have
> switched
> > to using it, so it can't be too bad.
> >
> > Maven has been proposed: http://maven.apache.org.  We have some
> > expertise offered from Brett Porter (a committer there).  It's an
> Apache
> > project, and it's very widely used.  We don't know much about it
> though,
> > outside of Brett.  I think that there's a legitimate criticism that
> it makes it
> > difficult for packagers, because it hides the system dependencies.
> That
> > might not be reason enough not to use it though.  I don't know if
> Gradle
> > suffers from the same problem.
> >
> > Scripting it all ourselves with Ant is a possibility.  David started
> this effort on
> > the deps-ctrl branch.  This puts us in control of our own destiny,
> and means
> > that we don't have to learn a new tool (or fit in with the way that
> the tool
> > works).  It does sound like we're just going to end up re-inventing
> half of
> > Maven that way though.
> >
> >
> > For me, it comes down to developer resources.  We need people who
> care
> > about this topic to volunteer some time to get the build working
> smoothly.  If
> > those people want to use Maven or Gradle, I don't think anyone else
> is going
> > to object too strongly, because no-one seems to know very much about
> > those tools.  At the least, we get the opportunity to try one of
> them, and see
> > if we like it.  Nothing is ever set in stone, and if we decide to
> rewrite the build
> > system in time for the 4.1 release, then we can do that.
> >
> > So that's the big question: is someone going to volunteer to get the
> build
> > system tamed for the 4.0 release?  It's a dirty job, but someone's
> got to do it,
> > and you'll earn significant kudos if you do!
> >
> Ivy has also been proposed.  It's also part of Apache and works within
> Ant and it uses the maven repository.  It is important that the build
> system doesn't tax the developer, such as making them setup in a
> certain way.

Yes, sorry, Ivy too.  Any volunteers to work on any of these tools?

Ewan.


RE: Build systems [was RE: [DISCUSS] Binaries (jars) in our source tree/source releases.]

Posted by Alex Huang <Al...@citrix.com>.

> -----Original Message-----
> From: Ewan Mellor [mailto:Ewan.Mellor@eu.citrix.com]
> Sent: Wednesday, August 08, 2012 3:31 PM
> To: cloudstack-dev@incubator.apache.org
> Subject: Build systems [was RE: [DISCUSS] Binaries (jars) in our source
> tree/source releases.]
> 
> We need to push this discussion on build systems to a conclusion.  Let me
> summarize so far.
> 
> * We need to be able to build CloudStack in various configurations (support
> for the range of hypervisors turned on or off, etc).
> 
> * We would like the build tool to be able to get the necessary dependencies
> automatically, to avoid the user having to download a dozen libraries from all
> sorts of different places.
> 
> * Many of our dependencies are version-dependent, so we would want to
> get a specific version or at the very least have greater-than constraints on the
> version that is downloaded.
> 
> * We want to be able to package CloudStack in RPMs and .debs that correctly
> depend on packages available on the target platform.
> 
> 
> Gradle has been proposed: http://www.gradle.org/.  No-one seems to know
> much about it, other than the fact that Hibernate and Spring have switched
> to using it, so it can't be too bad.
> 
> Maven has been proposed: http://maven.apache.org.  We have some
> expertise offered from Brett Porter (a committer there).  It's an Apache
> project, and it's very widely used.  We don't know much about it though,
> outside of Brett.  I think that there's a legitimate criticism that it makes it
> difficult for packagers, because it hides the system dependencies.  That
> might not be reason enough not to use it though.  I don't know if Gradle
> suffers from the same problem.
> 
> Scripting it all ourselves with Ant is a possibility.  David started this effort on
> the deps-ctrl branch.  This puts us in control of our own destiny, and means
> that we don't have to learn a new tool (or fit in with the way that the tool
> works).  It does sound like we're just going to end up re-inventing half of
> Maven that way though.
> 
> 
> For me, it comes down to developer resources.  We need people who care
> about this topic to volunteer some time to get the build working smoothly.  If
> those people want to use Maven or Gradle, I don't think anyone else is going
> to object too strongly, because no-one seems to know very much about
> those tools.  At the least, we get the opportunity to try one of them, and see
> if we like it.  Nothing is ever set in stone, and if we decide to rewrite the build
> system in time for the 4.1 release, then we can do that.
> 
> So that's the big question: is someone going to volunteer to get the build
> system tamed for the 4.0 release?  It's a dirty job, but someone's got to do it,
> and you'll earn significant kudos if you do!
> 
Ivy has also been proposed.  It's also part of Apache and works within Ant and it uses the maven repository.  It is important that the build system doesn't tax the developer, such as making them setup in a certain way.

--Alex