All,
I am not sold on maven at this point. Don't get me wrong I am a strong
advocate of Maven, and although it has its issues, I think it can be a big
benefit to a project. I am just not sold that it is worth the effort at the
moment. That being said here are a couple ways which maven could help the
wave in a box project:
Modularization
Maven encourages the construction of distinct modules for a project. As
the wave project grows, we will likely want to build a set of reusable
components that other projects could use. For example we may have
concurrency control stacks, protocol stacks, interface components, etc that
other projects might want to take advantage of. We have already gotten
request to use the editor, and we know people are interested in building new
UI's leveraging the client server protocol. With maven, the project can
easily be organized such that these modules are reference-able and usable by
others. Things like the Robot and Gadget API are obvious examples.
This could be done without maven, but maven makes this the norm rather than
the exception.
Build Time
Right now we have a pretty monolithic build. When I change anything, I
have to rebuild the whole project. With maven it's fairly easy to go in to
the sub module you are working on and make "local" changes there. You can
rebuild and deploy easily without having to do a whole build. Of course you
can run a full build if you like.
Testing
Right now we have an issue with testing. The unit tests are also
monolithic. Originally, they all run or they all didn't. Unfortunately, it
took too long to run all of them, so we now have a somewhat arbitrary
distinction between long running and short running tests. Only the short
ones run by default.
With maven, unit tests are scoped to the module which they test. If we had
a Robot API module, we would have tests in that module that just unit test
the Robot API. If you are working on the Robot API, you can make changes
and easily just run the Robot API unit tests. You can actually work on the
sub module, make changes, test and check in code, without having to run ALL
the unit tests for the whole project, assuming the tests are comprehensive
for the module. The job of running all the tests can be the job of the
integration build.
Standardization
When I was ramping up on wave, it took me over a month to become fully
comfortable with the nested ant scripts and how the project ultimately built
itself. As ant projects try to become more modular, the complexity of their
build scripts goes exponentially up. Furthermore, each project does it
differently. You have to study the build to figure it out. With maven, its
all pretty standard. Any developer experienced in maven can check out your
project, rigure out the module layout and how to build any specific module
they need within minutes.
Other Random Things
Great IDE integration.
Dependency management.
A standard mechanism to deploy our artifacts for other projects to use.
Lot's of reports that can be generated about dependancies, tests, etc.
Integration with SVN and standard mechanism to perform releases, tagging
and branches.
In my opinion maven would help us:
Organize the project better.
Make our codebase more testable.
Shorten our development and test cycles.
Make our project more accessible to new developers.
Make our project more reusable for others.
All of that said... it's not going to be easy to switch over, so I don't
take the decision lightly.
~Michael
On Apr 27, 2011, at 7:31 PM, Daniel Danilatos wrote:
It seems the real problem here is the large amount of jars checked
into the repository. What about something that just does dependency
management, for example apache ivy?
Does maven solve any other problems we have that make it worth going
all the way with it? I'm worried it could cause as much trouble as it
solves. I'm not fundamentally against it but I'd like to be confident
there are solid reasons/benefits to the waveprotocol project for
migrating to maven.
My 2c.
Dan
Στις 27 Απριλίου 2011 7:51 π.μ., ο χρήστης Michael MacFadden
<[email protected]> έγραψε:
Nelson,
I am all for this. I am very experienced with Maven and the proper ways
to set up multi module projects including dependency management, etc. I
think your assessment of the situation is correct. We would ultimately need
a code free to get this done.
However, before you go to far, I would try to get a real discussion
going on if this is something the team wants to do. If so we need to
discuss what the right modules are, what artifact and group ids we would
want, etc. I would hate to see you set up a lot of poms.xml that don't line
up with how we would ultimately do things.
As a side note, if we did want to go to maven, now would be the right
time. Everything else has to be migrated anyway, might as well take the hit
now. However, I would recommend that we get the current source migrated
over first.
~Michael
On Apr 26, 2011, at 10:12 AM, Nelson Silva wrote:
Hi all,
There have been several discussions on adding maven as a build tool.
I've decided to start working on this by adding a tools/maven
subdirectory with a multimodule project layout.
I pushed my ongoing work to a branch in my GitHub clone :
https://github.com/nelsonsilva/wave-protocol/tree/maven/tools/maven
While a complete port to maven as the main build tool would required a
code freeze and a healthy discussion of code reorganization into proper
modules we should be able to use a simple includes/excludes approach for now
and in the end, if everyone is ok with it, we could easily make the port
happen.
I would like to get some help with this from everyone who's interested.
There are lots of plugins and lots of ways to acomplish things in maven so
this should get the discussion started.
I chose to go with GitHub so everyone can fork it and send me pull
requests. If you'd rather go with the issue/patchset approach I'll open the
issue but the idea here was to have several people contribute and not a
single patch + review process.
Quickstart:
cd tools/maven
./install_deps.sh # This will install a couple of dependencies that
were either patched or were not found in existing maven repos
mvn install
cd server
ln -s ../../../war
mvn exec:java # This should start the server, although you'll have to
symlink the war directory for now
TO DO:
- Use profiles for dev/prod
- Remove as many ant tasks as possible and use pure maven plugins
- Use the assembly plugin to package the server with the webclient
- Setup the surefire tests for each module
- Rethink the modules, we should have pure api modules and the several
impl modules for persistence, websockets, auth, etc ...
- etc....
Looking forward to getting this going...
Regards,
Nelson Silva
