Developing Software in the 21st Century, versus using Maven (and Git)


Maven is case of something written by a bunch of past-their-sell-by-date c-shell script kiddies who think autoconf is leading edge code automation and don’t understand that object code requires an object environment to be in any way efficient, either for development or deployment. Ant is bad enough, but Maven combines all the worst features of Ant and Ivy. It doesn’t create an object environment, and it doesn’t play well with tools that do.  On top of that it aids in massive code bloat in Java app deployments by assisting the developer in using multiple versions of libraries rather than taking the small amount of time necessary to update code, or doing without whatever whiz-bang feature in the new library  if it isn’t whiz-bang enough to justify taking that time.

Simply, an object environment should have all class objects, i.e. the objects that determine the types of objects available in the system, live and available in the system at all times, whether for development or testing, and any deployment should be the same, with non-referenced development/test only objects removed.  From there I can do whatever I want as a developer: instantiate multiple objects of a class, set up various sequences and instantiation rules, etc. etc. Since the environment should be completely live, I shouldn’t need a build tool at all – as soon as I save it the project should be built and running.

In terms of deploying my app, it’s not difficult for the environment to simply throw out all the class objects that are never referenced by code in the namespaces that make up my app. The garbage collector in the JVM does almost the same thing on the fly today, as do all Smalltalk environments except for the bastardized GNU version that shouldn’t qualify as a real Smalltalk anyway. At that point I have a deployment environment made up of my objects and all the objects (primarily class objects) that my objects reference, i.e. my application and all dependencies. This is how virtual machines work. (that our VMs are so poorly written we need to run a Spring VM on a Java VM on a Linux VM on a VMWare VM on another Linux VM is another example of the idiocy of current software development and deployment paradigms). When dependencies get updated, it’s simple enough for the environment to prompt the developer to merge his old code to the new libs, merge the code using the new libs down to the older version, or keep both versions. Prompting encourages the developer to make the slight modifications that are sometimes necessary to avoid having twenty versions of every library, while tools like Maven hide the fact that you have twenty versions and results in the massive runtime bloat common in Java apps.

In the Java development space Eclipse comes closest to being a proper object environment, although granted there are plenty of plugins that break the paradigm in various ways. Most of the reasons given for using Maven fall apart when examined critically.

Netbeans and Idea are overblown text editors, not object environments, but if you do want to use their tools for something not covered by the thousands of Eclipse plugins, both can import and maintain Eclipse projects, your build will just be inordinately slow compared to developers using Eclipse, but then, they’d be that slow if they were pure Netbeans or Idea projects anyway.

Not a serious reason to use Maven.

The ease of exporting / importing settings in Eclipse (something every team should do in any IDE in any case) makes the different settings problem nothing more than laziness on the part of the development team (or a religious argument over spaces vs tabs, lol).

Again, not a serious reason to use Maven.

Team environment? Show me a team that doesn’t already use a repository like GIT or SVN. Why do we need to duplicate both the functionality and the maintenance headache by setting up Nexus repos as well?

That one’s actually a good reason NOT to use Maven.

Running a server build? Great idea, now, shouldn’t that be kicked off by code that’s actually checked in to the source repo rather than a random build that happens to get pushed to Nexus? This brings up a point against Git, particularly Git with Maven. Since in Git I don’t work on a branch, test locally, then commit (partly because my local test doesn’t prove the server build works due to differences in the Maven configuration in Jenkins and Eclipse) I have to commit my changes to a different branch in order to see that the server Maven build fails, then commit a further change to fix the problem, resulting in an unreadable source history in the repo. Checked in code should at the very least build and pass unit tests, which if Git and Maven were out of the picture should be guaranteed.

Exporting a headless build from Eclipse is trivial if you actually look into it – all you need is ant or Gradle, the developer build already maintained by Eclipse, and a few Eclipse jars (Eclipse will export all the necessary files for a headless build to a directory or zip file, or ftp them to the build server). Server build tools like Hudson/Jenkins can pull updated code from most source repos and call any build script, there’s no dependency on Maven. With Maven you either force developers to use a tool not suited to anybody but build engineers (the magnitudes longer it takes to build, even using M2E, is sufficient for that case to be made), or you live with the possibility that the server build doesn’t work quite like the workstation build, which is still true if you go through all the hassle of integrating the two using the plethora of M2E plugins. Either way you get a slower and more fragile workstation build for the sake of an equally slow and more fragile server build. On every Maven based project I’ve worked on I’ve seen transient Hudson/Jenkins errors that don’t show up in Eclipse unless you have absolutely every possible M2E plugin installed and correctly configured, and most developers never do.

Seems like another great reason to avoid Maven.

That doesn’t cover some of the most fundamental problems with Maven, such as:

  1. its namespaces breaking Java namespaces and XML namespaces,
  2. it’s build unit (the POM) having no relation to anything in the actual deployment environment (think about it, when you separate via POMs what are you actually accomplishing in the finished product? Nothing. All it accomplishes is a false sense that you’ve separated concerns and functionality into different build units that all run as one monolithic piece of code);
  3. the hassle of manually maintaining complex configuration files, which only gets worse if you happen to need to use OSGi or another container and have to maintain other config files that affect and are affected by the Maven config with very little obvious sense to it;
  4. the problems caused by trying to run unit tests without a full environment for the code to execute in;
  5. the myriad versions not only of dependencies but of Maven specific plugins (I’ve actually seen JAR hell in the Maven build itself where multiple Maven plugins were using conflicting dependencies – one of the problems Maven was supposed to solve).

Yes, you can build object code with Maven. You can also write pure object code in C or even assembler, but I don’t know why you’d want to.

The best reason to avoid Maven is the phenomenal amount of work required to de-mavenize a set of projects when you get sick of all the problems noted above (and numerous others not mentioned).

The mindset, inherited from C development, that the development cycle consists of write code, compile, assemble, build, deploy, test, do over again, is hopelessly outdated in an object environment. At some point we need to tell all the people with this mindset that they need to relearn how to develop, period. Doing so would remove any need for Maven, Git, and a host of other tools that do nothing but waste time.

Object development should be done in a live object environment, where a code change is tested as it is saved since the modified object is live. Deployment should consist of removing development only artefacts from that environment, creating a runtime that has everything used by the running app in development and test.

I’m currently dealing with a problem caused by creating deployment assemblies for an OSGi app using the maven-assembly plugin. The app works perfectly in the Eclipse environment, which hot deploys all code changes into a running OSGi container within the environment. However the configuration doesn’t survive intact through the maven-assembly process, despite having a very good configuration/build engineer whose sole job is to accomplish that process. If we got rid of Maven (very difficult now due to the amount of code, but possible) and used the BNDTOOLS Eclipse plugin we could simply export the Eclipse build as an Ant or Gradle headless build (note, the OSGi developers who write BND and BNDTOOLS don’t support Maven, and for good reason, the Maven plugin is written by the Felix developers who themselves use Netbeans and Maven, and no live environment other than at the end of the deploy cycle), where both tools set up the same environment as Eclipse, without the GUI objects that are only meant for developers anyway. The result would be an identical configuration and build for deployment. This would easily save 2-3 hours per day per developer currently spent watching slow Maven or M2E builds, and free up the config/build engineer to do more testing of the app on the deployment hosts.

Getting over the mindset of write/compile/assemble/build/deploy/test is the only major impediment. Pretending you’re coding on a 1979 VT100 terminal instead of a modern machine doesn’t make you a ‘real’ developer, it just demonstrates that your methods are 35 years out of date.

Of the developers on the team, none of the others properly understands a live object environment like Eclipse sufficiently to get it to work as a live environment with M2E and OSGi, and they are top developers, they just haven’t been exposed to it due to the prevalence of outdated command line development tools. They only realized it was possible to do so when we were pair programming to solve the configuration problem and I was sharing my screen, causing one of the other team members to exclaim “that’s how you write code so damn fast”, when he saw my code change instantly test itself in the background OSGi container.

Btw I can use a bash shell when I have to, such as when I’m looking at logs on a remote server, in fact I do so fairly efficiently precisely so I can get out of that environment as quickly as possible and return to the 21st century.

Advertisements

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s