Yet another software guy's blog

I am not sure how many people care about this anymore but UML still is a very effective way to communicate with other developers when the architecture is complex enough. Particularly when it comes to sequences of events that need to be described, one sequence diagram seems to make a big difference.

Recently, I was writing a document regarding a particular architecture that needed several class diagrams/sequence diagrams et al. I started out using “Enterprise architect”  which is okay but it is way too heavy weight and it comes with a lots of bells and whistles. To top it, at the end of it all, it generates crapload of stuff that dont contribute materially to my actual documentation of the architecture. And, It needed me to have windows box around to put that together. The next option was running visio or OmniGraffle but for both of them, it was way too much work to make the sequences to appear exactly the way I wanted it to show up.

So, I started looking for tools and I came across PlantUML. Between PlantUML and GraphViz, I can now do everything I want staying within vim. I  prefer using text based tools to generate images since it is easier to keep it up to date and archive and  edit with Vim. And PlantUML, is perfect for that.

It generates sequence diagrams without anything else and since it is a java application, it actually works on a mac like a charm. If you want to generate class diagrams and others, you need to install Graphviz. Note that on Mountain Lion, I had to install libtool before I could successfully run it but it was well worth it.

In PlantUML, it is extremely easy to create sequence diagram.

@startuml
Alice -> Bob: Authentication Request
Bob --> Alice: Authentication Response
@enduml

generates this image.

And it has so many plugins, It is amazing.

Syntax is simple enough and the documentation on the site is very easy to follow. Now, I can do my documentation within Vim between markdown and PlantUML/GraphViz.

I am one happy architect..

I have been wondering about copying classes (or cut/paste inheritance)  vs.Modularity frameworks such as OSGI.

To state the problem succinctly: If you have many modules in OSGI which provide independent services, what kind of reuse do you expect with utility classes? or specific functions that might be within some classes?

• Do you export the package from whichever bundle that has it to be usable anyone?
• Do you build a grab-all utility bundle with unrelated classes just so you can export them?
• Do you copy the classes that need to be reused so that modularity of the bundle is maintained with a “Shared nothing” approach?
• Do you build application specific utility bundles?

I get the concerns with editor inheritance.

(For those who want to know) Cut and Paste:

• Breaks DRY principle violently
• Potentially provides subtle but different behaviors within making it hard to maintain the code
• Goes against the grain for any self respecting software engineers

Instead you should

• Build a class hierarchy to share functionality
• Or better still, rethink why you need this look at the design patterns to provide you an optimization that would allow your to make sure that you are using the behavior you want.

Ok, I get that. This provides a ‘reuse’ of the behavior coded within a class or a function.

Creating a catchall utility bundle becomes a unwieldy mess all by itself with its own problems of dependency management and version management et al.

Exporting the packages directly to allow direct access to utility bundles seems to suffer the same issues of dependency and version management. Worse still, now we have arbitrary  module dependencies created and maintained just for utility classes.

The more I think about it, The question I am struggling with is: When do reuse concerns  outweigh modularity concerns? Should they ever? 

I think Reuse in the large, at the level of services provided by modules, should trump any reuse in the small, which occur by breaking the modularity concerns. But, I am not able to substantiate it with any good examples that would confirm or refute it. So, it is still an opinion that i am trying to figure out if anyone else shares the same views.

What do you think? Drop a comment.

BTW: I recently started blogging earnestly so comment on the style if you like. It is a bit of stream of consciousness and hopefully I improve over time.

I have been thinking about this quite a bit for the past few days. When you consider what an architect does in any established software company (look up Software architect), they seem to have the following characteristics.

1. They are able to work with stakeholders to generate the correct requirement definitions
2. They are able to generate design documents that will be fit in the overall architecture of the system.
3. They are able to (maybe) prototype their design.

This seems to make the architect part of a bygone era when the flow was clearly defined with no proper feedback.

Now there is a second school of thought which seems to re-map the architect to be a stand-in for the product owner (An Architect’s Place in Agile (randomactsofarchitecture.com).

This seems to indicate that architects are more or less trail blazers for the team which forages the forest ahead to come up with an incremental design approach that would address the next sprint’s stories.

I still think that there is a middle ground between those schools where they can be the change agents who instigate good patterns/tools/processes based on their experience and provide guidance of the same to other arms of the enterprise. But, they should still be able to dive into the local domain specific issues with code to show the same in action.

Now, that makes me wonder, Is an agile architect oxymoron? I think not. I think it depends on the viewpoint of the team that is observing the architect’s behavior. Within the team, s/he is an developer with deeper technical /domain expertise with non functional interests being considered and without, a technical product owner who can represent those non functional interests.

Well, I am still unconvinced though. Is there any need for an architect if you have really good developer with soft skills to represent the other interests or a truly technical product manager who can balance business requirements with non functional ones…

Still wondering..

That said, I am a senior software architect and I do play these roles (hopefully well) …