2012-03-01

Difficult decision

Considering the number of hits this post continues to have on a weekly basis, I should point out that a follow-up post can be found at A Difficult Decision - 2.

I like Go as a language. It is a rare, good balance between simplicity, power and expressiveness. I was so impressed with it as to begin the development of the next version of my chemistry product in it. There were minor hiccups, but I made progress, and wrote as many as about 12,000 lines of code (includes comments).

Then, I paused.

Now, about two months later, I have decided to shift the development of the new version of my product to … Java. If that sounds anti-climactic, so it is!

There are two significant reasons for this decision.

Plug-ins

The intended users of my product (and its API) are organic chemists and cheminformatics programmers. In practice, several organic chemists neither program themselves nor have programming assistants.

My product has several basic capabilities built into it. These are exposed as callable or configurable algorithms. What I also want to do is enable the user to extend the capabilities of the product through plug-ins.

In the Java land, this is well-studied problem, with more than one established way of solving. In the Go scenario, on the other hand, this is not straight-forward. Go programs are linked statically. There are requisites to enable plug-ins.

What would I need to do?

  1. Supply the statically-linked full product.
  2. Supply *.a files of my product's packages, and *.o files of the driver part.

What would my customer need to do?

  1. Install gcc and family and their dependencies. [Once; update: not needed after Go1. Thanks Andrew Gerrand!]
  2. Install (and maintain) the Go development environment. [Once, at least after Go1.]
  3. Download the plug-in package file .a, and place it in a designated directory.
  4. Build and install the application.[1]

Why will this not work?

It could, with programmers. It, unfortunately, will not, with chemists. For many chemists, the exertion of supplying input to a computer program and reading its output using a suitable viewer, is a considerable condescension. Should I ask them to compile code, and that too for each plug-in, typical chemists will laugh me out of court!

The API Consumption Issue

Most foot soldier programmers at pharmaceutical companies, biotechnology companies and informatics services providers have graduated in the last ten years, or so. Most of them learned Java as their first language, and have used it the most. Naturally, they tend to gravitate towards solutions with a published Java API.

Chemistry product companies acknowledge this. Almost all of them provide a Java API for their products and libraries.[2] Those that have remained with a non-Java API are certainly not the leaders today!

Therefore, for me to appeal to those programmers, I must provide a fully capable Java API. My prospects of commercial success depend on that factor in no minor way.

Conclusion

Accordingly, I have shifted the product to Java. My team and I have begun re-designing and re-implementing the legacy product — this time to suit a more Java-like style; to appeal to Java audience!



[1] goinstall does not help, since it needs source code to be available. Plug-in authors may not be willing to publish source code.

[2] If they cannot provide a proper Java API, they at least provide a JNI wrapper.

Post a Comment