Month: September 2008

Calliper, chalks and the axe!

Years ago, when I was still doing physics university in São Paulo, a friend biochemist stated one of the biggest truths about physics: Physicist is the one that measures with a calliper, marks with chalk and cuts with an axe!.

I didn’t get it until I got through some courses that teaches how to use the mathematical tools available, extrapolate to the most infamous case, than expand in a series, take the first argument and prove the theorem. If you get the second argument, you’re doing fine physics (but floating point precision will screw up anyway).

Only recently I’ve learnt that some scientists are really doing a lot by following in the opposite direction. While most molecular dynamics simulation are going to the quantum level, taking ages to get to an averagely reasonable result (by quantum standards), some labs are actually beating them in speed and quality of results by focusing on software optimizations rather than going berzerk on the physical model.

It’s not like the infamous Russian pen (which is a hoax, by the way), it’s only the normal over-engineering that we normally see when people are trying to impress the rest of the world. The Russians themselves can do some pretty dumb simplifications like the cucumber picker or over-engineering like the Screw Drive that, in the end, created more problems than solved.

Very clear, in software development, the situation can be as bad as that. The complexity of over-designed interfaces or over-engineered libraries can render a project useless in a matter of months. Working around would increase the big ball of mud and re-writing from scratch would take a long time, not to mention include more bugs than it solves.

Things that I’ve recently seen as over-engineering were:

  • Immutable objects (as arguments or on polymorphic lists): When you build some objects and feed them to polymorphic immutable lists (when creating a bigger object, for instance) and then need to change that afterwards you have to copy, change and then write back.
    This is not only annoying but is utterly ineffective when the list is big (and thousands of objects need to be copied back and forth). The way out of it is to use the bridge pattern and create several (RW) implementations of your objects and lists and whatever you have but that also increases a lot on code complexity and maintenance.
    My view of the matter is: protect your stuff from other people, not from yourself. As in “Library Consistent” or “Package-wise Consistent”.
  • Abuse of “Standard algorithms“: Ok, one of the important concepts in software quality is the use of standards. I’ve written it myself, over and over. But, like water, using no standards will kill your project the same way as abusing of them.
    So, if you create a std::set that gives you the power of log(N) searches, why the heck you’d use std::find_if ( begin(), end(), MyComparator() );, that gives you linear searches? Worse, that find was actually before each and every insert! std::set guarantees at least N.log(N) speed on insertion, but the “standard fail-safe assurance” was giving it N².log(N). For what? To assure no duplicated entries were ever inserted in the set, what was yet another thing guaranteed by the default container in question.
    All in all, the programmer was only trying to follow the same pattern over the entire code. A noble cause, indeed.

Now, I’m still defining what’s worse: over-engineering or under-engineering… Funny, though, both have very similar effects on our lives…