Productivity++ != Better


When it comes to productivity there seems to be an assumption that it’s a good thing, and that more equals better. I’d like to question that. What matters is producing the right stuff, whatever that is, and if you’re not sure, producing less would be better.

It can be hard to cancel work which is in progress or which has already been completed, even though, as software developers, version control can faithfully store and restore our work should we need it.

More tests does not equal better either. An increasing unit test count would seem uncontroversial, but let’s not forget we’re aiming for coverage and completeness whilst avoiding redundancy. Good unit tests and good design go hand in hand. Good unit tests avoid extended setup and good design reduces the need for combinatorial test suites. More system tests will slow down development, especially if they’re flaky.

I went to a wonderful and detailed talk at ACCU 2014 by Wojciech Seliga. He described his time as a development manager at Atlassian and the various steps taken to improve their CI builds: to speed them up and get them passing, that is. One particularly effective step was the daring elimination of an entire suite of fragile system tests. Late one evening, after a drink or two, a senior engineer just deleted them. Those tests haven’t failed since (!).

I wonder how that engineer felt. Nervous maybe, a bit reckless perhaps. But productive? Can destruction be productive?

On a similar note, one of the best changes made to the code base I work on has been to kill off unwanted configurations. We no longer support 32 bit builds: eliminating them has removed an entire class of integer conversion issues and slashed CI turnaround times. A reductive act, but one which allows us to produce more of what’s wanted.

What is wanted though?

The trouble is, if we’re not sure what to do, we like to do something, and we like to be seen to do something. We like to produce. We like to deliver. We don’t like to sit around in meetings talking about what to do. In these circumstances, though, production is questionable. More than once I have seen clean architecture crumble then decay, smothered by unwanted features.

Project dashboards can add to the problem. Graphics showing who added the most lines of code are open to misinterpretation and gaming. Who had the best ideas? Who reduced, removed or simplified? Does your dashboard display these metrics?

Remember, we are software developers, not software producers.