Hacker News Reader

> few developers like to write tests

I like how this statement completely skips the reason why developers don't like to write tests. The majority of managers refuse to allocate any time to write any tests. Hell, when we come up with a 6-week estimate for a project, they then tell us we must do it in 4. At 4 weeks, the complaints and blame game begins over why the project isn't completed. Why? Because it was a 6 week project, morons.

I've only had one company with management that wanted unit tests written for code. Yet there was no official time dedicated to this process. That 6-week project that would be estimated without unit tests? Still 4 weeks. Huh? What magical world do you live in?

Quality requires time. That 6 week project now needs 8-9 weeks. Not 4. Not 5. Find me management that understands this and actually cares about quality instead of impossible deadlines that fit their fantasy plan for the current quarter, and we'll talk.

To be clear, I love writing tests. I would always have them if the decisions for project deadlines were left to me. Such decisions are out of my control, so the company gets what they are willing to pay for.

That's not the only reason. Writing tests is hard.

Writing tests that are not highly coupled to implementation is hard. That means that every time you change the system, you have one or many test changes to make too.

How do you know what to test and what not to test? How many tests do you write? One line of code can, if you think about it exhaustively, require 3 or even more tests.

A lot of time you'll write code that facilitates testing. Do you test this code? What happens if the code silently fails and now you don't have proper test coverage of all the code paths said code is supporting? This is particularly true if you are mocking objects.

Learning how to test, properly, is almost as hard as learning how to code. My advice is, if what you're doing is boring, like most Rails apps, lean on the library as much as possible, break up anything that goes against the grain into unitary features, and test them each in turn. You don't need to test that ActiveRecord.belongs_to works.

It doesn't surprise me that there's developer reticence towards testing. I tried powering through that reticence, believing there's some "magical world" as you put it, where testing actually helps you code faster and more accurately, but I found even myself a TDD apostate, I don't even do it for my own projects anymore.

I agree. Nothing is as easy as it seems. There is a whole reason why someone is hired to do QA, and why there are research in software testing.

Technically these issues are solved with what I indicated is always missing: time. What you outlined is accurate: proper testing also requires proper planning for tests. For more complicate projects, you don't just start writing them without mapping out everything that needs to be done - and how to go about it. And yes, when code changes, more time must be invested into refactoring the tests.

This disastrous ecosystem exists with QA too (which is again, a form of testing). A lot of companies - usually the smaller ones - don't invest any resources at all for QA. Others just do point-and-click smokescreen tests that'll discover bugs that are blatantly obvious (also typically the ones the developer should never have committed to begin with). Real, proper QA? That requires investing serious time into test cases - completely figuring out every possible interaction and scenario that can take place. All before a single actual test is run. And similar to unit/functional tests, QA test cases need to be reviewed whenever the smallest bit of functionality changes.

The reality is that most employers who "want test cases" want there to be a magical solution that gives all the benefits, without any effort.

Maybe they actually believed the many developers who told them writing tests actually saves time?

Or the employer heard what they wanted to hear.

Employer, start of project: "Oh, it'll save time? Great!" Employer, end of project: "You said it'd save time, lies!" Developer: "I specifically explained how it saves time in the long run. Not the initial implementation." Employer: "..."

If you know your estimates are going to be squeezed by two weeks, add that in and say it'll take eight.

"It's a leader's job to get the team to understand why writing tests is important and have them write them."

And sometimes you shouldn't write tests, or perhaps no unit tests, or perhaps write the tests but throw most of them away, etc etc. It is all highly context specific.

The real issue for managers is even if they start out technical, within a year or two they lose touch.

Good discussion of this sort of thing in a series of recorded Hangouts between Martin Fowler, Kent Beck and DHH:

http://martinfowler.com/articles/is-tdd-dead/

The takeaway is that the usefulness of tests is the feedback it gives you about your code (and confidence, a subtype or result of feedback). All of the BDD vs. TDD vs. Test First vs. Test Whenever can get wrapped up in too much zealotry, and is ultimately secondary to the goal of getting feedback.

Sure. I don't disagree about not writing tests at times, etc. That's different than the point I'm trying to make. Leaders aren't necessarily tasked with doing any job a developer thinks is shit. The leaders job is to help the developer understand the why behind doing it.

Also, do you have anything quantitative to back your statement about being out of touch? That hasn't been my experience generally and I work quite hard to ensure I don't lose touch. That's obviously only data on my experiences - but I didn't make the assertion (I genuinely interested though).

Another recent engineer turned manager here. I can attest to losing touch, as I've had to work extra hard to see where things are trending in my spare time. If development wasn't also a personal hobby, this would be exceedingly difficult.

Fundamental shifts occur in development all the time, and if you're not actively coding while these are going on, you can definitely get left behind and be out of your element when it comes to decision making.

I have peers that cut their teeth on Enterprise Java and Oracle and now they are managing engineers doing Native Mobile, JavaScript Apps, Ruby on Rails, DevOps, Distributed Databases, and deploying on public clouds like AWS.

It's very easy for them to get lost in the jargon and they end up leaning on and trusting senior engineers to guide them a lot.

While this can be a good thing, it's very easy for the wrong senior engineer to lead them astray and start pushing their own agenda (rewrite all the things, resume driven development, etc).

> While this can be a good thing, it's very easy for the wrong senior engineer to lead them astray and start pushing their own agenda (rewrite all the things, resume driven development, etc).

This is a very good point, but how does an organization do this? Who can be at a management level for the majority of their time, yet still remain technically sharp enough in multiple technologies over time, to know when someone is outright lying to them? And it's not like you can just ask them to walk you through it - even ignoring the significant minority of egomaniac technical people who will more or less outright refuse such a request, in many cases even if they are relatively cooperative, often the subject matter is simply too low level or complex to be explained (or is deliberately explained in this manner), and sometimes the disagreement is not even a matter of fact, but rather a matter of opinion (everything XML, language choice, 100% test coverage, etc). It's a tough problem.

It's a good question, and a difficult one for me to answer as I'm still trying to figure it out.

One possible defense against this is to regularly refresh talent, and encourage high performing individuals to swap between contributor and manager roles without any career penalties in between stints.

Even if it doesn't work out for a contributor to manage people or a manager to write code, the cross pollination of talent could strengthen the technical capabilities across the organization. Contributors and managers would build trust with each other, and ultimately empathize enough with each other to act as trusted partners.

I've seen this play out in the startup ecosystem as you see VPEs leave to become early stage startup CTOs, get their hands dirty to launch ideas, and then ultimately bring in trusted contributors once the products have found product-market fit.

"Also, do you have anything quantitative to back your statement about being out of touch? That hasn't been my experience generally and I work quite hard to ensure I don't lose touch. That's obviously only data on my experiences - but I didn't make the assertion (I genuinely interested though)."

How do you know you haven't already lost touch? :p I joke, but the first time I switched into full time "architect" mode I did lose touch without even realising it.

I guess the key measure would be: could you jump into a senior developer role on your current project and be as productive as the other senior devs on the team? In my experience most team leads can, but very few managers can.

I doubt there is much in the way of hard data either way. I've personally met less than a handful of architects/managers that were clued in over my 15+ year career.

It's like looking at code written by a python developer who, a couple of years ago, used to be a c developer. The developer still understands memory management, but it's no longer forced to the front of their minds. And it shows in the code they write. Likewise because as a a manager you aren't forced to deal with the code your brain quickly abstracts it away. Your "atomic operations" when you think about a problem become a lot larger.

Your second comment, the one about autonomy, is the real insight that I wish more developers and managers would get. I almost quit reading the article upon finding the part you quoted. Thanks for pointing it out here!

Way too often, real world discussions about management in teams turn into discussions about technical detail ("If we only had tests / quicker builds / used language X then we'd get better results") and this basic, underlaying problem of room for autonomy in the process is overlooked in favour of topics that are more easily discussed.

The main reason I added this (mostly useless) comment was that all replies so far had followed the same pattern of only discussing the issue closest to technology; the tests.

Edit: removed duplicated word.

> Not all of the tests, necessarily, but enough to set an example, set the standard, and get the ball rolling.

I think you stopped your quote too soon. Writing a couple tests sets a pattern, an example o follow, and demonstrates how important they are because you are willing to take time to do them to begin with.

But just giving lip service to the importance of them does not change a culture of untested code.

Setting up code coverage reports, test harnesses, a few examples, and then talking about the test metrics shows it's important to management.