Hacker News Reader

Comment by bww | Hacker News Reader

It’s worth remembering that engineers don’t get paid to write tests, they get paid to produce software that supports some business need. In most circumstances, some amount of automated testing makes it faster to reliably meet those business needs.

If you’re building a lot of small, one-off tools for internal use, it may well be the case that some limited manual QA or UAT is sufficient to ensure that your work is good enough.

If you’re working on larger, more complex projects that are frequently updated, the shorter feedback loop that some amount of automated tests provide will probably save time and money by catching problems earlier, avoiding regressions, and reducing the need for repetitive, time-intensive manual testing.

But in any case, your testing needs will always be highly specific to the actual nature and needs of the project.

View on HN

> engineers don’t get paid to write tests, they get paid to produce software

I've heard this double-standard before, and it sounds plausible at first glance, but is quite demonstrably nonsense.

1) Automated tests are software

2) Engineers aren't paid to produce software (nobody would pay me to auto-generate a billion "hello world" programs per second). Engineers are paid to solve problems and/or create value. Tests solve problems (finding bugs, demonstrate usage, clarify thinking, document intent, etc.).

3) Software is a liability, not an asset. We should produce as little as needed, and delete it if possible.

4) Engineers aren't paid to perform manual testing (running commands over and over, clicking around web forms to see what happens, etc.)

I find #4, exploratory testing far more bang for buck than automated testing. I would argue engineers are paid to make services (software is a service) and exploratory testing is far more valuable towards that goal than automated tests.

a) users do crazy things. b) eating your own dogfood is worth it.

If by "exploratory testing" you mean things like clicking around on a site manually, then sure it can be useful to spend a couple of minutes looking for anything dodgy.

That doesn't mean engineers are paid to go through a list of regressions and edge-cases, look for broken links, double-check calculations with pen-and-paper, etc. All of that can be done across an entire site automatically in seconds; there's no excuse not to.

> a) users do crazy things

Again, thinking we can check such things better than a computer is misguided. Test frameworks like QuickCheck can stress-test a system with far more strange inputs/transitions than I could hope to think of.

Totally agree, any repeatable action can/should be automated. It hurts to see software engineers clicking login to test it still works.... Every day

If it's repeatable it isn't exploratory testing. The point of exploratory testing is to explore, not repeat ovee and over.

QuickCheck is a wonderful tool. Unfortunately most engineers go and write manual unit tests. But even with QuickCheck it is dumb to to test at the function level. Better to blackbox fuzz test.

I am dumb -- I test at the function level. I have a theory that the likelihood of catching a bug depends on the number of lines being tested, if each can be assumed to do its job. This implies that if a line calls a function that I wrote, I need to test that function. There is some work based on Kolmogorov complexity that seems to support this but didn't go very far (I did not see any follow up papers). Maybe it's all wrong. I am biased: I wrote quickcheck for R. (by the way, fuzz testing and randomized testing are two separate things).

> 1) Automated tests are software

Which is assumed to be free of defects because people are suckers for circular reasoning.

What are you talking about? The whole point of automated testing is to confront the empirical fact that all software tends to have bugs, so we shouldn't have much confidence that we got it right.

We can increase our confidence by trying a few examples, but only a little, since the scope for failure is so huge.

Hammering a system with thousands of edge-cases, interleavings, previous regressions, etc. can give us a bit more confidence; but takes a huge amount of time for a person to do manually. It's pretty easy to automate though, and there are a load of libraries and frameworks that make it even easier.

If we want even more confidence in a system, we can verify it against a machine-checkable specification. That's still too hard to automate in most cases.

In any particular situation we need to find an appropriate balance between the cost of increasing our confidence in the system (i.e. how much effort we're willing to spend on prove ourselves wrong) versus the cost of failure.

Automated tests (especially property checks) are so low-cost and high-yield that I find them always worth doing, unless the consequence of failure is near-zero.

What logically follows from the two statements below:

1. "Automated tests are software" 2. "all software tends to have bugs"

It follows that automated tests tend to have bugs. So what? That doesn't at all imply that we shouldn't use them.

You're presumably using a whole heap of software to post your comments, all of which will tend to have bugs; yet you still find it useful.

Boolean logic rarely captures the nuances of real-world situations, and the "rounding errors" accumulate at each step of such arguments, which can lead to some silly (but logically consistent) outcomes. I think you've fallen victim to this here.

In a similar way, it's a false dichotomy to think of software as being either buggy xor perfect; of testing (or any other QA) as proving brokenness xor correctness; or to consider bugginess as implying uselessness.

Notice that I framed things in terms of confidence (in a Bayesian sense) and costs; not absolutes.

If you want true confidence in your functional tests then yep, [you need to write tests for them](https://www.jvt.me/posts/2018/11/07/unit-test-functional-tes...) - may seem like tests all the way down, but huge benefits reaped in the team + projects I'm on

Oh I am so tired of clicking around web forms to see what happens...

Seems most of my days are made of this lately.

Tests are a great way to take away tedious, boring, uninteresting essential tasks that can and will wear down a team.

> It’s worth remembering that engineers don’t get paid to write tests, they get paid to produce software that supports some business need.

I view it differently.

We get paid to solve business problems, most often -- but not always -- with software.

Software that's broken in preventable ways doesn't solve business problems, it just creates new ones.

Outcomes, not output.

> Software that's broken in preventable ways doesn't solve business problems, it just creates new ones.

I think you may be missing OPs point. Software "that's broken in preventable ways" very frequently does solve business problems. And solves them well. That's why the industry works the way it does w.r.t. testing.

6 months to deliver 2 applications that each work 80% often provides more business value than 6 months to provide one application that works 100%. More functionality almost always is more valuable to a business - including warts and all.

I wasn't defining "broken" as "the engineers find it distasteful". I was defining it as "broken".

I agree with this:

> 6 months to deliver 2 applications that each work 80% often provides more business value than 6 months to provide one application that works 100%.

Though this is a question of product management prioritisation, not engineering practice. "It's faster to not write tests" depends entirely on a definition of completion that ignores the cost of repairing defects and coping with faults and failures in production.

> I wasn't defining "broken" as "the engineers find it distasteful". I was defining it as "broken".

If a piece of software literally will not launch period, then it is broken. If it can do pretty much anything more than that, then it "works to some degree", and just has some number of scenarios that are broken. And my point is that if it has 80% of scenarios that work and 20% of scenarios that are completely functionally broken (ie do not work at all for the customer) that still provides business value. That's the point I'm making. It's not ideal, but in the real world half broken software is immensely valuable.

> Software that's broken in preventable ways doesn't solve business problems, it just creates new ones.

So software that is broken in preventable ways, absolutely does solve business problems. And just about the entire software industry is proof of this.

To expand on the internal tool case from the GP:

If your software is designed to support, augment, or automate some sort of internal need for non-IT teams, there's a significant chance that there's already an inefficient, error-prone process in place as-is. Because even without your software, they still have to get their job done. And they're used to having a "good enough" mindset because they're used to making due with whatever tools/capabilities they have, regardless of fit.

As long as your software addresses their needs[1] and isn't any more error prone than whatever they're currently doing, they'll be happy. That said, whenever possible you do want to code your failure conditions or edge cases to fail hard and visibly or expose visible checks for the user. If it covers 80% of their needs, but they have to fall back to their old process for the other 20% of their needs, business users are generally happy with that. But if it pretends to work 100% of the time, but subtly and silently causes problems 20% of the time (and they don't know what 20%), then they'll be distrustful of it and resist (or resent) adopting it.

For externally-exposed software, I'd be a bit more wary of not having any tests at all. But for internal tools, I've seen far more systems without tests than with them.

Reply on news.ycombinator.com