allan kelly

One of the asides I made in “Programmers without TDD will be unemployable” which caused a bit of outrage in the testing community was my comment “Manual testing is a sin.” While I have been unfair to many testers, and somewhat simplistic, I still stand by the statement. Let me expand on why I still stand by the comment and why I am wrong.

It is all a question of context. Look at the full line the quote appeared in:

“Unit testing will be overwhelmingly automated. Manual testing is a sin. Manual unit testing doubly so.”

In the context of Unit Testing, as one who has undertaken formal and informal manual unit testing I believe unit testing can and should be fully automated. Informal unit testing is too ad hoc, prone to error, time consuming and difficult to replicate. Formal manual unit testing suffers all these problems and is hideously expensive.

In the context of unit testing I stand by “Manual testing is a sin.” I will also go further.

The vast majority of testing I see performed by professional testers – i.e. not unit testing – is manual, the vast majority of this testing could be automated. If you automate the testing the initial costs might go up – I say might because a good natural language test script is expensive to write itself – but the cost of executing the test will fall. More importantly the time it takes to run the test will fall dramatically.

The combination of low cost and fast turn execution means the tests become a very very effective feedback loop.

Failure to seize this amazing opportunity is something I consider sinful. It may be expensive but it is necessary. Even seizing 20% of this testing would be massively beneficial.

What I do not consider sinful is Exploratory Testing which almost by definition cannot be automated and therefore needs to be a manual process. Exploratory testing is about looking for the unusual, the unexpected, the thing you didn’t think of before, the thing you only think of because you now see it.

Automated exploratory testing might even be a contradiction in terms therefore manual exploratory testing is not sinful.

But, and this is a pretty big BUT: if the initial quality is low (e.g. unit testing is absent or poor) and the testing (e.g. system, integration, acceptance testing) that comes before exploratory testing is ineffective then exploratory testing will be effective at finding defects (because the earlier rounds didn’t find them) but it will be ineffective as exploratory testing.

The time and effort spent on “exploratory testing” is really being spent doing system testing. Real exploratory testing itself will not be possible when there are many defects because the exploration will continually be interrupted. 

Unfortunately the subtlety of this logic is lost on most organizations who just label everything that happens after writing actual production code “testing” and is made worse when testing is separated from the developers and, more importantly, those who understand what the system should do (the business, the BAs, the product managers, the users, etc.)

Worse still, the easy availability of cheap offshore testing capability means that rather than address the root cause of defects (low initial quality, system ineffective testing, misunderstanding of exploratory testing) means the whole thing can be hidden away cheaply.

There are probably some other examples of testing which cannot be automated and are not sinful. But I repeatedly find individuals and organizations who too ready to jump to “it cannot be automated” after very limited, or no, effort is spent on trying to automate it.

Similarly the fact that a professional tester cannot automate the testing of a piece of software system does not mean it cannot be automated. Automated testing – at least the setup – involves programming skills which many testers simply do not have, indeed it is wrong to assume they would have them.

Finally, there are a class of systems which as far as I know defy automated testing. Or rather defy automation at the moment. They defy automation because the creators of these systems have little or no interest in allowing automation. This set includes: SAP, Microsoft Sharepoint, most CRM systems such as Microsoft Dynamics, Murex (although I believe efforts are afoot here) and some other large packages.

I find it incredible that companies like SAP, Oracle and Microsoft can sell systems which cannot be effectively tested. But actually this is part of the sales-pitch. Because these systems are marketed and sold as “no programming required” to offer automated test tools would expose the lie. If you are involved with selecting one of these systems ask: how will we test it?

The inability to effectively test these systems is their Achilles heal, it is one of the reasons I don’t like these systems and it makes me wonder about whether organizations are right to use these systems.

The reaction to my last entry – Programmers without TDD will be unemployable by 2022 – took me a little by surprised. I expected there would be a reaction, perhaps a big reaction, but I frequently expect some reaction and usually I get none. This time I got a reaction and the scale of the reaction continues to surprise me.

Particularly gratifying are all the comments on my blog, on DZone, JavaCodeGeeks, and probably some other sites that syndicate my blog but haven’t bothered to tell me. The fact that people were moved enough to write anything, supportive or not, means I caused people to think and join the discussion.

I’ve come back on many, certainly not all, of those comments directly but I won’t come back to every point. In this post I’d like to touch on a few of points which I think need expansion or clarification. Sorry if you think this is me being defensive. Stop reading if you think so, consider this a blog post for one reader only, me!

(This post, and a couple of others coming out of the TDD post does mean it will be a little while before I can get back to my software economics series.)

First, the many cynical comments (e.g. “This seems unlikely”) don’t surprise me. I frequently come across such cynicism in my work. Indeed, were such cynicism not widespread then I don’t think there would have been much interest in the prediction or any point in making it.

I am rarely sympathetic to arguments that it “won’t work for me” or “won’t work in my domain” because overwhelmingly these arguments are at best a (flawed?) logical argument, but they are usually made without trying it, they are conjecture, . When someone comes and says “We tried it… we really tried it… we invested in it… we stuck at it for a few months… and it made us worse” then I’ll accept it as so. Usually I find people who argue against TDD have never tried is or tried it superficially.

Second a point of clarification. I am happy to acknowledge that there are other ways of reducing a defect count. Code reviews spring to mind. There is a lot of research that shows that when done well they can be very very effective – and also justify the time spent.

Over dinner the other night Phil Nash also suggested highly typed functional languages – or was it functional languages and strong typing? I’m sure Phil is right in both points eitherway. Although if you ever saw the any of the Lisp/Scheme I wrote in college you would know bugs are entire possible in functional languages!

The point is: there are other techniques, probably some I don’t even know about. I single TDD out because a) I know about it and b) I think it has achieved critical mass and will be adopted more widely in future.

Because of alternative approached there will still exists jobs where TDD is not a prerequisite but they will be few and far between, in particular niches. There will be a few jobs you can get without past TDD experience but you will be expected to learn it. The same way you can occasionally get a job as a Java programmer even if you have never programmed Java. But if I was still developing, I won’t like to bet my career on finding such jobs.

Next, and most gratifying, the discussion on Twitter and in blog comments that followed the publication flushed out several companies who already operate a policy of only hiring developers who practice TDD. 7 Digital and Compare the Market stick in the memory because a) I’ve known them for a while and more importantly, b) they are hiring! Thanks guys, consider yourself plugged.

A few other people tweeted or commented to say they personally operated such a policy but didn’t give company names.

So let me float a hypothesis – based on the fact that I regard TDD as a better way of developing code than not doing it:

The divide between productivity and value-add of companies which do TDD style development and those which don’t is growing. The first group are generally among the best, most productive, highest value add, and the second group which don’t TDD are a long way behind. The first group is accelerating onto BDD, Continuous Delivery and other “new” ideas, while the second aren’t.

Sure there are companies which say they do this, and there are companies where application is not consistent – some people do and some people don’t. For such companies to join the high performing ones they need to do more and do it more consistently.

There are networks effects to doing TDD: the more people who do it in an organization, and the more consistently it is done then the greater the benefits. If one developer starts doing TDD their code will be better and everyone will benefit, but with each extra developer doing it the benefit grows greater per person – there are economies of scale here. And when everyone does it then the benefits are far more than the sum of the parts.

Yes this is conjecture, or if you prefer, call it intuition.

I don’t know what form it will take – I think it will be mixed up, some will be methodology, some will be more relaxed, and it will differ from place to place, team to team.

Next there was some discussion of how the transmission mechanism, how would TDD being good actually get it inside corporate IT departments? That is something I’ll come back to in another post, I have a view on that which might poor petrol on this particular fire.

People also threw stones at me on Twitter because my company sells TDD training. Lets get this straight: we offer such services because we believe in it, because I believe in it I wrote blog posts like I did.

It is insulting and extremely cynical to suggest that I write blog posts like I did purely to sell training. It would be nice if life was that simple, you would see more blog posts like that and I would be a lot better off.

Now I’m not saying you have to have training to learn TDD – although I do recommend it. I am also happy read Steve and Nat’s book, Growing Object Oriented Software – and yes, that is an Amazon associates link, is that a crime? I make about £10 a year from such links.

But ask yourself, if you were to buy such training who would you rather buy it from? Someone who sells it as another course or someone who passionately believes in it?

Before I finish, two other things about the reactions post surprised me. That so many people railed against my aside comment about IDEs. Of course IDEs will still exists and will still be useful. I just suggested that would be less because they were debugging environments and thus we might see changes in IDEs.

Also, very few people mentioned my comments about University courses. In many ways those comments were the driving force behind writing the post. TDD is a proven technique and Universities should be teaching it on programming courses, and they should be teaching code reviewing and probably some more techniques too.

Finally, as someone else pointed out my prediction is a safe bet. If by 2022 I am proved wrong nobody will remember my prediction. And if I’m proved right nobody will care when I say “told you so.” Besides, I expect people more famous than me will make similar predictions – both for and against – and they are far more likely to be remembered than me.

In the last two posts I’ve taken two different views on what might happen to the Software Demand curve when we introduce Agile:

• The good Agile demand curve
• The bad Agile demand curve

When we discussed how Agile effects the software supply curve things were pretty clear cut. Agile, in the long term boosts supply, and may even flatten the curve – making it more responsive to price changes, i.e. more elastic. Unfortunately things are not so clear cut with the demand curve.

Good news: There is a story Agile people – OK me! – sometimes tell – about what “should” happen with demand in an Agile environment.

Bad news: The is another story about demand in an Agile environment, one that Classicist (i.e. believer in “Waterfall” style up front requirements and planning) fear. And I’ve seen that too. It ain’t pretty.

But, the key point is: Few of the Agile tools work directly on the demand curve. After all, Agile is about software and the demand for software is derived from something else so why would they?

Let me say that again in other words because it is important: Nobody wants software for the sake of software, the demand for software is derived from some other demand, e.g. a retailer wants to sell online, that means they need software.

The current Agile tool set (e.g. User Stories, Acceptance Criteria, Specification by Example, etc.) which touches on demand – call it need, want, requirements, specification, whatever you want – does not address the underlying demand. These tools operate with the demand as presented.

Sure there can be feedback from the tools and Agile deliveries but feedback is often absent. Even when feedback is present those responsible for the underlying demand may not want to hear – or act – on the feedback. And in an environment where success is measured by criteria like “On Budget, On Schedule, On Features” acting on feedback may be impossible.

Which brings us back to the 3 Styles of Agile: Iterative, Incremental and Evolutionary (part 1) and 3 Styles of Agile part 2. (At some point in the future I will reprise those three styles in the light of this analysis.)

One of the key factors in determining whether the feedback can affect underlying demand is whether those responsible (e.g. senior managers) who largely originate the underlying demand are open to feedback. Which brings us to another story I need to tell, another time…

Where does this leave us then?

If the current Agile toolset cannot address the underlying demand we need new tools. We might choose to add those tools to the thing we call Agile or we might choose to put them in another category.

Personally I believe many of those tools do exist but they exist outside the Agile space. They are fellow travellers if you will. For example: Beyond Budgeting is clearly one of those tools. However, many of those tools have their own names, they are not Agile, they are what they are (e.g. Beyond Budgetting). They fit very nicely with Agile but they are not.

However that is my personal view. I also think that Agile is a marketing bang-wagon and those tools will be conscripted to the Agile course and in many cases put under the Agile umbrella.

Either way the problem we need to recognise is: addressing the software demand curve Agile is necessary but not sufficient by itself.

Bit thank yo to the guys at ProgramCreek.com for including this blog on their list of “100 High-Quality Java Developers’ Blogs”.

Strictly speaking this is not a Java blog but most of the issues I talk about do effect Java developers.

It also seems that in listing this blog at number 55 (yeah, still room for improvement!) the ProgramCreek.com team have set up a little competition between me and my #NoProjects conspiracist Steve Smith who is just two places behind me 🙂

(Whoops, should have posted this when it was announced! – better late than never)

I’ll get back to software development economics, there are more posts in the works, but right now I need to say something….

I have blogged before on the subject of “What is Agile” – I’ve even expanded on that blog in an unfinished piece of writing called “What is Agile? Perspectives on Agile” – but sometimes I think its just sex….

(OK, this blog just got filtered out of lots of feeds and stopped by lots of firewalls but lets continue.)

In the software development world there are, broadly speaking, two groups of people: those who create the software (coders, testers, etc.) and those who manage the process (project managers, development managers, etc.). When discussing “Agile” I find that both sides think the problem is with the other.

To put it another way, if I’m talking to developers they think its managers who are the block to adopting more Agile techniques and improvement. But when I’m talking to managers they say its the developers who resist Agile and other improvements.

Which always reminds me of the old Philip Crosby quote:

“Quality has much in common with sex. Everyone is for it. (Under certain conditions, of course.) Everyone feels they understand it. (Even though they wouldn’t want to explain it.) Everyone thinks execution is only a matter of following natural inclinations. (After all, we do get along somehow.) And, of course, most people feel that all problems in these areas are caused by other people.” Philip Crosby

Lets bring it up to date and make it Agile specific, substitute the work ‘Agile’ for ‘Quality’:

“Agile has much in common with sex. Everyone is for it. (Under certain conditions, of course.) Everyone feels they understand it. (Even though they wouldn’t want to explain it.) Everyone thinks execution is only a matter of following natural inclinations. (After all, we do get along somehow.) And, of course, most people feel that all problems in these areas are caused by other people.” Allan Kelly after Philip Crosby

You see? It fits.

Agile isn’t someone else’s thing, it doesn’t belong to another person: Its yours. You are as likely to be the block as the person you think is the block.