An unreasonably high bar?

Some time ago Adam Goucher posted this response to my 5 questions interview with Michael Hunter. There's a few points in there that I want to come back to, but right now the one that's at the front of my mind is this:

Count tests to get a useless number; I can write a million tests that provide useless information but still shows 7 figures in the count.

Well yes, you could. But why would you? We seem to have a hankering in the industry for techniques that would give good results even when badly applied by malicious idiots. That seems unreasonable. And also pointless: I don't believe that the industry is populated by malicious idiots. On the other hand, the kind of answer one gets depends a lot on how a question is asked. 

There is (I read somewhere recently) a principle in economics that one cannot use one number as both a measure of and a target for the same thing and expect anything sensible to happen. [Allan tells me that this is Goodhart's Law --kb] In our world this is the route to the gaming of metrics. I also don't believe that gaming works by folks consciously sitting down and conspiring to fabricate results. I do believe that if we measure, say, test coverage at every check-in and publish it on our whizzy CI server dashboard thingy and have a trend line of coverage over time and we talk a lot about higher coverage being better, or even that test coverage has something to do with "quality" [that would be the "surrogate measure" part of Goodhart's Law --kb] then it is in fact the response of a smart and well intentioned team member to write more tests to get the number up. Even if those tests turn out not to be much use for anything else.

I think (certainly I hope so) that my recommendation to measure scope by counting tests doesn't fall into that trap. Don't write the tests so that you can measure scope. But observe that you can if you write the tests the right way. Of which I shall have a bit more to say later.

Soft-Craft

So, on Tuesday night the programme for Software Craftsmanship was worked out. Looks pretty good, I think: there's plenty of the doing sessions that Jason wanted with a reasonable counterbalance of more reflective activities.

I was quite amused by the metaphor collision that's taking place, though. It seems that some people find that the best way to move forward the craft of software development is to head down the dojo for some sparring. I'm far form convinced that either of those metaphors is much help by itself. To see them embedded in one another makes my head spin.

I look forward to the day when we feel secure enough as an industry to have a programming conference where people's ability to do programming is improved. 

Interview form Agile 2008

At Agile 2008 I was interviewed by Amr Elssamadisy for InfoQ, and the video is now up. 

Skill vs Mastery

Since it's the holidays I get to tourist around London at bit in a way that I tend not to much through the rest of the year. Today I took in the Rothko show at Tate Modern. It's the first time in a long time (maybe ever, or at least since Rothko gave them away) that so many of the colour field paintings have been in the same place at the same time. 

Although these paintings are (at first sight) relatively featureless the viewer is intended to get right up close to them, to feel as if they are in the painting, to be overwhelmed. Although they are not figurative or representational they do carry a lot of content, a lot of emotional impact. There were people sitting in the room where the Seagram murals are, completely transfixed by the paintings. People have been known to burst into tears at the sight of these works.

This sort of modern art is perhaps one of the easiest for the "my four-year-old could do better" school of art critic to avoid thinking about. Of course, their four-year-old couldn't. One room of the exhibition goes some way towards explaining why not. These paintings are on (very) close examination revealed to be extremely complicated objects. Rothko put a lot of technique into these paintings: a lot of careful choice of materials, a lot of careful surface treatments, a lot of careful brush work.  In the language of the curators most of them aren't a "painting" at all. They are mixed media on canvas, so complicated is the structure.

But none of this shows up when one stands before the works, only the emotional effect comes through. That's mastery. As with painting, as Steve says over here with reference to music (and code), it's all about the expressiveness.

Craftsmanship

If you read this blog you might be the kind of person who will be interested in the Software Craftsmanship conference being planned for next year in London. More proposals are needed to round out the programme, so get your thinking caps on.

XP Day London 08

Xp Day is done for another year.

I think that the incorporation of a large Open Space aspect worked, with some notes taken for doing it better in future. The programmed sessions seemed to be well received too.

Notes from many session are going up on the wiki, you might like to look, comment or contribute.

Deployment?

I'm sitting here in Zurich with some colleagues working on some guidelines for teams who are starting to use Agile in an evironment formely exclusively RUP-based. We have a stack of Agile books, all by big-name authors, from which we are farming references for them to use. Not one of these books has anything (so far as we can find) to say about deployment, actually getting the software into the hands of the user. The very word is not in the index of any of them. Oops.

"just work"

There's an idea in  the management world that some work is "just work". I once had a holiday job in a factory that involved a good deal of one kind of "just work"—unpack a box of bottles of nose spray originally destined for one market and re–label them for another, put them into a different box. Well defined tasks, repeated many, many times. At the limit "just work" comes close to the physicist's idea of work: it's what happens when a force displaces its point of application.

Over the last century or so "just work" has been the topic of intensive study. Techniques for organising the execution of "just work" with the goal of maximum output are very well developed. But what is "just work" for programmers? The question arose in the comments attached to this discussion of distributed development. It also relates to some thoughts I've been having recently about the rush to adopt Lean practices in some quarters.
 

Andres says that 

In knowledge work, “just work” is learning. [his emphasis]

If that's true then I think we have to say that a lot of contemporary development activity isn't knowledge work. Wiring up GUI controls? Transducing between HTTP requests and SQL queries? Mashing up web services? You have to learn how to do these things at all, it's true. But what do you learn while doing it once you know how? What learning takes place that is the work? 

I'm reluctant to accept "finding out how to hack your way around the deficiencies of the current framework" as a form of knowledge work. Meanwhile, this maybe does give me a way to think about a curious claim (see comments #3 and #4 on that posting) from the Lean camp:

A pair team has to produce twice as much output at a given level of quality as two solo programmers in order to be worthwhile.

What would we have to believe, in order to think that this claim were true? I think that we would have to believe that the paired programmers are engaged in quite a strong sense of "just work". What is it that a pair of programmers necessarily do half as much of because they are pairing? I can't think of anything but typing. Are keystrokes really what it is that programmers do that adds value? At a slightly higher level of sophistication we might say (as Corey in fact does) that the measure of value added is function points. That's a much better measure than typing (which implies code)

Measuring programming progress by lines of code is like measuring aircraft building progress by weight.—Bill Gates

but still gives me slight pause. Function points are still a measure from the technical solution space, not the business problem space. One of the things that makes me a bit nervous about the rush to Lean is the de–emphasis on conversations with the user/customer (which Leaners have a habit of dismissing as wasteful "planning") in favour of getting tons of code written as fast as possible. 

What I really want to see is getting users empowered to make money (or whatever it is they value) as fast as possible. That's knowledge work. And that requries learning.

Can you tell your POJO's from your NOJO's?

Ivan describes a nice distinction between Plain Old Java Objects in general and Non Object-oriented Java Objects in particular.

Certain (web, in particular) frameworks lead on in the direction of NOJO's. It begins with the DTO (just one kind of NOJO) and spreads from there. Is that good or bad? Can't say in the completely abstract, but if your current framework does lead you that way have you given thought to what design choices and architectural stances your are being lead to? Do you like them?

Another take on interviewing

Very nice posting here from Simon Johnson regarding the selection of candidates for programming jobs.

His blog doesn't allow comments, so I'll follow up here: The only real criticism I'd have of Simon's approach is that the list of desirable qualities in a  candidate is backwards. That is, it should read 

1. The person in front of you will complement your pre-existing team
2. The person in front of you can actually write a program
3. The person in front of you is smart

If the candidate won't complement your team it is of no interest whether or not they can program. Even if they are the semi–mythical 100x productive programmer (hint: those folks are not applying for your position). 

If they can't write a program (of the kind that your company finds valuable) it is of no interest how smart they are. Folk wisdom in many parts of the industry suggests that enough smart can outweigh any other shortfall in a programmer. This is only true in very, very limited and highly specific circumstances (hint: you are not in the business of writing hard real–time life–critical massively–parallel seven–nines embedded control software for the black hole at the centre of the galaxy).

If you can get along with them, and if they can code, smarter is preferable to dumber. Your only problem now is to find some way to recognise "smarter", which turns out to be very, very hard indeed.

Registration open for XP Day London 'O8

Registration for XPDay is now open. XPDay sells out each year, so please book now. We operate a waiting list once the conference is full but only a handful of places subsequently become available.

The full cost of the conference is £350 but the first 50 people to book get a special “Early Bird” price of £275.

The few programmed sessions are set, and we encourage registered attendees to think about the topics they'd like to see discussed in the Open Space sessions. Attendees have the option to prepare a 5 minute "lightening talk" to promote their preferred topics, and these will go into the Open Space topic selection..

I look forward to seeing you there.

Programming Interview Questions

There's a cottage industry devoted to devising questions to ask folks applying for "programming" jobs, to collating such questions and then posting answers to them on the internet. I was reminded of this recently when an announcement of a new Google tool, moderator, came along. The example issue being moderated was "Interview Questions for Software Engineers". At the time of writing the winner (with 4 upvotes and 0 down votes) is my own proposal:

Would you please sit down here at this workstation and write some code with me?

If you are reading this then you have almost certainly been through many an interview for a job that will require some programming to be done. Think back and ask yourself how many times you were offered such a job without ever having been in the same room as a computer and an interviewer.

In my own experience (I've had 9 jobs involving programming, and many more interviews) it's been quite rare. One has to wonder why that is. I suspect that having a candidate do some programming is considered very expensive—and yet hiring the wrong candidate is unbelievably expensive. Hence the questions, which seem to be aimed at finding out if someone will be able able to program without watching them do any.

The second most popular question on the moderator site right as I write is

Given a singly linked list, determine whether it contains a loop or not

I can see what's being aimed at here, and yet it feels as if, in the stress of an interview setting, all the answer given will tell you is whether or not the candidate has seen the textbook answer before. How does knowing that help?

By the way, this is a systems programming question. The vast majority of working programmers don't come anywhere near any such class of problem. They do applications programming. There seems to be a sort of folk wisdom in a large part of the industry that having the skills (and, see above, mere knowledge) of a systems programmer will suite them well to being an application programmer. It's far from clear to me that this is true: the two areas of programming are quite distinct and require different (albeit overlapping) skill sets. I wonder how many interviewers use questions like this to create the impression (maybe even to themselves) that their organisation is involved in sexy, impressive systems work, whether or not that's true?

A while ago we did an informal survey of folks who work for Zuhlke in the UK, asking what had made them apply to us and what had made them subsequently accept an offer. We (the managers) were surprised by how many said that the very recruitment process itself had encouraged them to accept an offer.

This is what we do: Line mangers (such as myself) read CVs. There is an HR department, they don't come anywhere near recruitment. CVs that seem to describe a credible technical career go to the Chairman of the company. The candidates he likes the sound of are invited in for a first interview (of around an hour) to see if they are firstly, someone we'd be prepared to put in front of a client under our brand and secondly, someone we would enjoy working alongside. There's also a short written test of programming knowledge.

Those who make it through that are invited back for an interview with a small panel of our consultants who explore their technical knowledge in depth. First the candidate is asked to give a technical presentation on an interesting piece of work they did recently: stand at a white-board, explain the architecture, the technology choices, how it worked, how the project worked, what they did etc.

Then we set to finding out what they know. We have a checklist of technologies that we've used in projects and the questions take the form "what do you know about $technology?" No tricks, no puzzles. We do a breadth first pass to eliminate the things that the candidate doesn't know (and the process rewards optimistic answers at this point with a lot of embarrassment later on) and then we drill down into each area until at least one of the candidate's or the interviewers' knowledge is exhausted. This goes on for several hours.

Those who make it through that, and people have been known to walk out, get invited back to do some pair programming. That's the point (and not before) at which we decide if we'd like to make an offer.

And yes, this is all unbelievably expensive (I mean, the Chairman does first interviews?) And very, very effective. Our false positive rate is essentially zero and best of all, clients love our engineers and ask for them back again and again. Which makes it all worth while. I can't think of a cheaper way that would actually work.

PS: yes we are hiring. If you'd like a go a this, drop me a line.

97 Things Every Software Architect Should Know

The 97 things project is approaching a milestone: having 97 things which, in the opinion of one self–selecting group (which includes myself), every software architect should know. The intention is that the "best" 97 things go into a book to be published by O'Reilly, but that the forum will continue to be active, to gather new so–called "axioms" and discussion of same. 

What do you think every software architect should know?

What do Android (and the rest) really mean?

If the FOSS rumblings from the mobile OS world are leaving you puzzled, you might benefit from a visit to Tim Joyce's blog. Tim's job is to understand what these things mean, and he's kindly embarking on a blog to share that understanding with the world.

Shameless self promotion dept.

I've been interviewed by Michael Hunter in his "five questions" column for Dr. Dobbs

Optimism

At the Tuesday Club the other week there was a bit of chat about what seems to be a foundational assumption of Agile: that in fact we can successfully execute software development projects, so we should run them in a way that brings about the greatest benefit.

This in contrast to what appears to be the (undeclared) foundational assumption of many other approaches: development project are so hard that we are most likely to fail, so we'd better find a way to do that as cheaply as possible.

This carried across into a panel discussion at the recent Unicom conference, where one of the attendees came up with this summary description of Agile:

Collaborative optimism to solve the right problems (and only the right problems) in an incremental way

I quite like that "collaborative optimism" bit.

Agile Adoption Patterns

I've just done a first pass through Amr Elssamadisy's cracking new book Agile Adoption Patterns. It's a timely work—enough teams have adopted enough Agile practices enough times that there should be patterns by now. And here they are.

There's a lot to like about the book. I especially appreciate the consistent message throughout that we do these practices because of their tie to business value. Not for truth and beauty, not for right versus wrong, but because they make our customers happy through the means of more money sooner, more often. There is an explicit mapping from various enhancements to business value onto clusters of the patterns. There is also mappings by business or process smell (of which there is a good, short catalogue presented). For each kind of business benefit the patterns within each cluster are logically related and given a partial order with respect to how well they help with that goal.

Separately, the clusters of patterns are described themselves as patterns, which is a nice organising principle. Agile Adoption Patterns is quite explicitly a handbook for finding out why your organisation might want to adopt Agile and then building a strategy to make that happen. A damn handy thing to have.

Scattered through the pattern catalogue there are little diagrams relating the patterns. I might wish for more of these. 

The patterns themselves contain a "Sketch" section, in which a surprisingly large cast of characters play out little vignettes. Dave Developer comes to realise the value of stand–up meetings because he sees Scott ScrumMaster [sic] removing his blockers, and so forth. I know that this style of thing is very fashionable these days, but it makes me cringe. Not only that, but I have a terrible memory for names, so keeping these fourteen characters lined up in my head is a strain.

The patterns have the usual "Therefore" section with the things–to–do in it, followed by and "Adoption" section telling you how to get there. They are all quite meaty, concrete and honest.  And best of all is the "But" section telling you common mis–steps, gotchas, misunderstandings and how to recognise same. This is outstanding! A common shorthand description of what a pattern is (in fact, it's in this book too) is words to the effect of a "solution to a problem in a context" Which is sort–of true a far as it goes but has its problems, mostly around the term ''solution". 

Oh, how we love "solutions" in the IT industry. A more accurate but less compelling description what patterns are is that they are mappings form one context to another, in which certain conflicting forces are more agreeably disposed (or resolved) in the latter than in the former. These "But" sections describe common scenarios (I've seen a lot of them, and so have you, I'll bet: Continuous Integration but continuously broken build, anyone?) that you could have and still make some sort of Jesuitical argument to be in agreement with the "Therefore", but actually are very wide of the mark. The next time I write any patterns, they will have a "But" section.

The highest level organisation of the patterns is into those concerned primarily with Feedback, technical practices, supporting practices and the cluster patterns. I was most interested to note that "Coach" is a supporting practice, but the role of ScrumMaster is assumed throughout.

I am one of the latter, but I'd rather be one of the former. Not so much of an option these days. Well, that's the power of marketing for you.

Meanwhile, one of the nicest things about having patterns is having a shared vocabulary. I think that Agile Adoption Patterns will become the standard reference for talking about Agile adoption, which will make my professional life easier and that of may others. A service to the community.

"Opportunities for Improvement"

The pattern form used is has nine sections and I'm not sure it helps to have each section shown in the table of contents. I feel the lack of a ready–to–hand one or two page list of all the patterns by name. Twelve pages of TOC for 320 pages of content seems a bit excessive. The same syndrome shows up in the index. If all the parts of a pattern are within a four page range do they really all need their own individual index entry? This is really a shame as the book otherwise benefits from the several other complementary ways to navigate it.

There is a stunningly egregious formatting botch on page 207. I'm sure this volume will go into multiple printings and I hope that someone at A–W gets that fixed.

The Non–functional in a Functional world?

This post shows an implementation of map/reduce in Haskell. It's a one–liner (if you ignore all the other lines) and begins with this claim:

the type says it all

p_map_reduce :: ([a] -> b) -> (b -> b -> b) -> [a] -> b

The type says a lot, that's for sure. I says that map/reduce results in a value of type b when given a list of values of type a, a function that gives one b given two b's and a function that gives one b given a list of a's. That is a lot of information and it's impressive that it (at least) can be stated on one line. Is it "all", though?

Even allowing for some inter–tubes posting hyperbole "all" is a very strong claim. A Haskell programmer should know that. If one already knows what map/reduce does then the type tells you a lot about how to use this implementation of it. And if you don't, maybe not so much.

But that's not what caught my attention. The point of map/reduce is to recruit large pools of computational resource to operate on very large data sets efficiently. The (14 line, in fact—which is still quite impressive) implementation uses Haskell's Control.Parallel features to chop the list of a's into 16 chunks. That's not in the type. The intention is to have enough lumps of work to 

bump all your cores to 100% usage, and linearly increase overall performance

which it may well do on a large enough data set, if I have fewer than seventeen cores. Currently, I do. There are nine cores in my flat right now. If my other laptop were here, it'd be eleven. But if I'm planning to use map/reduce in the first place, maybe I'm going to recruit all the machines in the office, which would currently be worth over twenty cores. This map/reduce isn't going to pin all of them.

Would Haskell's type system support giving p-map-reduce a dependent type with the number of worker processes in it? I don't know enough about it to tell. And why would we care? Well, that number 16 is a very significant part of the implementation of that function. 

The only reason we'd bother to use map/reduce is because of its non–functional characteristics (as the properties of software that make it actually useful are called) and these are no–where to be seen in its type as a function. 

Computer Weekly IT Blog Awards

The shortlist for the Computer Weekly IT Blog Awards 2008 is just out.

This blog was nominated (most kind, thanks), as were those of my colleagues Ben and Daryn, but Daryn's has made it onto the shortlist. This is fitting recognition for the fine work he has put into his blog, particularly his series of Ruby/Rails tutorials which are gathering some recongnition in that world. Feel free to go and vote for him!

XPDay London Session Submission Process

If you want to propose a session for XP Day London '08 (and I hope that you do), then here's what you need to do.

We want the conference to be largely self–organizing this year, so the submission process is suitably light weight.

Experience Reports

Please send a very brief outline of your report to submissions2008@xpday.org You will receive a URL to a google document. Make sure that you include an email address corresponding to a google account (or let us know if the address you send from is the one to use). The document will be writeable by you and the committee and all other submitters. Develop your report there. As in past years we invite the community to collaborate together to help maximize the quality of all submissions. After the close of submissions the committee will make a selection of experience reports to be presented at the conference. The selected reports will be shepherded between acceptance and the conference.

Conference Sessions

Most of the duration of the conference (other than experience reports) will be OpenSpace. There will be Lightning Talks early in the day where people can publicize topics that they would like to discuss. We encourage people to engage in this way. There will be optional rehearsals at XtC in the weeks running up to the conference. You do not need to attend these to make a lightening talk at the conference. If you would like to have such a rehearsal, send a title and a google account email address to submissions2008@xpday.org You will be allocated to a rehearsal slot. We invite groups outside London to schedule rehearsals too, and are happy to help with scheduling those. There will be a google calendar with the slots shown.

Some kinds of session can benefit from some preparation of materials and some logistics on the day. For example, a workshop involving some sort of server, or extensive materials. There will be a limited number of time slots during the conference for such sessions. Please submit a brief outline to submissions2008@xpday.org indicting an email address associated with a google account. You will be sent the URL of a google document within which to develop your proposal in collaboration with other submitters. After the close of submissions these proposals will be assessed by the committee and suitable sessions will be selected for the program.

We have decided to extend the submission period, so submissions for experience reports, rehearsals and programmed sessions now closes on Friday August 8th 2008. 

This year we want to de–emphasize sessions introducing Agile or Scrum or XP or TDD or... and promote topics of current interest to practitioners. We want the conference to be a forum in which the state of the art is advanced. That doesn't mean that only experts are welcome, or welcome to present. Experts have their failure modes, simple questions from journeymen often reveal the essence. 

All are welcome, so get your thinking caps on!