Slow and dirty with callouts

A RANT

Slow & Dirty

© Codemanship Ltd 2011

Please support Astrid Byro in her Everest
challenge to raise money for Bletchley Park

Twitter: @MsAsti

http://www.justgiving.com/Astrid-Byro


“Where have I
seen this graph
before?”

Relative Development Effort (person-hours)

Answer: You
haven’t. There’s
no credible body
of industry data
that supports it

Defects/KLOC


What data we do have
paints a compelling picture
of the opposite relationship
between quality and speed

When considering whether to
Steve McConnell, Software Quality At Top Speed, 1996 put more effort into quality or
less, you need to know what
side of this curve you’re on.
99%+ of us are to the left of it

Again, we have a mountain of
data that suggests that time
devoted to catching problems
earlier easily pays for itself in
time saved later

Barry Boehm, 2007


What makes a bug more
expensive to fix is the length of
the feedback cycle involved in
fixing it. Shorter feedback cycles
catch problems earlier when
they cost much less to fix

http://www.ambysoft.com/essays/whyAgileWorksFeedback.html


Bugs aren’t the only problems
that cost us more later. Time
invested early in making code
easier to maintain can also pay
big dividends as the code
rapidly evolves.

Full Team
Effort to
maintain

http://www.lalcrest.co.uk/cost.php


A leading software company
found that by the 8th version of
their only product, a line of code
cost 20x as much to
write/change

Market-leading Software Product Lifecycle
400

350

300
250

200
Cost/LOC
150

100

50

0
1 2 3 4 5 6 7 8
Major release

Engineering staffing costs
spiralled as they hired more and
more people to achieve less
and less

1400

1200

1000

800
Engineering Staff
600

400

200

0
1 2 3 4 5 6 7 8


The evolution of the product
slowed to a snail’s pace, and
today they are lagging far
behind their competitors

8000
Product
Size 7000
(KLOC)

6000

5000

4000

3000

2000

1000

0
1 2 3 4 5 6 7 8


Try this thought experiment: imagine
two teams A and B who compete to
guess a 4-digit number. Team A
guesses all 4 digits at a time, team B
guesses one digit at a time. Worst
case, team A may need 10,000
guesses, but team B would need max
40 guesses. Which team would you
bet on?

? ? ? ?

Now let’s make team A 10x as
“productive” as team B: for
every guess team B gets, team
A get 10. Which team would you
bet on now?


“It is not how fast we deliver that
defines the winners and losers, it is
how fast we learn from what we
deliver”
Me, Whenever


Consider the different way we
would start a marathon as
opposed to a 100m sprint


With the higher exertion of a sprint,
lactic acid builds up faster in our
muscles – until we can’t take in
oxygen fast enough to get rid of the
lactic acid. We call this anaerobic
exercise. Eventually, the lactic acid
builds up to the point where we
experience pain and cramps and
can’t go on.


When we run a marathon, we
must not exert ourselves to the
point where we’re doing anaerobic
exercise, as it’s not sustainable.
Start a marathon at a sprint, you
may take an early lead, but you’ll
end up being carried off on a
stretcher.


The vast majority of software
development sprints turn into
marathons – especially if that
first release is successful

Anaerobic Software Development
When teams write code at an unsustainable pace,
code smells build up faster, making progress
increasingly difficult and painful


This is a software development
marathon being run right now. The
winners in this race will be decided by
who sets the most sustainable pace of
innovation – not who sets the fastest
initial pace . The winner will not out-
deliver their competition. They will out-
learn them.


And we’re not just talking in the
long term, either. Experiments
like this show clearly that taking
more care over quality can pay
small dividends in very small,
short problems

Roman Numerals Kata
30
Time To
Completion
(mins) 29

28

27

26 With TDD
No TDD
25

24

23

22
1 2 3
Iterations
http://www.codemanship.co.uk/parlezuml/blog/?postid=1021


Do not be seduced by the illusion of “done”

Of course, one way to deliver
earlier without taking more care
is to widen the goal posts – e.g.,
testing less thoroughly, or just
ignoring problems


“The financial impact of
software quality problems is
in no way diminished by our
ability to ignore them”
Me, Just Now

The problem with this strategy is
that the business consequences
of quality issues have no
respect for your desire to ignore
them. The fiends!


But what if we’re
building the wrong
thing?!


Well, what of it? Let’s just accept
that we’re almost certainly going to
need to take a few swings at it
before we get the ball in the hole.
NOBODY gets it right first time.

I would say it’s a
given

So, given that we’re – to whatever extent
– building the wrong thing, and we can
only learn that by delivering something,
and given that it would take no more time
or money to build it right, and that we can
be pretty sure that we’ll need to evolve
our first release further and that our ability
to do that will depend on the quality of the
code…


…why would you choose to take
less care over quality?


Ah, but what
about…

But what about all these massively
successful start-ups who we know hacked
out their software during all-night pizza-
and-caffeine-fuelled code orgies?

Surely we should just get anything to
market quickly by any and all means, and
then we can fix all the problems with all
that lovely Web 2.0 bubble money that will
come flooding in?


And what about my
Great Aunt Doris, who
smoked 60 a day and
lived to be 103?


Or the guy who bet
his entire life savings
on one hand of poker
and became a
millionaire?


Successes like Facebook,
Twitter etc are statistical
aberrations, distorted
even more by the Web 2.0
bubble, which enables
them to buy their way out
of hugely expensive early
mistakes with other
people’s money

Mistakes that kill
less lucky
companies all the
time!


So, In Summary
• Business models learned from statistical aberrations
are Fool’s Gold
• For the overwhelming majority, we are on the left of
McConnell’s curve, and better = quicker
• Get to market sooner by doing a better job of
something simpler
• Expect to have to play more than one hand before you
win anything
• When it comes to discussing this topic with managers
and customers, grow a pair


Because start-ups
must eventually
become stay-ups


www.codemanship.com

@jasongorman


Slow and dirty with callouts

More Related Content

Slow and dirty with callouts