Tim Morgan

How Fighter Jets Lock On

Issue 55 December 2014

he end of procrastination
is the art of letting go.
Ive been a lifelong
procrastinator, at least until recent
years. I would put things off until
deadline, because I knew I could
come through. I came through on
tests after cramming last minute,
I turned articles in at the deadline
after waiting until the last hour, I
got things done.
Until I didnt. It turns out
procrastinating caused me to miss
deadlines, over and over. It stressed
me out. My work was less-thandesirable when I did it last minute.
Slowly, I started to realize that procrastination wasnt doing me any
favors. In fact, it was causing me a
lot of grief.
But I couldnt quit. I tried a lot of
things. I tried time boxing and goal
setting and accountability and the
Pomodoro Technique and Getting
Things Done. All are great methods,
but they only last so long. Nothing
really worked over the long term.
Thats because I wasnt getting to
the root problem.

and the false need to check my

email and news and other things
online were causing me problems. They were causing my
procrastination.
So I learned to let go of those too.
Heres the process I used to let go
of the distractions and false needs
that cause procrastination:

I paid attention to the pain

they cause me, later, instead
of only the temporary comfort/
pleasure they gave me right away.

I thought about the person

I want to be, the life I want
to live. I set my intentions to do the
good work I think I should do.

I watched my urges to check

things, to go to the comfort
of distractions. I saw that I wanted
to escape discomfort of something
hard, and go to the comfort of
something familiar and easy.

I realized I didnt need

that comfort. I could be in
discomfort and nothing bad would
happen. In fact, the best things
happen when Im in discomfort.
And then I smile, and breathe,
and let go.
And one step at a time, become
the person I want to be. n

I hadnt figured out the skill

that would save me from the
procrastination.
Until I learned about letting go.
Letting go first came to me when
I was quitting smoking. I had to let
go of the need to smoke, the use
You can only lose what you cling
of my crutch of cigarettes to deal
to. ~Buddha
with stress and problems.
Then I learned I needed to let go
of other false needs that were causLeo Babauta is the creator and writer at
Zen Habits. He is a former journalist and
ing me problems: sugar, junk food,
freelance writer of 18 years, a husband
meat, shopping, beer, possessions.
and father of six children, and lives on
Im not saying I can never do these
the island of Guam where he leads a very
things again once I let go of these
simple life.
needs, but I let go of the idea that
theyre really necessary. I let go of
an unhealthy attachment to them.
Then I learned that distractions
Get 50% off your first 6 months
circleci.com/?join=hm
2

and help change the future of search

Curator
Lim Cheng Soon

Contributors
Tim Morgan
Eric Lippert
Justin Kan
Steve Klabnik
James Hague
Matthew Plant
Julia Evans
Jonathan Katz
Philip Guo
Bryan Kennedy

Hacker Monthly is the print magazine version

of Hacker News news.ycombinator.com, a social news
website wildly popular among programmers and startup
founders. The submission guidelines state that content
can be anything that gratifies ones intellectual curiosity.
Every month, we select from the top voted articles on
Hacker News and print them in magazine format.
For more, visit hackermonthly.com

Advertising

Published by

[email protected]

Netizens Media
46, Taylor Road,
11600 Penang,
Malaysia.

Proofreader

Contact

Emily Griffin

[email protected]

Printer
Blurb

Cover: Milan Nykodym [flickr.com/photos/milannykodym]

Hacker Monthly is published by Netizens Media and not affiliated with Y Combinator in any way.

Contents
FEATURES

06 How Fighter Jets Lock On

By Tim

Morgan

10 Melting Aluminum
By Eric

Lippert
Photo: Milan Nykodym [flickr.com/photos/milannykodym]

STARTUP

SPECIAL

The Founders Guide To Selling Your

Company

34

14

By Justin Kan

The 1,000-Hour Rule

By Philip Guo

TDD Your API

38

Its Just Wood

By Bryan Kennedy

By Steve Klabnik

23

By Jonathan Katz

36

PROGRAMMING
18

Dont Become a Scientist!

Lost Lessons from 8-Bit BASIC

By James Hague

Writing a Simple Garbage Collector

in C

24

By Matthew Plant

30

How to Read an Executable

By JuliA Evans

For links to Hacker News dicussions, visit hackermonthly.com/issue-55

FEATURES

How Fighter Jets Lock On

By Tim Morgan

Photo: flickr.com/photos/milannykodym
he primary technology

that a military aircraft

uses to lock and track an
enemy aircraft is its onboard radar.
Aircraft radars typically have two
modes: search and track. In search
mode, the radar sweeps a radio
beam across the sky in a zig-zag
pattern. When the radio beam is
reflected by a target aircraft, an
indication is shown on the radar
display. In search mode, no single
aircraft is being tracked, but the
pilot can usually tell generally what
a particular radar return is doing
because with each successive sweep,
the radar return moves slightly.
This is an example of the fire
control radar display for an F-16
Fighting Falcon when the radar is in
a search mode:

6 FEATURES

Each white brick is a radar

return. Because the radar is only
scanning, not tracking, no other
information is available about the
radar targets. (There is one exception: The Doppler shift of the radar
return can be measured to estimate
how fast the aircraft is traveling
towards or away from you, much
like the pitch of an oncoming trains
whistle can tell you how fast its
coming at you. This is displayed as
the small white trend line originating from each brick.)

Note that the cursors are over

the bottom-most brick (closest to
our aircraft). The pilot is ready to
lock up this target. This will put the
radar into a track mode. In track
mode, the radar focuses its energy
on a particular target. Because the
radar is actually tracking a target
and not just displaying bricks when
it gets a reflection back, it can tell
the pilot a lot more about the
target. This is what the F-16s fire
control radar display looks like
when a target is locked:

Along the top we have a lot of

information about what our radar
target is doing:

Its aspect angle (angle between

its nose position and our nose
position) is 160 to the left.

Its heading is 190.

Its airspeed is 450 knots.

Our closure rate is 828 knots.

With this information, the pilot

gets a much better idea of what the
aircraft is doing, but at the expense
of information about other aircraft
in the area.
Note that in the above picture,
the bottom-most (closest) target is
locked (circle around it), the two
targets further away are tracked
(yellow squares), and there are two
radar returns even further away
(white bricks). This is demonstrating an advanced feature of modern
radars, situational awareness modes.
A radar in SAM combines both
tracking and scanning to allow a
pilot to track one or a small number
of interesting targets while not
losing the big picture of what other
targets are doing. In this mode, the
radar beam sweeps the sky, while
briefly and regularly pausing its
scan to check up on a locked target.
All of this comes with tradeoffs.
In the end, a radar is only as powerful as it is, and you can put a lot
of radar energy on one target, or
spread it out weakly throughout
the sky, or some compromise in
between. In the above photo you
can see two vertical bars spanning
the height of the display; these are
the azimuth scan limits. Its the
aircrafts way of telling you, "OK, I
can both track this target, and scan
for other targets, but in return, Im
only going to scan a 40 wide cone

in front of the aircraft, instead of

the usual 60. Radar, like life, is full
of tradeoffs.
An important thing to note
is that a radar lock is not always
required to launch weapons at a
target. For guns kills, if the aircraft
has a radar lock on a target, it can
accurately gauge range to the target
and provide the pilot with the
appropriate corrections for lead and
gravity drop to get an accurate guns
kill. Without the radar, the pilot
simply has to rely on his or her own
judgment.
As an example of that, lets take
a look at the F-16s HUD (heads-up
display) when in the process of
employing guns at a radar-locked
target:

It becomes incredibly simple:

that small circle labeled bullets at
target range is called the death
dot by F-16 pilots. Basically, it represents where the cannon rounds
would land if you fired right now,
and the rounds traveled the distance between you and the locked
target. In other words, if you want a
solid guns kill, simply fly the death
dot onto the airplane. Super simple.
But what if theres no radar lock?
Well now the HUD looks like this:

No death dot, but you still have

the funnel. The funnel represents
the path the cannon rounds would
travel out in front of you if you
fired right now. The width of the
funnel is equal to the apparent
width of a predetermined wingspan
at that particular range. So if you
didnt have a lock on your target,
but you knew it had a wingspan of
35 feet, you could dial in 35 feet,
then fly the funnel until the width
exactly lined up with the width of
the enemy aircrafts wings, then
squeeze the trigger.
And what about missiles? Again,
a radar lock is not required. For
heat-seeking missiles, a radar lock is
only used to train the seeker head
onto the target. Without a radar
lock, the seeker head scans the sky
looking for bright (hot) objects,
and when it finds one, it plays a distinctive whining tone to the pilot.
The pilot does not need radar in
this case, he just needs to maneuver
his aircraft until he has good tone,
and then fire the missile. The radar
only speeds up the process.

Photo: flickr.com/photos/milannykodym

Now, radar-guided missiles come

in two varieties: passive and active.
Passive radar missiles do require a
radar lock, because these missiles
use the aircrafts reflected radar
energy to track the target.
Active radar missiles, however,
have their own onboard radar,
which locks and tracks a target.
But this radar is on a one-way trip,
so its considerably less expensive (and less powerful) than the
aircrafts radar. So, these missiles
normally get some guidance help
from the launching aircraft until
they fly close enough to the target
where they can turn on their own
radar and go active. (This allows
the launching aircraft to turn away
and defend itself.) It is possible to
fire an active radar missile with no
radar lock (so-called maddog); in
this case, the missile will fly until

8 FEATURES

its nearly out of fuel, and then it

will turn on its radar and pursue
the first target it sees. This is not a
recommended strategy if there are
friendly aircraft in close proximity
to the enemy.
Finally, an aircraft can tell if a
radar is painting it or locked onto it.
Radar is just radio waves, and just as
your FM radio converts radio waves
into sound, so can an aircraft
analyze incoming radio signals to
figure out whos doing what. This is
called an RWR, or radar warning
receiver, and has both a video and
audio component. This is a typical
RWR display:

Although an aircrafts radar

can only scan out in front of the
aircraft, an aircraft can listen for
incoming radar signals in any direction, so the scope is 360. A digital
signal processor looks for recognizable radio chirps that correspond
to known radars, and displays their
azimuth on the scope. A chirp is a
distinctive waveform that a radio
uses. See, if two radios use the same
waveform simultaneously, theyll
confuse each other, because each
radio wont know which radar
returns are from its own transmitter. To prevent this, different radios
tend to use distinct waveforms.
This can also be used by the target
aircraft to identify the type of radar
being used, and therefore possibly,
the type of aircraft.

In this display, the RWR has

detected an F-15 (15 with a hat
on it indicating aircraft) at the
7-oclock position. The strength
of the radar is plotted as distance
from the center the closer to the
center, the stronger the detected
radar signal, and therefore possibly
the closer the transmitting aircraft.
Detected at the 12- to 1-oclock
position are two surface-to-air
missile (SAM) sites, an SA-5
Gammon, and an SA-6 Gainful.
These are Russian SAM launching
radars and represent a serious threat.
The RWR computer has determined
the SA-6 to be the highest priority threat in the area, and thus has
enclosed it with a diamond.
RWR also has an audio component. Each time a new radar signal
is detected, it is converted into an
audio wave and played for the pilot.
Because different radars sound different, pilots learn to recognize different airborne or surface threats by
their distinctive tones. The sound
is also an important cue to tell the
pilot what the radar is doing: If the
sound plays once, or intermittently,
it means the radar is only painting
our aircraft (in search mode). If a
sound plays continuously, the radar
has locked onto our aircraft and is
in track mode, and thus the pilots
immediate attention is demanded.
In some cases, the RWR can tell if
the radar is in launch mode (sending radar data to a passive radarguided missile), or if the radar
is that of an active radar-guided
missile. In either of these cases, a
distinctive missile launch tone is
played and the pilot is advised to
immediately act to counter the
threat. Note that the RWR has no
way of knowing if a heat-seeking
missile is on its way to our aircraft.

Aside from radar, there are other

technologies that are used to lock
onto enemy aircraft and ground
targets. A targeting pod is a very
powerful camera mounted on an
articulating swivel that allows it to
look in nearly every direction. This
camera is connected to an image
processor that is able to tell apart
vehicles and buildings from surrounding terrain and track moving
targets. This is the SNIPER XR
targeting pod:

Lastly, there are laser-guided

missiles as well. These beam riders
follow a laser beam emanating from
the aircraft to the target. Many
ground vehicles use laser rangefinders as well, and some aircraft
include a laser warning system
(LWS) that works similarly to an
RWR, but displays incoming laser
signals instead. n
Tim Morgan is a Ruby on Rails developer
at Square and holder of a commercial pilot
certificate. He has worked at a variety of
startups as a web developer over the past
eight years. Tim got his pilot certificate in
2007, and has been enjoying it as a hobby
ever since.
Reprinted with permission of the original author.
First appeared in hn.my/lockon (quora.com)

And this is what the pilot sees

when he operates it:

The pod is able to track vehicles

day and night, using visual or infrared cameras. Heat-seeking missiles
obviously use this same technology
to home in on aircraft, and electrooptical missiles use this technology
to track ground targets.

Melting Aluminum
By Eric Lippert

eres another install-

ment of my ongoing, seldom-updated

series of posts about building my
own backyard foundry. Today Ill
describe how the final step works:
actually melting and pouring the
metal.
Start by assembling all the equipment youll need in one place on a
day with no chance of rain.
Going from left to right:

Im wearing safety goggles, a

leather apron, cotton shirt and
trousers, and leather boots without laces with the trouser legs
around them. You want the boots
to protect you from accidentally
splashed molten metal, but also
want to be able to kick them off
in a hurry if theyre on fire. I wear
natural fibers for the rest of my
clothing; synthetic fibers can melt
onto you when exposed to high
heat and produce a nasty burn.
The mold is in a deep iron tray
an old baking pan. If the mold
separates then the aluminum will
spill into the tray, rather than
onto my concrete driveway. Concrete can explode when heated
rapidly.

10 FEATURES

My ingot tray, a rusty old muffin

tin, is also on an iron slab, this
one the former top of a table saw
that I found on the side of the
road.

The white can contains broken

up bits of scrap aluminum, old
sprues, and ingots to re-melt.

The furnace and lid; note that

the furnace is strapped to a
hand truck for ease of moving
it around, since it weighs 100
pounds.

A bag of charcoal. The grocery

store was out of wood charcoal,
which is preferable because it
leaves less clinker tiny bits of
stone in the ash than briquettes, which are compressed
wood and coal dust.

A bucket half full of sand. In the

event of a fire caused by spilling
liquid metal, you always want to
put it out with sand, not water.
Water will turn to steam and
launch the molten metal off in a
random direction.

Crucible lifters (in the bucket)

two iron rods with the ends bent
into hooks that fit into the loops
on the top of the crucible.

Crucible pourer (in the bucket)

a thick iron rod with a bolt
through it, so that I can lift the
crucible by its loops and then
pour it.

Stir stick (in the bucket) an

iron rod to move things around
inside the furnace.

A long-handled spoon, to skim

the dross off the top of the melt.

A pair of pliers. In an emergency

if I need to move the crucible in
a hurry I can pick it up with the
pliers.

The crucible a metal pipe with

loops bolted to the opening is
in the bucket right now as well.

Paper, lighter fluid, and matches

to get the furnace going.

A tub containing aluminumfoil-wrapped packets of KCl and

Na2CO3, used to flux the melt
and reduce dissolved hydrogen.
Today Im melting very clean
aluminum so Im not anticipating
that either will have much of an
effect, but it doesnt hurt.

Leather gloves and a face shield,

for later.

In the foreground, a thrift store

hair dryer. In retrospect I realize that Ive created a tripping
hazard by running the hose
between the furnace tools and
the mold; next time Ill remember to run the hose the long way
around. But I am never going to
be walking backwards carrying
the crucible, so it wont be too
bad.

I put a layer of charcoal with some lighter fluid and

paper in the furnace, get it lit, and a few minutes later,
turn the air on low:

Note that I have removed the lighter fluid from the

area. Lets keep the flammable liquids away from the
liquid metal, eh?
A few minutes later the layer of charcoal at the bottom
of the furnace is burning nicely, so I turn off the air and
put the crucible in the furnace.

11

I then surround the crucible with additional charcoal, put the lid on the furnace, turn the air back on,
and put an aluminum tube into the crucible through
the hole in the lid. Theres also a small amount of scrap
aluminum in the crucible already. You dont want to fill
the crucible with big pieces of aluminum as it will
expand as it heats, and possibly damage the crucible.
Its less of a concern with an iron crucible like mine;
ceramic crucibles are inelastic and can easily crack if
theyre full of expanding metal.

The fire will be smoky for some time as it is still

low-temperature and theres a lot of new charcoal to
catch fire. Note that I have removed the charcoal bag
from the area. Im planning on melting a lot of aluminum today part of the point of todays melt is to
find out just how much I can fit into the crucible so
I will need to add more charcoal halfway through. But
I dont want the bag cluttering up the area.
Ten minutes later the exhaust is much cleaner and
hotter. Aluminum is not like iron, which gradually
gets softer and more malleable as it heats up. Rather,
aluminum is more like water ice: as it approaches the
melting point it suddenly starts fracturing easily, then
it gets into a slushy state, and then it becomes a liquid.
Ten minutes in, the aluminum in the furnace is not
liquid but I could tear it with my stir stick more easily
than I could tear paper. The tube has started to collapse
into the furnace under its own weight.

12 FEATURES

I put on my gloves and face shield and add more

bits of scrap. Less than three minutes later the tubes
have melted completely. The liquid aluminum has
a huge surface area on the red-hot crucible bottom,
and transfers heat very quickly to the solid aluminum
tubes sinking into it. Eight minutes later I have melted
an entire lawn chair and am out of tubes, so I start on
sprues and ingots, which take longer because they have
much less surface area per unit mass. At this point the
charcoal has burned down quite a bit, so I add some
more around the crucible.
Aluminum tubing is easy to find as scrap, and in
fact for todays melt, I didnt even have to leave my
property; someone left a broken lawn chair on my
front lawn. Normally I am irritated when people leave
trash on my lawn, but I will gladly accept broken lawn
chairs! The alloy is designed for easy extrusion, not for
metalworking, so it is not ideal, but you cant beat the
price, and it usually is not painted. Paint (or vinyl coating) makes toxic fumes and causes the melt to take on
hydrogen gas, which then forms bubbles in the casting.
The Na2CO3 will help remove the hydrogen, but Id
rather not go there in the first place if I can avoid it.
Twenty minutes later I have melted enough sprues
and ingots that the crucible is almost completely full. I
take the lid off, skim the dross the bits of aluminum
oxide and impurities that float off the top and put it
in the ingot tray to cool; this will be trash. (Trying to
recover aluminum from aluminum oxide is not really
worth it when aluminum is so plentiful.) I add some
flux and degassing powder, and then turn the air blast
on to maximum to get the last bit of heat out of the
charcoal. I remove the cover from the mold. A minute
or so later, Im ready to pour; I remove the lid, take off
the last bits of dross, and get ready to take the very full

crucible out. Heres what the inside of the furnace

looks like:

This isnt quite as hot as I would like Id prefer

that the crucible be more obviously red-hot but this
was the largest melt Ive done, and its taken longer
than usual. As you can see, almost all the fuel is used
up. Its hot enough to pour, and so Im going for it. I
very carefully lift the crucible out of the furnace and
put it down in the sand bucket:

And then lift it right back up again with the pourer:

And then pour into the mold; the remainder goes in

the muffin tin.

Then wait for it to cool, dump the sand back into

the sand bin and extract the casting from the sand. The
result was 1632 grams of aluminum, so now I know
the maximum mass of aluminum I can melt with this
crucible and furnace. n
Eric Lippert designs C# analyzers at Coverity; previously he was
a member of the C# and JavaScript language design teams at
Microsoft. When not working on computers he enjoys playing
the piano, building things out of wood and metal, and trying to
keep his tiny sailboat upright. He writes about all of these fabulous
adventures and more at ericlippert.com
Reprinted with permission of the original author.
First appeared in hn.my/alu (ericlippert.com)

13

STARTUP

The Founders Guide To

Selling Your Company
By Justin Kan

or most founders, selling a

company is a life-changing
event that they have had
no training for. At Y Combinator,
one big thing we help our startups
with is navigating questions around
the acquisition process. Originally,
I wrote this guide for YC startups
outlining what Ive learned about
selling startups in my last ten years
as an entrepreneur. If you are going
through an acquisition, hopefully
this will be useful to you.

When to Sell
Similar to raising money, the best
time to sell your startup is when
you dont need to or want to. Paradoxically, you are probably thinking
about selling your startup as you
are experiencing a lack of traction,
tough competition, or difficult time
fundraising. However, this is a bad
time to sell your startup: you will
have few bidders and be more likely
to acquiesce to the demands of
anyone who does show up.
The best time to sell your startup
is when you have many options.
These options dont all have to be
acquisition offers, they can also be
venture term sheets for your next
round. You might even be operating

14 STARTUP

profitably and find yourself in the

enviable position of confidently
being able to turn down an offer.
Usually, you will have these options
because your startup is actually
experiencing great traction; counterintuitively, the best way to build
to flip is actually the same as building a successful company.
The following is a brief overview
of the steps that go into valuing
your company, garnering interest in
it and navigating through the acquisition process.

Starting Acquisition Talks

Do not enter acquisition talks
unless you are ready to sell your
company. Negotiating an acquisition is the most distracting thing
you can do in a startup: going
through M&A is an order of magnitude more distracting than raising
money. All of your ability to run the
day-to-day operations of your company will grind to a halt. You should
only enter an acquisition process if
you are certain 1) you want to sell
the company and 2) you are likely
to get a price you will accept. Dont
talk to potential acquirers just to
see what price you can get.

How Your Startup Will be Valued

Investors value companies based
on either their financial value or
their strategic value. A companys
financial value hinges on its profits
and model of its future cash flows.
For the vast majority of startups in
tech, this will be zero. It is much
more likely is that your startup will
be valued based on where it fits in
with the acquiring companys short
or long term strategy. Here are
some surprisingly common reasons
your startup has strategic value to
an acquirer:

The CEO finds it interesting,

or wants to keep it away from
another large tech company.

The executive that runs the

relevant division of the acquirer
needs to demonstrate big
moves.

A competitor to an acquirer is
out-executing it in a business
and you can help the acquirer
become better.

The acquirer doesnt have or

cant retain talent in an area
where you have employees.

Your businesses actually have

some synergies and combining them is theoretically
value-accretive.
The acquirer is running a similar
business but you are executing
much better. They are afraid of
you.

If some of these reasons seem

ridiculous and arbitrary, it is
because sometimes they are.
Remember that companies are
bought, not sold: in order for a
company to want to buy you, an
internal champion will have to
internalize one of these reasons.
This isnt something that can be
forced.
When it comes to setting price,
there is no right price for a company, there is only the price that
you can negotiate. Management
teams, investment bankers, and
corporate development people will
concoct an array of metrics like
cost per user to justify a number.
Ultimately, though, the clearing
price for a startup depends on what
the big company can justify to
the market (previous comparable
acquisitions are a good benchmark)
and the sale price agreed to by you
and your investors.
A potential acquirers first offer is
rarely its best offer. Dont be afraid
to say no the potential acquirer
isnt going anywhere. There are
many negotiation strategies, but in
order to extract the most value you
need to (1) be willing to walk away
and (2) initiate a competitive bidding process.

Getting Offers

Bullshit Offers

The best way to solicit acquisition

offers is to have ongoing conversations with potential acquirers
about ways you can work together.
These conversations usually involve
many different people within a big
company; youll only get an offer
once a sufficiently high-up decision
maker is convinced that buying you
is a better idea than partnering with
you.
When you have an offer, the
next thing to do is determine your
alternative options. You can do this
by letting other potential acquirers (including larger companies
you have partnered with and/or
competitors to the acquirer) know
that you have received a term
sheet from an acquirer and you are
considering selling yourself, but
would prefer a longer term future
with their company (find a reason).
Now would also be a good time to
call up VCs you have been talking
to and ask for a term sheet for your
next round.
Sometimes, a company you are
doing a critical business development partnership with will insist
on a cool down period in an
agreement. This is a period of time
during which you have to wait
once youve received an acquisition
offer before you can sign (for your
partner to theoretically prepare a
better counter offer). Cool down
clauses can actually work in your
favor as a way to acquire additional
offers once youve received a first
term sheet.

Most of the offers you will receive

to buy your startup will be bullshit
offers. It costs a company exactly
zero dollars to tell you: we want to
buy your startup. In fact, companies employ teams of people, under
the euphemism Corporate Development, to go around the Valley
and repeat this phrase to founders.
Bullshit offers are dangerous
because they can lull you into a
false sense that you are being successful. Like TechCrunch articles,
bullshit offers are a vanity metric,
not an actual measure of success.
You can tell if an offer is bullshit
because it will not be accompanied by an expiration date and/or
a promise of a term sheet delivery
within a very short period of time
(24-48 hours). When a sufficiently
high-up decision maker decides he/
she wants to buy your startup, he/
she will attempt to meet with you
constantly and put time pressure
on you, so as to prevent you from
shopping the deal and getting a
better offer. The absence of this
behavior indicates the other company is not serious about acquiring
your business.
Often the expectations of the
founders and corporate development people are very divergent.
Before proceeding far into conversations with big acquirers, you
should try to clarify valuation
expectations (and other important
consideration details, like retention
packages) as quickly as possible.
This strategy should help prevent
you from having half a dozen meetings, only to find out the potential
acquirer expects to pay $10 million
for your rapidly growing startup
that already has a term sheet for a
$15 million A round.

15

the startup, you have all the leverage

Asbefore
you sign a term sheet. Once you

sign, you have almost no leverage at all.

Hiring a Banker

Pre-Term Sheet Diligence

Like the world of venture capital, investment banking is a field

filled with a very small number of
extremely well-connected, analytical, and experienced people
and a much larger number people
pretending to be those things. It
is unlikely you will be able to find
someone in the former category
unless your startups realistic selling
price is in the mid-hundreds of millions or above. But the good news
is that you probably dont need an
investment banker at all unless your
selling price is that high, and maybe
not even then.
Investment bankers are expensive
(1 to 2% of the total deal value).
However, the good ones can help
you get a thorough understanding
of the competitive landscape, who
the individual decision makers are
at every potential acquirer, and
what buttons to push to maximize
your deal value. Also, the people
you are negotiating with in corporate development are professional negotiators: they spend all
day every day trying to pay less for
target companies. You have probably spent a lot of time doing things
that are not negotiating. Having a
professional negotiator on your side
is often extremely valuable, if you
can get someone good.

If you are committed to going

through an acquisition process, you
should be fairly free with your company data (under an NDA), as it
is better that the acquirer uncover
any red flags before youve signed a
term sheet and entered the closing
process. However, if the potential
acquirer asks to interview your
team you should absolutely refuse
(unless you are going through a
talent acquisition and have no
other options). Letting a potential
acquirer interview your team is
extremely distracting for them, and
signals to the acquirer that you are
willing to bend over.

16 STARTUP

Signing a Term Sheet

Once you have collected all your
term sheets, you can sign one.
Before you do, you should try to
negotiate the business and legal
points in as much detail as possible.
Feel free to push back on exploding
offer deadlines and other pressure
to sign immediately. After you sign,
you can expect any points that
werent previously negotiated will
end up with language in favor of
the acquirer.

As the startup, you have all the

leverage before you sign a term
sheet. Once you sign, you have
almost no leverage at all. This is
mental: before you sign a term
sheet, you havent yet decided to
sell. Once you do, you have committed to selling the deal to yourself, your employees and investors.
Once you get negotiation fatigue
and maybe even before you will
start to agree quickly to things you
wouldnt have considered at the
term sheet stage. Also, once youve
signed a term sheet you can no
longer shop your company to other
acquirers. If your deal falls apart,
other acquirers may have cooled off
or think that the deal fell through
because your company is damaged goods. It may be impossible to
resuscitate your other options.
When negotiating a term sheet,
push for a shortest possible closing
period (target 30 days) to avoid getting deal fatigue and to put pressure
on the acquirer (although, be aware
that sometimes a regulatory issue
will dictate the timing of the closing and that is outside of everyones
control). A short closing period will
also help you somewhat limit the
distraction from your main job: running your startup.

Before you sign a term sheet

that commits you to either working at the acquirer or accepting the
acquirers stock as consideration for
the acquisition, ask yourself if you
actually believe in the company.
The Valley is replete with cautionary tales of startups that sold themselves in exchange for the stock
of ultimately worthless acquirers.
Dont let yourself become another
one.
Just because you signed a term
sheet does not mean your deal is
done; in fact, it is very possible that
it will still fall apart. Despite a commitment to trying to close, companies change their minds all the time
during the diligence process. If your
deal fails and the result is that your
startup is collateral damage because
you were distracted, remember
that this is probably an acceptable
outcome for the would-have-been
acquirer.

Closing
The most dangerous stretch during
the acquisition process is the time
between when you decide you are
going to sell, and when the sale
actually occurs. Youve decided to
exchange stress for riches, and you
can already see that new house on
the horizon (and maybe one for
your parents).
What happens when the acquirer
comes back and changes the total
deal value?
What happens when the acquirer
changes its mind, and you have to go
back to the grind?
What happens when the acquirer
has talked to your senior management, and then decided your team
isnt good enough?
If you are running low on cash,
what happens when you run out of
cash before the deal closes, because

negotiating the deal documents took

twice as long as you expected?
These things happen. You should
be prepared to walk away from any
deal up until the point where you
are watching your bank account,
waiting for the wire transfer from
the acquirer to hit.
Getting to closing is a process
largely driven by lawyers. The
items to be negotiated are typically divided into legal points and
business points the lawyers will
resolve the legal points, but you are
expected to figure out the business issues. Remember that you
are ultimately responsible for the
outcome, and that the lawyers work
for you (and usually get paid if the
deal happens or not). You have
to stay on top of the lawyers to
make sure that they arent slowing
down the deal by getting bogged
down in details that dont actually
make any difference to you or your
stakeholders.

In Summary
Entering the acquisition process is
one of the most dangerous things
an early stage startup can do,
because the process is distracting,
demoralizing, and usually involves
giving your competition most of
your proprietary business data.
Founders who have been through
the process have said it is ten times
as distracting as fundraising. It often
cripples your ability to oversee the
business operations. Do not enter
into an acquisition process lightly. n
Justin Kan is a partner at Y Combinator.
Previously he founded Justin.tv, TwitchTV
and Socialcam.
Reprinted with permission of the original author.
First appeared in hn.my/sellingcompany (justinkan.com)

17

PROGRAMMING

TDD Your API

By Steve Klabnik

How do you TDD an API?

If youre not familiar, heres
the basic outline of Test Driven
Development:
1. Write a test for some behavior
youd like to introduce into your
system.
2. Run your test suite, and make
sure that test fails.
3. Write the simplest code that
implements the behavior.
4. Run your test suite, and make
sure that test passes.
5. Refactor, because the simplest
code often has undesirable
properties.
6. Commit, and GOTO 1.
To say that theres a large amount
of literature on the benefits of this
approach would be an understatement. I just want to focus on one
of the properties of this approach:
verifying behavior. TDD verifies
behavior in a few ways. The first
is in step two, with the failing
test. This failing test is important
because it verifies that our test
does test for a new behavior. If the
test was passing at this point, we
probably have a bug in our test.

18 PROGRAMMING

The second way that behavior is

verified in TDD is step four, when
the test passes. This verifies that our
new behavior satisfies the contract
weve laid out in the test, which is
incredibly important! Finally, when
software is changed, we use TDD
to verify that only the behavior we
were interested in changing has
changed. We dont want our good
change to break anything else.
So how is this different when
applied to the API context? Heres
how you to TDD for an API:
1. Create a separate project for
your API tests.
2. Write a test for some behavior
youd like to introduce into
your system. Because this test
suite is in a different project,
this will be an integration-style
test, which makes HTTP calls
against your customer-facing test
environment.
3. Run your test suite, and make
sure that test fails.
4. Implement this behavior in your
system, and push that code to
your test environment.
5. Run your test suite, and make
sure that test passes.

6. Refactor any of your test code

that may have gotten messy.
7. Commit, and GOTO 1.
Theres another aspect which
is important, though. This project
should be at least partially public
and included with all of your other
documentation. Because Balanced
is an open company, ours is 100%
public, but if you arent doing new
feature development in the open
the way we are, just have the latest
passing suite public. This aspect is
important because, in this model,
these API tests become the canonical source of truth for what your
API provides to customers. Your
customers can trust them because
theyre automatic and open. Does
the API work? Check the build
status.
Push to card [balancedpayments.
com/push-to-card] was the first real
feature weve built in this way. Its
been a long time coming, though.
Lets take a journey.

Origins
From the start, there was an idea that something
needed to be built to handle some API-related things.
With this dummy commit, the first question kicked off:
Markdown or reStructured Text.
There were three big questions to tackle:
1. How can we validate that our API is working as
intended?

* - ``name``
- The display name of the account
`optional`.

Some bits removed for brevity. And then creating a

card:
====================
cards.create
====================

2. Can we generate documentation from this?

3. How would all this be written, tool-wise?
While this was being decided, at least Issues provided a good forum to discuss things. From the first
version of the README:
The primary goal of this repo is to create more openness
behind the decisions driving the designs and functionality of the Balanced API. We reached out to existing
and potential customers when designing the API, but
that was a limited set of people we already knew. Weve
received tremendous growth in the last few months,
and our new customers have great feedback or at least
want to understand the reasoning behind the original
decisions.

An initial solution
It was settled that reStructured Text was the answer.
Heres what they looked like, roughly.
===============
accounts.create
===============
:uri: /v1/marketplaces/
(marketplace:marketplace)/accounts
:methods: POST
Fields
******
.. list-table::
:widths: 20 80
:header-rows: 1
* - Name
- Description
* - ``email_address``
- The email address of the account, unique
constraint.

:uri: /v1/marketplaces/
(marketplace:marketplace)/cards
:methods: POST

This was okay, but it wasnt great. For example, I

now have an account, and I have a card, but I cant add
that card to that account, because theyre two isolated
tests, not one larger test. These tests assume that each
URL/HTTP method combination can or should be
tested in isolation, and thats just not true. Because of
this, checking if tests passed or failed often involved
humans. The diff would show that a timestamp was a
fraction of a second off, or that an autogenerated ID
would appear which couldnt be predicted ahead of
time.
reStructured Text wasnt really designed to do this,
anyway. A new approach was sought.

The YAML year

The next iteration of this idea used YAML instead of
rST. The YAML tests looked like this:
require:
- ../card_fixtures.yml
- ../customer_fixtures.yml
scenarios:
- name: add_card_to_customer
request:
method: PATCH
href: "{card,cards.href}"
schema:
"$ref": "requests/_patch.json"
body: [{
"op": "replace",
"path": "/cards/0/links/customer",
"value": "{customer,customers.id}"
}]
response:
schema:

19

"$ref": "responses/cards.json"
matches: { "cards": [ { "links": { "customer": "{customer,customers.id}" } } ] }

Its almost a serialized HTTP request. Since they

have names, they can refer to each other, and you
can do things like interpolate variables, use a regular
expression to only match the parts you care about, and
make scenarios depend on each other for re-usability.
Speaking of which, how does that work? Well, each
scenario has a name, and if you refer to a previous
scenario, it will run that one first. So, in the href line
there, it refers to {card,cards.href}. This says run the
card scenario, and then get the value of cards.href in
the resulting JSON. We later do the same, twice, with
{customer,customers.id}.
Where are the card and customer scenarios? The first
line of the file requires two other YAML files, those are
probably good candidates. Here they are in their full
glory. Can you find the dependent scenarios?
customer_fixtures.yml:
scenarios:
- name: customer
request:
method: POST
href: /customers
schema:
"$ref": "../requests/customer.json"
body: {
"name": "Balanced testing"
}
response:
status_code: 201
schema:
"$ref": "../responses/customers.json"
- name: underwritten_merchant
request:
method: POST
href: /customers
schema:
"$ref": "requests/customer.json"
body: {
"name": "Henry Ford",
"dob_month": 7,
"dob_year": 1963,
"address": {
"postal_code": "48120"
}

20 PROGRAMMING

}
response:
status_code: 201
schema:
"$ref": "responses/customers.json"
matches: { "customers": [ { "merchant_
status": "underwritten" } ] }
- name: customer_with_card
request:
method: POST
href: /customers
schema:
"$ref": "requests/customer.json"
body: {
"name": "Darius the Great",
"email": "[email protected]",
"source": {
"name": "Darius the Great",
"number": "4111111111111111",
"expiration_month": 12,
"expiration_year": 2016,
"cvv": "123",
"address": {
"line1": "965 Mission St",
"line2": "Suite 425",
"city": "San Francisco",
"state": "CA",
"postal_code": "94103"
}
},
"meta": {
"ip_address": "174.240.15.249"
}
}
response:
status_code: 201
schema:
"$ref": "responses/customers.json"
- name: merchant_with_bank_account
request:
method: POST
href: /customers
schema:
"$ref": "requests/customer.json"
body: {
"name": "Henry Ford",
"dob_month": 7,

"dob_year": 1963,
"address": {
"postal_code": "48120"
},
"destination": {
"name": "Kareem Abdul-Jabbar",
"account_number": "9900000000",
"routing_number": "021000021",
"account_type": "checking"
}
}
response:
status_code: 201
schema:
"$ref": "responses/customers.json"
matches: { "customers": [ { "merchant_
status": "underwritten" } ] }

Whew! Thats a ton of stuff. grep can help you a bit

here, but its still not very fun. The same words repeat
quite a bit.
Heres card_fixtures.yml:
require:
- ./customer_fixtures.yml
scenarios:
- name: card
request:
method: POST
href: /cards
schema:
"$ref": "requests/card.json"
body: {
"number": "4111 1111 1111 1111",
"expiration_month": 12,
"expiration_year": 2016,
"cvv": "123",
"address": {
"line1": "965 Mission St",
"postal_code": "94103"
}
}
response:
status_code: 201
schema:
"$ref": "responses/cards.json"
- name: customer_card
request:
method: POST

href: /cards
schema:
"$ref": "requests/card.json"
body: {
"number": "4111 1111 1111 1111",
"expiration_month": 12,
"expiration_year": 2016,
}
response:
status_code: 201
schema:
"$ref": "responses/cards.json"
- name: associate_customer_card_with_customer
request:
method: PUT
href: "{customer,customers.href}"
schema:
"$ref": "requests/customer.json"
body: {
"card_uri": "{customer_card,cards.href}"
}
response:
schema:
"$ref": "responses/customers.json"
- name: card_processor_failure
request:
method: POST
href: /cards
schema:
"$ref": "requests/card.json"
body: {
"number": "4444444444444448",
"expiration_month": 12,
"expiration_year": 2018
}
response:
status_code: 201
schema:
"$ref": "responses/cards.json"

Did that make your eyes glaze over? Exactly.

Nobody actually wants to write YAML. Nobody actually wants to read YAML. Tests that refer to each other
help reusability but harm understanding. Since this was
entirely home-grown, people didnt really know how it
worked. This also meant that there was a large amount
of work to get up and running.

21

Time to throw it all out and do something again!

At this point, it was clear that building something
totally custom wasnt the right answer. The maintenance costs are just too high. And other people build
APIs, so they must do something like this, right? Well,
while there are some tools for doing things like this,
none of them are particularly great. They were all
missing at least one thing that we considered vital.
One of the bigger issues is that many of these tools
assume that youre simply doing Rails style REST, and
not using hypermedia. Or they require you to write a
WSDL or WADL or use RAML.

Eat your cucumbers!

Replacing this system was my first task here. I decided
to use Cucumber. The Cucumber tests look like this:
Feature: Credit cards
Scenario: Add a card to a customer
Given I have tokenized a card
And I have created a customer
When I make a PATCH request to /cards/:card_
id with the body:
"""
[{
"op": "replace",
"path": "/cards/0/links/customer",
"value": ":customer_id"
}]
"""
Then I should get a 200 OK status code
And I make a GET request to /cards/:card_id
Then the response is valid according to the
"cards" schema
And the fields on this card match:
"""
{
"links": { "customer": ":customer_id"
}
}
"""

Im often skeptical of Cucumber, and its often

misused, but I think this is a pretty great use case. Its
language agnostic, which is important to us, and its
got lots of room for explanations and clarifications. Its
clear that this needs a card and a customer to work,
and I can just find the step definitions (or ask Cucumber to tell me where they are) to see the details.

22 PROGRAMMING

Here are the specs [hn.my/p2cspec] for push-tocard. By using this approach, we caught problems
before the feature was even finished. The initial implementation contained a bug with a typo in one of the
response keys. There was some confusion in one corner
of the spec, and while some of the discussion happened
in person, the ways in which the spec failed while the
feature was being built out helped make sure that
everyone was on the same page.

Going forward
Now that we have a testable validation that our API
is working as intended, we can do all kinds of things.
Wed like to integrate this suite into our more general
continuous integration suite. Wed like to run this test
every hour, or every three hours, which can be a form
of alerting against regressions.
In order to do all that, we need to fix our build.
Thats what happens when you dont run the tests
automatically: regressions can creep in and you wont
realize it. These failures are all related to a bug which
only happens with a brand new marketplace, and
doesnt strictly affect customer behavior. I wanted the
fix to roll out before this post, but you dont always
get what you want. This is a good reminder that tests
are not perfect, and that theyre a tool to alert you that
something may have gone wrong, not proof that theres
an error. Test code can be fragile or have bugs, too. But
without these tests, we wouldnt have realized that
there was a regression, no matter how small.
Its hard to get any complex software system behavior correct. So far, weve found that these external tests
give us a really nice forum for working out these kinds
of issues. They also give us something to share with our
customers, and a way to ask for advice from experts
outside of the company. n
Steve Klabnik is a Rails committer, Rust enthusiast, and works for
Balanced Payments in San Francisco. He has authored Designing Hypermedia APIs, Rust for Rubyists, and Rails 4 in Action.
When hes not programming, he reads philosophy books and
plays Android: Netrunner.
Reprinted with permission of the original author.
First appeared in hn.my/tddapi (balancedpayments.com)

Lost Lessons from 8-Bit BASIC

By James Hague

nstructured program-

ming with GOTO is the

stuff of legend, as are
calling subroutines by line number
GOSUB 1000 and setting
global variables as a mechanism for
passing parameters.
The little language that fueled
the home computer revolution
has been long buried beneath an
avalanche of derision, or at least
disregarded as a relic from primitive
times. Thats too bad, because while
the language itself has serious shortcomings, the overall 8-bit BASIC
experience has high points that are
worth remembering.
Its hard to separate the language
and the computers it ran it on; flipping the power switch, even without a disk drive attached, resulted
in a BASIC prompt. If nothing else,
it could be treated as a calculator:

PRINT "seconds in a week:

",60*60*24*7

or
PRINT COS(2)/2

Notice how the cosine function is

always available for use. No importing a library. No qualifying it with
MATH.TRIG.
Or take advantage of this being a
full programming language:
T = 0
FOR I=1 TO 10:T=T+I*I:NEXT I
PRINT T

It wasnt just math. I remember seeing the Atari 800 on display in Sears, the distinctive blue

background and READY prompt

visible across the department. Id
switch to a bitmapped graphics
mode with a command window at
the bottom and dash off a program
that looped across screen coordinates displaying a multicolored
pattern. It would run as an in-store
demo for the rest of the day or until
some other know-it-all pressed the
BREAK key.
Theres a small detail that I
skipped over: entering a multiline program on a computer in a
department store. Without starting
an external editor. Without creating a file to be later loaded into the
BASIC interpreter (which wasnt
possible without a floppy drive).
Heres the secret. Take any line of
statements that would normally get
executed after pressing return:
PLOT 0,0:DRAWTO 39,0

and prefix it with a number:

10 PLOT 0,0:DRAWTO 39,0

The same commands, the same

editing keys, and yet its entirely different. It adds the line to the current
program as line number 10. Or if
line 10 already exists, it replaces it.
Lines are syntax checked as
entered. Well, each line is parsed
and tokenized so that previous
example turns into this:
Line #: 10
Bytes in line: 6
PLOT command
X: 0
Y: 0
DRAWTO command
X: 39
Y: 0

Thats how the line is stored in

memory, provided there arent any
errors. The displayed version is an
interpretation of those bytes. Code
formatting is entirely handled by
the system and not something you
think about.
All of this, from the alwaysavailable functions, to being able
to develop programs without
external tools, to code stored as
pre-parsed tokens, made BASIC not
just a language but a development
system. Compare that to most of
todays compilers which feed on
self-contained files of code. Sometimes theres a run-eval-print loop
so theres interactivity, but editing
real programs happens elsewhere.
And then there are what have come
to be known as Integrated Development Environments which tie
together file-oriented compilers
with text editors and sometimes
interactive command lines, but now
they get derided for reasons that
BASIC didnt: for being bulky and
cumbersome.
Did I mention that Atari BASIC
was contained in an 8kb ROM
cartridge?
How did IDEs go so wrong? n
James Hague has been Design Director
for Red Faction: Guerrilla, editor of Halcyon Days: Interviews with Classic Computer and Video Game Programmers,
co-founder of an indie game studio, and
a published photographer. He started his
blog Programming in the 21st Century,
in 2007.
Reprinted with permission of the original author.
First appeared in hn.my/8bitbasic (dadgum.com)

23

Writing a Simple Garbage

Collector in C
By Matthew Plant

think that
writing a garbage collector is really hard, a deep
magic understood by a few great
sages and Hans Boehm (et al)
[hn.my/boehm]. Well, its not. In
fact, its rather straight forward. I
claim that the hardest part of writing a GC is writing the memory
allocator, which is as hard to write
as it is to look up the malloc
example in K&R.
A few important things to note
before we begin. First, our code will
be dependent on the Linux kernel.
Not GNU/Linux, but the Linux
kernel. Second, our code will be
32-bit and not one bit more. Third,
please dont use this code. I did not
intend for it to be wholly correct,
and there may be subtle bugs I did
not catch. Regardless, the ideas
themselves are still correct. Now,
lets get started.
eople seem to

Making the Malloc

To begin, we need to write a
memory allocator, or as we will be
calling it, a malloc function. The
simplest malloc implementations
maintain a linked-list of free blocks
of memory that can be partitioned
and given out as needed. When a
user requests a chunk of memory,
a block of the right size is removed
from the free list and returned. If
no blocks of the right size exist,
either a block of a larger size is
partitioned into smaller blocks or
more memory is requested from the
kernel. Freeing a chunk of memory
simply adds it back to the free list.
Each chunk of memory in the
free list begins with a header
describing the block. Our header
will contain two fields, one indicating the size of the chunk and the
second pointing to the next free
block of memory:
typedef struct header {
unsigned int
size;
struct block
*next;
} header_t;

24 PROGRAMMING

Using headers that are embedded

in the memory we allocate is really
the only sensible way of doing this,
but it has the added benefit of automatically word-aligning the chunks,
which is important.
Because we will need to keep
track of the blocks of memory currently in use as well as the blocks
that are not, we will have a used list
in addition to a free list. Items will
be added to the used list when they
are removed from the free list, and
vice-versa.
We are almost ready to complete
the first step and write our malloc
implementation. Before we do that,
we first need to understand how to
request memory from the kernel.
Dynamically allocated memory
resides in the so-called heap, a
section of memory between the
stack and the BSS (uninitialized
data segment all your global
variables that have the default
value of zero). The heap starts at
a low address bordering the BSS
and ends at the program break,
which resides somewhere between
the BSS and the stack. Attempting to access any memory between

the stack and the break will cause an access violation

(unless you access within the amount the stack can be
extended by, but thats a whole separate conversation).
In order to obtain more memory from the kernel, we
simply extend the break, thus allowing us to access
more memory. To do this, we call the Unix sbrk system
call, which extends the break by its argument and
returns the address of the previous break on success,
thus giving the program more memory. On failure, sbrk
returns -1 casted to a void pointer, which is a terrible
convention that no one likes.
We can use this knowledge to create two functions:
morecore and add_to_free_list. In the case that we are
out of blocks in the free list, we will call morecore to
request more memory. Since requesting the kernel for
more memory is expensive, we will do it in page-size
chunks. Knowing what a page is isnt important right
now, but a terse explanation is that it is the smallest
unit of virtual memory that can be mapped to any
particular location in physical memory. We will use the
function add_to_free_list to do exactly what it sounds
like.
/*
* Scan the free list and look for a place to
put the block. Basically, we're
* looking for any block to be freed block might
have been partitioned from.
*/
static void
add_to_free_list(header_t *bp)
{
header_t *p;
for (p = freep; !(bp > p && bp < p->next); p
= p->next)
if (p >= p->next && (bp > p || bp <
p->next))
break;

} else
p->next = bp;
freep = p;
}
#define MIN_ALLOC_SIZE 4096 /* We allocate
blocks in page sized chunks. */
/*
* Request more memory from the kernel.
*/
static header_t *
morecore(size_t num_units)
{
void *vp;
header_t *up;
if (num_units < MIN_ALLOC_SIZE)
num_units = MIN_ALLOC_SIZE /
sizeof(header_t);
if ((vp = sbrk(num_units *
sizeof(header_t))) == (void *) -1)
return NULL;

up = (header_t *) vp;
up->size = num_units;
add_to_free_list (up);
return freep;
}

Now that we have our two helper functions, writing our malloc function is pretty straightforward. We
simply scan the free list and use the first block that is at
least as big as the chunk were trying to find. Because
we use the first block we find instead of trying to find a
better block, this algorithm is known as first fit.

if (bp + bp->size == p->next) {

bp->size += p->next->size;
bp->next = p->next->next;
} else
bp->next = p->next;
if (p + p->size == bp) {
p->size += bp->size;
p->next = bp->next;

25

A quick note to clarify: the size field in the header

struct is measured in header-sized blocks, and not bytes.
static header_t base;
/* Zero sized block to get us started. */
static header_t *usedp, *freep;
/*
* Find a chunk from the free list and put it in
* the used list.
*/
void *
GC_malloc(size_t alloc_size)
{
size_t num_units;
header_t *p, *prevp;
num_units = (alloc_size + sizeof(header_t) 1) / sizeof(header_t) + 1;
prevp = freep;
for (p = prevp->next;; prevp = p, p =
p->next) {
if (p->size >= num_units) {
/* Big enough. */
if (p->size == num_units)
/* Exact size. */
prevp->next = p->next;
else {
p->size -= num_units;
p += p->size;
p->size = num_units;
}
freep = prevp;

/* Add to p to the used list. */
if (usedp == NULL)
usedp = p->next = p;
else {
p->next = usedp->next;
usedp->next = p;
}
return (void *) (p + 1);
}
if (p == freep) {
/* Not enough memory. */
p = morecore(num_units);

26 PROGRAMMING

if (p == NULL)
/* Request for more memory failed. */
return NULL;
}
}
}

Note that the success of this function depends on

when we first use that freep = &base. Well make sure
this is the case in our init function.
Although this code isnt going to win any awards
for low fragmentation, itll work. And if it works, that
means we can finally get to the fun part: the garbage
collection!

Mark and Sweep

We did say that the garbage collector was going to
be simple, so we will be using the simplest algorithm
possible: stop the world naive mark and sweep. This
algorithm works in two parts:
First, we scan all the blocks of memory that could
possibly point to heap data and see if any do. To do this,
for each word-size chunk in the memory were looking
at, we look at each block in the used list. If the wordsized chunks value is within the range of a used block,
we mark the block.
Next, after all possible memory locations have been
searched, we go through the used list and add to the
free list all blocks that havent been marked.
Many people (or at least I did) get tripped up into
thinking that garbage collection is impossible in C,
because by writing a simple function like malloc, there
is no way of knowing many things about the outside
world. For example, there is no function in C that
returns a hash map to all the variables that have been
stack-allocated. But we can get by without this by realizing two important facts:
Firstly (gosh I say that a lot), in C, you can attempt
to access any memory location you want. There is no
chunk of memory that for some reason the compiler
can access but has an address that cannot be expressed
as an integer and then cast to a pointer. It isnt possible.
If memory is used in a C program, it can be accessed
by the program. This is a confusing notion for programmers unfamiliar to C, as many languages provide
restricted access to virtual memory addresses. C does
not.

Secondly, all variables are stored somewhere in

memory. Well, duh. But what that means is that if we
know generally where the variables are stored, we can
look through that memory and find all the possible
values of every variable. Additionally, because memory
access is generally only word-aligned, we only need to
look through every word in the memory regions.
Local variables can also be stored in registers, but
we wont worry about this because registers are usually dedicated to local variables, and by the time our
function is called theyll probably be saved on the stack
anyway.
Now we have a strategy for the marking phase of our
collector: look through a bunch of memory regions and
see if there is any memory that looks like it references
something in the used list. Writing a function to do
that is pretty clear cut:
#define UNTAG(p) (((unsigned int) (p)) &
0xfffffffc)
/*
* Scan a region of memory and mark any items in
* the used list appropriately.
* Both arguments should be word aligned.
*/
static void
mark_from_region(unsigned int *sp, unsigned int
*end)
{
header_t *bp;
for (; sp < end;
unsigned int
bp = usedp;
do {
if (bp +
bp +

sp++) {
v = *sp;

1 <= v &&
1 + bp->size > v) {
bp->next = ((unsigned int)

bp->next) | 1;
break;
}
} while ((bp = UNTAG(bp->next)) !=
usedp);
}
}

To ensure we only use two words in the header we

use a technique here called tagged pointers. Since our
next pointers will be word aligned, a few of the least
significant bits will always be zero. Thus, we mark the
least significant bit of the next pointer to indicate that
the current block (not the one pointed to by next!) has
been marked.
Now we can scan memory regions, but which
memory regions should we look through? There are
several:
1. The BSS (uninitialized data segment) and the initialized data segment. These contain all the global
and static variables in the program. Thus, they could
reference something in our heap.
2. The used chunks. Of course, if the user allocates a
pointer to another allocated chunk, we dont want
to free the pointed-to chunk.
3. The stack. Since the stack contains all the local
variables, this is arguably the most important place
to look.
But now theres the problem of knowing where
these memory regions start and end. We already know
where the used chunks are, but the other regions are
trickier.
We already know everything about the heap, so writing a mark_from_heap function is trivial:
/*
* Scan the marked blocks for references to
* other unmarked blocks.
*/
static void
mark_from_heap(void)
{
unsigned int *vp;
header_t *bp, *up;
for (bp = UNTAG(usedp->next); bp != usedp;
bp = UNTAG(bp->next)) {
if (!((unsigned int)bp->next & 1))
continue;
for (vp = (unsigned int *)(bp + 1);
vp < (bp + bp->size + 1);
vp++) {
unsigned int v = *vp;
up = UNTAG(bp->next);
do {

27

if (up != bp &&
up + 1 <= v &&
up + 1 + up->size > v) {
up->next = ((unsigned int)
up->next) | 1;
break;
}
} while ((up = UNTAG(up->next)) !=
bp);
}
}
}

Fortunately for the BSS and initialized data segments, most modern Unix linkers export the etext and
end symbols. The address of the etext symbol is the
start of the initialized data segment (the last address
past the text segment, which contains the programs
machine code), and the address of the end symbol is
the start of the heap. Thus, the BSS and initialized data
segment are located in between &etext and &end. This
is simple enough, but not platform independent.
The stack is a little harder. The top of the stack is
super simple to find using a little bit of inline assembly, as it is stored in the %esp register. However, well
be using the %ebp register as it ignores a few local
variables.
Finding the very bottom of the stack (where the
stack began) involves some trickery. Kernels tend to
randomize the starting point of the stack for security
reasons, so we cant hard code an address. To be honest,
Im not an expert on finding the bottom of the stack,
but I have a few rather poor ideas on how you can
make an accurate attempt. One possible way is you
could scan the call stack for the env pointer, which
would be passed as an argument to main. Another way
would be to start at the top of the stack and read every
subsequent address greater and handling the inexorable SIGSEGV. But were not going to do it either
way. Instead, were going to exploit the fact that linux
puts the bottom of the stack in a string in a file in the
processs entry in the proc directory (phew!). This
sounds silly and terribly indirect. Fortunately, I dont
feel ridiculous for doing it because its literally the
exact same thing Boehm GC does to find the bottom of the
stack!
Now we can make ourselves a little init function.
In it, we open the proc file on ourselves and find the
bottom of the stack. This is the 28th value printed so

28 PROGRAMMING

we discard the first 27. Boehm GC differs from us in

that they only use sys calls to do the file reading in
order to avoid the stdlib from using the heap, but we
dont really care.
/*
* Find the absolute bottom of the stack and set
* stuff up.
*/
void
GC_init(void)
{
static int initted;
FILE *statfp;
if (initted)
return;
initted = 1;
statfp = fopen("/proc/self/stat", "r");
assert(statfp != NULL);
fscanf(statfp,
"%*d %*s %*c %*d %*d %*d %*d %*d %*u
"
"%*lu %*lu %*lu %*lu %*lu %*lu %*ld
%*ld "
"%*ld %*ld %*ld %*ld %*llu %*lu %*ld
"
"%*lu %*lu %*lu %lu", &stack_bottom);
fclose(statfp);
usedp = NULL;
base.next = freep = &base;
base.size = 0;
}

Now we know the location of every memory region

we would need to scan, and thus, we can finally write
our explicitly-called collection function:
/*
* Mark blocks of memory in use and free the
* ones not in use.
*/
void
GC_collect(void)
{
header_t *p, *prevp, *tp;
unsigned long stack_top;

extern char end, etext;

/* Provided by the linker. */
if (usedp == NULL)
return;

/* Scan the BSS & initialized data segments. */
mark_from_region(&etext, &end);
/* Scan the stack. */
asm volatile ("movl
%%ebp, %0" : "=r"
(stack_top));
mark_from_region(stack_top, stack_bottom);
/* Mark from the heap. */
mark_from_heap();
/* And now we collect! */
for (prevp = usedp, p = UNTAG(usedp->next);;
prevp = p, p = UNTAG(p->next)) {
next_chunk:
if (!((unsigned int)p->next & 1)) {
/* The chunk hasn't been marked.
Thus, it must be set free. */
tp = p;
p = UNTAG(p->next);
add_to_free_list(tp);
if (usedp == tp) {
usedp = NULL;
break;
}
prevp->next = (unsigned int)p |
((unsigned int) prevp->next & 1);
goto next_chunk;
}
p->next = ((unsigned int) p->next) & ~1;
if (p == usedp)
break;
}
}

And that, my friends, is all there is. One simple

garbage collector written in C and for C. This code
isnt complete in itself, it needs a few little tweaks here
in there to get it working, but most of the code mostly
holds up on its own.

Conclusion
From elementary school through half of high school,
I played drums. Every Wednesday at around 4:30pm I
had a drum lesson from a teacher who was quite good.
Whenever I was having trouble learning a new
groove or beat or whatever, he would always give me
the same diagnosis: I was trying to do everything at
once. I looked at the sheet of music, and I simply tried
to play with all my hands. But I couldnt. And the
reason why is because I didnt know how to play the
groove yet, and simply trying to play the groove wasnt
how I was going to learn.
So my teacher would enlighten me as to how I could
learn: dont try playing everything at once. Learn to
play the high-hat part with your right hand. Once
youve got that down, learn to play the snare with
your left. Do the same with the bass, the tom-toms,
and whatever other parts there are. When you have all
the individual parts down, slowly begin to add them
together. Add them together in pairs, then in threes,
and eventually youll be able to play the entire thing.
I never got good at drums, but I did take these lessons to heart in my programming. Its really hard to
just start to type out an entire program. The only algorithm you need to write code is divide and conquer.
Write the function to allocate memory. Then, write
the function to look through memory. Then, write the
function that cleans up memory. Finally, add them all
together.
As soon as you get past this barrier as a programmer,
nothing practical becomes hard. You may not understand an algorithm, but anyone can understand an algorithm with enough time, paper, and the right book. If a
project seems daunting, break it up into its individual
parts. You may not know how to write an interpreter,
but you sure as hell can write a parser. Find out what
else you need to add, and do it. n
Matthew Plant is a sophomore undergraduate computer science
student at the University of Illinois at Urbana-Champaign. He was
born in San Francisco in 1995.
Reprinted with permission of the original author.
First appeared in hn.my/gcinc (web.engr.illinois.edu)

29

How to Read an Executable

How is a Binary Executable Organized?
By JuliA Evans

that executables were totally impenetrable. Id compile a C program, and then that was
it! I had a Magical Binary Executable that I could
no longer read.
It is not so! Executable file formats are regular file
formats that you can understand. Ill explain some
simple tools to start! Well be working on Linux, with
ELF binaries. (Binaries are kind of the definition of
platform-specific, so this is all platform-specific.) Well
be using C, but you could just as easily look at output
from any compiled language.
Lets write a simple C program, hello.c:
used to think

#include <stdio.h>
int main() {
printf("Penguin!\n");
}

Then we compile it (gcc -o hello hello.c), and

we have a binary called hello. This originally seems
impenetrable (how do we even binary?!), but lets see
how we can investigate it! Were going to learn what
symbols, sections, and segments are. At a high level:

Symbols are like function names, and are used to

answer If I call printf and its defined somewhere
else, how do I find it?

Symbols are organized into sections code lives

in one section (.text), and data in another (.data,
.rodata)

Sections are organized into segments

30 PROGRAMMING

Throughout, well use a tool called readelf to look

at these.
So, lets dive into our binary!

Step 1: open it in a text editor!

This is most naive possible way to view a binary. If I
run cat hello, I get something like this:
ELF>@@H@8
@@@@@@88@@@@ ((`(`
PP`P`Ptd@,,QtdRtd((`(`/lib64/ld-linuxx86-64.so.2GNUGNUn8wj7*oLh
__gmon_start__libc.so.6puts__libc_start_
mainGLIBC_2.2.5ui
1```HkH5 H[]fff.H=p
UHtH]H`]UH@]
Hl$Ld$H- L%
Ll$Lt$L|$H\$H8L)AIHIsH
t1@LLDAHH9uH\Hl$Ld$Ll$
Lt$(L|$0H8UHSHH
Ht(`DHHHuH[]
HoHPenguin!;,H

Theres text here, though! This was not a total

failure. In particular it says Penguin! and ELF. ELF
is the name of the binary format. So thats something!
Then there are a bunch of unprintable symbols, which
isnt a huge surprise because this is a binary.

Step 2: use readelf to see the symbol table

Throughout were going to use a tool called readelf
to explore our binary. Lets start by running readelf
--symbols on it. (Another popular tool to do this is
nm.)
$ readelf --symbols hello
Num:
Value
Size
Vis
Ndx Name
48: 0000000000000000
0
DEFAULT UND puts@@GLIBC_2.2.5
59: 0000000000400410
0
DEFAULT
13 _start
61: 00000000004004f4
16
DEFAULT
13 main

Type

Bind

FUNC

GLOBAL

FUNC

GLOBAL

FUNC

GLOBAL

Here we see three symbols: main is the address of

my main() function. puts looks like a reference to the
printf function I called in it (which I guess the compiler changed to puts as an optimization?). _start is
pretty important.
When the program starts running, you might think
it starts at main. It doesnt! It actually goes to _start.
This does a bunch of Very Important Things that I
dont understand very well, including calling main. So I
wont explain them.

Symbols
When you write a program, you might write a function called hello. When you compile the program, the
binary for that function is labelled with a symbol called
hello. If I call a function (like printf) from a library,
we need a way to look up the code for that function!
The process of looking up functions from libraries is
called linking. It can happen either just after we compile the program (static linking) or when we run the
program (dynamic linking).
So symbols are what allow linking to work! Lets find
the symbol for printf! Itll be in libc, where all the C
standard library functions are.
If I run nm on my copy of libc, it tells me no
symbols. But the internet tells me I can use objdump -tT instead! This works! objdump -tT /lib/
x86_64-linux-gnu/libc-2.15.so gives me this output.
[hn.my/objdump]
If you look at it, youll see sprintf, strlen, fork,
exec, and everything you might expect libc to have.
From here we can start to imagine how dynamic linking works we see that hello calls puts, and then we
can look up the location of puts in libcs symbol table.

Step 3: use objdump to see the binary, and learn

about sections!
Opening our binary in a text editor was a bad way to
open it. objdump is a better way. Heres an excerpt:
$ objdump -s hello
Contents of section .text:
400410 31ed4989 d15e4889 e24883e4
1.I..^H..H...PTI
400420 c7c0a005 400048c7 c1100540
[email protected][email protected]..
400430 f4044000 e8c7ffff fff49090
[email protected]...
Contents of section .interp:
400238 2f6c6962 36342f6c 642d6c69
lib64/ld-linux400248 7838362d 36342e73 6f2e3200
x86-64.so.2.
Contents of section .rodata:
4005f8 01000200 50656e67 75696e21
....Penguin!.

f0505449
0048c7c7
4883ec08

6e75782d

00

You can see that it shows us all the bytes in the file
as hex on the left, and a translation into ASCII on the
right.
There are a whole bunch of sections here. This
shows you all the bytes in your binary! Some sections
we care about:

.text is the programs actual code (the assembly).

_start and main are both part of the .text section.

.rodata is where some read-only data is stored (in

this case, our string Penguin!)

.interp is the filename of the dynamic linker!

The major difference between sections and segments

is that sections are used at link time (by ld) and segments are used at execution time. objdump shows us
the contents of the sections, which is nice, but doesnt
give us as much metadata about the sections as Id like.
Lets try readelf instead:

31

$ readelf --sections hello

Section Headers:
[Nr] Name
Type
Address
Offset
Size
EntSize
Flags
Link Info Align
[13] .text
PROGBITS
0000000000400410 00000410
00000000000001d8 0000000000000000 AX
0
0
16
[15] .rodata
PROGBITS
00000000004005f8 000005f8
000000000000000b 0000000000000000
A
0
0
4
[24] .data
PROGBITS
0000000000601010 00001010
0000000000000010 0000000000000000 WA
0
0
8
[25] .bss
NOBITS
0000000000601020 00001020
0000000000000010 0000000000000000 WA
0
0
8
[26] .comment
PROGBITS
0000000000000000 00001020
000000000000002a 0000000000000001 MS
0
0
1
Key to Flags:
W (write), A (alloc), X (execute), M (merge),
S (strings), l (large)
I (info), L (link order), G (group), T (TLS),
E (exclude), x (unknown)
O (extra OS processing required) o (OS specific), p (processor specific)

Neat! We can see .text is executable and read-only,

.rodata (read only data) is read-only, and .data is
read-write.

32 PROGRAMMING

Step 4: Look at some assembly!

We mentioned briefly that .text contains assembly
code. We can actually look at what it is really easily. If
we were magicians, we would already be able to read
and understand this:
Contents of section .text:
400410 31ed4989 d15e4889 e24883e4 f0505449
1.I..^H..H...PTI
400420 c7c0a005 400048c7 c1100540 0048c7c7
[email protected][email protected]..
400430 f4044000 e8c7ffff fff49090 4883ec08
[email protected]...

It starts with 31ed4989. Those are bytes that our

CPU interprets as code! And runs! However we are not
magicians (I dont know what 31 ed means!), and so
we will use a disassembler instead.
$ objdump -d ./hello
Disassembly of section .text:
0000000000400410 <_start>:
400410:
31 ed
%ebp,%ebp
400412:
49 89 d1
%rdx,%r9
400415:
5e
%rsi
400416:
48 89 e2
%rsp,%rdx
400419:
48 83 e4 f0
$0xfffffffffffffff0,%rsp

xor
mov
pop
mov
and

So we see that 31 ed is xoring two things. Neat!

Thats all the assembly well do for now.

Step 5: Segments!

Not magic!

Finally, a program is organized into segments or

program headers. Lets look at the segments for our
program using readelf --segments hello.

Executables arent magic. ELF is a file format like any

other! You can use readelf, nm, and objdump to inspect
your Linux binaries. Try it out! Have fun. n

Program Headers:
[... removed ...]
INTERP
0x0000000000000238
0x0000000000400238 0x0000000000400238
0x000000000000001c
0x000000000000001c R
1
[Requesting program interpreter: /lib64/
ld-linux-x86-64.so.2]
LOAD
0x0000000000000000
0x0000000000400000 0x0000000000400000
0x00000000000006d4
0x00000000000006d4 R E
200000
LOAD
0x0000000000000e28
0x0000000000600e28 0x0000000000600e28
0x00000000000001f8
0x0000000000000208 RW
200000
[... removed ...]

Julia Evans likes programming, playing with data, and finding out
why things that seem scary actually arent. She lives in Montreal
and works on Stripes machine learning team.
Reprinted with permission of the original author.
First appeared in hn.my/exe (jvns.ca)

Section to Segment mapping:

Segment Sections...
00
01
.interp
02 .interp .note.ABI-tag .note.gnu.build-id
.gnu.hash .dynsym
.dynstr .gnu.version .gnu.version_r
.rela.dyn .rela.plt .init .plt
.text .fini .rodata .eh_frame_hdr .eh_
frame
03
.ctors .dtors .jcr .dynamic .got .got.
plt .data .bss
04
.dynamic
05
.note.ABI-tag .note.gnu.build-id
06
.eh_frame_hdr
07
08
.ctors .dtors .jcr .dynamic .got

Segments are used to determine how to separate

different parts of the program into memory. The first
LOAD segment is marked R E (read / execute) and the
second is RW (read/write). .text is in the first segment
(we want to read it but never write to it), and .data,
.bss are in the second (we need to write to them, but
not execute them).

33

SPECIAL

Dont Become a Scientist!

By Jonathan Katz

re you thinking of
becoming a scientist?
Do you want to uncover
the mysteries of nature, perform
experiments, or carry out calculations to learn how the world works?
Forget it!
Science is fun and exciting. The
thrill of discovery is unique. If you
are smart, ambitious, and hardworking you should major in science as
an undergraduate. But that is as far
as you should take it. After graduation, you will have to deal with
the real world. That means that
you should not even consider going
to graduate school in science. Do
something else instead: medical
school, law school, computer science, engineering, or something else
which appeals to you.
Why am I (a tenured professor
of physics) trying to discourage you
from following a career path which
was successful for me? Because
times have changed (I received my
Ph.D. in 1973, and tenure in 1976).
American science no longer offers
a reasonable career path. If you
go to graduate school in science, it
is in the expectation of spending
your working life doing scientific
research, using your ingenuity
and curiosity to solve important

34 SPECIAL

and interesting problems. You will

almost certainly be disappointed,
probably when it is too late to
choose another career.
American universities train
roughly twice as many Ph.D.s
as there are jobs for them. When
something, or someone, is a glut
on the market, the price drops.
In the case of Ph.D. scientists, the
reduction in price takes the form
of many years spent in holding pattern postdoctoral jobs. Permanent
jobs dont pay much less than they
used to, but instead of obtaining a
real job two years after the Ph.D.
(as was typical 25 years ago) most
young scientists spend five, ten, or
more years as postdocs. They have
no prospect of permanent employment and often must obtain a new
postdoctoral position and move
every two years.
For example, consider two of
the leading candidates for a recent
Assistant Professorship in my
department. One was 37, ten years
out of graduate school (he didnt
get the job). The leading candidate,
whom everyone thinks is brilliant,
was 35, seven years out of graduate
school. Only then was he offered
his first permanent job (thats not
tenure, just the possibility of it six

years later, and a step off the treadmill of looking for a new job every
two years). The latest example is a
39 year old candidate for another
Assistant Professorship; he has
published 35 papers. In contrast, a
doctor typically enters private practice at 29, a lawyer at 25 and makes
partner at 31, and a computer
scientist with a Ph.D. has a very
good job at 27 (computer science
and engineering are the few fields
in which industrial demand makes
it sensible to get a Ph.D.). Anyone
with the intelligence, ambition, and
willingness to work hard to succeed
in science can also succeed in any of
these other professions.
Typical postdoctoral salaries
begin at $27,000 annually in the
biological sciences and about
$35,000 in the physical sciences
(graduate student stipends are less
than half these figures). Can you
support a family on that income?
It suffices for a young couple in a
small apartment, though I know of
one physicist whose wife left him
because she was tired of repeatedly
moving with little prospect of settling down. When you are in your
thirties you will need more: a house
in a good school district and all the
other necessities of ordinary middle

class life. Science is a profession,

not a religious vocation, and does
not justify an oath of poverty or
celibacy.
Of course, you dont go into
science to get rich. So you choose
not to go to medical or law school,
even though a doctor or lawyer
typically earns two to three times
as much as a scientist (one lucky
enough to have a good senior-level
job). I made that choice, too. I
became a scientist in order to have
the freedom to work on problems
which interest me. But you probably wont get that freedom. As a
postdoc you will work on someone
elses ideas, and may be treated as a
technician rather than as an independent collaborator. Eventually,
you will probably be squeezed out
of science entirely. You can get a
fine job as a computer programmer,
but why not do this at 22, rather
than putting up with a decade of
misery in the scientific job market
first? The longer you spend in science, the harder you will find it to
leave, and the less attractive you
will be to prospective employers in
other fields.
Perhaps you are so talented that
you can beat the postdoc trap;
some university (there are hardly
any industrial jobs in the physical
sciences) will be so impressed with
you that you will be hired into a
tenure track position two years out
of graduate school. Maybe. But the
general cheapening of scientific
labor means that even the most
talented stay on the postdoctoral
treadmill for a very long time; consider the job candidates described
above. And many who appear to be
very talented, with grades and recommendations to match, later find
that the competition of research is
more difficult, or at least different,

and that they must struggle with

the rest.
Suppose you do eventually
obtain a permanent job, perhaps a
tenured professorship. The struggle
for a job is now replaced by a
struggle for grant support, and again
there is a glut of scientists. Now
you spend your time writing proposals rather than doing research.
Worse, because your proposals are
judged by your competitors you
cannot follow your curiosity, but
must spend your effort and talents on anticipating and deflecting
criticism rather than on solving
the important scientific problems.
Theyre not the same thing: you
cannot put your past successes
in a proposal, because they are
finished work, and your new ideas,
however original and clever, are
still unproven. It is proverbial that
original ideas are the kiss of death
for a proposal; because they have
not yet been proved to work (after
all, that is what you are proposing to do) they can be, and will be,
rated poorly. Having achieved the
promised land, you find that it is
not what you wanted after all.
What can be done? The first
thing for any young person (which
means anyone who does not have
a permanent job in science) to do
is to pursue another career. This
will spare you the misery of disappointed expectations. Young Americans have generally woken up to the
bad prospects and absence of a reasonable middle class career path in
science and are deserting it. If you
havent yet, then join them. Leave
graduate school to people from
India and China, for whom the
prospects at home are even worse.
I have known more people whose
lives have been ruined by getting a
Ph.D. in physics than by drugs.

If you are in a position of leadership in science then you should try

to persuade the funding agencies
to train fewer Ph.D.s. The glut
of scientists is entirely the consequence of funding policies (almost
all graduate education is paid for by
federal grants). The funding agencies are bemoaning the scarcity of
young people interested in science
when they themselves caused this
scarcity by destroying science as
a career. They could reverse this
situation by matching the number
trained to the demand, but they
refuse to do so, or even to discuss
the problem seriously (for many
years the NSF propagated a dishonest prediction of a coming shortage of scientists, and most funding
agencies still act as if this were
true). The result is that the best
young people, who should go into
science, sensibly refuse to do so, and
the graduate schools are filled with
weak American students and with
foreigners lured by the American
student visa. n
Jonathan Katz is a professor of physics
at Washington University in St. Louis. He
has worked in many branches of physics,
including astrophysics, plasma physics and
applied physics, both in academia and at
National Laboratories. His present work is
in high energy astrophysics and rheology.
Reprinted with permission of the original author.
First appeared in hn.my/scientist (physics.wustl.edu)

35

The 1,000-Hour Rule

By Philip Guo

m not yet qualified to give

general life advice to kids, but I
would like to share one simple
piece of advice that I wouldve
liked to hear when I was a kid:
Find something you genuinely enjoy
doing for its own sake, stick with it,
keep learning more about it, and after
a decade or so, you cant help but get
good at it and feel proud of yourself.
This is my own personal take
on the popular 10,000-hour rule,
which claims that it takes around
10,000 hours of intense practice to
become an expert in a particular
topic. For instance, top-notch musicians, artists, athletes, scientists, and
other experts in their respective
fields all share a common experience: They practiced consistently
and with high intensity for over
10,000 hours, often starting at a
young age. That amounts to practicing for 4 hours every weekday for
a decade straight, which takes tremendous passion and perseverance.
I dont want to pressure you to
become an expert at anything, but
I do want you to become good at
something, which I think requires
far less time and dedication. So here

36 SPECIAL

is my 1,000-hour rule, a much gentler version of the 10,000-hour rule:

I claim that it takes roughly 1,000
hours of practice to get good at
something. 1,000 hours of practice
over a decade amounts to roughly
2 hours each week, which is far
more sustainable than 4 hours each
day. Surely you can spare 2 hours
each week for some hobby. So find
something you genuinely enjoy
doing for its own sake, stick with
it, keep learning more about it, and
after a decade or so, you cant help
but get good at it and feel proud of
yourself.

A Bit Of Narcissism
To give a personal example, the
website you are now viewing is the
primary hobby that I have maintained throughout the past decade.
I now consider myself fairly good at
creating personal websites and writing online articles.
I wasnt always proud of my
website design and writing. Ive
only begun receiving some praise
for my website in the past few
years, but Ive been making websites for over a decade, starting
back in 1997. My first few attempts
looked pretty horrendous (e.g., MS
Paint and stock clip art, 3-D naked
men). Some people have told me
that they think I have some sort
of talent for writing and website
design. I can guarantee that nobody
told me that when I first started at
age 13! I was a horrible writer back
in middle school: I got C and D
grades on some of my essays. And
I had absolutely no sense of visual
aesthetics.

from now, you will think

Athatdecade
its pretty cool that youve grown

fairly good at something that youve

consistently kept up over many years.

However, I learned to improve

my writing and visual design sense
throughout the past decade of
working on my website in my spare
time. I never remembered spending a tremendous amount of time
doing so, or forgoing my school or
work obligations. It was a hobby
that I could do whenever I felt in
the right mood. Thus, if you pursue
a hobby consistently over a decade,
even if you only put in a few hours
each week, I guarantee that you
cant help but get pretty good at it!
Its fairly easy to rack up your 1,000
hours when spread over an entire
decade.
I think Ive surpassed my 1,000
hours by now, but Im not nearly
at the 10,000 hours which are
indicative of true expertise. I dont
consider myself an expert writer
or website designer by any means.
I am merely a hobbyist, albeit one
who has pursued this hobby over
many years. However, even as a
hobbyist, Ive managed to hone
some deep intuitions about this
craft that simply cant be gleaned
from perusing a book or how-to
guide. There is simply no substitute for consistent practice over

time, and no easy way to get good

overnight. However, it is pretty easy
to get good if you just stick with
something you like doing over a
long period of time.

the potential and opportunity to

develop into world experts, but
simply becoming good at a skill
is well within almost everyones
reach. n

Take-Home Message

Philip Guo is an assistant professor of computer science at the University of Rochester. His main research interests are in
human-computer interaction (HCI), with a
focus on user interfaces for online learning.
He created a free Web-based tool for learning programming called Online Python
Tutor [pythontutor.com], which has been
used by over 500,000 people in over 165
countries. He has previously researched
online education at edX and Google.

But enough about me. You dont

care about how my hobby has been
personally fulfilling for me; you
only care about what you can gain
out of reading this article.
My advice is simple: Find something that you enjoy doing and then
keep doing it. Maintain a constant
curiosity to learn more about it,
and do it purely as a hobby without
any ulterior motives. A decade from
now, you will think that its pretty
cool that youve grown fairly good
at something that youve consistently kept up over many years.
And the best part is that it wont
ever be difficult or strenuous!
Becoming a world expert in
something takes 10,000 hours of
shedding blood, sweat, and tears, at
times even driving you to the brink
of insanity, but simply becoming
good at a hobby takes much less
time and emotional investment.
Very few people can have both

Reprinted with permission of the original author.

First appeared in hn.my/1000hr (pgbovine.net)

37

Its Just Wood

By Bryan Kennedy

few months back,

I decided to
start a vegetable garden. I have
fond memories of gardening with
my folks when I was a young boy, and I
figured a garden would be a good outlet from
my day to day life in tech.
Looking up and down the aisles at Lowes
and not finding a raised garden bed that met
my needs, I inquired with a staff member:
Sorry, we dont carry anything like that.
Darn. Ok, Ill try another store then, I
sighed in resignation.
You know, he said to me, its just wood.
It took me a moment to realize how
liberating this statement was. To him, woodworking held no mystery at all it was just
a thing you did. If he wanted a garden bed,
thered be no question in his mind that hed
make it himself. Why spend more on something suboptimal when you can build exactly
what you want?
So I did exactly that, and we now have a
wonderful, custom-built gardening bed growing some very tasty kale.

Most people let mystery stop them

We resign to not knowing how to change the
oil in our car, replace a busted light switch,
or develop that app idea into a real app. You
need to be an expert to do that, we say.
Thinking back to the times Ive challenged
mystery and learned to do something myself,
I recall it giving me great satisfaction. And
once demystified, those tasks that seemed
impossible before now seem trivial.
Did you know that there are over 110K
videos on YouTube about replacing a light
switch? Or that Hack Reactor claims 99%
job placement rate after graduating their
12-week developer program? The resources
you need to overcome the mysteries around
you are plentiful.
Challenge mystery
When you take it upon yourself to learn how
to do something, youve not only become
more self-reliant, but youve expanded your
mind to new possibilities. If I had settled for
a pre-built planter box, I wouldnt have been
able to construct exactly the kind of box I
wanted, nor improve the design along the
way to fit my needs. Now the next time I
want to construct something out of wood, I
have the confidence to know that I can figure
it out.
The best part? The more you challenge the
mysteries in your life, the better youll get at
it.
What will you demystify? n
Bryan Kennedy is the Co-Founder and CTO of Sincerely,
helping to scale thoughtfulness across the world.
Bryan is a YCombinator alum and an angel investor. On warm summer nights he runs MobMov.org, a
worldwide collective of guerrilla drive-ins.
Reprinted with permission of the original author.
First appeared in hn.my/wood (plusbryan.com)

38 SPECIAL

Metrics and monitoring for people

who know what they want
We know from experience that monitoring your servers and
applications can be painful, so we built the sort of service that
we would want to use. Simple to set up, responsive support
from people who know what they're talking about, and reliably
fast metric collection and dashboards.

Dashboards

StatsD

Happiness

Now with Grafana!

Why Hosted Graphite?

Hosted metrics and StatsD: Metric aggregation without the setup headaches
High-resolution data: See everything like some glorious mantis shrimp / eagle hybrid*
Flexible: Lots of sample code, available on Heroku
Transparent pricing: Pay for metrics, not data or servers
World-class support: We want you to be happy!

Promo code: HACKER

Grab a free trial at http://www.hostedgraphite.com

*Hosted Graphites mantis shrimp / eagle breeding program has been unsuccessful thus far

39