Understanding Computation (53 page)

Read Understanding Computation Online

Authors: Tom Stuart

Tags: #COMPUTERS / Programming / General

BOOK: Understanding Computation
2.19Mb size Format: txt, pdf, ePub
Coping with Uncomputability

The whole point of writing a program is to get a computer to do
something useful. As programmers, how are we supposed to cope with a world
where checking that a program works properly is an unsolvable
problem?

Denial is a tempting response:
Ignore the whole issue. It
would be nice if we could automatically verify program behavior, but we
can’t, so let’s just hope for the best and never try to check that a
program does its job correctly.

But that would be an overreaction, because the situation isn’t as
bad as it sounds. Rice’s theorem doesn’t mean that program analysis is
impossible, just that we can’t write a nontrivial analyzer that will
always
halt and produce the right answer. As we saw
in
Building a Halting Checker
, there’s nothing to stop
us from writing a tool that gives the right answer for
some
programs, as long as we can tolerate the fact
that there will always be
other
programs for which it
either gives the wrong answer or loops forever without returning
anything.

Here are some practical ways of analyzing and
predicting program behavior, in spite of
undecidability:

  • Ask undecidable questions, but give up if an answer can’t be
    found. For example, to check whether a program prints a particular
    string, we can run it and wait; if it doesn’t print that string within
    a particular period of time, say 10 seconds, we just terminate the
    program and assume it’s no good. We might accidentally throw out a
    program that produces the expected output after 11 seconds, but in
    many cases that’s an acceptable risk, especially since slow programs
    can be undesirable in their own right.

  • Ask several small questions whose answers, when taken together, provide empirical
    evidence for the answer to a larger question. When performing automated acceptance
    testing, we’re usually not able to check that a program does the right thing for every
    possible input, but we can try running it for a limited number of
    example
    inputs to see what happens. Each test run gives us
    information about how the program behaves in that specific case, and we can use that
    information to become more confident of how it’s likely to behave in general. This leaves
    open the possibility that there are other untested inputs that cause wildly different
    behavior, but we can live with that as long as our test cases do a good job of
    representing most kinds of realistic input.

    Another example of this approach is the use of unit tests to verify the behavior of
    small pieces of a program individually rather than trying to verify the program as a
    whole. A well-isolated unit test concentrates on checking the properties of a simple unit
    of code, and makes assumptions about other parts of the program by representing them with
    test doubles (i.e., stubs and mocks). Individual unit tests that exercise small pieces of
    well-understood code can be simple and fast, minimizing the danger that any one test will
    run forever or give a misleading answer.

    By unit testing all the program’s pieces in this way, we can set
    up a chain of assumptions and implications that resembles a
    mathematical proof: “if piece A works, then piece B works, and if
    piece B works, then piece C works.” Deciding whether all of these
    assumptions are justified is the responsibility of human reasoning
    rather than automated verification, although integration and
    acceptance testing can improve our confidence that the entire system
    does what it’s supposed to do.

  • Ask decidable questions by being conservative where necessary. The above suggestions
    involve actually running parts of a program to see what happens, which always introduces
    the risk of hitting an infinite loop, but there are useful questions that can be answered
    just by inspecting a program’s source code statically. The most obvious example is “does
    this program contain any syntax errors?,” but we can answer more interesting questions if
    we’re prepared to accept safe approximations in cases where the real answer is
    undecidable.

    A common analysis is to look through a program’s source to see if it contains
    dead code
    that computes values that are never used, or
    unreachable code
    that
    never gets evaluated. We can’t always tell whether code is truly dead or
    unreachable, in which case we have to be conservative and assume it isn’t, but there are
    cases where it’s obvious: in some languages, we know that an assignment to a local
    variable that’s never mentioned again is definitely dead, and that a statement that
    immediately follows a
    return
    is definitely unreachable.
    [
    84
    ]
    An optimizing compiler like GCC
    uses these techniques to identify and eliminate unnecessary code, making
    programs smaller and faster without affecting their behavior.

  • Approximate a program by converting it into something simpler,
    then ask decidable questions about the approximation. This important
    idea is the subject of the
    next
    chapter
    .

[
64
]
The smallest value
x
and
y
can reach is
1
, so they’ll meet there
if all else fails.

[
65
]
Ruby already has a built-in version of Euclid’s algorithm,
Integer#gcd
, but that’s beside
the point.

[
66
]
Lisp is really a family of programming languages—including
Common Lisp, Scheme, and Clojure—that share very similar
syntax.

[
67
]
It would be more useful to only assign numbers to the
syntactically valid
Ruby programs, but doing
that is more complicated.

[
68
]
Most of those numbers won’t represent syntactically valid Ruby programs, but we
can feed each potential program to the Ruby parser and reject it if it has any syntax
errors.

[
69
]
We’re using Unix shell syntax here. On Windows, it’s necessary to omit the single
quotes around the argument to
echo
, or to put the
text in a file and feed it to
ruby
with the
<
input redirection operator.

[
70
]
This is the shell command to run
does_it_say_no.rb
with its own source
code as input.

[
71
]
We can only get away with this because parts B and C happen
not to contain any difficult characters like backslashes or
unbalanced curly brackets. If they did, we’d have to escape them
somehow and then undo that escaping as part of assembling the
value of
program
.

[
72
]
Douglas Hofstadter coined the name
quine
for a program that
prints itself.

[
73
]
Can you resist the urge to write a Ruby program that can
perform this transformation on any Ruby program? If you use
%q{}
to quote
data
’s value, how will you handle
backslashes and unbalanced curly brackets in the original
source?

[
74
]
Briefly, for the curious: every pushdown automaton has an equivalent context-free
grammar and vice versa; any CFG can be rewritten in
Chomsky normal
form
; and any CFG in that form must take exactly 2
n
− 1
steps to generate a string of length
n
. So we can turn the original PDA
into a CFG, rewrite the CFG into Chomsky normal form, and then turn that CFG back into a
PDA. The resulting pushdown automaton recognizes the same language as the original, but
now we know exactly how many steps it’ll take to do it.

[
75
]
The choice of the empty string is unimportant; it’s just an arbitrary fixed input.
The plan is to run
does_it_halt.rb
on
self-contained programs that don’t read anything from standard input, so it doesn’t
matter what
input
is.

[
76
]
Faber’s million-dollar prize expired in 2002, but anyone who
produced a proof today would still be in for some serious fortune
and glory on the rock star mathematician circuit.

[
77
]
This is reminiscent of
#evaluate_on_itself
from
Universal Systems Can Loop Forever
, with
#halts?
in place of
#evaluate
.

[
78
]
Or equivalently: what does
#halts_on_itself?
return if we call it
with
do_the_opposite.rb
’s
source code as its argument?

[
79
]
Surely “responsible software engineering professionals”?

[
80
]
Again, that input might be irrelevant if the program doesn’t actually read anything
from
$stdin
, but we’re including it for the sake of
completeness and consistency.

[
81
]
Total programming languages are a potential solution to this
problem, but so far they haven’t taken off, perhaps because they can
be more difficult to understand than conventional languages.

[
82
]
Stephen Wolfram coined the name
computational
irreducibility
for the idea that a program’s
behavior can’t be predicted without running it.

[
83
]
This is roughly the content of
Gödel’s
first incompleteness theorem
.

[
84
]
The Java language specification requires the compiler to reject any program that
contains unreachable code. See
http://docs.oracle.com/javase/specs/jls/se7/html/jls-14.html#jls-14.21
for a
lengthy explanation of how a Java compiler is meant to decide which parts of a program
are potentially reachable without running any of it.

Chapter 9. Programming in Toyland

Programming is about using syntax to communicate ideas to a machine. When we write a
program, we have an idea of what we want the machine to do when it executes that program, and
knowing the semantics of our programming language gives us some confidence that the machine is
going to understand what each small piece of the program means.

But a complex computer program is greater than the sum of its
individual statements and expressions. Once we’ve plugged together many
small parts to make a larger whole, it would be useful to be able to check
whether the overall program actually does what we originally wanted it to
do. We might want to know that it always returns certain results, for
example, or that running it will have certain side effects on the filesystem
or network, or just that it doesn’t contain obvious bugs that will make it
crash on unexpected inputs.

In fact, there are all sorts of properties that we might want our programs to have, and it
would be really convenient if we could just check the syntax of a particular program to see
whether or not it has those properties, but we know from Rice’s theorem
that predicting a program’s behavior by looking at its source code can’t always
give us the right answer. Of course, the most direct way to find out what a program will do is
just to run it, and sometimes that’s okay—a lot of software testing is done by running programs
on known inputs and checking the actual outputs against the expected ones—but there are a few
reasons why running code might not be an acceptable way of investigating it either.

For one thing, any useful program is likely to deal with information
that won’t be known until run time: interactive input from the user, files
passed in as arguments, data read from the network, that sort of thing. We
can certainly try running a program with dummy inputs to get some sense of
what it does, but that just tells us about the program’s behavior for those
inputs; what happens when the real inputs are different? Running a program
on all possible combinations of inputs is often impractical or impossible,
and trying the program once with a specific set of inputs, however realistic
they are, doesn’t necessarily tell us very much about how it will behave in
general.

Another problem, which we
explored in
Universal Systems Can Loop Forever
, is that programs written
in sufficiently powerful
[
85
]
languages are perfectly capable of running forever without
ever producing a result. This makes it impossible to reliably investigate
arbitrary programs by running them, because it’s sometimes impossible to
tell in advance whether a program is going to run indefinitely (see
The Halting Problem
), so any automatic checker that tried to run a
candidate program would be at risk of never coming back with an
answer.

And lastly, even in the case of a program that does have all its input data available in
advance and that will, for whatever reason, always eventually terminate instead of looping
forever, it might just be expensive or inconvenient to run that program to see what happens. It
could take a long time to finish, or have irreversible side effects—sending emails, transferring
money, launching missiles—which are undesirable for testing purposes.

All these reasons make it useful to be able to discover information
about a program without actually running it. One way of doing this is to use
abstract interpretation
, an analysis technique in
which we execute a simplified version of the program and use the results to
deduce properties of the original.

Abstract Interpretation

Abstract
interpretation gives us a way of approaching a problem that’s somehow too
difficult to handle, perhaps because it’s too large, too complex, or has too many unknowns to
be tackled directly. The main idea of abstract interpretation is to use an
abstraction
, a model of the real problem that discards enough detail
to make it manageable—perhaps by making it smaller, simpler, or by eliminating unknowns—but
that also retains enough detail to make its solution relevant to the original problem.

To make this vague idea more concrete, let’s look at a simple
application of abstract interpretation.

Route Planning

Suppose
you’re a tourist in an unfamiliar country and want to plan
a road trip to another town. How will you decide which route to take? A
direct solution is to jump into your rental car and drive in whichever
direction seems the most promising. Depending on how lucky you are, and
on how much help the foreign road signs give you, this brute-force
exploration of an unknown road network might eventually get you to your
destination. But it’s an expensive strategy, and it’s likely that you’ll
just get more and more lost until you give up completely.

Using a map to plan your trip is a much more sensible idea. A printed road atlas is an
abstraction that sacrifices a lot of the detail of the physical road network. It doesn’t
tell you what the traffic is like, which roads are currently closed, where individual
buildings are, or anything at all about the third dimension; it is, crucially, much smaller
and flatter than the real thing. But a map does retain the most important information
required for journey planning: the relative positions of all the towns, which roads lead to
each town, and how those roads are connected to each other.

Despite all of this missing detail, an accurate map is useful, because the route you
plan with it is likely to be valid in reality, not just in the abstract world of the map. A
cartographer has done the expensive work of creating a model of reality, giving you the
opportunity to perform a computation on that model by looking at a simplified representation
of the road network and planning your route on it. You can then transfer the result of that
computation back into the real world when you actually get in your car and drive to your
destination, allowing cheap decisions made in the abstract world of the map to prevent you
from having to make expensive decisions on the physical roads.

The approximations used by a map make navigational computations much easier without
fatally compromising the fidelity of their results. There are plenty of circumstances in
which decisions made with a map might turn out to be wrong—there is no guarantee that a map
has told you absolutely
everything
you need to know about your
journey—but planning your route in advance lets you rule out particular kinds of mistakes
and makes the whole problem of getting from one place to another much more
manageable.

Abstraction: Multiplying Signs

Planning a route
with a printed map is a real application of abstract interpretation, but it’s
also very informal. For a more formal example, we can look at the multiplication of numbers;
although it’s still only a toy example, multiplication gives us a chance to start writing
code to investigate these ideas.

Pretend for a moment that multiplying two numbers is a difficult
or expensive operation, and that we’re interested in finding out some
information about the result of a multiplication without having to
actually perform it. Specifically: what is the
sign
of the result? Is it a negative number, zero, or a positive
number?

The notionally expensive way of finding out is to compute in the
concrete
world, using the
standard
interpretation
of multiplication: multiply the numbers for
real, look at the resulting number, and decide whether that result is
negative, zero, or positive. In Ruby, for example:

>>
6
*
-
9
=> -54

-54
is negative, so we’ve
learned that the product of
6
and
-9
is a negative number. Job
done.

However, it’s also possible to discover the same information by
computing in an
abstract
world, using an
abstract interpretation
of multiplication. Just
as a map uses lines on a flat page to represent roads in the real world,
we can use abstract values to represent numbers; we can plan a route on
a map instead of finding our way by trial and error on real roads, and
we can define an abstract multiplication operation on abstract values
instead of using concrete multiplication on concrete numbers.

To do this, we need to design abstract values that make the calculation simpler while
still retaining enough information to be a useful answer. We can take advantage of the fact
that the
absolute values
[
86
]
of two multiplied numbers don’t make any difference to the sign of the
result:

>>
(
6
*
-
9
)
<
0
=> true
>>
(
1000
*
-
5
)
<
0
=> true
>>
(
1
*
-
1
)
<
0
=> true

As young children, we’re taught that it’s only the signs of the arguments that matter:
the product of two positive numbers, or two negative numbers, is always a positive number;
the product of one positive and one negative number is always negative; and the product of
zero with any other number is always zero.

So let’s use these different kinds of number—“negative,” “zero,”
and “positive”—as our abstract values. We can do this in Ruby by
defining a
Sign
class and creating
three instances of it:

class
Sign
<
Struct
.
new
(
:name
)
NEGATIVE
,
ZERO
,
POSITIVE
=
[
:negative
,
:zero
,
:positive
].
map
{
|
name
|
new
(
name
)
}
def
inspect
"##{
name
}
>"
end
end

This gives us Ruby objects that we can use as our abstract values:
Sign::NEGATIVE
represents “any
negative number,”
Sign::ZERO
represents “the number zero,” and
Sign::POSITIVE
represents “any positive
number.” These three
Sign
objects
make up the tiny abstract world where we’ll perform abstract
calculations, while our concrete world consists of the
virtually unlimited supply of Ruby integers.
[
87
]

We can define abstract multiplication for
Sign
values by implementing just the
sign-related aspect of concrete multiplication:

class
Sign
def
*
(
other_sign
)
if
[
self
,
other_sign
].
include?
(
ZERO
)
ZERO
elsif
self
==
other_sign
POSITIVE
else
NEGATIVE
end
end
end

Instances of
Sign
can now be
“multiplied” together just like numbers, and our implementation of
Sign#*
produces answers that are
consistent with multiplication of actual numbers:

>>
Sign
::
POSITIVE
*
Sign
::
POSITIVE
=> #
>>
Sign
::
NEGATIVE
*
Sign
::
ZERO
=> #
>>
Sign
::
POSITIVE
*
Sign
::
NEGATIVE
=> #

For example, the last line above asks the question: what do we get
when we multiply any positive number by any negative number? The answer
comes back: a negative number. This is still a kind of multiplication,
but it’s much simpler than the kind we’re used to, and it only works on
“numbers” that have had almost all of their identifying information
removed. If we’re still imagining that real multiplication is expensive,
this seems like a cut-down version of multiplication that could be
considered cheap.

Armed with our abstract world of numbers and an abstract
interpretation of multiplication for those numbers, we can tackle the
original problem in a different way. Rather than multiplying two numbers
directly to find out the sign of their result, we can convert the
numbers into their abstract counterparts and multiply those instead.
First we need a way to convert concrete numbers into abstract
ones:

class
Numeric
def
sign
if
self
<
0
Sign
::
NEGATIVE
elsif
zero?
Sign
::
ZERO
else
Sign
::
POSITIVE
end
end
end

Now we can convert two numbers and do the multiplication in the
abstract world:

>>
6
.
sign
=> #
>>
-
9
.
sign
=> #
>>
6
.
sign
*
-
9
.
sign
=> #

Again we’ve calculated that
6 * -9
produces a
negative number, but this time we’ve done it without any multiplication of actual numbers.
Stepping up into the abstract world gives us an alternative way of performing the
computation, and crucially, the abstract result can be translated back down into the
concrete world so we can make sense of it, although we can only get an approximate concrete
answer because of the detail we sacrificed in making the abstraction. In this case, the
abstract result
Sign::NEGATIVE
tells us that any of the
concrete numbers
-1
,
-2
,
-3
, etc., might be the answer to
6 * -9
, but that the answer is definitely not
0
or any positive number like
1
or
500
.

Note that, because Ruby values are objects—data structures that
carry their operations with them—we can use the same Ruby expression to
perform either a concrete or an abstract computation depending on
whether we provide concrete (
Fixnum
)
or abstract (
Sign
) objects as
arguments. Take a
#calculate
method
that multiplies three numbers in a particular way:

def
calculate
(
x
,
y
,
z
)
(
x
*
y
)
*
(
x
*
z
)
end

If we call
#calculate
with
Fixnum
objects, the calculation will
be done by
Fixnum#*
and we’ll get a
concrete
Fixnum
result. If we call it
with
Sign
instances instead, the
Sign#*
operation will be used and
produce a
Sign
result.

>>
calculate
(
3
,
-
5
,
0
)
=> 0
>>
calculate
(
Sign
::
POSITIVE
,
Sign
::
NEGATIVE
,
Sign
::
ZERO
)
=> #

This gives us a limited opportunity to perform abstract
interpretation within real Ruby programs by replacing concrete arguments
with their abstract counterparts and running the rest of the code
without
modification.

Note

This technique is reminiscent of the way that
test doubles
(e.g.,
stubs
and
mocks
) are used in
automated unit testing. Test doubles are special placeholder objects that get injected
into code as a way to control and verify its behavior. They’re useful in any situation
where using more realistic objects as test data would be too inconvenient or
expensive.

Other books

Ghost in the Wind by E.J. Copperman
Floralia by Farris, J. L.
The Seduction of Sarah Marks by Kathleen Bittner Roth
Shadowplay by Laura Lam
Violent Spring by Gary Phillips
Nobody Girl by Leslie Dubois