The document discusses design by contract (DbC), which involves specifying function contracts that define pre-conditions, post-conditions, and error conditions. It provides an example of specifying contracts for a square root function in Racket. It then addresses common questions about DbC, such as what to do if contracts are specified incorrectly, performance impacts, and how DbC relates to testing and static type checking. The document advocates for hybrid approaches that combine contracts with testing and/or static typing.
HTML Injection Attacks: Impact and Mitigation Strategies
Introduction to Contracts and Functional Contracts
1. Daniel Prager @agilejitsu agilejitsu.blogspot.com
Melbourne Functional User Group, September 6, 2013
There are two ways of constructing a software design: One way is to make it
so simple that there are obviously no deficiencies, and the other way is to
make it so complicated that there are no obvious deficiencies.
The first method is far more difficult. It demands the same skill, devotion,
insight, and even inspiration as the discovery of the simple physical laws
which underlie the complex phenomena of nature.
– Tony Hoare, in his 1980 Turing Award lecture
Introduction to Contracts and
Functional Contracts
2. A personal Preamble
● 1998: I discover Design by Contract and start using it
– Time spent debugging: ↓ 90%
– Joy ↑
● I've used variants ever since
– Technical pay-off:
– Hit rate in persuading colleagues to join in: variable
● In this talk I
– Introduce the basics of Design by Contract
– Show how it can work in combination with functional programming
– Share some hard-won advice
– Ideally chat with you about where DbC sits today in relation to automated testing and static
type-checking
3. Overview
● The Problem of Software Quality
● Contracts: The basic idea
● Design By Contract (DbC)
● Questions and Concerns
● Functional Contracts
● Comparison with Tests and with Static Type-checking
● Some Hybrid Approaches
● Conclusion
4. The Problem of Software Quality
● Problem: How to design and program correct,
robust, maintainable, etc., etc. software?
● Assumption: BUGS happen!
● Prevention: Simplicity and clarity of design;
modularity; concision
● Treatment: Rapid isolation and diagnosis of
problems
5. Contracts: a metaphor
● Commercial Contract
– The client pays a supplier
to provide a good or service
– The contract
●
makes explicit the
expectations on both parties
● specifies who is to blame –
client or supplier – if
something goes wrong.
● Programming Contract
– The calling code (the client) calls a function
(the supplier)
– The contract
● defines explicit pre-conditions on
arguments passed by the calling code (and
on the initial state)
● defines explicit post-conditions on the
result of the function (and the final state)
● specifies which code is to blame – calling
code or function – if something goes wrong.
6. The standard example in Eiffel
square_root (x: REAL): REAL is
-- Returns the positive square root of x
require
-- Pre-condition section
positive_argument: x >= 0.0
do
-- Implementation section
...
ensure
-- Post-condition section
correct_root: abs(Result * Result – x) <= 0.000001
positive_root: Result >= 0
end
7. Same example in Racket
with hand-rolled contract
(define (square-root x)
(require (non-negative-real? x) "real, non-negative argument")
(let ([result ...])
(ensure (non-negative-real? result) "real, non-negative result")
(ensure (approx-equal? (* result result) x) "correct root")
result))
(define (require test [message “”])
(unless test (error 'pre-condition-violation message)))
(define (ensure test [message “”])
(unless test (error 'post-condition-violation message)))
(define (non-negative-real? x) (and (real? x) (>= x 0)))
(define (approx-equal? x y [tol 0.000001]) (<= (abs (- x y)) tol))
8. Quiz: Test your understanding
(define (square-root x)
(require (non-negative-real? x) "real, non-negative argument")
(let ([result (/ x 2)])
(ensure (non-negative-real? result) "real, non-negative result")
(ensure (approx-equal? (* result result) x) "correct root")
result))
● Which errors would the following calls induce?
– (square-root “five”)
– (square-root -8.0)
– (square-root 9)
10. Square root with hand-rolled
contract and static types
#lang typed/racket
(define: (square-root [x : Nonnegative-Real]) : Nonnegative-Real
(let ([result ...])
(ensure (approx-equal? (* result result) x) "correct root")
result))
(define: (require [test : Boolean] [message : String]) : Symbol
(if test 'ok (error 'pre-condition-violation message)))
(define: (ensure [test : Boolean] [message : String]) : Symbol
(if test 'ok (error 'post-condition-violation message)))
(define: (approx-equal? [x : Real] [y : Real]) : Boolean
(<= (abs (- x y)) 0.000001))
11. Design by Contract
● Design By Contract (DbC) is the discipline of writing out the contracts before
implementation.
– Steps of Design By Contract:
1)Declare a function and write-down its contract
2)Write an implementation
3)Manually test and fix any breakages
4)Refactor the contract and the code for concision and precision
5)Rinse, repeat.
– The mechanics are similar to Test-Driven Design/Development (TDD):
1)Write a failing test
2)Make it pass
3)Refactor to remove duplication, etc.
4)Rinse, repeat.
●
12. Let's DbC together!
● Scenario: Insert a string into another string
– Signature: (define (string-insert str other-str pos) …)
– Usage:
● (string-insert "stuff" "FOO" 0) → "FOOstuff"
● (string-insert "stuff" "FOO" 3) → "stuFOOff"
● (string-insert "stuff" "FOO" 5) → “stuffFOO”
● Challenge:
– Write down as many pre-conditions as you can
– Write down some post-conditions [usually harder]
13. Design by Contract: a solution
(define (string-insert str other-str pos)
(require (string? str) "str is a string")
(require (string? other-str) "other-str is a string")
(require (integer? pos) "pos is an integer")
(require (<= 0 pos (string-length str)) "0 <= pos <= length(str)")
(let ([result ...]
[other-len (string-length other-str)])
(ensure (string? result))
(ensure (= (string-length result) (+ (string-length str) other-len))
"len(result) = len(str) + len(other-str)")
(ensure (string=? (substring result pos (+ pos other-len)) other-str)
"other-str is spliced into the correct spot")
result))
15. How are we doing so far?
● Questions and Concerns?
16. How are we doing so far?
● Questions and Concerns
– What if I make a mistake in specifying the contract?
– Do contracts slow down execution speed?
– Do I need to fully specify every contract?
– Can I use contracts with language X?
– Do contracts replace documentation?
– What about state?
– What about object-oriented programming?
– What about functional programming?
17. Questions and Concerns
● What if I make a mistake in specifying the contract?
● This happens, occasionally.
● Running the program with contracts in place checks
consistency between contracts and code, not absolute
correctness.
● Usually the contracts are simpler than the code, so most of the
time the problem is in the client or supplier code.
● Advice:
– Try to keep your contracts clear and concise
– If the problem is unclear, review the contract
18. Questions and Concerns
● Do contracts slow down execution speed?
– Yes, but …
● Pre-conditions are usually very cheap to test
● Complex post-conditions (and especially invariants) can be expensive
● Advice:
– Make the level of checking configurable.
– Turn everything on in testing; turn painfully slow checks off selectively
– Leave pre-condition checking on in production, augmented by recovery strategies
● Do I need to fully specify every contract?
– No, but …
● High contract coverage helps with design, defect-detection, and overall effectiveness
● Advice:
– Specify pre-conditions fully
– Keep post-conditions simple initially; jot down complex ones as comments and implement as needed
19. Questions and Concerns
● Can I use contracts with language X?
– Almost certainly, Yes
– You can usually roll your own support using asserts,
exceptions, or even pop-ups.
– Many languages have built-in support or libraries:
● Eiffel, Racket, Clojure, D, .NET languages (via Code
Contracts add-on), etc.
20. Questions and Concerns
● Do contracts replace documentation?
– Partly. Contracts can help reduce the documentation
burden and help keep technical documentation up-to-
dated.
– Ideally, language support should include a tool for
summarizing source code, by
● omitting implementation details
● retaining signatures, doc-strings and contracts
● formatting the output appropriately, and adding hyperlinks
21. Questions and Concerns
● What about state?
– State adds a bit of complexity, but Contracts can cope.
– Pre-conditions can check initial state (as well as arguments):
● E.g. A pre-condition on a routine that reads a token from a file should
check that the file is open and that the end hasn't been reached.
– Post-conditions can check final state (as well as the result)
● E.g. A setter can check that the desired effect has occurred.
– Advice:
● Favour pure functions over ones with side-effects
● Favour immutable data structures
22. Questions and Concerns
● What about object-oriented programming?
– OO design and programming with contracts is a major
focus of the Eiffel language, but not the focus of this talk.
– Besides object state, OO-progamming with contracts
involves invariant checking, and support for inheritance
when methods are redefined in sub-types.
– Advice:
● Read up on it elsewhere, e.g.
– Bertrand Meyer's Object-oriented Software Construction, 2nd ed.
23. Questions and Concerns
● What about functional programming?
– That's the focus of the next part of this presentation!
– The material so far applied equally to both the imperative
and functional paradigms.
– Now we switch to some more functional aspects
● Out-of-the-box support in Racket for contracts
● Higher-order functions and contracts
● Checking contracts at module boundaries
24. Square root reprised, using
Racket's contract combinators
; The simple (-> domain range) contract combinator is concise,
; but limited:
;
(define/contract (real-sqrt-1 x)
(-> non-negative-real? non-negative-real?)
...)
; The “indy” (->i ...) contract combinator gives names to the
; argument(s) and to the result: greater richness, less concision
;
(define/contract (real-sqrt x)
(->i ([x non-negative-real?])
(r non-negative-real?)
#:post (r x) (approx-equal? (* r r) x))
...)
25. Contract combinators provide richer
error messages
● Hand-rolled:
– (square-root 'foo)
● Combinator:
– (real-sqrt 'foo)
pre-condition-violation: real,
non-negative argument
real-sqrt: contract violation
expected: non-negative-real?
given: 'foo
in: the x argument of
(->i
((x non-negative-real?))
(r non-negative-real?)
#:post
(r x)
...)
contract from: .../contracts.rkt
Blaming: .../contracts.rkt
At: .../contracts.rkt: [line/col of the contract]
26. Higher-order functions and
Contracts
● Higher-order functions can't be checked
immediately for conformance to a predicate. E.g.
(define/contract (make-indenter n)
(->i ([n natural-number/c])
[r (n) (->i ([s string?])
[result string?]
#:post (s result) (= (string-length result)
(+ n (string-length s))))])
(λ (s) (string-append (make-string n #space) s)))
Usage: ((make-indenter 4) “foo”) → “ foo”
27. Higher order functions and Contracts
● Racket wraps the higher-order functions in a guard and
checks what's passed in and returned at the time of
function application.
● Failures are deciphered in the error message. E.g.
((make-indenter 4) 'foo) → make-indenter: contract violation
expected: string?
given: 'foo
in: the s argument of
the r result of
(->i ((n natural-number/c)) (r (n) ...))
28. Attaching contracts at module
boundaries
● Traditionally, contracts are enforced at function
boundaries, but other choices are possible.
● In Racket, contracts are commonly wrapped around
existing functions (and data) when they are
exported from modules.
● Contract checking only occurs across module
boundaries, useful e.g. in highly recursive
scenarios.
30. Additional facilities in Racket
● Racket includes many more features for working
with contracts, including:
– Support for variable-arity and keyword arguments
– Contracts on structures
– Additional contract combinators
● See the Racket docs for details
31. Interlude: What do you do now?
● Questions for fans of automated tests:
– TDD? BDD? CI?
– What do you like about automated tests?
– Pains?
● Questions for fans of static typing:
– How sophisticated is your type-system?
– Is anyone using dependent types?
– Is anyone automatically generating tests from types?
– Likes? Pains?
● Any other approaches?
32. Automated Tests smackdown
● Positives of tests
– Popular
– Concrete and relatively
easy to understand
– Automated tests exercise
the code
– No performance impact on
production code
– TDD encourages good
design
● Negatives of tests:
– Requires discipline
– Tests don't apportion precise
blame (although TDD / rolling
back can help isolate issues)
– Not helpful in production
– A big test suite can be time-
consuming to run and take a
lot of work to maintain
33. Static Type-Checking smackdown
● Positives
– Type checks are at compile
time: early feedback
– Faster execution
– Simplifies contracts
– Enforces discipline
● Limitation
– You can't check everything
statically: Rice's theorem
● Negatives
– Demands discipline
– Can break programmer flow
– Can result in lots of boilerplate (Java,
C#, etc.)
– Large systems can be slow to compile
– Simpler type systems can be unduly
restrictive and inflexible
– Languages with superior type systems
are fairly challenging to learn (Haskell
is hard)
– Language / tool support is essential
34. Contracts smackdown
● Positives
– Assigns blame accurately
– Crushes debugging time
– Simplifies code and tests
– Can express elaborate checks
– Simplifies documentation
– Helps clarify design and
improve modularity
– Most of the benefit can be
achieved without out-of-the-
box language support
● Negatives
– Demands discipline
– Demands skill
– Not widely known or used
– Can slow down
performance, e.g. by
inflating algorithmic
complexity if you're not
careful
35. Hybrid Approach #1
● Contracts + Static Type-Checking
– Maximum concision
– This was the original mix designed into Eiffel
– Challenges?
● High-ish discipline factor
● No out-of-the-box contracts yet in Typed Racket
● For functional programming, rolling your own
higher-order contract support is non-trivial.
36. Hybrid Approach #2
● Tests + Contracts:
– Option 1:
● use explicit pre-conditions
● don't use explicit post-conditions; use tests instead
– Option 2 (my favourite):
● use scenario tests to exercise code and drive Continuous Integration
● Use pre-conditions and post-conditions instead of unit tests
– Option 3:
● automatically generate random unit tests from contracts
●
37. Hybrid Approach #3
● Tests + Contracts + Types:
● Worthwhile once you're comfortable with all three, especially
for large and complex software
● Challenges
– High discipline approach
– Difficult to get everyone in a team on board
38. Conclusion
● If you consistently write automated tests, you have the
necessary self-discipline to try contracts.
● If you are a fan of static type-checking, you can approach
contracts as a logical, dynamic extension.
● Using contracts consistently should
– greatly reduce the time you spend debugging
– make your code, tests, and documentation more concise and
readable
– clarify and simplify your design choices
– change the way you think!
Acknowledgements: Thanks to Matthias Felleisen, Greg Hendershott, Robby
Findler and Russell Sim, respectively for helpful critique and suggestions,
encouragement, a just-in-time correction, and for suggesting I give a talk.
Hinweis der Redaktion
I first learned about Design by Contract in 1998 and started using it on a major commercial project that I was designing and implementing in Visual Basic 6 (ugh!). To my joy, the time I spent debugging dropped by 90%. Since then I've experimented with several variants of programming with Contracts in a wide variety of domains and using diverse languages: Eiffel, VB.NET, C#, Python, ActionScript, and – most recently – Racket. The technical pay-off has always been very good, but my ability to persuade colleagues to try this approach has varied from point-blank refusal to reasonable uptake. Since then, the only thing that has had a comparable impact on my approach to software design and quality has been learning and then applying functional programming concepts (starting with SICP). To my mind, both Design by Contract and functional programming have an almost mathematical cleanliness, and indeed the two can be applied together in some interesting ways. With this talk I'd like to introduce you to the basics , share some of what I've learned, and perhaps discuss with you where Design by Contract sits nowadays as a pragmatic approach to Quality in a landscape also populated by TDD and increasingly advanced type systems.
In the first two cases, the blame lies with the client – the calling code Notice that the reasons are intermingled; separate require statements could fix this, but it's a trade-off between concision and precision In the last case the blame lies with the supplier – the implementation code In all cases the stack trace helps locate the problem We could improve the error messages, by including dynamic information about the nature of the violation, and refining the wording, or make use of out-of-the-box or 3 rd party facilities.
Checking types, then numerical relationships, then data is a common pattern You do not have to fully check the result in the post-condition E.g. The final ensure clause in the above example. Some sort of consistency check is often enough
Checking types, then numerical relationships, then data is a common pattern You do not have to fully check the result in the post-condition E.g. The final ensure clause in the above example. Some sort of consistency check is often enough