algebra unifies calculi of programming

33
Algebra unifies calculi of programming Tony Hoare Feb 2012

Upload: lowell

Post on 25-Feb-2016

56 views

Category:

Documents


1 download

DESCRIPTION

Algebra unifies calculi of programming. Tony Hoare Feb 2012. With Ideas from. Ian Wehrman John Wickerson Stephan van Staden Peter O’Hearn Bernhard Moeller Georg Struth Rasmus Petersen …and others. and Calculi from. Robin Milner Edsger Dijkstra Ralph Back Carroll Morgan - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Algebra unifies calculi of programming

Algebra unifies calculi of programming

Tony Hoare

Feb 2012

Page 2: Algebra unifies calculi of programming

With Ideas from

• Ian Wehrman• John Wickerson• Stephan van Staden• Peter O’Hearn• Bernhard Moeller• Georg Struth• Rasmus Petersen• …and others

Page 3: Algebra unifies calculi of programming

and Calculi from

• Robin Milner• Edsger Dijkstra• Ralph Back• Carroll Morgan• Gilles Kahn,• Gordon Plotkin• Cliff Jones• Tony Hoare

Page 4: Algebra unifies calculi of programming

but in this talk, only four

• Robin Milner x• Edsger Dijkstra x• Ralph Back

• Carroll Morgan x• Gilles Kahn,• Gordon Plotkin• Cliff Jones

• Tony Hoare x

Page 5: Algebra unifies calculi of programming

Subject matter: specs

• variables (p, q, r) stand for computer programs, designs, contracts, specifications,…

• they all describe what happens inside/around a computer that executes a given program.

• The program itself is the most precise description– giving all the excruciating detail.

• The user specification is the most abstract– describing only interactions with environment.

• Designs come in between.

Page 6: Algebra unifies calculi of programming

Example specs

• Postcondition:– execution ends with array A sorted

• Conditional correctness:– if execution ends, it ends with A sorted

• Precondition: – execution starts with x even

• Program: x := x+1 – the final value of x is one greater than the initial

Page 7: Algebra unifies calculi of programming

More examples of specs

• Safety:– There are no buffer overflows

• Termination:– execution is finite (ie., always ends)

• Liveness:– no infinite internal activity (livelock)

• Fairness:– no infinite waiting

• Probability:– the ration of a’s to b’s tends to 1 with time

Page 8: Algebra unifies calculi of programming

Also

• Security– low programs do not access high variables

• Separation– threads do not assign to shared variables

• Communication– outputs on channel c are in alphabetical order

• Predictability– there are no race conditions

• timing– interval between request and response is short

Page 9: Algebra unifies calculi of programming

Advantages of unification

• Same laws for programs, designs, specs• Same laws for many forms of correctness • Tools based on the laws serve many purposes– and communicate by sound interfaces

• Scientific controversy is resolved– and engineers confidently apply the science

Page 10: Algebra unifies calculi of programming

Operators on specs

• or \/ disjunction• and /\ conjunction• then ; sequential composition• while* concurrent composition

Constants• skip I terminates immediately• true ⊤ does anything• false ⊥ never starts

Page 11: Algebra unifies calculi of programming

The language model• a language is a set of strings of characters • /\ is set intersection• \/ is set union• ; is pointwise concatenation of strings• * is pointwise interleaving of strings

• I is the language with only the empty string• ⊤ is set of all strings• ⊥ is the empty set.

Page 12: Algebra unifies calculi of programming

Laws

\/ /\ ; *assoc yes yes yes yescomm yes yes no yes yes yes nonounit T zero T

Page 13: Algebra unifies calculi of programming

Distribution axioms

• ; distributes through \/• (p*q) ; (p’ * q’) => (p;p’)*(q;q’)– wherep => q =def q = p \/ q (refinement)

• in the language model– there are less interleavings on the left of =>

• remember the exchange law in categories?

Page 14: Algebra unifies calculi of programming

Theorems

• => is a partial order• All the operators are monotonic• (p*q);q’ => p*(q;q’)– Proof: substitute for p’ in exchange

• (p/\q);(p’/\q’) => (p;p’)/\(q;q’)– Proof: lhs => p;p’ by monotonicity of

;lhs => q;q’ similarly.

The result follows from Boolean algebra.

Page 15: Algebra unifies calculi of programming

Hoare triple: {p} q {r} •defined as p;q => r – starting in the final state of any execution of p,

q ends in the final state of some execution of r– (p and r may be arbitrary specs).

•example: {..x+1 ≤ n} x:= x + 1 {..x ≤ n} • where ..b (finally b) describes all executions that end in a state

satisfying a single-state predicate b .

Page 16: Algebra unifies calculi of programming

Milner triple: r - p -> q

• defined as p;q => r (same as Hoare!)– r may be executed by first executing p

and then executing q .– p is usually restricted to atomic actions.•example: (<a>;q) – <a> -> q

where <a> is an atomic action

– r -> p = def. p => r• an execution step may reduce non-determinism

Page 17: Algebra unifies calculi of programming

Theorems

• Using these definitions, the rules of the Hoare calculus and of the Milner calculus

are derivable by simple algebraic proofs from the laws of the algebra.

Page 18: Algebra unifies calculi of programming

Rule of consequence

• p => p’ {p’} q {r’} r’ => r{p} q {r}

• r -> r’ r’ –q-> p’ p’ -> pr –q-> p

• Proof: ; is monotonic, => is transitive.• These two rules are not only similar,

they are the same!

Page 19: Algebra unifies calculi of programming

Sequential composition

{p} q {s} {s} q’ {r} {p} q;q’ {r}

r –q-> s s –q’-> p r –q;q’-> p

– Proof: associativity of ;

Page 20: Algebra unifies calculi of programming

Small-step rule

p –<a>-> r (r;q’) –<a>->(q;q’)

{p} q {r} .{p} q;q’ {r;q’}

Proof: monotonicity of ;

Page 21: Algebra unifies calculi of programming

Choice

• {p} q {r} {p} q’ {r}{p} (q \/ q’) {r}

– both choices must be correct

• r –q-> p (r \/ r’) –q-> p

– so the execution may make either choice

Page 22: Algebra unifies calculi of programming

Frame Law

• {p} q {r}{p*f} q {r*f}

– adapts a rule to a wider environment f

• r –q-> p(r*f) –q-> (p*f)

– a step that is possible for a single threadis still possible in a wider environment f

Page 23: Algebra unifies calculi of programming

Concurrency (and Conjunction)

• {p} q {r} {p’} q’ {r’} {p*p’} (q * q’) {r*r’}

– permits modular proof of concurrent programs.

• {p} q {r} {p’} q’ {r’} {p/\p’} (q /\ q’) {r/\r’}

– in Floyd’s rule of conjunction, q = q’.

Page 24: Algebra unifies calculi of programming

Concurrency

r –p-> q r’ -p’-> q’ (r*r) -(p*p’)-> (q*q’)

– provided p*p’ = τ– where τ is the unobserved transition,(which occurs (in CCS) when p and p’ are an input and an output on the same channel).

Page 25: Algebra unifies calculi of programming

Dijkstra triple: p => q\r

• usually written: p => wlp(q,r) ‘defined’ by: p;q => r (again) wp(q,r) = wlp(q, r/\ terminates)

q\r specifies the weakest program which can be executed before q to achieve the overall effect of r .

Page 26: Algebra unifies calculi of programming

Morgan triple: q => r/p

• ‘defined’ by: p;q => r (again)

• r/p is the weakest program which can be executed after p to achieve r.

Page 27: Algebra unifies calculi of programming

Theorems

• q\(r /\ r’) = (q\r) /\ (q\r’)• (r /\ r’)/p = (r/p) /\ (r’/p)

• (q \/ q’)\r = (q\r) /\ (q’\r)• r/(p \/ p’) = (r/p) /\ (r/p’)

Page 28: Algebra unifies calculi of programming

Theorems

• (q;q’)\r = q\(q’\r)• r/(p;p’) = (r/p)/p’

• (q\r)*(q’\r’) => (q*q’)\(r*r’)exchange law

Page 29: Algebra unifies calculi of programming

Unification

is the goal of every branch of pure science,because it increases conviction in theory

Specialisationis needed for each application, e.g.,

Hoare logic: proofs of correctness,Milner: implementation,Dijkstra: program analysis,Morgan: programs from specifications …

Page 30: Algebra unifies calculi of programming

Algebra

• is modular, incremental, comprehensible, abstract and beautiful.

• An algebraic law can say as little as you like– using any other concepts that you need.– new properties can be added by new laws.

• The definition of a concept is allowed only once– so it must describe all the needed properties,– using only previously defined concepts.

• Inductive clauses restrict incrementality• Algebra is good for unification

Page 31: Algebra unifies calculi of programming

Summary: p;q => ris written

p {q} r{p} q {r}r –p-> qp => wlp(q,r)q => [p, r]

byHoare (triple)Wirth (triple)Milner (transition)Dijkstra (weakest precondition)Morgan (specification statement)

Milner restricts p to atomic actions (small-step version).The others restrict p and r to descriptions of single states.

Page 32: Algebra unifies calculi of programming

Conclusions• All the calculi are derived from the same algebra of

programming.

• The algebra is simpler than each of the calculi,

• and stronger than all of them combined.

Page 33: Algebra unifies calculi of programming

Isaac Newton

Communication with Richard Gregory (1694)

“Our specious algebra [of fluxions] is fit enough to find out, but entirely unfit to consign to writing and commit to posterity.”