Bertrand Meyer's technology+ blog

_{About this article: it originated as a series of posts on the Communications of the ACM blog. I normally repost such articles here. (Even though copy-paste is usually not good, there are three reasons for this duplication: the readership seems to be largely disjoint; I can use better formatting, since their blog software is more restrictive than WordPress; and it is good to have a single repository for all my articles, including both those who originated on CACM and those who did not.) The series took the form of nine articles, where each of the first few ended with a quiz, to which the next one, published a couple of days later, provided an answer. Since all these answers are now available it would make no sense to use the same scheme, so I am instead publishing the whole thing as a single article  with nine sections, slightly adapted from the original.}

_{I was too lazy so far to collect all the references into a single list, so numbers such as [1] refer to the list at the end of the corresponding section.}

A colleague recently asked me to present a short overview of  axiomatic semantics as a guest lecture in one of his courses. I have been teaching courses on software verification for a long time (see e.g. here), so I have plenty of material; but instead of just reusing it, I decided to spend a bit of time on explaining why it is good to have a systematic approach to software verification. Here is the resulting tutorial.

1. Introduction and attempt #1

Say “software verification” to software professionals, or computer science students outside of a few elite departments, and most of them will think  “testing”. In a job interview, for example, show a loop-based algorithm to a programmer and ask “how would you verify it?”: most will start talking about devising clever test cases.

Far from me to berate testing [1]; in fact, I have always thought that the inevitable Dijkstra quote about testing — that it can only show the presence of errors, not their absence [2] — which everyone seems to take as an indictment and dismissal of testing (and which its author probably intended that way) is actually a fantastic advertisement for testing: a way to find bugs? Yes! Great! Where do I get it?  But that is not the same as verifying the software, which means attempting to ascertain that it has no bugs.

Until listeners realize that verification cannot just mean testing, the best course material on axiomatic semantics or other proof techniques will not attract any interest. In fact, there is somewhere a video of a talk by the great testing and public-speaking guru James Whittaker where he starts by telling his audience not to worry, this won’t be a standard boring lecture, he will not start talking about loop invariants [3]! (Loop invariants are coming in this article, in fact they are one of its central concepts, but in later sections only, so don’t bring the sleeping bags yet.) I decided to start my lecture by giving an example of what happens when you do not use proper verification. More than one example, in fact, as you will see.

A warning about this article: there is nothing new here. I am using an example from my 1990 book Introduction to the Theory of Programming Languages (exercise 9.12). Going even further back, a 1983 “Programming Pearls” Communications of the ACM article by Jon Bentley [4] addresses the same example with the same basic ideas. Yet almost forty years later these ideas are still not widely known among practitioners. So consider these articles as yet another tutorial on fundamental software engineering stuff.

The tutorial is a quiz. We start with a program text:

from

i := 1 ; j := n              — Result initialized to 0.

until i = j loop

m := (i + j) // 2         — Integer division

if t [m] ≤ x then i := m  else  j := m end

end

if x = t [i] then Result := i end

All variables are of integer type. t is an up-sorted array of integers, indexed from 1 to n . We do not let any notation get between friends. A loop from p until e loop q end executes p then, repeatedly: stops if e (the exit condition) is true, otherwise executes q. (Like {p ; while not e do {q}} in some other notations.) “:=” is assignment, “=” equality testing.  “//” is integer division, e.g. 6 //3 = 7 //3 = 2. Result is the name of a special variable whose final value will be returned by this computation (as part of a function, but we only look at the body). Result is automatically initialized to zero like all integer variables, so if execution does not assign anything to Result the function will return zero.

First question: what is this program trying to do?

OK, this is not the real quiz. I assume you know the answer: it is an attempt at “binary search”, which finds an element in the array, or determines its absence, in a sequence of about log₂ (n) steps, rather than n if we were use sequential search.  (Remember we assume the array is sorted.) Result should give us a position where x appears in the array, if it does, and otherwise be zero.

Now for the real quiz: does this program meet this goal?

The answer should be either yes or no. (If no, I am not asking for a correct version, at least not yet, and in any case you can find some in the literature.) The situation is very non-symmetric, we might say Popperian:

• To justify a no answer it suffices of a single example, a particular array t and a particular value x, for which the program fails to set Result as it should.
• To justify a yes answer we need to provide a credible argument that for every t and  x the program sets Result as it should.

Notes to section 1

[1] The TAP conference series (Tests And Proofs), which Yuri Gurevich and I started, explores the complementarity between the two approaches.

[2] Dijkstra first published his observation in 1969. He did not need consider the case of infinite input sets: even for a trivial finite program that multiplies two 32-bit integers, the number of cases to be examined, 2⁶⁴, is beyond human reach. More so today with 64-bit integers. Looking at this from a 2020 perspective, we may note that exhaustive testing of a finite set of cases, which Dijkstra dismissed as impossible in practice, is in fact exactly what the respected model checking verification technique does; not on the original program, but on a simplified — abstracted — version precisely designed to keep the number of cases tractable. Dijkstra’s argument remains valid, of course, for  the original program if non-trivial. And model-checking does not get us out of the woods: while we are safe if its “testing” finds no bug, if it does find one we have to ensure that the bug is a property of the original program rather than an artifact of the abstraction process.

[3] It is somewhere on YouTube, although I cannot find it right now.

[4] Jon Bentley: Programming Pearls: Writing Correct Programs, in Communications of the ACM, vol. 26, no. 12, pp. 1040-1045, December 1983, available for example here.

2. Attempt #2

Was program #1 correct? If so it should yield the correct answer. (An answer is correct if either Result is the index in t of an element equal to x, or Result = 0 and x does not appear in t.)

This program is not correct. To prove that it is not correct it suffices of a single example (test case) for which the program does not  “yield the correct answer”. Assume x = 1 and the array t has two elements both equal to zero (n = 2, remember that arrays are indexed from 1):

t = [0   0]

The successive values of the variables and expressions are:

                                            m       i          j            i + j + 1

After initialization:                   1         2                3

i ≠ j, so enter loop:           1       1        2                 6         — First branch of “if” since t [1] ≤ x
— so i gets assigned the value of m

But then neither of the values of i and j has changed, so the loop will repeat its body identically (taking the first branch) forever. It is not even that the program yields an incorrect answer: it does not yield an answer at all!

Note (in reference to the famous Dijkstra quote mentioned in the first article), that while it is common to pit tests against proofs, a test can actually be a proof: a test that fails is a proof that the program is incorrect. As valid as the most complex mathematical proof. It may not be the kind of proof we like most (our customers tend to prefer a guarantee that the program is correct), but it is a proof all right.

We are now ready for the second attempt:

—  Program attempt #2.

from

i := 1 ; j := n

until i = j or Result > 0  loop

m := (i + j) // 2         — Integer division

if t [m] ≤ x then

i := m  + 1

elseif t [m] = x then

Result := m

else                         — In this case t [m] > x

j := m – 1

end

end

Unlike the previous one this version always changes i or j, so we may hope it does not loop forever. It has a nice symmetry between i and j.

Same question as before: does this program meet its goal?

3. Attempt #3

The question about program #2, as about program #1: was: it right?

Again no.  A trivial example disproves it: n = 1, the array t contains a single element t [1] = 0, x = 0. Then the initialization sets both i and j to 1, i = j holds on entry to the loop which stops immediately, but Result is zero whereas it should be 1 (the place where x appears).

Here now is attempt #3, let us see it if fares better:

—  Program attempt #3.

from

i := 1 ; j := n

until i = j loop

m := (i + j + 1) // 2

if t [m] ≤ x then

i := m  + 1

else

j := m

end

end

if 1  ≤ i  and i ≤ n then Result := i end
       — If not, Result remains 0.

What about this one?

3. Attempt #4 (also includes 3′)

The first two program attempts were wrong. What about the third?

I know, you have every right to be upset at me, but the answer is no once more.

Consider a two-element array t = [0 0] (so n = 2, remember that our arrays are indexed from 1 by convention) and a search value x = 1. The successive values of the variables and expressions are:

                                                  m          i          j            i + j + 1

After initialization:                            1        2           4

i ≠ j, so enter loop:               2           3        2          6                  — First branch of “if” since t [2] < x

i ≠ j,  enter loop again:        3           ⚠                                       — Out-of-bounds memory access!
— (trying to access non-existent t [3])

Oops!

Note that we could hope to get rid of the array overflow by initializing i to 0 rather than 1. This variant (version #3′) is left as a bonus question to the patient reader. (Hint: it is also not correct. Find a counter-example.)

OK, this has to end at some point. What about the following version (#4): is it right?

—  Program attempt #4.

from

i := 0 ; j := n + 1

until i = j loop

m := (i + j) // 2

if t [m] ≤ x then

i := m  + 1

else

j := m

end

end

if 1 ≤ i  and i ≤ n then Result := i end

5. Attempt #5

Yes, I know, this is dragging on. But that’s part of the idea: witnessing how hard it is to get a program right if you just judging by the seat of your pants. Maybe we can get it right this time?

Are we there yet? Is program attempt #4 finally correct?

Sorry to disappoint, but no. Consider a two-element array t = [0 0], so n = 2, and a search value x = 1 (yes, same counter-example as last time, although here we could also use x = 0). The successive values of the variables and expressions are:

                                                 m          i          j            i + j

After initialization:                           0        3           3

i ≠ j, so enter loop:               1           2       3          5            — First branch of “if”

i ≠ j, enter loop again:         2         3        3         6            — First branch again

i = j, exit loop

The condition of the final “if” is true, so Result gets the value 3. This is quite wrong, since there is no element at position 3, and in any case x does not appear in t.

But we are so close! Something like this should work, should it not?

So patience, patience, let us tweak it just one trifle more, OK?

—  Program attempt #5.

from

i := 0 ; j := n

until i ≥ j or Result > 0 loop

m := (i + j) // 2

if t [m] < x then

i := m + 1

elseif  t [m] > x then

j := m

else

Result := m

end

end

Does it work now?

6. Attempt #6

The question about program #5  was the same as before: is it right, is it wrong?

Well, I know you are growing more upset at me with each section, but the answer is still that this program is wrong. But the way it is wrong is somewhat specific; and it applies, in fact, to all previous variants as well.

This particular wrongness (fancy word for “bug”) has a history. As I pointed out in the first article, there is a long tradition of using binary search to illustrate software correctness issues. A number of versions were published and proved correct, including one in the justly admired Programming Pearls series by Jon Bentley. Then in 2006 Joshua Bloch, then at Google, published a now legendary blog article [2] which showed that all these versions suffered from a major flaw: to obtain m, the approximate mid-point between i and j, they compute

(i + j) // 2

which, working on computer integers rather than mathematical integers, might overflow! This in a situation in which both i and j, and hence m as well, are well within the range of the computer’s representable integers, 2^-n to 2ⁿ (give or take 1) where n is typically 31 or, these days, 63, so that there is no conceptual justification for the overflow.

In the specification that I have used for this article, i starts at 1, so the problem will only arise for an array that occupies half of the memory or more, which is a rather extreme case (but still should be handled properly). In the general case, it is often useful to use arrays with arbitrary bounds (as in Eiffel), so we can have even a small array, with high indices, for which the computation will produce an overflow and bad results.

The Bloch gotcha is a stark reminder that in considering the correctness of programs we must include all relevant aspects and consider programs as they are executed on a real computer, not as we wish they were executed in an ideal model world.

(Note that Jon Bentley alluded to this requirement in his original article: while he did not explicitly mention integer overflow, he felt it necessary to complement his proof by the comment that that  “As laborious as our proof of binary search was, it is still unfinished by some standards. How would you prove that the program is free of runtime errors (such as division by zero, word overflow, or array indices out of bounds)?” Prescient words!)

It is easy to correct the potential arithmetic overflow bug: instead of (i + j) // 2, Bloch suggested we compute the average as

i + (j – i) // 2

which is the same from a mathematician’s viewpoint, and indeed will compute the same value if both variants compute one, but will not overflow if both i and j are within range.

So we are ready for version 6, which is the same as version 5 save for that single change:

—  Program attempt #6.

from

i := 0 ; j := n

until i ≥ j or Result > 0 loop

m := i + (j – i) // 2

if t [m] < x then

i := m + 1

elseif  t [m] > x then

j := m

else

Result := m

end

end

Now is probably the right time to recall the words by which Donald Knuth introduces binary search in the original 1973 tome on Sorting and Searching of his seminal book series The Art of Computer Programming:

Although the basic idea of binary search is comparatively straightforward, the details can be somewhat tricky, and many good programmers have done it wrong the first few times they tried.

Do you need more convincing? Be careful what you answer, I have more variants up my sleeve and can come up with many more almost-right-but-actually-wrong program attempts if you nudge me. But OK, even the best things have an end. This is not the last section yet, but that was the last program attempt. To the naturally following next question in this running quiz,  “is version 6 right or wrong”, I can provide the answer: it is, to the best of my knowledge, a correct program. Yes! [3].

But the quiz continues. Since answers to the previous questions were all  that the programs were not correct, it sufficed in each case to find one case for which the program did not behave as expected. Our next question is of a different nature: can you find an argument why version #6 is correct?

References for section 6

[1] (In particular) Jon Bentley: Programming Pearls — Writing Correct Programs, in Communications of the ACM, vol. 26, no. 12, December 1983, pages 1040-1045, available here.

[2] Joshua Bloch: Extra, Extra — Read All About It: Nearly All Binary Searches and Mergesorts are Broken, blog post, on the Google AI Blog, 2 June 2006, available here.

[3] A caveat: the program is correct barring any typos or copy-paste errors — I am starting from rigorously verified programs (see the next posts), but the blogging system’s UI and text processing facilities are not the best possible for entering precise technical text such as code. However carefully I check, I cannot rule out a clerical mistake, which of course would be corrected as soon as it is identified.

7. Using a program prover

Preceding sections presented candidate binary search algorithms and asked whether they are correct. “Correct” means something quite precise: that for an array t and a value x, the final value of the variable Result is a valid index of t (that is to say, is between 1 and n, the size of t) if and only if x appears at that index in t.

The last section boldly stated that program attempt #6 was correct. The question was: why?

In the case of the preceding versions, which were incorrect, you could prove that property, and I do mean prove, simply by exhibiting a single counter-example: a single t and x for which the program does not correctly set Result. Now that I asserting the program to be correct, one example, or a million examples, do not suffice. In fact they are almost irrelevant. Test as much as you like and get correct results every time, you cannot get rid of the gnawing fear that if you had just tested one more time after the millionth test you would have produced a failure. Since the set of possible tests is infinite there is no solution in sight [1].

We need a proof.

I am going to explain that proof in the next section, but before that I would like to give you an opportunity to look at the proof by yourself. I wrote in one of the earlier articles that most of what I have to say was already present in Jon Bentley’s 1983 Programming Pearls contribution [2], but a dramatic change did occur in the four decades since: the appearance of automated proof system that can handle significant, realistic programs. One such system, AutoProof, was developed at the Chair of Software engineering at ETH Zurich [3] (key project members were Carlo Furia, Martin Nordio, Nadia Polikarpova and Julian Tschannen, with initial contributions by Bernd Schoeller) on the basis of the Boogie proof technology from Microsoft Research).

AutoProof is available for online use, and it turns out that one of the basic tutorial examples is binary search. You can go to the corresponding page and run the proof.

I am going to let you try this out (and, if you are curious, other online AutoProof examples as well) without too many explanations; those will come in the next section. Let me simply name the basic proof technique: loop invariant. A loop invariant is a property INV associated with a loop, such that:

• A. After the loop’s initialization, INV will hold.
• B. One execution of the loop’s body, if started with INV satisfied (and the loop’s exit condition not satisfied, otherwise we wouldn’t be executing the body!), satisfies INV again when it terminates.

This idea is of course the same as that of a proof by induction in mathematics: the initialization corresponds to the base step (proving that P (0) holds) and the body property to the induction step (proving that from P (n) follows P (n + 1). With a traditional induction proof we deduce that the property (P (n)) holds for all integers. For the loop, we deduce that when the loop finishes its execution:

• The invariant still holds, since executing the loop means executing the initialization once then the loop body zero or more times.
• And of course the exit condition also holds, since otherwise we would still be looping.

That is how we prove the correctness of a loop: the conjunction of the invariant and the exit condition must yield the property that we seek (in the example, the property, stated above of Result relative to t and x).

We also need to prove that the loop does terminate. This part involves another concept, the loop’s variant, which I will explain in the next section.

For the moment I will not say anything more and let you look at the AutoProof example page (again, you will find it here), run the verification, and read the invariant and other formal elements in the code.

To “run the verification” just click the Verify button on the page. Let me emphasize (and emphasize again and again and again) that clicking Verify will not run the code. There is no execution engine in AutoProof, and the verification does not use any test cases. It processes the text of the program as it appears on the page and below. It applies mathematical techniques to perform the proof; the core property to be proved is that the proposed loop invariant is indeed invariant (i.e. satisfies properties A and B above).

The program being proved on the AutoProof example page is version #6 from the last section, with different variable names. So far for brevity I have used short names such as i, j and m but the program on the AutoProof site applies good naming practices with variables called low, up, middle and the like. So here is that version again with the new variable names:

—  Program attempt #7  (identical to #6 with different variable names) .

from

low := 0 ; up := n

until low ≥ up or Result > 0 loop

middle := low + ((up – low) // 2)

if a [middle] < value then      — The array is now called a rather than t

low := middle + 1

elseif  a [middle] > value then

up := middle

else

Result := middle

end

end

This is exactly the algorithm text on the AutoProof page, the one that you are invited to let AutoProof verify for you. I wrote “algorithm text” rather than “program text” because the actual program text (in Eiffel) includes variant and invariant clauses which do not affect the program’s execution but make the proof possible.

Whether or not these concepts (invariant, variant, program proof) are completely new to you, do try the prover and take a look at the proof-supporting clauses. In the next article I will remove any remaining mystery.

Note and references for section 7

[1] Technically the set of possible [array, value] pairs is finite, but of a size defying human abilities. As I pointed out in the first section, the “model checking” and “abstract interpretation” verification techniques actually attempt to perform an exhaustive test anyway, after drastically reducing the size of the search space. That will be for some other article.

[2]  Jon Bentley: Programming Pearls: Writing Correct Programs, in Communications of the ACM, vol. 26, no. 12, pp. 1040-1045, December 1983, available for example here.

[3] The AutoProof page contains documentations and numerous article references.

8. Understanding the proof

The previous section invited you to run the verification on the AutoProof tutorial page dedicated to the example. AutoProof is an automated proof system for programs. This is just a matter of clicking  “Verify”, but more importantly, you should read the annotations added to the program text, particularly the loop invariant, which make the verification possible. (To avoid any confusion let me emphasize once more that clicking “Verify” does not run the program, and that no test cases are used; the effect is to run the verifier, which attempts to prove the correctness of the program by working solely on the program text.)

Here is the program text again, reverting for brevity to the shorter identifiers (the version on the AutoProof page has more expressive ones):

from

i := 0 ; j := n

until i ≥ j or Result > 0 loop

m := i + (j – i) // 2

if t [m] < x then

i := m + 1

elseif  t [m] > x then

j := m

else

Result := m

end

end

Let us now see what makes the proof possible. The key property is the loop invariant, which reads

A:   1  ≤ i  ≤ j  ≤ n + 1
B:   0  ≤ Result  ≤ n
C:   ∀ k: 1 .. i –1  |  t [k] < x
D:   ∀ k: j .. n  |  t [k] > x
E:    (Result > 0)   ⇒   (t [Result] = x)

The notation is slightly different on the Web page to adapt to the Eiffel language as it existed at the time it was produced; in today’s Eiffel you can write the invariant almost as shown above. Long live Unicode, allowing us to use symbols such as ∀ (obtained not by typing them but by using smart completion, e.g. you start typing “forall” and you can select the ∀ symbol that pops up), ⇒ for  “implies” and many others

Remember that the invariant has to be established by the loop’s initialization and preserved by every iteration. The role of each of its clauses is as follows:

• A: keep the indices in range.
• B: keep the variable Result, whose final value will be returned by the function, in range.
• C and D: eliminate index intervals in which we have determined that the sought value, x, does not appear. Before i, array values are smaller; starting at j, they are greater. So these two intervals, 1..i and j..n, cannot contain the sought value. The overall idea of the algorithm (and most other search algorithms) is to extend one of these two intervals, so as to narrow down the remaining part of 1..n where x may appear.
• E: express that as soon as we find a positive (non-zero) Result, its value is an index in the array (see B) where x does appear.

Why is this invariant useful? The answer is that on exit it gives us what we want from the algorithm. The exit condition, recalled above, is

i ≥ j or Result > 0

Combined with the invariant, it tells us that on exit one of the following will hold:

• Result > 0, but then because of E we know that x appears at position Result.
• i < j, but then A,  C and D  imply that x does not appear anywhere in t. In that case it cannot be true that Result > 0, but then because of B Result must be zero.

What AutoProof proves, mechanically, is that under the function’s precondition (that the array is sorted):

• The initialization ensures the invariant.
• The loop body, assuming that the invariant is satisfied but the exit condition is not, ensures the loop invariant again after it executes.
• The combination of the invariant and the exit condition ensures, as just explained, the postcondition of the function (the property that Result will either be positive and the index of an element equal to x, or zero with the guarantee that x appears nowhere in t).

Such a proof guarantees the correctness of the program if it terminates. We (and AutoProof) must prove separately that it does terminate. The technique is simple: find a “loop variant”, an integer quantity v  which remains non-negative throughout the loop (in other words, the loop invariant includes or implies v ≥ 0) and decreases on each iteration, so that the loop cannot continue executing forever. An obvious variant here is j – i + 1 (where the + 1 is needed because j – i may go down to -1 on the last iteration if x does not appear in the array). It reflects the informal idea of the algorithm: repeatedly decrease an interval i .. j – 1 (initially, 1 .. n) guaranteed to be such that x appears in t if and only if it appears at an index in that interval. At the end, either we already found x or the interval is empty, implying that x does not appear at all.

A great reference on variants and the techniques for proving program termination is a Communications of the ACM article of 2011: [3].

The variant gives an upper bound on the number of iterations that remain at any time. In sequential search, j – i + 1 would be our best bet; but for binary search it is easy to show that  log₂  (j – i + 1) is also a variant, extending the proof of correctness with a proof of performance (the key goal of binary search being to ensure a logarithmic rather than linear execution time).

This example is, I hope, enough to highlight the crucial role of loop invariants and loop variants in reasoning about loops. How did we get the invariant? It looks like I pulled it out of a hat. But in fact if we go the other way round (as advocated in classic books [1] [2]) and develop the invariant and the loop together the process unfolds itself naturally and there is nothing mysterious about the invariant.

Here I cannot resist quoting (thirty years on!) from my own book Introduction to the Theory of Programming Languages [4]. It has a chapter on axiomatic semantics (also known as Hoare logic, the basis for the ideas used in this discussion), which I just made available: see here [5]. Its exercise 9.12 is the starting point for this series of articles. Here is how the book explains how to design the program and the invariant [6]:

In the general case [of search, binary or not] we aim for a loop body of the form

m := ‘‘Some value in 1.. n such that i ≤ m < j’’;

if t [m] ≤ x then

i := m + 1

else

j := m

end

It is essential to get all the details right (and easy to get some wrong):

• The instruction must always decrease the variant j – i, by increasing i or decreasing j. If the the definition of m specified just m ≤ j rather than m < j, the second branch would not meet this goal.
•  This does not transpose directly to i: requiring i < m < j would lead to an impossibility when j – i is equal to 1. So we accept i ≤ m but then we must take m + 1, not m, as the new value of i in the first branch.
•  The conditional’s guards are tests on t [m], so m must always be in the interval 1 . . n. This follows from the clause 0 ≤ i ≤ j ≤ n + 1 which is part of the invariant.
•  If this clause is satisfied, then m ≤ n and m > 0, so the conditional instruction indeed leaves this clause invariant.
• You are invited to check that both branches of the conditional also preserve the rest of the invariant.
• Any policy for choosing m is acceptable if it conforms to the above scheme. Two simple choices are i  and j – 1; they lead to variants of the sequential search algorithm [which the book discussed just before binary search].

For binary search, m will be roughly equal to the average of i and j.

“Roughly” because we need an integer, hence the // (integer division).

In the last section, I will reflect further on the lessons we can draw from this example, and the practical significance of the key concept of invariant.

References and notes for section 8

[1] E.W. Dijkstra: A Discipline of Programming, Prentice Hall, 1976.

[2] David Gries: The Science of Programming, Springer, 1989.

[3] Byron Cook, Andreas  Podelski and Andrey Rybalchenko: Proving program termination, in Communications of the ACM, vol. 54, no. 11, May 2011, pages 88-98, available here.

[4] Bertrand Meyer, Introduction to the Theory of Programming Languages, Prentice Hall, 1990. The book is out of print but can be found used, e.g. on Amazon. See the next entry for an electronic version of two chapters.

[5] Bertrand Meyer Axiomatic semantics, chapter 9 from [3], available here. Note that the PDF was reconstructed from an old text-processing system (troff); the figures could not be recreated and are missing. (One of these days I might have the patience of scanning them from a book copy and adding them. Unless someone wants to help.) I also put online, with the same caveat, chapter 2 on notations and mathematical basis: see here.

[6] Page 383 of [4] and [5]. The text is verbatim except a slight adaptation of the programming notation and a replacement of the variables: i in the book corresponds to i – 1 here, and j to j – 1. As a matter of fact I prefer the original conventions from the book (purely as a matter of taste, since the two are rigorously equivalent), but I changed here to the conventions of the program as it appears in the AutoProof page, with the obvious advantage that you can verify it mechanically. The text extract is otherwise exactly as in the 1990 book.

9. Lessons learned

What was this journey about?

We started with a succession of attempts that might have “felt right” but were in fact all wrong, each in its own way: giving the wrong answer in some cases, crashing (by trying to access an array outside of its index interval) in some cases, looping forever in some cases. Always “in some cases”,  evidencing the limits of testing, which can never guarantee that it exercises all the problem cases. A correct program is one that works in all cases. The final version was correct; you were able to prove its correctness with an online tool and then to understand (I hope) what lies behind that proof.

To show how to prove such correctness properties, I have referred throughout the series to publications from the 1990s (my own Introduction to The Theory of Programming Languages), the 1980s (Jon Bentley’s Programming Pearls columns, Gries’s Science of Programming), and even the 1970s (Dijkstra’s Discipline of Programming). I noted that the essence of my argument appeared in a different form in one of Bentley’s Communications articles. What is the same and what has changed?

The core concepts have been known for a long time and remain applicable: assertion, invariant, variant and a few others, although they are much better understood today thanks to decades of theoretical work to solidify the foundation. Termination also has a more satisfactory theory.

On the practical side, however, the progress has been momentous. Considerable engineering has gone into making sure that the techniques scaled up. At the time of Bentley’s article, binary search was typical of the kind of programs that could be proved correct, and the proof had to proceed manually. Today, we can tackle much bigger programs, and use tools to perform the verification.

Choosing binary search again as an example today has the obvious advantage that everyone can understand all the details, but should not be construed as representative of the state of the art. Today’s proof systems are far more sophisticated. Entire operating systems, for example, have been mechanically (that is to say, through a software tool) proved correct. In the AutoProof case, a major achievement was the proof of correctness [1] of an entire data structure (collections) library, EiffelBase 2. In that case, the challenge was not so much size (about 8,000 source lines of code), but the complexity of both:

• The scope of the verification, involving the full range of mechanisms of a modern object-oriented programming language, with classes,  inheritance (single and multiple), polymorphism, dynamic binding, generics, exception handling etc.
• The code itself, using sophisticated data structures and algorithms, involving in particular advanced pointer manipulations.

In both cases, progress has required advances on both the science and engineering sides. For example, the early work on program verification assumed a bare-bones programming language, with assignments, conditionals, loops, routines, and not much more. But real programs use many other constructs, growing ever richer as programming languages develop. To cover exception handling in AutoProof required both theoretical modeling of this construct (which appeared in [2]) and implementation work.

More generally, scaling up verification capabilities from the small examples of 30 years ago to the sophisticated software that can be verified today required the considerable effort of an entire community. AutoProof, for example, sits at the top of a tool stack relying on the Boogie environment from Microsoft Research, itself relying on the Z3 theorem prover. Many person-decades of work make the result possible.

Beyond the tools, the concepts are esssential. One of them, loop invariants, has been illustrated in the final version of our program. I noted in the first article the example of a well-known expert and speaker on testing who found no better way to announce that a video would not be boring than  “relax, we are not going to talk about loop invariants.” Funny perhaps, but unfair. Loop invariants are one of the most beautiful concepts of computer science. Not so surprisingly, because loop invariants are the application to programming of the concept of mathematical induction. According to the great mathematician Henri Poincaré, all of mathematics rests on induction; maybe he exaggerated, maybe not, but who would think of teaching mathematics without explaining induction? Teaching programming without explaining loop invariants is no better.

Below is an illustration (if you will accept my psychedelic diagram) of what a loop is about, as a problem-solving technique. Sometimes we can get the solution directly. Sometimes we identify several steps to the solution; then we use a sequence (A ; B; C). Sometimes we can find two (or more) different ways of solving the problem in different cases; then we use a conditional (if c then A else B end). And sometimes we can only get a solution by getting closer repeatedly, not necessarily knowing in advance how many times we will have to advance towards it; then, we use a loop.

We identify an often large (i.e. very general) area where we know the solution will lie; we call that area the loop invariant. The solution or solutions (there may be more than one) will have to satisfy a certain condition; we call it the exit condition. From wherever we are, we shoot into the invariant region, using an appropriate operation; we call it the initialization. Then we execute as many times as needed (maybe zero if our first shot was lucky) an operation that gets us closer to that goal; we call it the loop body. To guarantee termination, we must have some kind of upper bound of the distance to the goal, decreasing each time discretely; we call it the loop variant.

This explanation is only an illustration, but I hope it makes the ideas intuitive. The key to a loop is its invariant. As the figure suggests, the invariant is always a generalization of the goal. For example, in binary search (and many other search algorithms, such as sequential search), our goal is to find a position where either x appears or, if it does not, we can be sure that it appears nowhere. The invariant says that we have an interval with the same properties (either x appears at a position belonging to that interval or, if it does not, it appears nowhere). It obviously includes the goal as a special case: if the interval has length 1, it defines a single position.

An invariant should be:

1. Strong enough that we can devise an exit condition which in the end, combined with the invariant, gives us the goal we seek (a solution).
2. Weak enough that we can devise an initialization that ensures it (by shooting into the yellow area) easily.
3. Tuned so that we can devise a loop body that, from a state satifying the invariant, gets us to a new one that is closer to the goal.

In the example:

1. The exit condition is simply that the interval’s length is 1. (Technically, that we have computed Result as the single interval element.) Then from the invariant and the exit condition, we get the goal we want.
2. Initialization is easy, since we can just take the initial interval to be the whole index range of the array, which trivially satisfies the invariant.
3. The loop body simply decreases the length of the interval (which can serve as loop variant to ensure termination). How we decrease the length depends on the search strategy; in sequential search, each iteration decreases the length by 1, correct although not fast, and binary search decreases it by about half.

The general scheme always applies. Every loop algorithm is characterized by an invariant. The invariant may be called the DNA of the algorithm.

To demonstrate the relevance of this principle, my colleagues Furia, Velder, and I published a survey paper [6] in ACM Computing Surveys describing the invariants of important algorithms in many areas of computer science, from search algorithms to sorting (all major algorithms), arithmetic (long integer addition, squaring), optimization and dynamic programming  (Knapsack, Levenshtein/Edit distance), computational geometry (rotating calipers), Web (Page Rank)… I find it pleasurable and rewarding to go deeper into the basis of loop algorithms and understand their invariants; like a geologist who does not stop at admiring the mountain, but gets to understand how it came to be.

Such techniques are inevitable if we want to get our programs right, the topic of this article. Even putting aside the Bloch average-computation overflow issue, I started with 5 program attempts, all kind of friendly-looking but wrong in different ways. I could have continued fiddling with the details, following my gut feeling to fix the flaws and running more and more tests. Such an approach can be reasonable in some cases (if you have an algorithm covering a well-known and small set of cases), but will not work for non-trivial algorithms.

Newcomers to the concept of loop invariant sometimes panic: “this is all fine, you gave me the invariants in your examples, how do I find my own invariants for my own loops?” I do not have a magic  recipe (nor does anyone else), but there is no reason to be scared. Once you have understood the concept and examined enough examples (just a few of those in [6] should be enough), writing the invariant at the same time as you are devising a loop will come as a second nature to you.

As the fumbling attempts in the first few sections should show, there is not much of an alternative. Try this approach. If you are reaching these final lines after reading what preceded them, allow me to thank you for your patience, and to hope that this rather long chain of reflections on verification will have brought you some new insights into the fascinating challenge of writing correct programs.

References

[1] Nadia Polikarpova, Julian Tschannen, and Carlo A. Furia: A Fully Verified Container Library, in Proceedings of 20th International Symposium on Formal Methods (FM 15), 2015. (Best paper award.)

[2] Martin Nordio, Cristiano Calcagno, Peter Müller and Bertrand Meyer: A Sound and Complete Program Logic for Eiffel, in Proceedings of TOOLS 2009 (Technology of Object-Oriented Languages and Systems), Zurich, June-July 2009, eds. M. Oriol and B. Meyer, Springer LNBIP 33, June 2009.

[3] Boogie page at MSR, see here for publications and other information.

[4] Z3 was also originally from MSR and has been open-sourced, one can get access to publications and other information from  its Wikipedia page.

[5] Carlo Furia, Bertrand Meyer and Sergey Velder: Loop invariants: Analysis, Classification and Examples, in ACM Computing Surveys, vol. 46, no. 3, February 2014. Available here.

[6] Dynamic programming is a form of recursion removal, turning a recursive algorithm into an iterative one by using techniques known as “memoization” and  “bottom-up computation” (Berry). In this transformation, the invariant plays a key role. I will try to write this up some day as it is a truly elegant and illuminating explanation.

The page for the 2020 LASER summer school (31 May to 7 June) now has the basic elements (some additions still forthcoming) and registration at the early price is open. The topic is DevOps, Microservices and Software Development for the Age of the Web with both conceptual lectures and contributions from industry, by technology leaders from Amazon, Facebook and ServiceNow. The confirmed speakers are:

• Fabio Casati, ServiceNow and University of Trento, and Kannan Govindarajan from ServiceNow on Taking AI from research to production – at scale.
• Adrian Cockcroft, Amazon Web Services, on Building and Operating Modern Applications.
• Elisabetta Di Nitto, Politecnico di Milano.
• Valérie Issarny, INRIA, on The Web for the age of the IoT.
• Erik Meijer, Facebook, on Software Development At Scale.
• Me, on Software from beginning to end: a comprehensive method.

As always, the setup is the incomparable environment of the Hotel del Golfo in Procchio, Elba Island off the coast of Tuscany, ideal at that time of year (normally good weather, warm but not hot, few tourists). The school is intensive but there is time to enjoy the beach, the hotel’s amenities and the wonderful of environment of Elba (wake up your inner Napoleon). The school has a fairly small size and everyone lives under the same (beautiful) roof, so there is plenty of time for interaction with the speakers and other participants.

About these participants: the school is intended for engineers and managers in industry as well as researchers and PhD student. In fact it’s a mix that one doesn’t find that often, allowing for much cross-learning.

Another way to put it is that this is now the 16th edition of the school (it started in 2004 but we skipped one year), so it cannot be doing everything wrong.

Some important concepts of software engineering, established over the years, are not widely known in the community. One use of this blog is to provide tutorials on such overlooked ideas. An earlier article covered one pertaining to project management: the Shortest Possible Schedule property . Here is another, this time in the area of requirements engineering, also based on a publication that I consider to be a classic (it is over 40 years old) but almost unknown to practitioners.

Practitioners are indeed, as in most of my articles, the intended audience. I emphasize this point right at the start because if you glance at the rest of the text you will see that it contains (horror of horrors) some mathematical formulae, and might think “this is not for me”. It is! The mathematics is very simple and my aim is practical: to shed light on an eternal question that faces anyone writing requirements (whatever the style, traditional or agile): how can I be sure that a requirements specification is complete?

To a certain extent you cannot. But there is better answer, a remarkably simple one which, while partial, helps.

Defining completeness

The better answer is called “sufficient completeness” and comes from the theory of abstract data types. It was introduced in a 1978 article by Guttag and Horning [1]. It is also implicit in a more down-to-earth document, the 1998 IEEE standard on how to write requirements [2].

There is nothing really new in the present article; in fact my book Object-Oriented Software Construction [3] contains an extensive discussion of sufficient completeness (meant to be more broadly accessible than Guttag and Horning’s scholarly article). But few people know the concepts; in particular very few practitioners have heard of sufficient completeness (if they have heard at all of abstract data types). So I hope the present introduction will be useful.

The reason the question of determining completeness of requirements seems hopeless at first is the natural reaction: complete with respect to what? To know that the specification is complete we would need a more general description of all that our stakeholders want and all the environment constraints, but this would only push the problem further: how do we know that such description itself is complete?

That objection is correct in principle: we can never be sure that we did not forget something someone wanted, or some property that the environment imposes. But there also exist more concrete and assessable notions of completeness.

The IEEE standard gives three criteria of completeness. The first states that “all requirements” have been included, and is useless, since it  runs into the logical paradox mentioned above, and is tautological anyway (the requirements are complete if they include all requirements, thank you for the information!). The second is meaningful but of limited interest (a “bureaucratic” notion of completeness): every element in the requirements document is numbered, every cross-reference is defined and so on. The last criterion is the interesting one: “Definition of the responses of the software to all realizable classes of input data in all realizable classes of situations”. Now this is meaningful. To understand this clause we need to step back to sufficient completeness and, even before that, to abstract data types.

Abstract data types will provide our little mathematical excursion (our formal picnic in the words of an earlier article) in our study of requirements and completeness. If you are not familiar with this simple mathematical theory, which every software practitioner should know, I hope you will benefit from the introduction and example. They will enable us to introduce the notion of sufficient completeness formally before we come back to its application to requirements engineering.

Specifying an abstract data type

 Abstract data types are the mathematical basis for object-oriented programming. In fact, OO programming but also OO analysis and OO design are just a realization of this mathematical concept at various levels of abstraction, even if few OO practitioners are aware of it. (Renewed reference to [3] here if you want to know more.)

An ADT (abstract data type) is a set of objects characterized not by their internal properties (what they are) but by the operations applicable to them (what they have), and the properties of these operations. If you are familiar with OO programming you will recognize that this is exactly, at the implementation level, what a class is. But here we are talking about mathematical objects and we do not need to consider implementation.

An example  of a type defined in this way, as an ADT, is a notion of POINT on a line. We do not say how this object is represented (a concept that is irrelevant at the specification level) but how it appears to the rest of the world: we can create a new point at the origin, ask for the coordinate of a point, or move the point by a certain displacement. The example is the simplest meaningful one possible, but it gives the ideas.

An ADT specification has three part: Functions, Preconditions and Axioms. Let us see them (skipping Preconditions for the moment) for the definition of the POINT abstract data type.

The functions are the operations that characterize the type. There are three kinds of function, defined by where the ADT under definition, here POINT, appears:

• Creators, where the type appears only among the results.
• Queries, where it appears only among the arguments.
• Commands, where it appears on both sides.

There is only one creator here:

new: → POINT

new is a function that takes no argument, and yields a point (the origin). We will write the result as just new (rather than using empty parentheses as in new ()).

Creators correspond in OO programming to constructors of a class (creation procedures in Eiffel). Like constructors, creators may have arguments: for example instead of always creating a point at the origin we could decide that new creates a point with a given coordinate, specifying it as INTEGER → POINT and using it as new (i) for some integer i (our points will have integer coordinates). Here for simplicity we choose a creator without arguments. In any case the new type, here POINT, appears only on the side of the results.

Every useful ADT specification needs at least one creator, without which we would never obtain any objects of the type (here any points) to work with.

There is also only one query:

x: POINT → INTEGER

 which gives us the position of a point, written x (p) for a point p. More generally, a query enables us to obtain properties of objects of the new type. These properties must be expressed in terms of types that we have already defined, like INTEGER here. Again there has to be at least one query, otherwise we could never obtain usable information (information expressed in terms of what we already know) about objects of the new type. In OO programming, queries correspond to fields (attributes) of a class and functions without side effects.

And we also have just one command:

move: POINT × INTEGER → POINT

a function that for any point p and integer i and yields a new point, move (p, i).  Again an ADT specification is not interesting unless it has at least one command, representing ways to modify objects. (In mathematics we do not actually modify objects, we get new objects. In imperative programming we will actually update existing objects.) In the classes of object-oriented programming, commands correspond to procedures (methods which may change objects).

You see the idea: define the notion of POINT through the applicable operations.

Listing their names and the types of their arguments types results (as in POINT × INTEGER → POINT) is not quite enough to specify these operations: we must specify their fundamental properties, without of course resorting to a programming implementation. That is the role of the second component of an ADT specification, the axioms.

For example I wrote above that new yields the origin, the point for which x = 0,  but you only had my word for it. My word is good but not good enough. An axiom will give you this property unambiguously:

x (new) = 0                                    — A0

The second axiom, which is also the last, tells us what move actually does. It applies to any point p and any integer m:

x (move (p, m)) = x (p) + m       — A1

In words: the coordinate of the point resulting from moving p by m is the coordinate of p plus m.

That’s it! (Except for the notion of precondition, which will wait a bit.) The example is trivial but this approach can be applied to any number of  data types, with any number of applicable operations and any level of complexity. That is what we do, at the design and implementation level, when writing classes in OO programming.

Is my ADT sufficiently complete?

Sufficient completeness is a property that we can assess on such specifications. An ADT specification for a type T (here POINT) is sufficiently complete if the axioms are powerful enough to yield the value of any well-formed query expression in a form not involving T. This definition contains a few new terms but the concepts are very simple; I will explain what it means through an example.

With an ADT specification we can form all kinds of expressions, representing arbitrarily complex specifications. For example:

x (move (move (move (new, 3), x (move (move (new, -2), 4))), -6))

This expression will yield an integer (since function x has INTEGER as its result type) describing the result of a computation with points. We can visualize this computation graphically; note that it involves creating two points (since there are two occurrences of new) and moving them, using in one case the current coordinate of one of them as displacement for the other. The following figure illustrates the process.

The result, obtained informally by drawing this picture, is the x of P5, that is to say -1. We will derive it mathematically below.

Alternatively, if like most programmers (and many other people) you find it more intuitive to reason operationally than mathematically, you may think of the previous expression as describing the result of the following OO program (with variables of type POINT):

create p                                — In C++/Java syntax: p = new POINT();
create q
p.move (3)
q.move (-2)
q.move (4)
p.move (q.x)
p.move (-6)

Result := p.x

You can run this program in your favorite OO programming language, using a class POINT with new, x and move, and print the value of Result, which will be -1.

Here, however, we will stay at the mathematical level and simplify the expression using the axioms of the ADT, the same way we would compute any other mathematical formula, applying the rules without needing to rely on intuition or operational reasoning. Here is the expression again (let’s call it i, of type INTEGER):

i = x (move (move (move (new, 3), x (move (move (new, -2), 4))), -6))

A query expression is one in which the outermost function being applied, here x, is a query function. Remember that a query function is one which the new type, here POINT, appears only on the left. This is the case with x, so the above expression i is indeed a query expression.

For sufficient completeness, query expressions are the ones of interest because their value is expressed in terms of things we already know, like INTEGERs, so they are the only way we can concretely obtain directly usable information the ADT (to de-abstract it, so to speak).

But we can only get such a value by applying the axioms. So the axioms are “sufficiently complete” if they always give us the answer: the value of any such query expression.

 Let us see if the above expression i satisfies this condition of sufficient completeness. To make it more tractable let us write  it in terms of simpler expressions (all of type POINT), as illustrated by the figure below:

p1 = move (new, 3)
p2= move (new, -2)
p3= move (p2, 4)
p4= move (p1, x (p3))
p5= move (p4, -6)
i = x (p5)

(You may note that the intermediate expressions roughly correspond to the steps in the above interpretation of the computation as a program. They also appear in the illustrative figure repeated below.)

Now we start applying the axioms to evaluating the expressions. Remember that we have two axioms: A0 tells us that x (new) = 0 and A1 that x (move (p, m)) = x (p) + m. Applying A1 to the definition the expression i yields

i = x (p4) – 6
= i4 – 6

if we define

i4 = x (p4)      — Of type INTEGER

We just have to compute i4. Applying A1 to the definion of p4 tells us that

i4 = x (p1) + x (p3)

To compute the two terms:

• Applying A1 again, we see that the first term x (p1) is x (new) + 3, but then A0 tells us that x (new) is zero, so x (p1) is 3.
• As to x (p3), it is, once more from A1, x (p2) + 4, and x (p2) is (from A1 then A0), just -2, so x (p3) is 2.

In the end, then, i4 is 5, and the value of the entire expression i = i4 – 6 is -1. Good job!

Proving sufficient completeness

The successful computation of i was just a derivation for one example, showing that in that particular case the axioms yield the answer in terms of an INTEGER. How do we go from one example to an entire specification?

The bad news first: like all interesting problems in programming, sufficient completeness of an ADT specification is theoretically undecidable. There is no general automatic procedure that will process an ADT specification and print out ““sufficiently complete” or “not sufficiently complete”.

Now that you have recovered from the shock, you can share the computer scientist’s natural reaction to such an announcement: so what. (In fact we might define the very notion of computer scientist as someone who, even before he brushes his teeth in the morning — if he brushes them at all — has already built the outline of a practical solution to an undecidable problem.) It is enough that we can find a way to determine if a given specification is sufficiently complete. Such a proof is, in fact, the computer scientist’s version of dental hygiene: no ADT is ready for prime time unless it is sufficiently complete.

The proof is usually not too hard and will follow the general style illustrated for our simple example.

We note that the definition of sufficient completeness said: “the axioms are powerful enough to yield the value of any well-formed query expression in a form not involving the type”. I have not defined “well-formed” yet. It simply means that the expressions are properly structured, with the proper syntax (basically the correct matching of parentheses) and proper number and types of arguments. For example the following are not well-formed (if p is an expression of type POINT):

move (p, 55(     — Bad use of parentheses.
move (p)            — Wrong number of arguments.
move (p, p)       — Wrong type: second argument should be an integer.

Such expressions are nonsense, so we only care about well-formed expressions. Note that in addition to new, x and move , an expression can use integer constants as in the example (although we could generalize to arbitrary integer expressions). We consider an integer constant as a query expression.

We have to prove that with the two axioms A0 and A1 we can determine the value of any query expression i. Note that since the only query functions is x, the only possible form for i, other than an integer constant, is x (p) for some expression p of type POINT.

The proof proceeds by induction on the number n of parenthesis pairs in a query expression i.

There are two base steps:

• n = 0: in that case i can only be an integer constant. (The only expression with no parentheses built out of the ADT’s functions is new, and it is not a query expression.) So the value is known. In all other cases i will be of the form x (p) as noted.
• n = 1: in that case p  can only be new, in other words i = x (new), since the only function that yields points, other than new, is move, and any use of it would add parentheses. In this case axiom A0 gives us the value of i: zero.

For the induction step, we consider i with n + 1 parenthesis pairs for n > 1. As noted, i is of the form x (p), so p has exactly n parenthesis pairs. p cannot be new (which would give 0 parenthesis pairs and was taken care of in the second base step), so p has to be of the form

p =  move (p’, i’)    — For expressions p’ of type POINT and i’ of type INTEGER.

implying (since i = x (p)) that by axiom A1, the value of i is

x (p’) + i’

So we will be able to determine the value of i if we can determine the value of both x (p’) and i’. Since p has n parenthesis pairs and p =  move (p’, i’), both p’ and i’ have at most n – 1 parenthesis pairs. (This use of n – 1 is legitimate because we have two base steps, enabling us to assume n > 1.) As a consequence, both x (p’) and i’ have at most n parenthesis pairs, enabling us to deduce their values, and hence the value of i, by the induction hypothesis.

Most proofs of sufficient completeness in my experience follow this style: induction on the number of parenthesis pairs (or the maximum nesting level).

Preconditions

I left until now the third component of a general ADT specification: preconditions. The need for preconditions arises because most practical specifications need some of their functions to be partial. A partial function from X to Y is a function that may not yield a value for some elements of X. For example, the inverse function on real numbers, which yields 1 / a for x, is partial  since it is not defined for a = 0 (or, on a computer, for non-zero but very small a).

Assume that in our examples we only want to accept points that lie in the interval [-4, +4]:

 We can simply model this property by turning move into a partial function. It was specified above as

move: POINT × INTEGER → POINT

The ordinary arrow → introduces a total (always defined) function. For a partial function we will use a crossed arrow ⇸, specifying the function as

move: POINT × INTEGER ⇸ POINT

Other functions remain unchanged. Partial functions cause trouble: for f in X ⇸ Y we can no longer cheerfully use f (x) if f is a partial function, even for x of the appropriate type X. We have to make sure that x belongs to the domain of f, meaning the set of values for which f is defined. There is no way around it: if you want your specification to be meaningful and it uses partial functions, you must specify explicitly the domain of each of them. Here is how to do it, in the case of move:

move (p: POINT; d: INTEGER) require |x (p) + d | < 5    — where |…| is absolute value

To adapt the definition (and proofs) of sufficient completeness to the possible presence of partial functions:

• We only need to consider (for the rule that axioms must yield the value of query expressions) well-formed expressions that satisfy the associated preconditions.
• The definition must, however, include the property that axioms always enable us to determine whether an expression satisfies the associated preconditions (normally a straightforward part of the proof since preconditions are themselves query expressions).

Updating the preceding proof accordingly is not hard.

Back to requirements

The definition of sufficient completeness is of great help to assess the completeness of a requirements document. We must first regretfully note that for many teams today requirements stop at  “use cases” (scenarios) or  “user stories”. Of course these are not requirements; they only describe individual cases and are to requirements what tests are to programs. They can serve to check requirements, but do not suffice as requirements. I am assuming real requirements, which include descriptions of behavior (along with other elements such as environment properties and project properties). To describe behaviors, you will define operations and their effects. Now we know what the old IEEE standard is telling us by stating that complete requirements should include

definition of the responses of the software to all realizable classes of input data in all realizable classes of situations

Whether or not we have taken the trouble to specify the ADTs, they are there in the background; our system’s operations reflect the commands, and the effects we can observe reflect the queries. To make our specification complete, we should draw as much as possible of the (mental or explicit) matrix of possible effects of all commands on all queries. “As much as possible” because software engineering is engineering and we will seldom be able to reach perfection. But the degree of fullness of the matrix tells us a lot (possible software metric here?) about how close our requirements are to completeness.

I should note that there are other aspects to completeness of requirements. For example the work of Michael Jackson, Pamela Zave and Axel van Lamsweerde (more in some later article, with full references) distinguishes between business goals, environment constraints and system properties, leading to a notion of completeness as how much the system properties meet the goals and obey the constraints [4]. Sufficient completeness operates at the system level and, together with its theoretical basis, is one of those seminal concepts that every practicing software engineer or project manager should master.

References and notes

[1] John V. Guttag, Jim J. Horning: The Algebraic Specification of Abstract Data Types, in Acta Informatica, vol. 10, no. 1, pages 27-52, 1978, available here from the Springer site. This is a classic paper but I note that few people know it today; in Google Scholar I see over 700 citations but less than 100 of them in the past 8 years.

[2]  IEEE: Recommended Practice for Software Requirements Specifications, IEEE Standard 830-1998, 1998. This standard is supposed to be obsolete and replaced by newer ones, more detailed and verbose, but it remains the better reference: plain, modest and widely applied by the industry. It does need an update, but a good one.

[3] Bertrand Meyer, Object-Oriented Software Construction, 2nd edition, Prentice Hall, 1997. The discussion of sufficient completeness was in fact already there in the first edition from 1988.

[4] With thanks to Elisabetta Di Nitto from Politecnico di Milano for bringing up this notion of requirements completeness.

A version of this article was first published on the Communications of the ACM blog.

Over breakfast at your hotel you read an article berating banks about the fraudulent credit card transactions they let through. You proceed to check out and bang! Your credit card is rejected because (as you find out later) the bank thought [1] it couldn’t possibly be you in that exotic place. Ah, those banks! They accept too much. Ah, those banks! They reject too much. Finding the right balance is a case of soundness versus precision.

Similar notions are essential to the design of tools for program analysis, looking for such suspicious cases as  dead code (program parts that will never be executed). An analysis can be sound, or not; it can be complete, or not.

These widely used concepts are sometimes misunderstood.  The first answer I get when innocently asking people whether the concepts are clear is yes, of course, everyone knows! Then, as I bring up such examples as credit card rejection or dead code detection, assurance quickly yields to confusion. One sign that things are not going well is when people start throwing in terms like “true positive” and “false negative”. By then any prospect of reaching a clear conclusion has vanished. I hope that after reading this article you will never again (in a program analysis context) be tempted to use them.

Now the basic idea is simple. An analysis is sound if it reports all errors, and complete if it only reports errors. If not complete, it is all the more precise that it reports fewer non-errors.

You can stop here and not be too far off [2]. But a more nuanced and precise discussion helps.

1. A relative notion

As an example of common confusion, one often encounters attempts to help through something like Figure 1, which cannot be right since it implies that all sound methods are complete. (We’ll have better pictures below.)

Figure 1: Naïve (and wrong) illustration

Perhaps this example can be dismissed as just a bad use of illustrations [3] but consider the example of looking for dead code. If the analysis wrongly determines that some reachable code is unreachable, is it unsound or incomplete?

With this statement of the question, the only answer is: it depends!

It depends on the analyzer’s mandate:

• If it is a code checker that alerts programmers to cases of bad programming style, it is incomplete: it reports as an error a case that is not. (Reporting that unreachable code is reachable would cause unsoundness, by missing a case that it should have reported.)
• If it is the dead-code-removal algorithm of an optimizing compiler, which will remove unreachable code, it is unsound: the compiler will remove code that it should not. (Reporting that unreachable code is reachable would cause incompleteness, by depriving the compiler of an optimization.)

As another example, consider an analyzer that finds out whether a program will terminate. (If you are thinking “but that can’t be done!“, see the section “Appendix: about termination” at the very end of this article.) If it says a program does not terminates when in fact it does, is it unsound or incomplete?

Again, that depends on what the analyzer seeks to establish. If it is about the correctness of a plain input-to-output program (a program that produces results and then is done), we get incompleteness: the analyzer wrongly flags a program that is actually OK. But if it is about verifying that continuously running programs, such as the control system for a factory, will not stop (“liveness”), then the analyzer is unsound.

Examples are not limited to program analysis. A fraud-indentification process that occasionally rejects a legitimate credit card purchase is, from the viewpoint of preserving the bank from fraudulent purchases, incomplete. From the viewpoint of the customer who understands a credit card as an instrument enabling payments as long as you have sufficient credit, it is unsound.

These examples suffice to show that there cannot be absolute definitions of soundness and precision: the determination depends on which version of a boolean property we consider desirable. This decision is human and subjective. Dead code is desirable for the optimizing compiler and undesirable (we will say it is a violation) for the style checker. Termination is desirable for input-output programs and a violation for continuously running programs.

Once we have decided which cases are desirable and which are violations, we can define the concepts without any ambiguity: soundness means rejecting all violations, and completeness means accepting all desirables.

While this definition is in line with the unpretentious, informal one in the introduction, it makes two critical aspects explicit:

• Relativity. Everything depends on an explicit decision of what is desirable and what is a violation. Do you want customers always to be able to use their credit cards for legitimate purchases, or do you want to detect all frauds attempts?
• Duality. If you reverse the definitions of desirable and violation (they are the negation of each other), you automatically reverse the concepts of soundness and completeness and the associated properties.

We will now explore the consequences of these observations.

2. Theory and practice

For all sufficiently interesting problems, theoretical limits (known as Rice’s theorem) ensure that it is impossible to obtain both soundness and completeness.

But it is not good enough to say “we must be ready to renounce either soundness or completeness”. After all, it is very easy to obtain soundness if we forsake completeness: reject every case. A termination-enforcement analyzer can reject every program as potentially non-terminating. A bank that is concerned with fraud can reject every transaction (this seems to be my bank’s approach when I am traveling) as potentially fraudulent. Dually, it is easy to ensure completeness if we just sacrifice soundness: accept every case.

These extreme theoretical solutions are useless in practice; here we need to temper the theory with considerations of an engineering nature.

The practical situation is not as symmetric as the concept of duality theoretically suggests. If we have to sacrifice one of the two goals, it is generally better to accept some incompleteness: getting false alarms (spurious reports about cases that turn out to be harmless) is less damaging than missing errors. Soundness, in other words, is essential.

Even on the soundness side, though, practice tempers principle. We have to take into account the engineering reality of how tools get produced. Take a program analyzer. In principle it should cover the entire programming language. In practice, it will be built step by step: initially, it may not handle advanced features such as exceptions, or dynamic mechanisms such as reflection (a particularly hard nut to crack). So we may have to trade soundness for what has been called  “soundiness” [4], meaning soundness outside of cases that the technology cannot handle yet.

If practical considerations lead us to more tolerance on the soundness side, on the completeness side they drag us (duality strikes again) in the opposite direction. Authors of analysis tools have much less flexibility than the theory would suggest. Actually, close to none. In principle, as noted, false alarms do not cause catastrophes, as missed violations do; but in practice they can be almost as bad.  Anyone who has ever worked on or with a static analyzer, going back to the venerable Lint analyzer for C, knows the golden rule: false alarms kill an analyzer. When people discover the tool and run it for the first time, they are thrilled to discover how it spots some harmful pattern in their program. What counts is what happens in subsequent runs. If the useful gems among the analyzer’s diagnostics are lost in a flood of irrelevant warnings, forget about the tool. People just do not have the patience to sift through the results. In practice any analysis tool has to be darn close to completeness if it has to stand any chance of adoption.

Completeness, the absence of false alarms, is an all-or-nothing property. Since in the general case we cannot achieve it if we also want soundness, the engineering approach suggests using a numerical rather than boolean criterion: precision. We may define the precision pr as 1 – im where im is the imprecision:  the proportion of false alarms.

The theory of classification defines precision differently: as pr = tp / (tp + fp), where tp is the number of false positives and fp the number of true positives. (Then im would be fp / (tp + fp).) We will come back to this definition, which requires some tuning for program analyzers.

From classification theory also comes the notion of recall: tp / (tp + fn) where fn is the number of false negatives. In the kind of application that we are looking at, recall corresponds to soundness, taken not as a boolean property (“is my program sound?“) but  a quantitative one (“how sound is my program?“). The degree of unsoundness un would then be fn / (tp + fn).

3. Rigorous definitions

With the benefit of the preceding definitions, we can illustrate the concepts, correctly this time. Figure 2 shows two different divisions of the set of U of call cases (universe):

• Some cases are desirable (D) and others are violations (V).
• We would like to know which are which, but we have no way of finding out the exact answer, so instead we run an analysis which passes some cases (P) and rejects some others (R).

Figure 2: All cases, classified

The first classification, left versus right columns in Figure 2, is how things are (the reality). The second classification, top versus bottom rows, is how we try to assess them. Then we get four possible categories:

• In two categories, marked in green, assessment hits reality on the nail:  accepted desirables (A), rightly passed, and caught violations (C), rightly rejected.
• In the other two, marked in red, the assessment is off the mark: missed violations (M), wrongly passed; and false alarms (F), wrongly accepted.

The following properties hold, where U (Universe) is the set of all cases and  ⊕ is disjoint union [5]:

— Properties applicable to all cases:
U = D ⊕ V
U = P ⊕ R
D = A ⊕ F
V = C ⊕ M
P = A ⊕ M
R = C ⊕ F
U = A ⊕M ⊕ F ⊕ C

We also see how to define the precision pr: as the proportion of actual violations to reported violations, that is, the size of C relative to R. With the convention that u is the size of U and so on, then  pr = c / r, that is to say:

• pr = c / (c + f)      — Precision
• im = f / (c + f)      — Imprecision

We can similarly define soundness in its quantitative variant (recall):

• so = a / (a + m)      — Soundness (quantitative)
• un = m / (a + m)   — Unsoundness

These properties reflect the full duality of soundness and completeness. If we reverse our (subjective) criterion of what makes a case desirable or a violation, everything else gets swapped too, as follows:

Figure 3: Duality

We will say that properties paired this way “dual” each other [6].

It is just as important (perhaps as a symptom that things are not as obvious as sometimes assumed) to note which properties do not dual. The most important examples are the concepts of  “true” and “false” as used in “true positive” etc. These expressions are all the more confusing that the concepts of True and False do dual each other in the standard duality of Boolean algebra (where True duals False,  Or duals And, and an expression duals its negation). In “true positive” or “false negative”,  “true” and “false” do not mean True and False: they mean cases in which (see figure 2 again) the assessment respectively matches or does not match the reality. Under duality we reverse the criteria in both the reality and the assessment; but matching remains matching! The green areas remain green and the red areas remain red.

The dual of positive is negative, but the dual of true is true and the dual of false is false (in the sense in which those terms are used here: matching or not). So the dual of true positive is true negative, not false negative, and so on. Hereby lies the source of the endless confusions.

The terminology of this article removes these confusions. Desirable duals violation, passed duals rejected, the green areas dual each other and the red areas dual each other.

4. Sound and complete analyses

If we define an ideal world as one in which assessment matches reality [7], then figure 2 would simplify to just two possibilities, the green areas:

Figure 4: Perfect analysis (sound and complete)

This scheme has the following properties:

— Properties of a perfect (sound and complete) analysis as in Figure 4:
M = ∅              — No missed violations
F = ∅               — No false alarms
P = D                — Identify  desirables exactly
R = V                –Identify violations exactly

As we have seen, however, the perfect analysis is usually impossible. We can choose to build a sound solution, potentially incomplete:

Figure 5: Sound desirability analysis, not complete

In this case:

— Properties of a sound analysis (not necessarily complete) as in Figure 5:
M = ∅              — No missed violations
P = A                — Accept only desirables
V = C                — Catch all violations
P ⊆ D               — Under-approximate desirables
R ⊇ V               — Over-approximate violations

Note the last two properties. In the perfect solution, the properties P = D and R = V mean that the assessment, yielding P and V, exactly matches the reality, D and V. From now on we settle for assessments that approximate the sets of interest: under-approximations, where the assessment is guaranteed to compute no more than the reality, and over-approximations, where it computes no less. In all cases the assessed sets are either subsets or supersets of their counterparts. (Non-strict, i.e. ⊆ and ⊇ rather than ⊂ and ⊃; “approximation” means possible approximation. We may on occasion be lucky and capture reality exactly.)

We can go dual and reach for completeness at the price of possible unsoundness:

Figure 6: Complete desirability analysis, not sound

The properties are dualled too:

— Properties of a complete analysis (not necessarily sound), as in Figure 6:
F = ∅              — No false alarms
R = C               — Reject only violations
D = A               — Accept all desirables
P ⊇ D               — Over-approximate desirables
R ⊆ V              — Under-approximate violations

5. Desirability analysis versus violation analysis

We saw above why the terms “true positives”, “false negatives” etc., which do not cause any qualms in classification theory, are deceptive when applied to the kind of pass/fail analysis (desirables versus violations) of interest here. The definition of precision provides further evidence of the damage. Figure 7 takes us back to the general case of Figure 2 (for analysis that is guaranteed neither sound nor complete)  but adds these terms to the respective categories.

Figure 7: Desirability analysis (same as fig. 2 with added labeling)

The analyzer checks for a certain desirable property, so if it wrongly reports a violation (F) that is a false negative, and if it misses a violation (M) it is a false positive. In the  definition from classification theory (section 2, with abbreviations standing for True/False Positives/Negatives): TP = A, FP = M, FN =  F, TN = C, and similarly for the set sizes: tp = a, fp = m, fn = f, tn = c.

The definition of precision from classification theory was pr = tp / (tp + fp), which here gives a / (a + m). This cannot be right! Precision has to do with how close the analysis is to completeness, that is to day, catching all violations.

Is classification theory wrong? Of course not. It is simply that, just as Alice stepped on the wrong side of the mirror, we stepped on the wrong side of duality. Figures 2 and 7 describe desirability analysis: checking that a tool does something good. We assess non-fraud from the bank’s viewpoint, not the stranded customer’s; termination of input-to-output programs, not continuously running ones; code reachability for a static checker, not an optimizing compiler. Then, as seen in section 3, a / (a + m) describes not precision but  soundness (in its quantitative interpretation, the parameter called “so” above).

To restore the link with classification theory , we simply have to go dual and take the viewpoint of violation analysis. If we are looking for possible violations, the picture looks like this:

Figure 8: Violation analysis (same as fig. 7 with different positive/negative labeling)

Then everything falls into place:  tp = c, fp = f, fn =  m, tn = a, and the classical definition of  precision as pr = tp / (tp + fp) yields c / (c + f) as we are entitled to expect.

In truth there should have been no confusion since we always have the same picture, going back to Figure 2, which accurately covers all cases and supports both interpretations: desirability analysis and violation analysis. The confusion, as noted, comes from using the duality-resistant “true”/”false” opposition.

To avoid such needless confusion, we should use the four categories of the present discussion:  accepted desirables, false alarms, caught violations and missed violations [8]. Figure 2 and its variants clearly show the duality, given explicitly in Figure 3, and sustains  interpretations both for desirability analysis and for violation analysis. Soundness and completeness are simply special cases of the general framework, obtained by ruling out one of the cases of incorrect analysis in each of Figures 4 and 5. The set-theoretical properties listed after Figure 2 express the key concepts and remain applicable in all variants. Precision c / (c + f) and quantitative soundness a / (a + m) have unambiguous definitions matching intuition.

The discussion is, I hope, sound. I have tried to make it complete. Well, at least it is precise.

Notes and references

[1] Actually it’s not your bank that “thinks” so but its wonderful new “Artificial Intelligence” program. ⤣

[2] For a discussion of these concepts as used in testing see Mauro Pezzè and Michal Young, Software Testing and Analysis: Process, Principles and Techniques, Wiley, 2008. ⤣

[3] Edward E. Tufte: The Visual Display of Quantitative Information, 2nd edition, Graphics Press, 2001.⤣

[4] Michael Hicks,What is soundness (in static analysis)?, blog article available here, October 2017. ⤣

[5] The disjoint union property X = Y ⊕ Z means that Y ∩ Z = ∅ (Y and Z are disjoint) and X = Y ∪ Z (together, they yield X). ⤣

[6] I thought this article would mark the introduction into the English language of “dual” as a verb, but no, it already exists in the sense of turning a road from one-lane to two-lane (dual). ⤣

[7] As immortalized in a toast from the cult movie The Prisoner of the Caucasus: “My great-grandfather says: I have the desire to buy a house, but I do not have the possibility. I have the possibility to buy a goat, but I do not have the desire. So let us drink to the matching of our desires with our possibilities.” See 6:52 in the version with English subtitles. ⤣

[8] To be fully consistent we should replace the term “false alarm” by rejected desirable. I is have retained it because it is so well established and, with the rest of the terminology as presented, does not cause confusion. ⤣

[9] Byron Cook, Andreas Podelski, Andrey Rybalchenko: Proving Program Termination, in Communications of the ACM, May 2011, Vol. 54 No. 5, Pages 88-98. ⤣

Background and acknowledgments

This reflection arose from ongoing work on static analysis of OO structures, when I needed to write formal proofs of soundness and completeness and found that the definitions of these concepts are more subtle than commonly assumed. I almost renounced writing the present article when I saw Michael Hicks’s contribution [4]; it is illuminating, but I felt there was still something to add. For example, Hicks’s set-based illustration is correct but still in my opinion too complex; I believe that the simple 2 x 2 pictures used above convey the ideas  more clearly. On substance, his presentation and others that I have seen do not explicitly mention duality, which in my view is the key concept at work here.

I am grateful to Carlo Ghezzi for enlightening discussions, and benefited from comments by Alexandr Naumchev and others from the Software Engineering Laboratory at Innopolis University.

Appendix: about termination

With apologies to readers who have known all of the following from kindergarten: a statement such as (section 1): “consider an analyzer that finds out whether a program will terminate” can elicit no particular reaction (the enviable bliss of ignorance) or the shocked rejoinder that such an analyzer is impossible because termination (the “halting” problem) is undecidable. This reaction is just as incorrect as the first. The undecidability result for the halting problem says that it is impossible to write a general termination analyzer that will always provide the right answer, in the sense of both soundness and completeness, for any program in a realistic programming language. But that does not preclude writing termination analyzers that answer the question correctly, in finite time, for given programs. After all it is not hard to write an analyzer that will tell us that the program from do_nothing until True loop do_nothing end will terminate and that the program from do_nothing until False loop do_nothing end will not terminate. In the practice of software verification today, analyzers can give such sound answers for very large classes of programs, particularly with some help from programmers who can obligingly provide variants (loop variants, recursion variants). For a look into the state of the art on termination, see the beautiful survey by Cook, Podelski and Rybalchenko [9].

Also appears in the Communications of the ACM blog

(Originally published on CACM blog.)

Most of the world programs in a very strange way. Strange to me. I usually hear the reverse question: people ask us, the Eiffel community, to explain why we program our way. I hardly understand the question, because the only mystery is how anyone can even program in any other way.

The natural reference is the beginning of One Flew Over the Cuckoo’s Nest: when entering an insane asylum and wondering who is an inmate and who a doctor, you may feel at a loss for objective criteria. Maybe the rest of the world is right and we are the nut cases. Common sense suggests it.

But sometimes one can go beyond common sense and examine the evidence. So lend me an ear while I explain my latest class invariant. Here it is, in Figure 1. (Wait, do not just run away yet.)

Figure 1: From the invariant of class MULTIGRAPH

This is a program in progress and by the time you read this note the invariant and enclosing class will have changed. But the ideas will remain.

Context: multigraphs

The class is called MULTIGRAPH and describes a generalized notion of graph, illustrated in Figure 2. The differences are that: there can be more than one edge between two nodes, as long as they have different tags (like the spouse and boss edges between 1 and 2); and there can be more than one edge coming out of a given node and with a given tag (such as the two boss edges out of 1, reflecting that 1’s boss might be 2 in some cases and 3 in others). Some of the nodes, just 1 here, are “roots”.

The class implements the notion of multigraph and provides a wide range of operations on multigraphs.

Figure 2: A multigraph

Data structures

Now we turn to the programming and software engineering aspects. I am playing with various ways of accessing multigraphs. For the basic representation of a multigraph, I have chosen a table of triples:

                triples_table: HASH_TABLE [TRIPLE, TUPLE [source: INTEGER; tag: INTEGER; target: INTEGER]]  — Table of triples, each retrievable through its `source’, `tag’ and `target’.

where the class TRIPLE describes [source, tag, target] triples, with a few other properties, so they are not just tuples. It is convenient to use a hash table, where the key is such a 3-tuple. (In an earlier version I used just an ARRAY [TRIPLE], but a hash table proved more flexible.)

Sources and targets are nodes, also called “objects”; we represent both objects and tags by integers for efficiency. It is easy to have structures that map symbolic tag names such as “boss” to integers.

triples_table is the core data structure but it turns out that for the many needed operations it is convenient to have others. This technique is standard: for efficiency, provide different structures to access and manipulate the same underlying information, with some redundancy. So I also have:

 triples_from:  ARRAYED_LIST [LIST [TRIPLE]]
               — Triples starting from a given object. Indexed by object numbers.

  triples_with:  HASH_TABLE [LIST [TRIPLE], INTEGER]
               — Triples labeled by a given tag. Key is tag number.

 triples_to:  ARRAYED_LIST [LIST [TRIPLE]]
               — Triples leading into a given object. Indexed by object numbers.

Figure 3 illustrates triples_from and Figures 4 illustrates triples_with. triples_to is similar.

Figure 3: The triples_from array of lists and the triples_table

Figure 4: The triples_with array of lists and the triples_table

It is also useful to access multigraphs through yet another structure, which gives us the targets associated with a given object and tag:

successors: ARRAY [HASH_TABLE [LIST [TRIPLE], INTEGER]]
               — successors [obj] [t] includes all o such that there is a t- reference from obj to o.

For example in Figure 1 successors [1] [spouse] is {2, 3}, and in Figures 3 and 4 successors [26] [t] is {22, 55, 57}. Of course we can obtain the “successors” information through the previously defined structures, but since this is a frequently needed operation I decided to include a specific data structure (implying that every operation modifying the multigraph must update it). I can change my mind later on and decide to make “successors” a function rather than a data structure; it is part of the beauty of OO programming, particularly in Eiffel, that such changes are smooth and hardly impact client classes.

There is similar redundancy in representing roots:

                roots:  LINKED_SET [INTEGER]
                              — Objects that are roots.

                is_root:  ARRAY [BOOLEAN]
                              — Which objects are roots? Indexed by object numbers.

If o is a root, then it appears in the “roots” set and is_root [o] has value True.

Getting things right

These are my data structures. Providing such a variety of access modes is a common programming technique. From a software engineering perspective ― specification, implementation, verification… ― it courts disaster. How do we maintain their consistency? It is very easy for a small mistake to slip into an operation modifying the graph, causing one of the data structures to be improperly updated, but in a subtle and rare enough way that it will not manifest itself during testing, coming back later to cause strange behavior that will be very hard to debug.

For example, one of the reasons I have a class TRIPLE and not just 3-tuples is that a triple is not exactly  the same as an edge in the multigraph. I have decided that by default the operation that removes and edge would not remove the corresponding triple from the data structure, but leave it in and mark it as “inoperative” (so class TRIPLE has an extra “is_inoperative” boolean field). There is an explicit GC-like mechanism to clean up deleted edges occasionally. This approach brings efficiency but makes the setup more delicate since we have to be extremely careful about what a triple means and what removal means.

This is where I stop understanding how the rest of the world can work at all. Without some rigorous tools I just do not see how one can get such things right. Well, sure, spend weeks of trying out test cases, printing out the structures, manually check everything (in the testing world this is known as writing lots of “oracles”), try at great pains to find out the reason for wrong results, guess what program change will fix the problem, and start again. Stop when things look OK. When, as Tony Hoare once wrote, there are no obvious errors left.

Setting aside the minuscule share of projects (typically in embedded life-critical systems) that use some kind of formal verification, this process is what everyone practices. One can only marvel that systems, including many successful ones, get produced at all. To take an analogy from another discipline, this does not compare to working like an electrical engineer. It amounts to working like an electrician.

For a short time I programmed like that too (one has to start somewhere, and programming methodology was not taught back then). I no longer could today. Continuing with the Hoare citation, the only acceptable situation is to stop when there are obviously no errors left.

How? Certainly not, in my case, by always being right the first time. I make mistakes like everyone else does. But I have the methodology and tools to avoid some, and, for those that do slip through, to spot and fix them quickly.

Help is available

First, the type system. Lots of inconsistencies, some small and some huge, which in an untyped language would only hit during execution, do not make it past compilation. We are not just talking here about using REAL instead of INTEGER. With a sophisticated type system involving multiple inheritance, genericity, information hiding and void safety, a compiler error message can reflect a tricky logical mistake. You are using a SET as if it were a LIST (some operations are common, but others not). You are calling an operation on a reference that may be void (null) at run time. And so on.

By the way, about void-safety: for a decade now, Eiffel has been void-safe, meaning a compile-time guarantee of no run-time null pointer dereferencing. It is beyond my understanding how the rest of the world can still live with programs that run under myriad swords of Damocles: x.op (…) calls that might any minute, without any warning or precedent, hit a null x and crash.

Then there is the guarantee of logical consistency, which is where my class invariant (Figure 1) comes in. Maybe it scared you, but in reality it is all simple concepts, intended to make sure that you know what you are doing, and rely on tools to check that you are right. When you are writing your program, you are positing all kinds, logical assumptions, large and (mostly) small, all the time. Here, for the structure triples_from [o] to make sense, it must be a list such that:

• It contains all the triples t in the triples_table such that t.source = o.
•  It contains only those triples!

You know this when you write the program; otherwise you would not be having a “triples_from” structure. Such gems of knowledge should remain an integral part of the program. Individually they may not be rocket science, but accumulated over the lifetime of a class design, a subsystem design or a system design they collect all the intelligence that makes the software possible.  Yet in the standard process they are gone the next minute! (At best, some programmers may write a comment, but that does not happen very often, and a comment has no guarantee of precision and no effect on testing or correctness.)

Anyone who takes software development seriously must record such fundamental properties. Here we need the following invariant clause:

across triples_from as tf all

across tf.item as tp all tp.item.source = tf.cursor_index end

end

(It comes in the class, as shown in Figure 1, with the label “from_list_consistent”. Such labels are important for documentation and debugging purposes. We omit them here for brevity.)

What does that mean? If we could use Unicode (more precisely, if we could type it easily with our keyboards) we would write things like “∀ x: E | P (x)”:  for all x in E, property P holds of x. We need programming-language syntax and write this as across E as x all P (x.item) end. The only subtlety is the “.item” part, which gives us generality beyond the ∀ notation: x in the across is not an individual element of E but a cursor that moves over E. The actual element at cursor position is x.item, one of the properties of that cursor. The advantage is that the cursor has more properties, for example x.cursor_index, which gives its position in E. You do not get that with the plain “ of mathematics.

If instead of ∀ you want ∃ (there exists), use some instead of all. That is pretty much all you need to know to understand all the invariant clauses of class MULTIGRAPH as given in Figure 1.

So what the above invariant clause says is: take every position tf in triples_from; its position is tf.cursor_index and its value is tf.item. triples_from is declared as ARRAYED_LIST [LIST [TRIPLE]], so tf.cursor_index is an integer representing an object o, and tf.item is a list of triples. That list should  consist of the triples having tf.cursor_index as their source. This is the very property that we are expressing in this invariant clause, where the innermost across says: for every triple tp.item in the list, the source of that triple is the cursor index (of the outside across). Simple and straightforward, I think (although such English explanations are so much more verbose than formal versions, such as the Eiffel one here, and once you get the hang of it you will not need them any more).

How can one ever include a structure such as triples_from without expressing such a property? To put the question slightly differently: am I inside the asylum looking out, or outside the asylum looking in? Any clue would be greatly appreciated.

More properties

For the tag ( “with_”) and target lists, the properties are similar:

across triples_with as tw all across tw.item as tp all tp.item.tag = tw.key end end

across triples_to as tt all across tt.item as tp all tp.item.target = tt.cursor_index end end 

We also have some properties of array bounds:

 is_root.lower = 1 and is_root.upper = object_count

triples_from.lower = 1 and triples_from.upper = object_count

triples_to.lower = 1 and triples_to.upper = object_count

where object_count is the number of objects (nodes), and for an array a (whose bounds in Eiffel are arbitrary, not necessarily 0 or 1, and set on array creation), a.lower and a.upper are the bounds. Here we number the arrays from 1.

There are, as noted, two ways to represent rootness. We must express their consistency (or risk trouble). Two clauses of the invariant do the job:

across roots as t all is_root [t.item] end

across is_root as t all (t.item = roots.has (t.cursor_index)) end

The first one says that if we go through the list “roots” we only find elements whose “is_root” value is true; the second, that if we go through the array “is_root” we find values that are true where and only where the corresponding object, given by the cursor index, is in the “roots” set. Note that the “=” in that second property is between boolean values (if in doubt, check the type instantly in the EIffelStudio IDE!), so it means “if and only if”.

Instead of these clauses, a more concise version, covering them both, is just

roots ~ domain (is_root)

with a function domain that gives the domain of a function represented by a boolean array. The ~ operator denotes object equality, redefined in many classes, and in particular in the SET classes (“roots” is a LINKED_SET) to cover equality between sets, i.e. the property of having the same elements.

The other clauses are all similarly self-explanatory. Let us just go through the most elaborate one, successors_consistent, involving three levels of across:

across successors as httpl all                   — httpl.item: hash table of list of triples

        across httpl.item as tpl all                — tpl.item: list of triples (tpl.key: key (i.e. tag) in hash table (tag)

                  across tpl.item as tp all            — tp.item: triple

                         tp.item.tag = tpl.key

and tp.item.source = httpl.cursor_index

                   end

          end

end

You can see that I struggled a bit with this one and made provisions for not having to struggle again when I would look at the code again 10 minutes, 10 days or 10 months later. I chose (possibly strange but consistent) names such as httpl for hash-table triple, and wrote comments (I do not usually need any in invariant and other contract clauses) to remind me of the type of everything. That was not strictly needed since once again the IDE gives me the types, but it does not cost much and could help.

What this says: go over “successors”; which as you remember is an ARRAY, indexed by objects, of HASH_TABLE, where each entry of such a hash table has an element of type [LIST [TRIPLE] and a key of type INTEGER, representing the tag of a number of outgoing edges from the given object. Go over each hash table httpl. Go over the associated list of triples tpl. Then for each triple tp in this list: the tag of the triple must be the key in the hash table entry (remember, the key does denote a tag); and the source of the triple must the object under consideration, which is the current iteration index in the array of the outermost iteration.

I hope I am not scaring you at this point. Although the concepts are simple, this invariant is more sophisticated than most of those we typically write. Many invariant clauses (and preconditions, and postconditions) are very simple properties, such as x > 0 or x ≠ y. The reason this one is more elaborate is not that I am trying to be fussy but that without it I would be the one scared to death. What is elaborate here is the data structure and programming technique. Not rocket science, not anything beyond programmers typically do, but elaborate. The only way to get it right is to buttress it by the appropriate logical properties. As noted, these properties are there anyway, in the back of your head, when you write the program. If you want to be more like an electrical engineer than an electrician, you have to write them down.

There is more to contracts

Invariants are not the only kind of such “contract” properties. Here for example, from the same class, is a (slightly abbreviated) part of the postcondition (output property) of the operation that tells us, through a boolean Result, if the multigraph has an edge of given components osource, t (the tag) and otarget :

Result =

(across successors [osource] [t] as tp some

not tp.item.is_inoperative and tp.item.target = otarget

end)

In words, this clause expresses the compatibility of the operation with the “successors” view: it must answer yes if and only if otarget appears in the successor set of osource for t, and the corresponding triple is not marked inoperative.

The concrete benefits

And so? What do we get out of making these logical properties explicit? Just the intellectual satisfaction of doing things right, and the methodological guidance? No! Once you have done this work, it is all downhill. Turn on the run-time assertion monitoring option (tunable separately for preconditions, postconditions, invariants etc., and on by default in development mode), and watch your tests run. If you are like almost all of us, you will have made a few mistakes, some which will seem silly when or rather if you find them in time (but there is nothing funny about a program that crashes during operation) and some more subtle. Sit back, and just watch your contracts be violated. For example if I change “<=” to “<” in the invariant property “tw.key <= max_tag”, I get the result of Figure 5. I see the call stack that I can traverse, the object run-time structure that I can explore, and all the tools of a modern debugger for an OO language. Finding and correcting the logical flaw will be a breeze.

Figure 5: An invariant violation brings up the debugger

The difference

It will not be a surprise that I did not get all the data structures and algorithms of the class MULTIGRAPH  right the first time. The Design by Contract approach (the discipline of systematically expressing, whenever you write any software element, the associated logical properties) does lead to fewer mistakes, but everyone occasionally messes up. Everyone also looks at initial results to spot and correct mistakes. So what is the difference?

Without the techniques described here, you execute your software and patiently examine the results. In the example, you might output the content of the data structures, e.g.

List of outgoing references for every object:

        1: 1-1->1|D, 1-1->2|D, 1-1->3|D, 1-2->1|D, 1-2->2|D,  1-25->8|D, 1-7->1|D, 1-7->6|D,

1-10->8|D, 1-3->1|D, 1-3->2|D, 1-6->3|D, 1-6->4|D, 1-6->5|D

        3: 3-6->3, 3-6->4, 3-6->5, 3-9->14, 3-9->15,   3-9->16, 3-1->3, 3-1->2, 3-2->3, 3-2->2,

                  3-25->8, 3-7->3, 3-7->6, 3-10->8, 3-3->3,  3-3->2    

List of outgoing references for every object:

        1: 1-1->1|D, 1-1->2|D, 1-1->3|D, 1-2->1|D, 1-2->2|D, 1-25->8|D, 1-7->1|D, 1-7->6|D,

1-10->8|D, 1-3->1|D,  1-3->2|D, 1-6->3|D, 1-6->4|D, 1-6->5|D

        3: 3-6->3, 3-6->4, 3-6->5, 3-9->14, 3-9->15,  3-9->16, 3-1->3, 3-1->2, 3-2->3, 3-2->2,

                                 3-25->8, 3-7->3, 3-7->6, 3-10->8, 3-3->3,  3-3->2

and so on for all the structures. You check the entries one by one to ascertain that they are as expected. The process nowadays has some automated support, with tools such as JUnit, but it is still essentially manual, tedious and partly haphazard: you write individual test oracles for every relevant case. (For a more automated approach to testing, taking advantage of contracts, see [1].) Like the logical properties appearing in contracts, these oracles are called “assertions” but the level of abstraction is radically different: an oracle describes the desired result of one test, where a class invariant, or routine precondition, or postcondition expresses the properties desired of all executions.

Compared to the cost of writing up such contract properties (simply a matter of formalizing what you are thinking anyway when you write the code), their effect on testing is spectacular. Particularly when you take advantage of “across” iterators. In the example, think of all the checks and crosschecks automatically happening across all the data structures, including the nested structures as in the 3-level across clause. Even with a small test suite, you immediately get, almost for free, hundreds or thousands of such consistency checks, each decreasing the likelihood that a logical flaw will survive this ruthless process.

Herein lies the key advantage. Not that you will magically stop making mistakes; but that the result of such mistakes, in the form of contract violations, directly points to logical properties, at the level of your thinking about the program. A wrong entry in an output, whether you detect it visually or through a Junit clause, is a symptom, which may be far from the cause. (Remember Dijkstra’s comment, the real point of his famous Goto paper, about the core difficulty of programming being to bridge the gap between the static program text, which is all that we control, and its effect: the myriad possible dynamic executions.) Since the cause of a bug is always a logical mistake, with a contract violation, which expresses a logical inconsistency, you are much close to that cause.

(About those logical mistakes: since a contract violation reflects a discrepancy between intent, expressed by the contract, and reality, expressed by the code, the mistake may be on either side. And yes, sometimes it is the contract that is wrong while the implementation in fact did what is informally expected. There is partial empirical knowledge [1] of how often this is the case. Even then, however, you have learned something. What good is a piece of code of which you are not able to say correctly what it is trying to do?)

The experience of Eiffel programmers reflects these observations. You catch the mistakes through contract violations; much of the time, you find and correct the problem easily. When you do get to producing actual test output (which everyone still does, of course), often it is correct.

This is what has happened to me so far in the development of the example. I had mistakes, but converging to a correct version was a straightforward process of examining violations of invariant violations and other contract elements, and fixing the underlying logical problem each time.

By the way, I believe I do have a correct version (in the sense of the second part of the Hoare quote), on the basis not of gut feeling or wishful thinking but of solid evidence. As already noted it is hard to imagine, if the code contains any inconsistencies, a test suite surviving all the checks.

Tests and proofs

Solid evidence, not perfect; hard to imagine, not impossible. Tests remain only tests; they cannot exercise all cases. The only way to achieve demonstrable correctness is to rely on mathematical proofs performed mechanically. We have this too, with the AutoProof proof system for Eiffel, developed in recent years [1]. I cannot overstate my enthusiasm for this work (look up the Web-based demo), its results (automated proof of correctness of a full-fledged data structures and algorithms library [2]) and its potential, but it is still a research effort. The dynamic approach (meaning test-based rather than proof-based) presented above is production technology, perfected over several decades and used daily for large-scale mission-critical applications. Indeed (I know you may be wondering) it scales up without difficulty:

• The approach is progressive. Unlike fully formal methods (and proofs), it does not require you to write down every single property down to the last quantifier. You can start with simple stuff like x > 0. The more you write, the more you get, but it is the opposite of an all-or-nothing approach.
• On the practical side, if you are wondering about the consequences on performance of a delivered system: there is none. Run-time contract monitoring is a compilation option, tunable for different kinds of contracts (invariants, postconditions etc.) and different parts of a system. People use it, as discussed here, for development, testing and debugging. Most of the time, when you deliver a debugged system, you turn it off.
• It is easy to teach. As a colleague once mentioned, if you can write an if-then-else you can write a precondition. Our invariants in the above example where a bit more sophisticated, but programmers do write loops (in fact, the Eiffel loop for iterating over a structure also uses across, with “loop” and instructions instead of “all” or “some” and boolean expressions). If you can write a loop over an array, you can write a property of the array’s elements.
• A big system is an accumulation of small things. In a blog article [5] I recounted how I lost a full day of producing a series of technical diagrams of increasing complexity, using one of the major Web-based collaborative development tools. A bug of the system caused all the diagrams to reproduce the first, trivial one. I managed to get through to the developers. My impression (no more than an educated guess resulting from this interaction) is that the data structures involved were far simpler than the ones used in the above discussion. One can surmise that even simple invariants would have uncovered the bug during testing rather than after deployment.
• Talking about deployment and tools used directly on the cloud: the action in software engineering today is in DevOps, a rapid develop-deploy loop scheme. This is where my perplexity becomes utter cluelessness. How can anyone even consider venturing into that kind of exciting but unforgiving development model without the fundamental conceptual tools outlined above?

We are back then to the core question. These techniques are simple, demonstrably useful, practical, validated by years of use, explained in professional books (e.g. [6]), introductory programming textbooks (e.g. [7]), EdX MOOCs (e.g. [8]), YouTube videos, online tutorials at eiffel.org, and hundreds of articles cited thousands of times. On the other hand, most people reading this article are not using Eiffel. On reflection, a simple quantitative criterion does exist to identify the inmates: there are far more people outside the asylum than inside. So the evidence is incontrovertible.

What, then, is wrong with me?

References

(Nurse to psychiatrist: these are largely self-references. Add “narcissism” to list of patient’s symptoms.)

1.    Ilinca Ciupa, Andreas Leitner, Bertrand Meyer, Manuel Oriol, Yu Pei, Yi Wei and others: AutoTest articles and other material on the AutoTest page.

2. Bertrand Meyer, Ilinca Ciupa, Lisa (Ling) Liu, Manuel Oriol, Andreas Leitner and Raluca Borca-Muresan: Systematic evaluation of test failure results, in Workshop on Reliability Analysis of System Failure Data (RAF 2007), Cambridge (UK), 1-2 March 2007 available here.

3.    Nadia Polikarpova, Ilinca Ciupa and Bertrand Meyer: A Comparative Study of Programmer-Written and Automatically Inferred Contracts, in ISSTA 2009: International Symposium on Software Testing and Analysis, Chicago, July 2009, available here.

4.    Carlo Furia, Bertrand Meyer, Nadia Polikarpova, Julian Tschannen and others: AutoProof articles and other material on the AutoProof page. See also interactive web-based online tutorial here.

5.    Bertrand Meyer, The Cloud and Its Risks, blog article, October 2010, available here.

6.    Bertrand Meyer: Object-Oriented Software Construction, 2^nd edition, Prentice Hall, 1997.

7.    Bertrand Meyer: Touch of Class: Learning to Program Well Using Objects and Contracts, Springer, 2009, see touch.ethz.ch and Amazon page.

8.    MOOCs (online courses) on EdX : Computer: Art, Magic, Science, Part 1 and Part 2. (Go to “archived versions” to follow the courses.)

We “core” computer scientists and software engineers always whine that our research themes forever prevent us, to the delight of our physicist colleagues but unjustly, from reaching the gold standard of academic recognition: publishing in Nature. I think I have broken this barrier now by disproving the old, dusty laws of physics! Brace yourself for my momentous discovery: I have evidence of negative speeds.

My experimental setup (as a newly self-anointed natural scientist I am keen to offer the possibility of replication) is the Firefox browser. I was downloading an add-on, with a slow connection, and at some point got this in the project bar:

Negative speed! Questioning accepted wisdom! Nobel in sight! What next, cold fusion?

I fear I have to temper my enthusiasm in deference to more mundane explanations. There’s the conspiracy explanation: the speed is truly negative (more correctly, it is a “velocity”, a vector of arbitrary direction, hence in dimension 1 possibly negative); Firefox had just reversed the direction of transfer, surreptitiously dumping my disk drive to some spy agency’s server.

OK, that is rather far-fetched. More likely, it is a plain bug. A transfer speed cannot be negative; this property is not just wishful thinking but should be expressed as an integral part of the software. Maybe someone should tell Firefox programmers about class invariants.

(Reposted from the CACM blog [*].)

Software engineering was never a popular subject. It started out as “programming methodology”, evoking the image of bearded middle-aged men telling you with a Dutch, Swiss-German or Oxford accent to repent and mend your ways. Consumed (to paraphrase Mark Twain) by the haunting fear that someone, somewhere, might actually enjoy coding.

That was long ago. With a few exceptions including one mentioned below, to the extent that anyone still studies programming methodology, it’s in the agile world, where the decisive argument is often “I always say…”. (Example from a consultant’s page:  “I always tell teams: `I’d like a [user] story to be small, to fit in one iteration but that isn’t always the way.’“) Dijkstra did appeal to gut feeling but he backed it through strong conceptual arguments.

The field of software engineering, of which programming methodology is today just a small part, has enormously expanded in both depth and width. Conferences such as ICSE and ESEC still attract a good crowd, the journals are buzzing, the researchers are as enthusiastic as ever about their work, but… am I the only one to sense frustration? It is not clear that anyone outside of the community is interested. The world seems to view software engineering as something that everyone in IT knows because we all develop software or manage people who develop software. In the 2017 survey of CS faculty hiring in the U.S., software engineering accounted, in top-100 Ph.D.-granting universities, for 3% of hires! (In schools that stop at the master’s level, the figure is 6%; not insignificant, but not impressive either given that these institutions largely train future software engineers.) From an academic career perspective, the place to go is obviously  “Artificial Intelligence, Data Mining, and Machine Learning”, which in those top-100 universities got 23% of hires.

Nothing against our AI colleagues; I always felt “AI winter” was an over-reaction [1], and they are entitled to their spring. Does it mean software engineering now has to go into a winter of its own? That is crazy. Software engineering is more important than ever. The recent Atlantic  “software apocalypse” article (stronger on problems than solutions) is just the latest alarm-sounding survey. Or, for just one recent example, see the satellite loss in Russia [2] (juicy quote, which you can use the next time you teach a class about the challenges of software testing: this revealed a hidden problem in the algorithm, which was not uncovered in decades of successful launches of the Soyuz-Frigate bundle).

Such cases, by the way, illustrate what I would call the software professor’s dilemma, much more interesting in my opinion than the bizarre ethical brain-teasers (you see what I mean, trolley levers and the like) on which people in philosophy departments spend their days: is it ethical for a professor of software engineering, every morning upon waking up, to go to cnn.com in the hope that a major software-induced disaster has occurred,  finally legitimizing the profession? The answer is simple: no, that is not ethical. Still, if you have witnessed the actual state of ordinary software development, it is scary to think about (although not to wish for) all the catastrophes-in-waiting that you suspect are lying out there just waiting for the right circumstances .

So yes, software engineering is more relevant than ever, and so is programming methodology. (Personal disclosure: I think of myself as the very model of a modern methodologist [3], without a beard or a Dutch accent, but trying to carry, on today’s IT scene, the torch of the seminal work of the 1970s and 80s.)

What counts, though, is not what the world needs; it is what the world believes it needs. The world does not seem to think it needs much software engineering. Even when software causes a catastrophe, we see headlines for a day or two, and then nothing. Radio silence. I have argued to the point of nausea, including at least four times in this blog (five now), for a simple rule that would require a public auditing of any such event; to quote myself: airline transportation did not become safer by accident but by accidents. Such admonitions fall on deaf ears. As another sign of waning interest, many people including me learned much of what they understand of software engineering through the ACM Risks Forum, long a unique source of technical information on software troubles. The Forum still thrives, and still occasionally reports about software engineering issues, but most of the traffic is about privacy and security (with a particular fondness for libertarian rants against any reasonable privacy rule that the EU passes). Important topics indeed, but where do we go for in-depth information about what goes wrong with software?

Yet another case in point is the evolution of programming languages. Language creation is abuzz again with all kinds of fancy new entrants. I can think of one example (TypeScript) in which the driving force is a software engineering goal: making Web programs safer, more scalable and more manageable by bringing some discipline into the JavaScript world. But that is the exception. The arguments for many of the new languages tend to be how clever they are and what expressive new constructs they introduce. Great. We need new ideas. They would be even more convincing if they addressed the old, boring problems of software engineering: correctness, robustness, extendibility, reusability.

None of this makes software engineering less important, or diminishes in the least the passion of those of us who have devoted our careers to the field. But it is time to don our coats and hats: winter is upon us.

Notes

[1] AI was my first love, thanks to Jean-Claude Simon at Polytechnique/Paris VI and John McCarthy at Stanford.

[2] Thanks to Nikolay Shilov for alerting me to this information. The text is in Russian but running it through a Web translation engine (maybe this link will work) will give the essentials.

[3] This time borrowing a phrase from James Noble.

[*] I am reposting these CACM blog articles rather than just putting a link, even though as a software engineer I do not like copy-paste. This is my practice so far, and it might change since it raises obvious criticism, but here are the reasons: (A) The audiences for the two blogs are, as experience shows, largely disjoint. (B) I like this site to contain a record of all my blog articles, regardless of what happens to other sites. (C) I can use my preferred style conventions.

The full lineup of speakers at the 2018 LASER summer school on Software for Blockchains, Bitcoin and Distributed Trust is now ready, with the announcement of a new speaker, Primavera De Filippi from CNRS and Harvard on social and legal aspects.

The other speakers are Christian Cachin (IBM), Maurice Herlihy (Brown), Christoph Jentzsch (slock.it), me, Emil Gun Sirer (Cornell) and Roger Wattenhofer (ETH).

The school is the 14th in the LASER series and takes place June 2-10, 2018, on the island of Elba in Italy.

Early-fee registration deadline is February 10. The school’s page is here.

One of the most significant recent developments in software engineering is the concept of Devops*. Dismissing the idea as “just the latest buzzword” would be wrong. It may be a buzzword but it reflects a fundamental change in the way we structure system development; with web applications in particular the traditional distinctions between steps of development, V&V** and deployment fade out. If you are using Microsoft Word, you know or can easily find out the version number; but which version of your search engine are you using?

With the new flexibility indeed come new risks, as when a bug in the latest “devopsed”  version of Google Docs caused me to lose a whole set of complex diagrams irretrievably; an earlier article on this blog (“The Cloud and Its Risks“, October 2010) told the story.

In the new world of continuous integrated development/V&V/deployment, software engineering principles are more necessary than ever, but their application has to undergo a profound adaptation.

With Jean-Michel Bruel (Toulouse), Elisabetta Di Nitto (Milan) and Manuel Mazzara (Innopolis), we are organizing a workshop on the topic, DEVOPS 18, on 5-6 March 2018 near Toulouse. The Call for Papers is available here, with Springer LNCS proceedings. The submission deadline is January 15, but for that date a 2-page extended abstract is sufficient. I hope that the event will help the community get a better grasp of the software engineering techniques and practices applicable to this new world of software development.

Notes

*I know, it’s supposed to be DevOps (I am not a great fan of upper case in the middle of words).
** Validation & Verification.

Many programs take “execution arguments” which the program users provide at the start of execution. In EiffelStudio you can enter them under Execution -> Execution parameters.

The program can access them through the Kernel Library class ARGUMENTS. Typically, the root class of the system inherits from ARGUMENTS and its creation procedure will include something like

if argument_count /= N then
……..print (“XX expects exactly N arguments: AA, BB, …%N”)
else
……..u := argument (1) ; v := argument (2) ; …
……..“Proceed with normal execution, using u, v, …”
end

where N is the number of expected arguments, XX is the name of the program, and AA, …. are the roles of arguments. u, v, … are local variables. The criterion for acceptance could be “at least N” instead of exactly N. The features argument_count and arguments come from class ARGUMENTS.

In all but trivial cases this scheme (which was OK years ago, in a less sophisticated state of the language) does not work! The reason is that the error branch will fail to initialize attributes. Typically, the “Proceed with…” part in the other branch is of the form

               attr1 := u
                attr2 := v
                …
                create obj1.make (attr1, …)
                create obj2.make (attr2, …)
                “Work with obj1, obj2, …”

If you try to compile code of this kind, you will get a compilation error:

Eiffel is void-safe: it guarantees that no execution will ever produce null-pointer dereference (void call). To achieve this guarantee, the compiler must make sure that all attributes are “properly set” to an object reference (non-void) at the end of the creation procedure. But the error branch fails to initialize obj1 etc.

You might think of replacing the explicit test by a precondition to the creation procedure:

               require
                                argument_count = N

but that does not work; the language definition explicit prohibits preconditions in a root creation procedure. The Ecma-ISO standard (the official definition of the language, available here) explains the reason for the corresponding validity rule (VSRP, page 32):

A routine can impose preconditions on its callers if these callers are other routines; but it makes no sense to impose a precondition on the external agent (person, hardware device, other program…) that triggers an entire system execution, since there is no way to ascertain that such an agent, beyond the system’s control, will observe the precondition.

The solution is to separate the processing of arguments from the rest of the program’s work. Add a class CORE which represents the real core of the application and separate it from the root class, say APPLICATION. In APPLICATION, all the creation procedure does is to check the arguments and, if they are fine, pass them on to an instance of the core class:

                note
                                description: “Root class, processes execution arguments and starts execution”
                class APPLICATION create make feature
                                core: CORE
                                                — Application’s core object
                                make
……..……..……..……..……..……..— Check arguments and proceed if they make sense.
                                                do
                                                             if argument_count /= N then
                                                                                print (“XX expects exactly N arguments: AA, BB, …%N”)
                                                                else
                                                                                create core.make (argument (1), argument (2) ; …)
                                                                                                — By construction the arguments are defined!
                                                                                core.live
                                                                                                — Perform actual work
                                                                                               — (`live’ can instead be integrated with `make’ in CORE.)

                                                                end
                                                end
                 end
 
We may call this little design pattern “Split the Root”. Nothing earth-shattering; it is simply a matter of separating concerns (cutting off the Model from the View). It assumes a system that includes text-based output, whereas many applications are graphical. It is still worth documenting, for two reasons.

First, in its own modest way, the pattern is useful for simple programs; beginners, in particular, may not immediately understand why the seemingly natural way of processing and checking arguments gets rejected by the compiler.

The second reason is that Split the Root illustrates the rules that preside over a carefully designed language meant for carefully designed software. At first it may be surprising and even irritating to see code rejected because, in a first attempt, the system’s root procedure has a precondition, and in a second attempt because some attributes are not initialized — in the branch where they do not need to be initialized. But there is a reason for these rules, and once you understand them you end up writing more solid software.

Much of the progress in robotics is due to software advances, and software issues remain at the heart of the formidable challenges that remain. The 2017 LASER summer school, held in September in Elba, brings together some of the most prestigious international experts in the area.

The LASER school has established itself as one of the principal forums to discussed advanced software issues. The 2017 school takes place from 9 to 17 September in the idyllic setting of the Hotel del Golfo in Procchio, Elba Island, Italy.

Robotics is progressing at an amazing pace, bringing improvements to almost areas of human activity. Today’s robotics systems rely ever more fundamentally on complex software, raising difficult issues. The LASER 2017 summer school covers both the current state of robotics software technology and open problems. The lecturers are top international experts with both theoretical contributions and major practical achievements in developing robotics systems.
The LASER school is intended for professionals from the industry (engineers and managers) as well as university researchers, including PhD students. Participants learn about the most important software technology advances from the pioneers in the field. The school’s focus is applied, although theory is welcome to establish solid foundations. The format of the school favors extensive interaction between participants and speakers.

We have lined up an impressive roster of speakers from the leading edge of both industry and academia:

• Rodolphe Gélin, Aldebaran Robotics
• Ashish Kapoor, Microsoft Research
• Davide Brugali, University of Bergamo, on Managing software variability in robotic control systems
• Nenad Medvidovic, University of Southern California, on Software Architectures of Robotics Systems
• Bertrand Meyer, Politecnico di Milano & Innopolis University, on Concurrent Object-Oriented Robotics Software
• Issa Nesnas, NASA Jet Propulsion Laboratory, on Experiences from robotic software development for research and planetary flight robots
• Hiroshi (“Gitchang”) Okuno, Waseda University & Kyoto University, on Open-Sourced Robot Audition Software HARK: Capabilities and Applications

The school takes place at the magnificent Hotel del Golfo in the Gulf of Procchio, Elba. Along with an intensive scientific program, participants will have time to enjoy the countless natural and cultural riches of this wonderful, history-laden jewel of the Mediterranean.

For more information about the school, the speakers and registration see the LASER site.

One of the most delicate aspects of design is feature interaction. As users, we suffer daily from systems offering features that individually make sense but clash with each other. In my agile book [1] I explained in detail, building on the work of Pamela Zave, why this very problem makes one of the key ideas of agile methods,  the reliance on “user stories” for requirements, worthless and damaging.

A small recent incident reminded me of the perils of feature interaction. I used my Lenovo W540 laptop without power for a short while, then reached a sedentary location and plugged it in. Hence my surprise when, some hours later, it started beeping to alert me that it was running out of battery. The natural reactions — check the outlet and the power cord — had no effect. I found the solution, but just in time: otherwise, including if I had not heard the warning sound, I would have been unable to use the laptop any further. That’s right: I would not have been able to restart the computer at all, even with access to a power outlet, and even though it was perfectly functional and so was its (depleted) battery. The reason is that the problem arose from a software setting, which (catch-22 situation) I could not correct without starting the computer [2].

The only solution would have been to find another, non-depleted battery. That is not a trivial matter if you have traveled with your laptop outside of a metropolis: the W540 has a special battery which ordinary computer shops do not carry [3].

The analysis of what made such a situation possible must start with the list of relevant hardware and software product features.

Hardware:

• HA. This Lenovo W series includes high-end laptops with high power requirements, which the typical 65-watt airplane power jack does not satisfy.
• HB. With models prior to the W540, if you tried to connect a running laptop to the power supply in an airplane, it would not charge, and the power indicator would start flickering.  But you could still charge it if you switched it off.
• HC. The W540 effectively requires 135 watts and will not take power from a 65-watt power source under any circumstances.

Software:

• SA. The operating system (this discussion assumes Windows) directly reflects HC by physically disabling charging if the laptop is in the “Airplane” power mode.
• SB. If you disable wireless, the operating system automatically goes into the “Airplane” power mode.
• SC. In the “Airplane” power mode, the laptop, whether or not connected through a charger to a power outlet of any wattage, will not charge. The charging function is just disabled.
• SD. One can edit power modes to change parameters, such as time to automatic shutoff, but the no-charging property in Airplane mode is not editable and not even mentioned in the corresponding UI dialog. It seems to be a behind-the-scenes property magically attached to the power-mode name “Airplane”.
• SE. There is a function key for disabling wireless: F8. As a consequence of SB it also has the effect of switching to “Airplane” mode.
• SF. Next to F8 on the keyboard is F7.
• SG. F7 serves to display the screen content on another monitor (Windows calls it a “projector”). F7 offers a cyclic set of choices: laptop only, laptop plus monitor etc.
• SH. In the old days (like five years ago), such function keys setting important operating system parameters on laptops used to be activated only if you held them together with a special key labeled “Fn”. For some reason (maybe the requirement was considered too complicated for ordinary computer users) the default mode on Lenovo laptops does not use the “Fn” key anymore: you just press the desired key, such as F7 or F8.
• SI. You can revert to the old mode, requiring pressing “Fn”, by going into the BIOS and performing some not-absolutely-trivial steps, making this possibility the preserve of techies. (Helpfully, this earlier style is called “Legacy mode”, as a way to remind you that your are an old-timer, probably barely graduated from MS-DOS and still using obsolete conventions. In reality, the legacy mode is the right one to use, whether for techies or novices: it is all too easy to hit a function key by mistake and get totally unexpected results. The novice, not the techie, is the one who will be completely confused and panicked as a result. The first thing I do with a new laptop is to go to the BIOS and set legacy mode.)

By now you have guessed what happened in my case, especially once you know that I had connected the laptop to a large monitor and had some trouble getting that display to work. In the process I hit Fn-F7 (feature SG) several times.  I must have mistakenly (SF) pressed F8 instead of F7 at some point. Normally, Legacy mode (SI) should have made me immune to the effects of hitting a function key by mistake, but I did use the neighboring key F7 for another purpose. Hitting F8 disabled wireless (SE) and switched on Airplane power mode (SB). At that point the laptop, while plugged in correctly, stopped charging (SC, SD).

How did I find out? Since I was looking for a hardware problem I could have missed the real cause entirely and ended up with a seemingly dead laptop. Fortunately I opened the Power Options dialog to see what it said about the battery. I noticed that among the two listed power plans the active one was not “Power Saver”, to which I am used, but “Airplane”. I did not immediately pay  attention to that setting; since I had not used the laptop for a while I just thought that maybe the last time around I had switched on “Airplane”, even though that made little sense since I was not even aware of the existence of that option. After trying everything else, though, I came back to that intriguing setting, changed to the more usual “Power Saver”, and the computer started to charge again. I was lucky to have a few percent of battery still left at that point.

Afterwards I found a relevant discussion thread on a Lenovo user forum.

As is often the case in such feature-interaction mishaps, most of the features make sense individually [4]. What causes trouble is some unforeseen combination of features.

There is no sure way to avoid such trouble, but there is a sure way to cause it: design a system feature by feature, as with user stories in agile development. The system must do this and it must do that. Oh, by the way, it must also do that. And that. User stories have one advantage: everyone understands them. But that is also their limitation. Good requirements and design require professionals who can see the whole beyond the parts.

A pernicious side of this situation is that many people believe that use cases and user stories are part of object-oriented analysis, whereas the OO approach to requirements and design is the reverse: rise above individual examples to uncover the fundamental abstractions.

As to my laptop, it is doing well, thanks. And I will be careful with function keys.

Reference and notes

[1] Bertrand Meyer: Agile! The Good, the Hype and the Ugly, Springer, 2014,  Amazon page: here, book page: here. A description of the book appeared here on this blog at the time of publication.

[2] Caveat: I have not actually witnessed this state in which a plugged-in laptop will not restart. The reason is simply that I do not have an alternate battery at the moment so I cannot perform the experiment with the almost certain result of losing the use of my laptop. I will confirm the behavior as soon as I have access to a spare battery.

[3] It has been my systematic experience over the past decade and a half that Lenovo seems to make a point, every couple of years, to introduce new models with incompatible batteries and docking stations. (They are also ever more incredibly bulky, with the one for the W540 almost as heavy as the laptop itself. On the other hand the laptops are good, otherwise I would not be bothering with them.)

[4] One exception here is feature SB: switching wireless off does not necessaril y mean you want to select a specific power mode! It is a manifestation of the common syndrome  of software tools that think they are smarter than you, and are not. Another exception is SE: to let a simple key press change fundamental system behavior is to court disaster. But I had protected myself by using legacy mode and was hit anyway.

The AutoProof technology pursues the goal of “Verification As a Matter Of Course”, integrated into the EVE development environment. (The AutoProof  project page here; see particularly the online interactive tutorial.) A one-day workshop devoted to the existing AutoProof and current development will take place on October 1 near Toulouse in France. It is an informal event (no proceedings planned at this point, although based on the submissions we might decide to produce a volume), on a small scale, designed to bring together people interested in making the idea of practical verification a reality.

The keynote will be given by Rustan Leino from Microsoft Research, the principal author of the Boogie framework on which the current implementation of AutoProof relies.

For submissions (or to attend without submitting) see the workshop page here. You are also welcome to contact me for more information.

Many robotics applications are by nature concurrent; in his ongoing PhD work, Andrey Rusakov [1] is building a comprehensive concurrent robot programming framework, Roboscoop [2], based on the SCOOP model for simple concurrent object-oriented programming [3] and the Ros operating system. As part of this work it is important to know how much robotics applications use concurrency, stay away from concurrency — perhaps because programmers are afraid of the risks — and could benefit from more concurrency. Rusakov has prepared a questionnaire [4] to help find out. If you have experience in robot programming, please help him by answering the questionnaire, which takes only a few minutes.

References

[1] Rusakov’s home page here.

[2] Roboscoop project page, here,

[3] Simple Concurrent Object-Oriented Programming, see here.

[4] The questionnaire is here.

A third ACM webinar this year (after two on agile methods): I will be providing a general introduction to Design by Contract. The date is this coming Thursday, September 17, and the time is noon New York (18 Paris/Zurich, 17 London, 9 Los Angeles, see here for hours elsewhere). Please tune in! The event is free but requires registration here.

Programming, wrote Dijkstra many years ago, is a branch of applied mathematics. That is only half of the picture: the other half is engineering, and this dual nature of programming is part of its attraction.

Descriptions of the mathematical side are generally, in my view, too complicated. This article [1] presents a mathematical theory of programs and programming based on concepts taught in high school: elementary set theory. The concepts covered include:

• Programming.
• Specification.
• Refinement.
• Non-determinism.
• Feasibility.
• Correctness.
• Programming languages.
• Kinds of programs: imperative, functional, object-oriented.
• Concurrency (small-step and large-step)
• Control structures (compound, if-then-else and Dijkstra-style conditional, loop).
• State, store and environment.
• Invariants.
• Notational conventions for building specifications and programs incrementally.
• Loop invariants and variants.

One of the principal ideas is that a program is simply the description of a mathematical relation. The program text is a rendering of that relation. As a consequence, one may construct programming languages simply as notations to express certain kinds of mathematics. This approach is the reverse of the usual one, where the program text and its programming languages are the starting point and the center of attention: theoreticians develop techniques to relate them to mathematical concepts. It is more effective to start from the mathematics (“unparsing” rather than parsing).

All the results (74 properties expressed formally, a number of others in the text) are derived as theorems from rules of elementary set theory; there are no new axioms whatsoever.

The paper also has a short version [2], omitting proofs and many details.

References

[1] Theory of Programs, available here.
[2] Theory of Programs, short version of [1] (meant for quick understanding of the ideas, not for publication), available here.

Among the open problems of verification, particularly the verification of object-oriented programs, one of the most vexing is framing: how to specify and verify what programs element do not change. Continuing previous work, this article presents a “double frame inference” method, automatic on both sides the specification and verification sides. There is no need to write frame specifications: they will be inferred from routine postconditions. For verification, the method computes the set of actually changed properties through a “change calculus”, itself based on the previously developed alias calculus.

Some verification techniques, such as Hoare-style proofs, require significant annotation effort and potentially yield full functional verification; others, such as model checking and abstract interpretation, have more limited goals but seek full automation. Framing, in my opinion, should be automatic, freeing the programmer-verifier to devote the annotation effort to truly interesting properties.

Reference

[1] Bertrand Meyer: Framing the Frame Problem, in Dependable Software Systems, Proceedings of August 2014 Marktoberdorf summer school, eds. Alexander Pretschner, Manfred Broy and Maximilian Irlbeck, NATO Science for Peace and Security, Series D: Information and Communication Security, Springer, 2015 (to appear), pages 174-185; preprint available here.

In support of his view of software methodology, Leslie Lamport likes to use the example of non-recursive Quicksort. Independently of the methodological arguments, his version of the algorithm should be better known. In fact, if I were teaching “data structures and algorithms” I would consider introducing it first.

As far as I know he has not written down his version in an article, but he has presented it in lectures; see [1]. His trick is to ask the audience to give a non-recursive version of Quicksort, and of course everyone starts trying to remove the recursion, for example by making the stack explicit or looking for invertible functions in calls. But his point is that recursion is not at all fundamental in Quicksort. The recursive version is a specific implementation of a more general idea.

Lamport’s version — let us call it Lampsort —is easy to express in Eiffel. We may assume the following context:

a: ARRAY [G -> COMPARABLE]        — The array to be sorted.
pivot: INTEGER                                      —  Set by partition.
picked: INTEGER_INTERVAL            — Used by the sorting algorithm, see below.
partition (i, j: INTEGER)
……..require      — i..j is a sub-interval of the array’s legal indexes:
……..……..i < j
……..……..i >= a.lower
……..……..j <= a.upper
……..do
……..……..… Usual implementation of partition …
……..ensure     — The expected effect of partition:
……..……..pivot >= i
……..……..pivot < j
……..……..— a [i..j] has been reshuffled so that elements in i..pivot are less than
……..……..— or equal to those in pivot+1 .. j.
……..end

We do not write the implementation of partition since the point of the present discussion is the overall algorithm. In the usual understanding, that algorithm consists of doing nothing if the array has no more than one element, otherwise performing a partition and then recursively calling itself on the two resulting intervals. The implementation can take advantage of parallelism by forking the recursive calls out to different processors. That presentation, says Lamport, describes only a possible implementation. The true Quicksort is more general. The algorithm works on a set not_sorted of integer intervals i..j such that the corresponding array slices a [i..j] are the only ones possibly not sorted; the goal of the algorithm is to make not_sorted empty, since then we know the entire array is sorted. In Eiffel we declare this set as:

not_sorted: SET [INTEGER_INTERVAL]

The algorithm initializes not_sorted to contain a single element, the entire interval; at each iteration, it removes an interval from the set, partitions it if that makes sense (i.e. the interval has more than one element), and inserts the resulting two intervals into the set. It ends when not_sorted is empty. Here it is:

……..from                                 — Initialize interval set to contain a single interval, the array’s entire index range:
……..…..create not_sorted.make_one (a.lower |..| a.upper)….         ..……..
……..invariant
……..…..— See below
……..until
……..…..not_sorted.is_empty                                                            — Stop when there are no more intervals in set
……..loop
……..…..picked := not_sorted.item                                                     — Pick an interval from (non-empty) interval set.
……..……if picked.count > 1 then                                                      — (The precondition of partition holds, see below.)
……..……..…..partition (picked.lower, picked.upper)                 — Split, moving small items before & large ones after pivot.
……..……..…..not_sorted.extend (picked.lower |..| pivot)            — Insert new intervals into the set of intervals: first
……..……....…not_sorted.extend (pivot + 1 |..| picked.upper)     — and second.
……..……end
……..…...not_sorted.remove (picked)                                               — Remove interval that was just partitioned.
…….end

Eiffel note: the function yielding an integer interval is declared in the library class INTEGER using the operator |..| (rather than just  ..).

The query item from SET, with the precondition not is_empty,  returns an element of the set. It does not matter which element. In accordance with the Command-Query Separation principle, calling item does not modify the set; to remove the element you have to use the command remove. The command extend adds an element to the set.

The abstract idea behind Lampsort, explaining why it works at all, is the following loop invariant (see [2] for a more general discussion of how invariants provide the basis for understanding loop algorithms). We call “slice” of an array a non-empty contiguous sub-array; for adjacent slices we may talk of concatenation; also, for slices s and t,  s <= t means that every element of s is less than or equal to every element of t. The invariant is:

a is the concatenation of the members of a set slices of disjoint slices, such that:
– The elements of a are a permutation of its original elements.
– The index range of any member  of slices having more than one element is in not_sorted.
– For any adjacent slices s and t (with s before t), s <= t.

The first condition (conservation of the elements modulo permutation) is a property of partition, the only operation that can modify the array. The rest of the invariant is true after initialization (from clause) with slices made of a single slice, the full array. The loop body maintains it since it either removes a one-element interval from not_sorted (slices loses the corresponding slice) or performs partition with the effect of partitioning one slice into two adjacent ones satisfying s <= t, whose intervals replace the original one in not_sorted. On exit, not_sorted is empty, so slices is a set of one-element slices, each less than or equal to the next, ensuring that the array is sorted.

The invariant also ensures that the call to partition satisfies that routine’s precondition.

The Lampsort algorithm is a simple loop; it does not use recursion, but relies on an interesting data structure, a set of intervals. It is not significantly longer or more difficult to understand than the traditional recursive version

sort (i, j: INTEGER)
……..require
……..……..i <= j
……..……..i >= a.lower
……..……..j <= a.upper
……..do
……..……if j > i then                    — Note that precondition of partition holds.
……..……..…..partition (i, j)         — Split into two slices s and t such that s <= t.
……..……..…..sort (i, pivot)          — Recursively sort first slice.
……..……..…..sort (pivot+1, j)      — Recursively sort second slice.
……..……end……..…..
……..end

Lampsort, in its author’s view, captures the true idea of Quicksort; the recursive version, and its parallelized variants, are only examples of possible implementations.

I wrote at the start that the focus of this article is Lampsort as an algorithm, not issues of methodology. Let me, however, give an idea of the underlying methodological debate. Lamport uses this example to emphasize the difference between algorithms and programs, and to criticize the undue attention being devoted to programming languages. He presents Lampsort in a notation which he considers to be at a higher level than programming languages, and it is for him an algorithm rather than a program. Programs will be specific implementations guided in particular by efficiency considerations. One can derive them from higher-level versions (algorithms) through refinement. A refinement process may in particular remove or restrict non-determinism, present in the above version of Lampsort through the query item (whose only official property is that it returns an element of the set).

The worldview underlying the Eiffel method is almost the reverse: treating the whole process of software development as a continuum; unifying the concepts behind activities such as requirements, specification, design, implementation, verification, maintenance and evolution; and working to resolve the remaining differences, rather than magnifying them. Anyone who has worked in both specification and programming knows how similar the issues are. Formal specification languages look remarkably like programming languages; to be usable for significant applications they must meet the same challenges: defining a coherent type system, supporting abstraction, providing good syntax (clear to human readers and parsable by tools), specifying the semantics, offering modular structures, allowing evolution while ensuring compatibility. The same kinds of ideas, such as an object-oriented structure, help on both sides. Eiffel as a language is the notation that attempts to support this seamless, continuous process, providing tools to express both abstract specifications and detailed implementations. One of the principal arguments for this approach is that it supports change and reuse. If everything could be fixed from the start, maybe it could be acceptable to switch notations between specification and implementation. But in practice specifications change and programs change, and a seamless process relying on a single notation makes it possible to go back and forth between levels of abstraction without having to perform repeated translations between levels. (This problem of change is, in my experience, the biggest obstacle to refinement-based approaches. I have never seen a convincing description of how one can accommodate specification changes in such a framework without repeating the whole process. Inheritance, by the way, addresses this matter much better.)

The example of Lampsort in Eiffel suggests that a good language, equipped with the right abstraction mechanisms, can be effective at describing not only final implementations but also abstract algorithms. It does not hurt, of course, that these abstract descriptions can also be executable, at the possible price of non-optimal performance. The transformation to an optimal version can happen entirely within the same method and language.

Quite apart from these discussions of software engineering methodology, Lamport’s elegant version of Quicksort deserves to be known widely.

References

[1] Lamport video here, segment starting at 0:32:34.
[2] Carlo Furia, Bertrand Meyer and Sergey Velder: Loop invariants: Analysis, Classification and Examples, in ACM Computing Surveys, September 2014, preliminary text here.

The French National Research Center (CNRS) has just awarded [1] its annual gold medal to Gérard Berry, a great recognition for an outstanding computer scientist. I first discovered Berry’s work through a brilliant 1976 article, Bottom-Up Computation of Recursive Programs [2], which explained recursion using methods of denotational semantics, and should really figure in collections of classic CS articles. One can hardly understand dynamic programming algorithms, for example, without realizing that they are bottom-up versions of recursive algorithms, through transformations described in the article. I relied extensively on its techniques for my own textbooks, from the advanced concepts of Introduction to the Theory of Programming Languages all the way down to the introductory presentation of recursion in Touch of Class.

Berry then went on to play a founding role in synchronous languages, explaining (in a paper whose reference I don’t have right now) that he went through a revelation of sorts after realizing that the automatic-control industry needed languages completely different from those computer scientists typically love. He created the Estérel language and shepherded it all the way to industrialization, serving as Chief Scientist of Estérel Technologies in the 2000 decade, before coming back to research.

He is also well known in France as a popularizer of informatics (computer science) through a number of successful books aiming at a large audience. He was appointed the first professor of informatics at the Collège de France, and in his inaugural lecture series presented a sweeping description of how information technology advances, based on powerful scientific concepts, permeate today’s world and raise ever new challenges. Hardly could the CNRS have chosen to distinguish a better representative of our discipline.

References

[1] CNRS announcement: here.

[2] Gérard Berry, Bottom-Up Computation of Recursive Programs, in RAIRO (Revue Française d’Automatique, Informatique, Recherche Opérationnelle, Informatique Théorique), tome 10, no R1 (1976), pp. 47-82, available here.

My colleagues and I have just finished recording our new MOOC (online course), an official ETH offering on the EdX platform. The preview is available [1] and the course will run starting in September.

As readers of this blog know, I  have enthusiastically, under the impulsion of Marco Piccioni at ETH, embraced MOOC technology to support and spread our courses. The particular target has been the introduction to programming that I have taught for over a decade at ETH based on the Touch of Class textbook [2]. In February this blog announced [3] the release of our first MOOC, embodying the essentials of our ETH course and making it available not only to ETH students but to the whole world. The course does not just include video lectures: it also supports active student participation through online exercises and programs that can be compiled and tested on the cloud, with no software installation. These advanced features result from our research on support for distributed software development (by Christian Estler and Martin Nordio, with Carlo Furia and others).

This first course was a skunkworks project, which we did entirely on our own without any endorsement from ETH or any of the main MOOC players. We and our students have very much benefited from the consequent flexibility, and the use of homegrown technology relying on the MOODLE framework. We will keep this course for our own students and for any outside participant who prefers a small-scale, “boutique” version. But the EdX brand and EdX’s marketing power will enable us to reach a much broader audience. We want to provide the best introductory computing course on the market and the world needs to know about it. In addition, the full support of media services at ETH  helped us reach a higher standard on the technical side. (For our first course, the home-brewed one, we did not have a studio, so that every time an ambulance drove by — our offices are close to the main Zurich hospital — we had to restart the current take.)

The course’s content is not exactly the same: we have broadened the scope from just programming to computing, although it retains a strong programming component. We introduced additional elements such as an interview with Professor Peter Widmayer of ETH on the basics of computer science theory. For both new material and the topics retained from the first version we have adapted to the accepted MOOC practice of short segments, although we did not always exactly meet the eight-minute upper limit that was suggested to us.

We hope that you, and many newcomers, will like the course and benefit from it.

References

[3] EdX course: Computing: Art, Magic, Science, preview available here.

[2] Bertrand Meyer: Touch of Class: Learning how to Program Well, with Objects and Contracts, Springer Verlag, revised printing, 2013, book page here.

[3] Learning to Program, Online: article from this blog, 3 February 2014, available here.

It is remarkable that almost half a century after Dijkstra’s goto article, and however copiously and reverently it may be cited, today’s programs (other than in Eiffel) are still an orgy of gotos. There are not called gotos, being described as constructs that break out of a loop or exit a routine in multiple places, but they are gotos all the same. Multiple routine exits are particularly bad since they are in effect interprocedural gotos.

Ian Joyner has just released a simple and cogent summary of why routines should always have one entry and one exit.

References

[1] Ian Joyner: Single-entry, single-exit (SESE) heuristic, available here.

The SCOOP concurrency model has a clear division of objects into “regions”, improving the clarity and reliability of concurrent programs by establishing a close correspondence between the object structure and the process structure. Each region has an associated “processor”, which executes operations on the region’s objects. A literal application of this rule implies, however, a severe performance penalty. As part of the work for his PhD thesis (defended two weeks ago), Benjamin Morandi found out that a mechanism for specifying certain processors as “passive” yields a considerable performance improvement. The paper, to be published at COORDINATION, describes the technique and its applications.

Reference

Benjamin Morandi, Sebastian Nanz and Bertrand Meyer: Safe and Efficient Data Sharing for Message-Passing Concurrency, to appear in proceedings of COORDINATION 2014, 16th International Conference on Coordination Models and Languages, Berlin, 3-6 June 2014, draft available here.
.

The ETH introductory programming course, which I have taught since 2003 and used as the basis for the Springer Touch of Class textbook, is now available as a MOOC: an online course, open to anyone interested [1]. The project was started and led by Marco Piccioni.

The MOOC was released in September; although it was “open” (the other “O”) from the start, we have not publicized it widely until now, since we used it first for the benefit of students taking the course at ETH, and took advantage of this experience to polish it. If you follow the acronym buzz you may say it was first a “SPOC” (Small Personal Online Course”). The experience with our students has been extremely encouraging: they took it as a supplement to the lectures and widely praised its value.We hope that many others will find it useful as well.

MOOCs are hot but they have attracted as much criticism as hype. We have seen the objections: low completion rates, lack of direct human contact, threats to traditional higher education. Two things are clear, though: MOOCs are more than a passing fad; and they have their own pedagogical advantages.

MOOCs are here to stay; one ignores them at one’s own peril. For courses on a popular topic, I believe that in a few years almost everyone will be teaching from a MOOC. Not in the sense of telling students “just follow this course on the Web and come back for the exam“, but as a basis for individual institutions’ courses. For example the students might be told to take the lessons online, then come to in-person lectures (“Flip The Classroom“) or discussion sessions. For any given topic, such as introduction to programming, only a handful of MOOCs will emerge in this role. We would like the ETH course to be one of them.

The question is not just to replace courses and textbooks with an electronic version. MOOCs enrich the learning experience. Introduction to Programming is a particularly fertile application area for taking advantage of technology: the presentation of programming methods and techniques becomes even more effective if students can immediately try out the ideas, compile the result, run it, and see the results on predefined tests. Our course offers many such interactive exercises, thanks to the E4Mooc (Eiffel for MOOCs) framework developed by Christian Estler [2]. This feature has proved to be a key attraction of the course for ETH students. Here for example is an exercise asking you to write a function that determines whether a string is a palindrome (reads identically in both directions):

The program area is pre-filled with a class skeleton where all that remains for you to enter is the algorithm for the relevant function. Then you can click “Compile” and, if there are no compilation errors, “Run” to test your candidate code against a set of predefined test values. One of the benefits for users taking the course is that there is no software to install: everything runs in the cloud, accessible from the browser. Here we see the MOOC not just as a technique for presenting standard material but as an innovative learning tool, opening up pedagogical techniques that were not previously possible.

Besides E4Mooc, our course relies on the Moodle learning platform. Our experience with Moodle has been pleasant; we noticed, for example, that students really liked the Moodle feature enabling them to gain virtual “badges” for good answers, to the point of repeating exercises until they got the badges. For instructors preparing the course, building a MOOC is a huge amount of work (that was not a surprise, people had told us); but it is worth the effort.

We noted that attendance at the lectures increased as compared to previous years. The matter is a natural concern: other than the cold November mornings in Zurich (one of the lectures is at 8 AM) there are many reasons not to show up in class:  the textbook covers much of the material; all the slides are online; so are the slides for exercise sessions (tutorials) and texts of exercises and some earlier exams; lectures were video-recorded in previous years, and students can access the old recordings. Our feeling is that the MOOC makes the course more exciting and gives students want to come to class and hear more.

The MOOC course is not tied to a particular period; you can take it whenever you like. (The current practice of offering MOOCs at fixed times is disappointing: what is the point of putting a course online if participants are forced to fit to a fixed schedule?)

Marco Piccioni and I are now off to our second MOOC, which will be a generalization of the first, covering the basics not just of programming but of computer science and IT overall, and will be available on one of the major MOOC platforms. We will continue to develop the existing MOOC, which directly supports our in-person course, and which we hope will be of use to many other people.

Take the MOOC, or tell a beginner near you to take it, and tell us what you think.

References

[1] Bertrand Meyer, Marco Piccioni and other members of the ETH Chair of Software Engineering: ETH Introduction to Programming MOOC, available here.

[2] H-Christian Estler, E4Mooc demo, available on YouTube: here.

As part of our “Concurrency Made Easy” ERC Advanced Investigator Grant project (2012-2017), we are offering PhD positions at the Chair of Software Engineering of ETH Zurich. The goal of the project is to build a sophisticated programming and verification architecture to make concurrent and distributed programming simple and reliable, based on the ideas of Eiffel and particularly the SCOOP concurrency model. Concurrency in its various forms (particularly multithreading) as well as distributed computing are required for most of today’s serious programs, but programming concurrent applications remains a challenge. The CME project is determined to break this complexity barrier.  Inevitably, achieving simplicity for users (in this case, application programmers) requires, under the hood, a sophisticated infrastructure, both conceptual (theoretical models) and practical (the implementation). We are building that infrastructure.

ETH offers an outstanding research and education environment and competitive salaries for “assistants” (PhD students), who are generally expected in addition to their research to participate in teaching, in particular introductory programming, and other activities of the Chair.  The candidates we seek have: a master’s degree in computer science or related field from a recognized institution (as required by ETH); a strong software engineering background, both practical and theoretical, and more generally a strong computer science and mathematical culture; a good knowledge of verification techniques (e.g. Hoare-style, model-checking, abstract interpretation); some background in concurrency or distribution; and a passion for high-quality software development. Prior publications, and experience with Eiffel, are pluses. In line with ETH policy, particular attention will be given to female candidates.

Before applying, you should become familiar with our work; see in particular the research pages at se.ethz.ch including the full description of the CME project at cme.ethz.ch.

Candidates should send (in PDF or text ) to se-open-positions@lists.inf.ethz.ch a CV and a short cover letter describing their view of the CME project and ideas about their possible contribution.

There will be two talks in the Software Engineering Seminar at ITMO, 18:00 local time, Tuesday, January 14, 2014. Please arrive 10 minutes early for registration.

Place: ITMO, Sytninskaya Ulitsa, Saint Petersburg.

Andrey Terekhov (SPBGU): Programming crystals

(I do not know whether this talk will be in Russian or English. An abstract follows but the talk is meant as the start of a discussion rather than a formal lecture.)

В течение последних 20-30 лет основными языками программирования кристаллов были VHDL и Verilog. Эти языки изначально проектировались как средства создания проектной документации, потом они стали использоваться в качестве инструмента моделирования и только сравнительно недавно для этих языков появились средства генерации кода уровня RTL (Register Transfer Language). Тексты на  VHDL и Verilog очень громоздки, трудно читаемые, плохо стандартизованы (одна и та же программа может синтезироваться на одном инструменте и не поддаваться синтезу на другом. Лет 10 назад появился язык SystemC – это С++ с огромным набором библиотек. С одной стороны, любая программа на SystemC может транслироваться стандартными трансляторами С++ , есть удобные средства потактного моделирования и приличные средства генгерации RTL, с другой стороны, требование совместимости с С++ не прошло даром – если в базовом языке нет средств описания параллелизма и конвейеризации, их приходится добавлять весьма искусственными приемами через приставные библиотеки. Буквально в прошлом году фирма Xilinx выпустила продукт Vivado, в рекламе которого утверждается, что он способен автоматически транслировать обычные программы на С/C++ в RTL промышленного качества.

Мы выполнили несколько экспериментов по использованию этого продукта, оказалось, что обещанной автоматизации там нет, пользователь должен писать на С, постоянно думая о том, как его код будет выглядеть в финальном RTL,  расставлять огромное количество прагм, причем не всегда очевидных.

Основной тезис доклада – такая важная область, как проектирование кристаллов, нуждается в специализированных языковых и инструментальных средствах, обеспечивающих  создание компактных и  легко читаемых программ, которые могут быть использованы как для симуляции, так и для генерации эффективного RTL. В докладе будут приведены примеры программ на языке HaSCoL (Hardware and Software Codesign Language), разработанном на кафедре системного программирования СПбГУ, и даны некоторые сравнительные характеристики.

Sergey Velder (ITMO): Alias graphs

(My summary – BM.) In the ITMO SEL work on automatic alias analysis, a new model has been developed: alias graphs, an abstraction of the object structure. This short talk will compare it to previously used approaches.

In honor of Niklaus Wirth’s 80-th birthday we are organizing a symposium at ETH on February 20, 2014. This is a full-day event with invited talks by:

• Vint Cerf
• Hans Eberlé
• Michael Franz
• me
• Carroll Morgan
• Martin Odersky
• Clemens Szyperski
• Niklaus Wirth himself

From the symposium’s web page:

Niklaus Wirth was a Professor of Computer Science at ETH Zürich, Switzerland, from 1968 to 1999. His principal areas of contribution were programming languages and methodology, software engineering, and design of personal workstations. He designed the programming languages Algol W, Pascal, Modula-2, and Oberon, was involved in the methodologies of structured programming and stepwise refinement, and designed and built the workstations Lilith and Ceres. He published several text books for courses on programming, algorithms and data structures, and logical design of digital circuits. He has received various prizes and honorary doctorates, including the Turing Award, the IEEE Computer Pioneer, and the Award for outstanding contributions to Computer Science Education.

Participation is free (including breaks, lunch and the concluding “Apéro”) but space is strictly limited and we expect to run out of seats quickly. So if you are interested (but only if you are certain to attend) please register right away.

Symposium page and access to registration form: here.

The first law of branching is: don’t. There is no other law.

The only sane way to develop software in a group, whether collocated or distributed, is to have a single branch (“trunk”) to which everyone commits changes, with constant running of the regression test suite to make sure that any breaking change is detected and corrected right away.

To allow branching, that is to say the emergence of separate lines of development with the expectation that they will be merged back later on, is to guarantee disaster. It is easy to diverge, but hard to converge; not only hard, but unpredictable. It can take days, weeks or more to reconcile changes and resolve conflicts, when the reason for the changes is no longer fresh in the developers’ memories, and the developers themselves may even no longer be there. Conflicts should be detected right away, and corrected immediately.

The EiffelStudio development never uses branches. A related development, EVE (Eiffel Verification Environment), maintained at ETH, includes all research tools, and is reconciled every Friday with the EiffelStudio trunk, so it is never allowed to diverge into a separate branch. Most other successful teams I know apply the first law of branching too. Solve conflicts before they have had the time to become conflicts.

A new version of my Touch of Class [1] programming textbook is available. It is not quite a new edition but more than just a new printing. All the typos that had been reported as of a few months ago have been corrected.

The format is also significantly smaller. This change is more than a trifle. When а  reader told me for the first time “really nice book, pity it is so heavy!”, I commiserated and did not pay much attention. After twenty people said that, and many more after them, including professors looking for textbooks for their introductory programming classes, I realized it was a big deal. The reason the book was big and heavy was not so much the size of the contents (876 is not small, but not outrageous for a textbook introducing all the fundamental concepts of programming). Rather, it is a technical matter: the text is printed in color, and Springer really wanted to do a good job, choosing thick enough paper that the colors would not seep though. In addition I chose a big font to make the text readable, resulting in a large format. In fact I overdid it; the font is bigger than necessary, even for readers who do not all have the good near-reading sight of a typical 19-year-old student.

We kept the color and the good paper,  but reduced the font size and hence the length and width. The result is still very readable, and much more portable. I am happy to make my contribution to reducing energy consumption (at ETH alone, think of the burden on Switzerland’s global energy bid of 200+ students carrying the book — as I hope they do — every morning on the buses, trains and trams crisscrossing the city!).

Springer also provides electronic access.

Touch of Class is the textbook developed on the basis of the Introduction to Programming course [2], which I have taught at ETH Zurich for the last ten years. It provides a broad overview of programming, starting at an elementary level (zeros and ones!) and encompassing topics not usually covered in introductory courses, even a short introduction to lambda calculus. You can get an idea of the style of coverage of such topics by looking up the sample chapter on recursion at touch.ethz.ch. Examples of other topics covered include a general introduction to software engineering and software tools. The presentation uses full-fledged object-oriented concepts (including inheritance, polymorphism, genericity) right from the start, and Design by Contract throughout. Based on the “inverted curriculum” ideas on which I published a number of articles, it presents students with a library of reusable components, the Traffic library for graphical modeling of traffic in a city, and builds on this infrastructure both to teach students abstraction (reusing code through interfaces including contracts) and to provide them models of high-quality code for imitation and inspiration.

For more details see the article on this blog that introduced the book when it was first published [3].

References

[1] Bertrand Meyer, Touch of Class: An Introduction to Programming Well Using Objects and Contracts, Springer Verlag, 2nd printing, 2013. The Amazon page is here. See the book’s own page (with slides and other teaching materials, sample chapter etc.) here. (Also available in Russian, see here.)

[2] Einführung in die Programmierung (Introduction to Programming) course, English course page here.

[3] Touch of Class published, article on this blog, 11 August 2009, see [1] here.

I have mentioned this paper before but as a draft. It has now been accepted by ACM’s Computing Surveys and is scheduled to appear in September 2014; the current text, revised from the previous version, is available [1].

Here is the abstract:

Software verification has emerged as a key concern for ensuring the continued progress of information technology. Full verification generally requires, as a crucial step, equipping each loop with a “loop invariant”. Beyond their role in verification, loop invariants help program understanding by providing fundamental insights into the nature of algorithms. In practice, finding sound and useful invariants remains a challenge. Fortunately, many invariants seem intuitively to exhibit a common flavor. Understanding these fundamental invariant patterns could therefore provide help for understanding and verifying a large variety of programs.

We performed a systematic identification, validation, and classification of loop invariants over a range of fundamental algorithms from diverse areas of computer science. This article analyzes the patterns, as uncovered in this study,governing how invariants are derived from postconditions;it proposes a taxonomy of invariants according to these patterns, and presents its application to the algorithms reviewed. The discussion also shows the need for high-level specifications based on “domain theory”. It describes how the invariants and the corresponding algorithms have been mechanically verified using an automated program prover; the proof source files are available. The contributions also include suggestions for invariant inference and for model-based specification.

Reference

[1] Carlo Furia, Bertrand Meyer and Sergey Velder: Loop invariants: Analysis, Classification and Examples, in ACM Computing Surveys, to appear in September 2014, preliminary text available here.

This report continues the series of ICSE 2013 article previews (see the posts of these last few days, other than the DOSE announcement), but is different from its predecessors since it talks about a paper from our group at ETH, so you should not expect any dangerously delusional,  disingenuously dubious or downright deceptive declaration or display of dispassionate, disinterested, disengaged describer’s detachment.

The paper [1] (mentioned on this blog some time ago) is entitled How good are software specifications? and will be presented on Wednesday by Nadia Polikarpova. The basic result: stronger specifications, which capture a more complete part of program functionality, cause only a modest increase in specification effort, but the benefits are huge; in particular, automatic testing finds twice as many faults (“bugs” as recently reviewed papers call them).

Strong specifications are specifications that go beyond simple contracts. A straightforward example is a specification of a push operation for stacks; in EiffelBase, the basic Eiffel data structure library, the contract’s postcondition will read

item = x                                          /A/
count = old count + 1

where x is the element being pushed, item the top of the stack and count the number of elements. It is of course sound, since it states that the element just pushed is now the new top of the stack, and that there is one more element; but it is also  incomplete since it says nothing about the other elements remaining as they were; an implementation could satisfy the contract and mess up with these elements. Using “complete” or “strong” preconditions, we associate with the underlying domain a theory [2], or “model”, represented by a specification-only feature in the class, model, denoting a sequence of elements; then it suffices (with the convention that the top is the first element of the model sequence, and that “+” denotes concatenation of sequences) to use the postcondition

model = <x> + old model         /B/

which says all there is to say and implies the original postconditions /A/.

Clearly, the strong contracts, in the  /B/ style, are more expressive [3, 4], but they also require more specification effort. Are they worth the trouble?

The paper explores this question empirically, and the answer, at least according to the criteria used in the study, is yes.  The work takes advantage of AutoTest [5], an automatic testing framework which relies on the contracts already present in the software to serve as test oracles, and generates test cases automatically. AutoTest was applied to both to the classic EiffelBase, with classic partial contracts in the /A/ style, and to the more recent EiffelBase+ library, with strong contracts in the /B/ style. AutoTest is for Eiffel programs; to check for any language-specificity in the results the work also included testing a smaller set of classes from a C# library, DSA, for which a student developed a version (DSA+) equipped with strong model-based contracts. In that case the testing tool was Microsoft Research’s Pex [7]. The results are similar for both languages: citing from the paper, “the fault rates are comparable in the C# experiments, respectively 6 . 10^-3 and 3 . 10^-3 . The fault complexity is also qualitatively similar.”

The verdict on the effect of strong specifications as captured by automated testing is clear: the same automatic testing tools applied to the versions with strong contracts yield twice as many real faults. The term “real fault” comes from excluding spurious cases, such as specification faults (wrong specification, right implementation), which are a phenomenon worth studying but should not count as a benefit of the strong specification approach. The paper contains a detailed analysis of the various kinds of faults and the corresponding empirically determined measures. This particular analysis is for the Eiffel code, since in the C#/Pex case “it was not possible to get an evaluation of the faults by the original developers“.

In our experience the strong specifications are not that much harder to write. The paper contains a precise measure: about five person-weeks to create EiffelBase+, yielding an “overall benefit/effort ratio of about four defects detected per person-day“. Such a benefit more than justifies the effort. More study of that effort is needed, however, because the “person” in the person-weeks was not just an ordinary programmer. True, Eiffel experience has shown that most programmers quickly get the notion of contract and start applying it; as the saying goes in the community, “if you can write an if-then-else, you can write a contract”. But we do not yet have significant evidence of whether that observation extends to model-based contracts.

Model-based contracts (I prefer to call them “theory-based” because “model” means so many other things, but I do not think I will win that particular battle) are, in my opinion, a required component of the march towards program verification. They are the right compromise between simple contracts, which have proved to be attractive to many practicing programmers but suffer from incompleteness, and full formal specification à la Z, which say everything but require too much machinery. They are not an all-or-nothing specification technique but a progressive one: programmers can start with simple contracts, then extend and refine them as desired to yield exactly the right amount of precision and completeness appropriate in any particular context. The article shows that the benefits are well worth the incremental effort.

According to the ICSE program the talk will be presented in the formal specification session, Wednesday, May 22, 13:30-15:30, Grand Ballroom C.

References

[1] Nadia Polikarpova, Carlo A. Furia, Yu Pei, Yi Wei and Bertrand Meyer: What Good Are Strong Specifications?, to appear in ICSE 2013 (Proceedings of 35th International Conference on Software Engineering), San Francisco, May 2013, draft available here.

[2] Bertrand Meyer: Domain Theory: the forgotten step in program verification, article on this blog, see here.

[3] Bernd Schoeller, Tobias Widmer and Bertrand Meyer: Making Specifications Complete Through Models, in Architecting Systems with Trustworthy Components, eds. Ralf Reussner, Judith Stafford and Clemens Szyperski, Lecture Notes in Computer Science, Springer-Verlag, 2006, available here.

[4] Nadia Polikarpova, Carlo Furia and Bertrand Meyer: Specifying Reusable Components, in Verified Software: Theories, Tools, Experiments (VSTTE ‘ 10), Edinburgh, UK, 16-19 August 2010, Lecture Notes in Computer Science, Springer Verlag, 2010, available here.

[5] Bertrand Meyer, Ilinca Ciupa, Andreas Leitner, Arno Fiva, Yi Wei and Emmanuel Stapf: Programs that Test Themselves, IEEE Computer, vol. 42, no. 9, pages 46-55, September 2009, also available here.

[6] Bertrand Meyer, Ilinca Ciupa, Andreas Leitner, Arno Fiva, Yi Wei and Emmanuel Stapf: Programs that Test Themselves, in IEEE Computer, vol. 42, no. 9, pages 46-55, September 2009, also available here.

[7] Nikolai Tillman and Peli de Halleux, Pex: White-Box Generation for .NET, in Tests And Proofs (TAP 2008), pp. 134-153.