Publication announcement: survey on requirements techniques, formal and non-formal

There is a new paper out, several years in the making:

The Role of Formalism in System Requirements
Jean-Michel Bruel, Sophie Ebersold, Florian Galinier, Manuel Mazzara, Alexander Naumchev, Bertrand Meyer
Computing Surveys (ACM), vol. 54, no. 5, June 2021, pages 1-36
DOI: https://doi.org/10.1145/3448975
Preprint available here.

The authors are from the Schaffhausen Institute of Technology in Switzerland, the University of Toulouse in France and Innopolis University in Russia. We make up a cross-institutional (and unofficial) research group which has for several years now been working on improving the state of software requirements, with both an engineering perspective and an interest in taking advantage of formal methods.

The article follows this combined formal-informal approach by reviewing the principal formal methods in requirements but also taking into consideration non-formal ones — including techniques widely used in industry, such as DOORS — and studying how they can be used in a more systematic way. It uses a significant example (a “Landing Gear System” or LGS for aircraft) to compare them and includes extensive tables comparing the approaches along a number of systematic criteria.

Here is the abstract:

A major determinant of the quality of software systems is the quality of their requirements, which should be both understandable and precise. Most requirements are written in natural language, which is good for understandability but lacks precision.

To make requirements precise, researchers have for years advocated the use of mathematics-based notations and methods, known as “formal.” Many exist, differing in their style, scope, and applicability.

The present survey discusses some of the main formal approaches and compares them to informal methods.The analysis uses a set of nine complementary criteria, such as level of abstraction, tool availability, and traceability support. It classifies the approaches into five categories based on their principal style for specifying requirements: natural-language, semi-formal, automata/graphs, mathematical, and seamless (programming-language-based). It includes examples from all of these categories, altogether 21 different approaches, including for example SysML, Relax, Eiffel, Event-B, and Alloy.

The review discusses a number of open questions, including seamlessness, the role of tools and education, and how to make industrial applications benefit more from the contributions of formal approaches.

For me, of course, this work is the continuation of a long-running interest in requirements and specifications and how to express them using the tools of mathematics, starting with a 1985 paper, still being cited today, with a strikingly similar title: On Formalism in Specifications.

Trivia: the “response to referees” (there were no fewer than eight of them!) after the first review took up 85 pages. Maybe not for the Guinness Book, but definitely a personal record. (And an opportunity to thank the referees for detailed comments that considerably helped shape the final form of the paper.)

Correction (20 July 2021): I just noted that I had forgotten to list myself among the authors! Not a sign of modesty (I don’t have any), more of absent-mindedness. Now corrected.

On beauty and software (online talk on Wednesday, 17 CET / 11 EDT / 8 PDT)

This Wednesday (still “tomorrow” as I am writing this), 10 March 2021, I am giving a talk on “The Beauty of Software” on the occasion of the graduation ceremony of the first students of the Schaffhausen Institute of Technology. The event starts at 17 Schaffhausen/Zurich/Paris etc. time (11 AM New York, 8 AM San Francisco) and my own talk, starting half an hour later, will take about one hour.

The talk is (surprise!) given online. Registration is free but required: you can find the registration form on the announcement page here.

The abstract appears below. It is rather ambitious-sounding and I cannot promise the talk will live up to the promise, but I feel it necessary at least to attempt some initial steps towards a better understanding of beauty in software, which might help understand beauty in general.

The Beauty of Software

Software runs the world and delivers riches. Every passionate software engineer or computer scientist is also attuned to another of its features: the study and practice of software construction reveal gems of utter beauty.

While the concept of beauty is most naturally associated with art, scientists and engineers of all fields often invoke it. Beauty is a strong guiding principle in searching for solutions to scientific and technical problems, and arbitrating between rival candidate solutions. The reaction is often instinctive: “What an elegant theory!” “This technique is too ugly to be a viable approach”.

What do such appeals to beauty really mean? Do they pertain to the same concept of beauty as found in nature and art? Is beauty only “in the eye of the beholder”, is it conditioned by cultural prejudices, or does it submit to an objective definition?

In this talk, an initial step towards a more extensive study of what beauty means for software, I will present a few artifacts from software engineering and computer science which I find strikingly beautiful and – at the risk of breaking the charm – analyze what might make them so. This analysis will lead to a tentative definition of a notion both alluring and elusive: beauty.

Some contributions

Science progresses through people taking advantage of others’ insights and inventions. One of the conditions that makes the game possible is that you acknowledge what you take. For the originator, it is rewarding to see one’s ideas reused, but frustrating when that happens without acknowledgment, especially when you are yourself punctilious about citing your own sources of inspiration.

I have started to record some concepts that are widely known and applied today and which I believe I originated in whole or in part, whether or not their origin is cited by those who took them. The list below is not complete and I may update it in the future. It is not a list of ideas I contributed, only of those fulfilling two criteria:

  • Others have built upon them.  (If there is an idea that I think is great but no one paid attention to it, the list does not include it.)
  • They have gained wide visibility.

There is a narcissistic aspect to this exercise and if people want to dismiss it as just showing I am full of myself so be it. I am just a little tired of being given papers to referee that state that genericity was invented by Java, that no one ever thought of refactoring before agile methods, and so on. It is finally time to state some facts.

Facts indeed: I back every assertion by precise references. So if I am wrong — i.e. someone preceded me — the claims of precedence can be refuted; if so I will update or remove them. All articles by me cited in this note are available (as downloadable PDFs) on my publication page. (The page is up to date until 2018; I am in the process of adding newer publications.)

Post-publication note: I have started to receive some comments and added them in a Notes section at the end; references to those notes are in the format [A].

Final disclaimer (about the narcissistic aspect): the exercise of collecting such of that information was new for me, as I do not usually spend time reflecting on the past. I am much more interested in the future and definitely hope that my next contributions will eclipse any of the ones listed below.

Programming concepts: substitution principle

Far from me any wish to under-represent the seminal contributions of Barbara Liskov, particularly her invention of the concept of abstract data type on which so much relies. As far as I can tell, however, what has come to be known as the “Liskov Substitution Principle” is essentially contained in the discussion of polymorphism in section 10.1 of in the first edition (Prentice Hall, 1988) of my book Object-Oriented Software Construction (hereafter OOSC1); for example, “the type compatibility rule implies that the dynamic type is always a descendant of the static type” (10.1.7) and “if B inherits from A, the set of objects that can be associated at run time with an entity [generalization of variable] includes instances of B and its descendants”.

Perhaps most tellingly, a key aspect of the substitution principle, as listed for example in the Wikipedia entry, is the rule on assertions: in a proper descendant, keep the invariant, keep or weaken the precondition, keep or strengthen the postcondition. This rule was introduced in OOSC1, over several pages in section 11.1. There is also an extensive discussion in the article Eiffel: Applying the Principles of Object-Oriented Design published in the Journal of Systems and Software, May 1986.

The original 1988 Liskov article cited (for example) in the Wikipedia entry on the substitution principle says nothing about this and does not in fact include any of the terms “assertion”, “precondition”, “postcondition” or “invariant”. To me this absence means that the article misses a key property of substitution: that the abstract semantics remain the same. (Also cited is a 1994 Liskov article in TOPLAS, but that was many years after OOSC1 and other articles explaining substitution and the assertion rules.)

Liskov’s original paper states that “if for each object o1 of type S there is an object o2 of type T such that for all programs P defined in terms of T, the behavior of P is unchanged when o1 is substituted for oz, then S is a subtype of T.” As stated, this property is impossible to satisfy: if the behavior is identical, then the implementations are the same, and the two types are identical (or differ only by name). Of course the concrete behaviors are different: applying the operation rotate to two different figures o1 and o2, whose types are subtypes of FIGURE and in some cases of each other, will trigger different algorithms — different behaviors. Only with assertions (contracts) does the substitution idea make sense: the abstract behavior, as characterized by preconditions, postconditions and the class invariants, is the same (modulo respective weakening and strengthening to preserve the flexibility of the different version). Realizing this was a major step in understanding inheritance and typing.

I do not know of any earlier (or contemporary) exposition of this principle and it would be normal to get the appropriate recognition.

Software design: design patterns

Two of the important patterns in the “Gang of Four” Design Patterns book (GoF) by Gamma et al. (1995) are the Command Pattern and the Bridge Pattern. I introduced them (under different names) in the following publications:

  • The command pattern appears in OOSC1 under the name “Undo-Redo” in section 12.2. The solution is essentially the same as in GoF. I do not know of any earlier exposition of the technique. See also notes [B] and [C].
  • The bridge pattern appears under the name “handle technique” in my book Reusable Software: The Base Component Libraries (Prentice Hall, 1994). It had been described several years earlier in manuals for Eiffel libraries. I do not know of an earlier reference. (The second edition of Object-Oriented Software Construction — Prentice Hall, 1997, “OOSC2” –, which also describes it, states that a similar technique is described in an article by Josef Gil and Ricardo Szmit at the TOOLS USA conference in the summer of 1994, i.e. after the publication of Reusable Software.)

Note that it is pointless to claim precedence over GoF since that book explicitly states that it is collecting known “best practices”, not introducing new ones. The relevant questions are: who, pre-GoF, introduced each of these techniques first; and which publications does the GoF cites as “prior art”  for each pattern. In the cases at hand, Command and Bridge, it does not cite OOSC1.

To be concrete: unless someone can point to an earlier reference, then anytime anyone anywhere using an interactive system enters a few “CTRL-Z” to undo commands, possibly followed by some “CTRL-Y” to redo them (or uses other UI conventions to achieve these goals), the software most likely relying on a technique that I first described in the place mentioned above.

Software design: Open-Closed Principle

Another contribution of OOSC1 (1988), section 2.3, reinforced in OOSC2 (1997) is the Open-Closed principle, which explained one of the key aspects of inheritance: the ability to keep a module both closed (immediately usable as is) and open to extension (through inheritance, preserving the basic semantics. I am mentioning this idea only in passing since in this case my contribution is usually recognized, for example in the Wikipedia entry.

Software design: OO for reuse

Reusability: the Case for Object-Oriented Design (1987) is, I believe, the first publication that clearly explained why object-oriented concepts were (and still are today — in Grady Booch’s words, “there is no other game in town”) the best answer to realize the goal of software construction from software components. In particular, the article:

  • Explains the relationship between abstract data types and OO programming, showing the former as the theoretical basis for the latter. (The CLU language at MIT originated from Liskov’s pioneering work on abstract data types, but was not OO in the full sense of the term, missing in particular a concept of inheritance.)
  • Shows that reusability implies bottom-up development. (Top-down refinement was the mantra at the time, and promoting bottom-up was quite a shock for many people.)
  • Explains the role of inheritance for reuse, as a complement to Parnas’s interface-based modular construction with information hiding.

Software design: Design by Contract

The contribution of Design by Contract is one that is widely acknowledged so I don’t have any point to establish here — I will just recall the essentials. The notion of assertion goes back to the work of Floyd, Hoare and Dijkstra in the sixties and seventies, and correctness-by-construction to Dijktra, Gries and Wirth, but Design by Contract is a comprehensive framework providing:

  • The use of assertions in an object-oriented context. (The notion of class invariant was mentioned in a paper by Tony Hoare published back in 1972.)
  • The connection of inheritance with assertions (as sketched above). That part as far as I know was entirely new.
  • A design methodology for quality software: the core of DbC.
  • Language constructs carefully seamed into the fabric of the language. (There were precedents there, but in the form of research languages such as Alphard, a paper design only, not implemented, and Euclid.)
  • A documentation methodology.
  • Support for testing.
  • Support for a consistent theory of exception handling (see next).

Design by Contract is sometimes taken to mean simply the addition of a few assertions here and there. What the term actually denotes is a comprehensive methodology with all the above components, tightly integrated into the programming language. Note in particular that preconditions and postconditions are not sufficient; in an OO context class invariants are essential.

Software design: exceptions

Prior to the Design by Contract work, exceptions were defined very vaguely, as something special you do outside of “normal” cases, but without defining “normal”. Design by Contract brings a proper perspective by defining these concepts precisely. This was explained in a 1987 article, Disciplined Exceptions ([86] in the list), rejected by ECOOP but circulated as a technical report; they appear again in detail in OOSC1 (sections 7.10.3 to 7.10.5).

Other important foundational work on exceptions, to which I know no real precursor (as usual I would be happy to correct any omission), addressed what happens to the outcome of an exception in a concurrent or distributed context. This work was done at ETH, in particular in the PhD theses  of B. Morandi and A. Kolesnichenko, co-supervised with S. Nanz. See the co-authored papers [345] and [363].

On the verification aspect of exceptions, see below.

Software design: refactoring

I have never seen a discussion of refactoring that refers to the detailed discussion of generalization in both of the books Reusable Software (1994, chapter 3) and Object Success (Prentice Hall, 1995, from page 122 to the end of chapter 6). These discussions describe in detail how, once a program has been shown to work, it should be subject to a posteriori design improvements. It presents several of the refactoring techniques (as they were called when the idea gained traction several years later), such as moving common elements up in the class hierarchy, and adding an abstract class as parent to concrete classes ex post facto.

These ideas are an integral part of the design methodology presented in these books (and again in OOSC2 a few later). It is beyond me why people would present refactoring (or its history, as in the Wikipedia entry on the topic) without referring to these publications, which were widely circulated and are available for anyone to inspect.

Software design: built-in documentation and Single-Product principle

Another original contribution was the idea of including documentation in the code itself and relying on tools to extract the documentation-only information (leaving implementation elements aside). The idea, described in detail in OOSC1 in 1988 (sections 9.4 and 9.5) and already mentioned in the earlier Eiffel papers, is that code should be self-complete, containing elements of various levels of abstraction; some of them describe implementation, but the higher-level elements describe specification, and are distinguished syntactically in such a way that tools can extract them to produce documentation at any desired level of abstraction.

The ideas were later applied through such mechanisms as JavaDoc (with no credit as far as I know). They were present in Eiffel from the start and the underlying principles, in particular the “Single Product principle” (sometimes “Self-Documentation principle”, and also generalized by J. Ostroff and R. Paige as “Single-Model principle”). Eiffel is the best realization of these principles thanks to:

  • Contracts (as mentioned above): the “contract view” of a class (called “short form” in earlier descriptions) removes the implementations but shows the relevant preconditions, postconditions and class invariants, given a precise and abstract specification of the class.
  • Eiffel syntax has a special place for “header comments”, which describe high-level properties and remain in the contract view.
  • Eiffel library class documentation has always been based on specifications automatically extracted from the actual text of the classes, guaranteeing adequacy of the documentation. Several formats are supported (including, from 1995 on, HTML, so that documentation can be automatically deployed on the Web).
  • Starting with the EiffelCase tool in the early 90s, and today with the Diagram Tool of EiffelStudio, class structures (inheritance and client relationships) are displayed graphically, again in an automatically extracted form, using either the BON or UML conventions.

One of the core benefits of the Single-Product principle is to guard against what some of my publications called the “Dorian Gray” syndrome: divergence of an implementation from its description, a critical problem in software because of the ease of modifying stuff. Having the documentation as an integral part of the code helps ensure that when information at some level of abstraction (specification, design, implementation) changes, the other levels will be updated as well.

Crucial in the approach is the “roundtripping” requirement: specifiers or implementers can make changes in any of the views, and have them reflected automatically in the other views. For example, you can graphically draw an arrow between two bubbles representing classes B and A in the Diagram Tool, and the code of B will be updated with “inherit A”; or you can add this Inheritance clause textually in the code of class B, and the diagram will be automatically updated with an arrow.

It is important to note how contrarian and subversive these ideas were at the time of their introduction (and still to some extent today). The wisdom was that you do requirements then design then implementation, and that code is a lowly product entirely separate from specification and documentation. Model-Driven Development perpetuates this idea (you are not supposed to modify the code, and if you do there is generally no easy way to propagate the change to the model.) Rehabilitating the code (a precursor idea to agile methods, see below) was a complete change of perspective.

I am aware of no precedent for this Single Product approach. The closest earlier ideas I can think of are in Knuth’s introduction of Literate Programming in the early eighties (with a book in 1984). As in the Single-product approach, documentation is interspersed with code. But the literate programming approach is (as presented) top-down, with English-like explanations progressively being extended with implementation elements. The Single Product approach emphasizes the primacy of code and, in terms of the design process, is very much yoyo, alternating top-down (from the specification to the implementation) and bottom-up (from the implementation to the abstraction) steps. In addition, a large part of the documentation, and often the most important one, is not informal English but formal assertions. I knew about Literate Programming, of course, and learned from it, but Single-Product is something else.

Software design: from patterns to components

Karine Arnout’s thesis at ETH Zurich, resulting in two co-authored articles ([255] and [257], showed that contrary to conventional wisdom a good proportion of the classical design patterns, including some of the most sophisticated, can be transformed into reusable components (indeed part of an Eiffel library). The agent mechanism (see below) was instrumental in achieving that result.

Programming, design and specification concepts: abstract data types

Liskov’s and Zilles’s ground-breaking 1974 abstract data types paper presented the concepts without a mathematical specification, using programming language constructs instead. A 1976 paper (number [3] in my publication list, La Description des Structures de Données, i.e. the description of data structures) was as far as I know one of the first to present a mathematical formalism, as  used today in presentations of ADTs. John Guttag was taking a similar approach in his PhD thesis at about the same time, and went further in providing a sound mathematical foundation, introducing in particular (in a 1978 paper with Jim Horning) the notion of sufficient completeness, to which I devoted a full article in this blog  (Are My Requirements Complete?) about a year ago. My own article was published in a not very well known journal and in French, so I don’t think it had much direct influence. (My later books reused some of the material.)

The three-level description approach of that article (later presented in English for an ACM workshop in the US in 1981, Pingree Park, reference [28]) is not well known but still applicable, and would be useful to avoid frequent confusions between ADT specifications and more explicit descriptions.

When I wrote my 1976 paper, I was not aware of Guttag’s ongoing work (only of the Liskov and Zilles paper), so the use of a mathematical framework with functions and predicates on them was devised independently. (I remember being quite happy when I saw what the axioms should be for a queue.) Guttag and I both gave talks at a workshop organized by the French programming language interest group in 1977 and it was fun to see that our presentations were almost identical. I think my paper still reads well today (well, if you read French). Whether or not it exerted direct influence, I am proud that it independently introduced the modern way of thinking of abstract data types as characterized by mathematical functions and their formal (predicate calculus) properties.

Language mechanisms: genericity with inheritance

Every once in a while I get to referee a paper that starts “Generics, as introduced in Java…” Well, let’s get some perspective here. Eiffel from its introduction in 1985 combined genericity and inheritance. Initially, C++ users and designers claimed that genericity was not needed in an OO context and the language did not have it; then they introduced template. Initially, the designers of Java claimed (around 1995) that genericity was not needed, and the language did not have it; a few years later Java got generics. Initially, the designers of C# (around 1999) claimed that genericity was not needed, and the language did not have it; a few years later C# and .NET got generics.

Genericity existed before Eiffel of course; what was new was the combination with inheritance. I had been influenced by work on generic modules by a French researcher, Didier Bert, which I believe influenced the design of Ada as well; Ada was the language that brought genericity to a much broader audience than the somewhat confidential languages that had such a mechanism before. But Ada was not object-oriented (it only had modules, not classes). I was passionate about object-oriented programming (at a time when it was generally considered, by the few people who had heard of it as an esoteric, academic pursuit). I started — in the context of an advanced course I was teaching at UC Santa Barbara — an investigation of how the two mechanisms relate to each other. The results were a paper at the first OOPSLA in 1986, Genericity versus Inheritance, and the design of the Eiffel type system, with a class mechanism, inheritance (single and multiple), and genericity, carefully crafted to complement each other.

With the exception of a Trellis-Owl, a  design from Digital Equipment Corporation also presented at the same OOPSLA (which never gained significant usage), there were no other OO languages with both mechanisms for several years after the Genericity versus Inheritance paper and the implementation of genericity with inheritance in Eiffel available from 1986 on. Eiffel also introduced, as far as I know, the concept of constrained genericity, the second basic mechanism for combining genericity with inheritance, described in Eiffel: The Language (Prentice Hall, 1992, section 10.8) and discussed again in OOSC2 (section 16.4 and throughout). Similar mechanisms are present in many languages today.

It was not always so. I distinctly remember people bringing their friends to our booth at some conference in the early nineties, for the sole purpose of having a good laugh with them at our poster advertising genericity with inheritance. (“What is this thing they have and no one else does? Generi-sissy-tee? Hahaha.”). A few years later, proponents of Java were pontificating that no serious language needs generics.

It is undoubtedly part of of the cycle of invention (there is a Schopenhauer citation on this, actually the only thing from Schopenhauer’s philosophy that I ever understood [D]) that people at some point will laugh at you; if it did brighten their day, why would the inventor deny them one of the little pleasures of life? But in terms of who laughs last, along the way C++ got templates, Java got generics, C# finally did too, and nowadays all typed OO languages have something of the sort.

Language mechanisms: multiple inheritance

Some readers will probably have been told that multiple inheritance is a bad thing, and hence will not count it as a contribution, but if done properly it provides a major abstraction mechanism, useful in many circumstances. Eiffel showed how to do multiple inheritance right by clearly distinguishing between features (operations) and their names, defining a class as a finite mapping between names and features, and using renaming to resolve any name clashes.

Multiple inheritance was made possible by an implementation innovation: discovering a technique (widely imitated since, including in single-inheritance contexts) to implement dynamic binding in constant time. It was universally believed at the time that multiple inheritance had a strong impact on performance, because dynamic binding implied a run-time traversal of the class inheritance structure, already bad enough for single inheritance where the structure is a tree, but prohibitive with multiple inheritance for which it is a directed acyclic graph. From its very first implementation in 1986 Eiffel used what is today known as a virtual table technique which guarantees constant-time execution of routine (method) calls with dynamic binding.

Language mechanisms: safe GC through strong static typing

Simula 67 implementations did not have automatic garbage collection, and neither had implementations of C++. The official excuse in the C++ case was methodological: C programmers are used to exerting manual control of memory usage. But the real reason was a technical impossibility resulting from the design of the language: compatibility with C precludes the provision of a good GC.

More precisely, of a sound and complete GC. A GC is sound if it will only reclaim unreachable objects; it is complete if it will reclaim all unreachable objects. With a C-based language supporting casts (e.g. between integers and pointers) and pointer arithmetic, it is impossible to achieve soundness if we aim at a reasonable level of completeness: a pointer can masquerade as an integer, only to be cast back into a pointer later on, but in the meantime the garbage collector, not recognizing it as a pointer, may have wrongly reclaimed the corresponding object. Catastrophe.

It is only possible in such a language to have a conservative GC, meaning that it renounces completeness. A conservative GC will treat as a pointer any integer whose value could possibly be a pointer (because it lies between the bounds of the program’s data addresses in memory). Then, out of precaution, the GC will refrain from reclaiming the objects at these addresses even if they appear unreachable.

This approach makes the GC sound but it is only a heuristics, and it inevitably loses completeness: every once in a while it will fail to reclaim some dead (unreachable) objects around. The result is a program with memory leaks — usually unacceptable in practice, particularly for long-running or continuously running programs where the leaks inexorably accumulate until the program starts thrashing then runs out of memory.

Smalltalk, like Lisp, made garbage collection possible, but was not a typed language and missed on the performance benefits of treating simple values like integers as a non-OO language would. Although in this case I do not at the moment have a specific bibliographic reference, I believe that it is in the context of Eiffel that the close connection between strong static typing (avoiding mechanisms such as casts and pointer arithmetic) and the possibility of sound and complete garbage collection was first clearly explained. Explained in particular around 1990 in a meeting with some of the future designers of Java, which uses a similar approach, also taken over later on by C#.

By the way, no one will laugh at you today for considering garbage collection as a kind of basic human right for programmers, but for a long time the very idea was quite sulfurous, and advocating it subjected you to a lot of scorn. Here is an extract of the review I got when I submitted the first Eiffel paper to IEEE Transactions on Software Engineering:

Systems that do automatic garbage collection and prevent the designer from doing his own memory management are not good systems for industrial-strength software engineering.

Famous last words. Another gem from another reviewer of the same paper:

I think time will show that inheritance (section 1.5.3) is a terrible idea.

Wow! I wish the anonymous reviewers would tell us what they think today. Needless to say, the paper was summarily rejected. (It later appeared in the Journal of Systems and Software — as [82] in the publication list — thanks to the enlightened views of Robert Glass, the founding editor.)

Language mechanisms: void safety

Void safety is a property of a language design that guarantees the absence of the plague of null pointer dereferencing.

The original idea came (as far as I know) from work at Microsoft Research that led to the design of a research language called C-omega; the techniques were not transferred to a full-fledged programming language. Benefiting from the existence of this proof of concept, the Eiffel design was reworked to guarantee void safety, starting from my 2005 ECOOP keynote paper (Attached Types) and reaching full type safety a few years later. This property of the language was mechanically proved in a 2016 ETH thesis by A. Kogtenkov.

Today all significant Eiffel development produces void-safe code. As far as I know this was a first among production programming languages and Eiffel remains the only production language to provide a guarantee of full void-safety.

This mechanism, carefully crafted (hint: the difficult part is initialization), is among those of which I am proudest, because in the rest of the programming world null pointer dereferencing is a major plague, threatening at any moment to crash the execution of any program that uses pointers of references. For Eiffel users it is gone.

Language mechanisms: agents/delegates/lambdas

For a long time, OO programming languages did not have a mechanism for defining objects wrapping individual operations. Eiffel’s agent facility was the first such mechanism or among the very first together the roughly contemporaneous but initially much more limited delegates of C#. The 1999 paper From calls to agents (with P. Dubois, M. Howard, M. Schweitzer and E. Stapf, [196] in the list) was as far as I know the first description of such a construct in the scientific literature.

Language mechanisms: concurrency

The 1993 Communications of the ACM paper on Systematic Concurrent Object-Oriented Programming [136] was certainly not the first concurrency proposal for OO programming (there had been pioneering work reported in particular in the 1987 book edited by Tokoro and Yonezawa), but it innovated in offering a completely data-race-free model, still a rarity today (think for example of the multi-threading mechanisms of dominant OO languages).

SCOOP, as it came to be called, was implemented a few years later and is today a standard part of Eiffel.

Language mechanisms: selective exports

Information hiding, as introduced by Parnas in his two seminal 1972 articles, distinguishes between public and secret features of a module. The first OO programming language, Simula 67, had only these two possibilities for classes and so did Ada for modules.

In building libraries of reusable components I realized early on that we need a more fine-grained mechanism. For example if class LINKED_LIST uses an auxiliary class LINKABLE to represent individual cells of a linked list (each with a value field and a “right” field containing a reference to another LINKABLE), the features of LINKABLE (such as the operation to reattach the “right” field) should not be secret, since LINKED_LIST needs them; but they should also not be generally public, since we do not want arbitrary client objects to mess around with the internal structure of the list. They should be exported selectively to LINKED_LIST only. The Eiffel syntax is simple: declare these operations in a clause of the class labeled “feature {LINKED_LIST}”.

This mechanism, known as selective exports, was introduced around 1989 (it is specified in full in Eiffel: The Language, from 1992, but was in the Eiffel manuals earlier). I think it predated the C++ “friends” mechanism which serves a similar purpose (maybe someone with knowledge of the history of C++ has the exact date). Selective exports are more general than the friends facility and similar ones in other OO languages: specifying a class as a friend means it has access to all your internals. This solution is too coarse-grained. Eiffel’s selective exports make it possible to define the specific export rights of individual operations (including attributes/fields) individually.

Language mechanisms and implementation: serialization and schema evolution

I did not invent serialization. As a student at Stanford in 1974 I had the privilege, at the AI lab, of using SAIL (Stanford Artificial Intelligence Language). SAIL was not object-oriented but included many innovative ideas; it was far ahead of its time, especially in terms of the integration of the language with (what was not yet called) its IDE. One feature of SAIL with which one could fall in love at first sight was the possibility of selecting an object and having its full dependent data structure (the entire subgraph of the object graph reached by following references from the object, recursively) stored into a file, for retrieval at the next section. After that, I never wanted again to live without such a facility, but no other language and environment had it.

Serialization was almost the first thing we implemented for Eiffel: the ability to write object.store (file) to have the entire structure from object stored into file, and the corresponding retrieval operation. OOSC1 (section 15.5) presents these mechanisms. Simula and (I think) C++ did not have anything of the sort; I am not sure about Smalltalk. Later on, of course, serialization mechanisms became a frequent component of OO environments.

Eiffel remained innovative by tackling the difficult problems: what happens when you try to retrieve an object structure and some classes have changed? Only with a coherent theoretical framework as provided in Eiffel by Design by Contract can one devise a meaningful solution. The problem and our solutions are described in detail in OOSC2 (the whole of chapter 31, particularly the section entitled “Schema evolution”). Further advances were made by Marco Piccioni in his PhD thesis at ETH and published in joint papers with him and M. Oriol, particularly [352].

Language mechanisms and implementation: safe GC through strong static typing

Simula 67 (if I remember right) did not have automatic garbage collection, and neither had C++ implementations. The official justification in the case of C++ was methodological: C programmers are used to exerting manual control of memory usage. But the real obstacle was technical: compatibility with C makes it impossible to have a good GC. More precisely, to have a sound and complete GC. A GC is sound if it will only reclaim unreachable objects; it is complete if it will reclaim all unreachable objects. With a C-based language supporting casts (e.g. between integers and pointers) and pointer arithmetic, it is impossible to achieve soundness if we aim at a reasonable level of completeness: a pointer can masquerade as an integer, only to be cast back into a pointer later on, but in the meantime the garbage collector, not recognizing it as a pointer, may have wrongly reclaimed the corresponding object. Catastrophe. It is only possible in such a language to have a conservative GC, which will treat as a pointer any integer whose value could possibly be a pointer (because its value lies between the bounds of the program’s data addresses in memory). Then, out of precaution, it will not reclaim the objects at the corresponding address. This approach makes the GC sound but it is only a heuristics, and it may be over-conservative at times, wrongly leaving dead (i.e. unreachable) objects around. The result is, inevitably, a program with memory leaks — usually unacceptable in practice.

Smalltalk, like Lisp, made garbage collection possible, but was not a typed language and missed on the performance benefits of treating simple values like integers as a non-OO language would. Although in this case I do not at the moment have a specific bibliographic reference, I believe that it is in the context of Eiffel that the close connection between strong static typing (avoiding mechanisms such as casts and pointer arithmetic) and the possibility of sound and complete garbage collection was first clearly explained. Explained in particular to some of the future designers of Java, which uses a similar approach, also taken over later on by C#.

By the way, no one will laugh at you today for considering garbage collection as a kind of basic human right for programmers, but for a long time it was quite sulfurous. Here is an extract of the review I got when I submitted the first Eiffel paper to IEEE <em>Transactions on Software Engineering:

Software engineering: primacy of code

Agile methods are widely and properly lauded for emphasizing the central role of code, against designs and other non-executable artifacts. By reading the agile literature you might be forgiven for believing that no one brought up that point before.

Object Success (1995) makes the argument very clearly. For example, chapter 3, page 43:

Code is to our industry what bread is to a baker and books to a writer. But with the waterfall code only appears late in the process; for a manager this is an unacceptable risk factor. Anyone with practical experience in software development knows how many things can go wrong once you get down to code: a brilliant design idea whose implementation turns out to require tens of megabytes of space or minutes of response time; beautiful bubbles and arrows that cannot be implemented; an operating system update, crucial to the project which comes five weeks late; an obscure bug that takes ages to be fixed. Unless you start coding early in the process, you will not be able to control your project.

Such discourse was subversive at the time; the wisdom in software engineering was that you need to specify and design a system to death before you even start coding (otherwise you are just a messy “hacker” in the sense this word had at the time). No one else in respectable software engineering circles was, as far as I know, pushing for putting code at the center, the way the above extract does.

Several years later, agile authors started making similar arguments, but I don’t know why they never referenced this earlier exposition, which still today I find not too bad. (Maybe they decided it was more effective to have a foil, the scorned Waterfall, and to claim that everyone else before was downplaying the importance of code, but that was not in fact everyone.)

Just to be clear, Agile brought many important ideas that my publications did not anticipate; but this particular one I did.

Software engineering: the roles of managers

Extreme Programming and Scrum have brought new light on the role of managers in software development. Their contributions have been important and influential, but here too they were for a significant part prefigured by a long discussion, altogether two chapters, in Object Success (1995).

To realize this, it is enough to read the titles of some of the sections in those chapters, describing roles for managers (some universal, some for a technical manager): “risk manager”, “interface with the rest of the world” (very scrummy!), “protector of the team’s sanity”, “method enforcer” (think Scrum Master), “mentor and critic”. Again, as far as I know, these were original thoughts at the time; the software engineering literature for the most part did not talk about these issues.

Software engineering: outsourcing

As far as I know the 2006 paper Offshore Development: The Unspoken Revolution in Software Engineering was the first to draw attention, in the software engineering community, to the peculiar software engineering challenges of distributed and outsourced development.

Software engineering: automatic testing

The AutoTest project (with many publications, involving I. Ciupa, A. Leitner, Y. Wei, M. Oriol, Y. Pei, M. Nordio and others) was not the first to generate tests automatically by creating numerous instances of objects and calling applicable operations (it was preceded by Korat at MIT), but it was the first one to apply this concept with Design by Contract mechanisms (without which it is of little practical value, since one must still produce test oracles manually) and the first to be integrated in a production environment (EiffelStudio).

Software engineering: make-less system building

One of the very first decisions in the design of Eiffel was to get rid of Make files.

Feldman’s Make had of course been a great innovation. Before Make, programmers had to produce executable systems manually by executing sequences of commands to compile and link the various source components. Make enabled them to instead  to define dependencies between components in a declarative way, resulting in a partial order, and then performed a topological sort to produce the sequence of comments. But preparing the list of dependencies remains a tedious task, particularly error-prone for large systems.

I decided right away in the design of Eiffel that we would never force programmers to write such dependencies: they would be automatically extracted from the code, through an exhaustive analysis of the dependencies between modules. This idea was present from the very the first Eiffel report in 1985 (reference [55] in the publication list): Eiffel programmers never need to write a Make file or equivalent (other than for non-Eiffel code, e.g. C or C++, that they want to integrate); they just click a Compile button and the compiler figures out the steps.

Behind this approach was a detailed theoretical analysis of possible relations between modules in software development (in many programming languages), published as the “Software Knowledge Base” at ICSE in 1985. That analysis was also quite instructive and I would like to return to this work and expand it.

Educational techniques: objects first

Towards an Object-Oriented Curriculum ( TOOLS conference, August 1993, see also the shorter JOOP paper in May of the same year) makes a carefully argued case for what was later called the Objects First approach to teaching programming. I would be interested to know if there are earlier publications advocating starting programming education with an OO language.

The article also advocated for the “inverted curriculum”, a term borrowed from work by Bernie Cohen about teaching electrical engineering. It was the first transposition of this concept to software education. In the article’s approach, students are given program components to use, then little by little discover how they are made. This technique met with some skepticism and resistance since the standard approach was to start from the very basics (write trivial programs), then move up. Today, of course, many introductory programming courses similarly provide students from day one with a full-fledged set of components enabling them to produce significant programs.

More recent articles on similar topics, taking advantage of actual teaching experience, are The Outside-In Method of Teaching Programming (2003) and The Inverted Curriculum in Practice (at ICSE 2006, with Michela Pedroni). The culmination of that experience is the textbook Touch of Class from 2009.

Educational techniques: Distributed Software Projects

I believe our team at ETH Zurich (including among others M. Nordio, J. Tschannen, P. Kolb and C. Estler and in collaboration with C. Ghezzi, E. Di Nitto and G. Tamburrelli at Politecnico di Milano, N. Aguirre at Rio Cuarto and many others in various universities) was the first to devise,  practice and document on a large scale (see publications and other details here) the idea of an educational software project conducted in common by student groups from different universities. It yielded a wealth of information on distributed software development and educational issues.

Educational techniques: Web-based programming exercises

There are today a number of cloud-based environments supporting the teaching of programming by enabling students to compile and test their programs on the Web, benefiting from a prepared environment (so that they don’t have to download any tools or prepare control files) and providing feedback. One of the first — I am not sure about absolute precedence — and still a leading one, used by many universities and applicable to many programming languages, is Codeboard.

The main developer, in my chair at ETH Zurich, was Christian Estler, supported in particular by M. Nordio and M. Piccioni, so I am only claiming a supporting role here.

Educational techniques: key CS/SE concepts

The 2001 paper Software Engineering in the Academy did a good job, I think, of defining the essential concepts to teach in a proper curriculum (part of what Jeannette Wing’s 2006 paper called Computational Thinking).

Program verification: agents (delegates etc.)

Reasoning about Function Objects (ICSE 2010, with M. Nordio, P. Müller and J. Tschannen) introduced verification techniques for objects representing functions (such as agents, delegates etc., see above) in an OO language. Not sure whether there were any such techniques before.

Specification languages: Z

The Z specification language has been widely used for formal development, particularly in the UK. It is the design of J-R Abrial. I may point out that I was a coauthor of the first publication on Z in English (1980),  describing a version that preceded the adaptation to a more graphical-style notation done later at Oxford. The first ever published description of Z, pertaining to an even earlier version, was in French, in my book Méthodes de Programmation (with C. Baudoin), Eyrolles, 1978, running over 15 pages (526-541), with the precise description of a refinement process.

Program verification: exceptions

Largely coming out of the PhD thesis of Martin Nordio, A Sound and Complete Program Logic for Eiffel (TOOLS 2009) introduces rules for dealing with exceptions in a Hoare-style verification framework.

Program verification: full library, and AutoProof

Nadia Polikarpova’s thesis at ETH, aided by the work of Carlo Furia and Julian Tschannen (they were the major contributors and my participation was less important), was as far as I know the first to produce a full functional verification of an actual production-quality reusable library. The library is EiffelBase 2, covering fundamental data structures.

AutoProof — available today, as a still experimental tool, through its Web interface, see here — relied on the AutoProof prover, built by the same team, and itself based on Microsoft Research’s Boogie and Z3 engines.

More

There are more concepts worthy of being included here, but for today I will stop here.

Notes

[A] One point of divergence between usual presentations of the substitution principle and the view in OOSC and my other publications is the covariance versus contravariance of routine argument types. It reflects a difference of views as to what the proper policy (both mathematically sound and practically usable) should be.

[B]  The GoF book does not cite OOSC for the command or bridge patterns. For the command pattern it cites (thanks to Adam Kosmaczewski for digging up the GoF text!) a 1985 SIGGRAPH paper by Henry Lieberman (There’s More to Menu Systems than Meets the Screen). Lieberman’s paper describes the notion of command object and mentions undoing in passing, but does not include the key elements of the command pattern (as explained in full in OOSC1), i.e. an abstract (deferred) command class with deferred procedures called (say) do_it and undo_it, then specific classes for each kind of command, each providing a specific implementation of those procedures, then a history list of commands supporting multiple-level undo and redo as explained in OOSC1. (Reading Lieberman’s paper with a 2021 perspective shows that it came tantalizingly close to the command pattern, but doesn’t get to it. The paper does talk about inheritance between command classes, but only to “define new commands as extensions to old commands”, not in the sense of a general template that can be implemented in many specific ways. And it does mention a list of objects kept around to enable recovery from accidental deletions, and states that the application can control its length, as is the case with a history list; but the objects in the list are not command objects, they are graphical and other objects that have been deleted.)

[C] Additional note on the command pattern: I vaguely remember seeing something similar to the OOSC1 technique in an article from a supplementary volume of the OOPSLA proceedings in the late eighties or early nineties, i.e. at the same time or slightly later, possibly from authors from Xerox PARC, but I have lost the reference.

[D] Correction: I just checked the source and learned that the actual Schopenhauer quote (as opposed to the one that is usually quoted) is different; it does not include the part about laughing. So much for my attempts at understanding philosophy.

 

The right forms of expression

If you want to know whether your_string has at least one upper-case character, you will write this in Eiffel:

if  ∃ c: your_string ¦ c.is_upper then

Such predicate-calculus boolean expressions, using a quantifier (“for all”) or (“there exists”) are becoming common in Eiffel code. They are particularly useful in Design by Contract assertions, making it possible to characterize deep semantic properties of the code and its data structures. For example a class invariant clause in a class I wrote recently states

from_lists_exist: ∀ tf: triples_from ¦ tf Void                        — [1]

meaning that all the elements, if any, of the list triples_from  are non-void (non-null). The notation is the exact one from mathematics. (Mathematical notation sometimes uses a dot in place of the bar, but the bar is clearer, particularly in an OO context where the dot has another use.)

Programming languages should support time-honored notations from mathematics. Reaching this goal has been a driving force in the evolution of Eiffel, but not as a concession to “featurism” (the gratuitous piling up of language feature upon feature). The language must remain simple and consistent; any new feature must find its logical place in the overall edifice.

The design of programming languages is a constant search for the right balance between rigor, simplicity, consistency, formal understanding, preservation of existing code, innovation and expressiveness. The design of Eiffel has understood the last of these criteria as implying support for established notations from mathematics, not through feature accumulation but by re-interpreting these notations in terms of the language’s fundamental concepts. A typical example is the re-interpretation of the standard mathematical notation a + b as as simply an operator-based form for the object-oriented call a.plus (b), obtained by declaring “+” as an operator alias for the function plus in the relevant classes. There are many more such cases in today’s Eiffel. Quantifier expressions using and  are the latest example.

 They are not a one-of-a-kind trick but just as a different syntax form for loops. Expressed in a more verbose form, the only one previously available, [1] would be:

across triples_from is tf all tf /= Void end                         — [2]

It is interesting to walk back the history further. [2] is itself a simplification of

across triples_from as tf all tf.item /= Void end               — [3]

where the “.item” has a good reason for being there, but that reason is irrelevant to a beginner. The earlier use of as in [3] is also the reason for the seemingly bizarre use of is in [2], which is only explainable by the backward compatibility criterion (code exists that uses as , which has a slightly different semantics from is), and will go away. But a few years ago the across loop variant did not exist and you would have had to write the above boolean expressions as

all_non_void (triples_from)

after defining a function

all_non_void (l: LIST [T]): BOOLEAN                                    — [4]
                         — Are all the elements of `l’, if any, non-void?
          local
pos: INTEGER
do
from
pos := l.index
l.start
Result := True
until not Result or l.after loop
l.forth
end
go_ith (pos)
end

The road traveled from [4] to [1] is staggering. As we introduced new notations in the history of Eiffel the reaction of the user community has sometimes been between cautious and negative. With the exception of a couple of quickly discarded ideas (such as the infamous and short-lived “!!” for creation), they were generally adopted widely because they simplify people’s life without adding undue complexity to the language. The key has been to avoid featurism and choose instead to provide two kinds of innovation:

  • Major conceptual additions, which elevate the level of abstraction of the language. A typical introduction was the introduction of agents, which provide the full power of functional programming in an object-oriented context; another was the SCOOP concurrency mechanism. There have been only a few such extensions, all essential.
  • Syntactical variants for existing concepts, allowing more concise forms obtained from traditional mathematical notation. The use of quantifier expressions as in [1] is the latest example.

Complaints of featurism still occasionally happen when people first encounter the new facilities, but they fade away quickly as people start using them. After writing a few expressions such as [1], no one wants to go back to any of the other forms.

These quantifier expressions using and , as well as the “” not-equal sign for what used to be (and still commonly is) written “/=”, rely on Unicode. Eiffel started out when ASCII was the law of the land. (Or 8-bit extended ASCII, which does not help much since the extensions are rendered differently in different locales, i.e. the same 8-bit character code may mean something different on French and Swedish texts.) In recent years, Eiffel has made a quiet transition to full Unicode support. (Such support extends to manifest strings and operators, not to identifiers. The decision, which could be revisited, has been to keep the ASCII-only  policy for identifiers to favor compatible use by programmers regardless of their mother tongues.) The use of Unicode considerably extends the expressive power of the language, in particular for scientific software which can — thanks to Eiffel’s mechanism for defining free operators — rely on advanced mathematical notations.

Unicode is great, but I hear the question: how in the world can we enter the corresponding symbols, since our keyboards are still ASCII plus some extensions?

It would be tedious to have to select from a list of special symbols (as you do when inserting a mathematical symbol in Microsoft Word or, for that matter, as I did when inserting the phrase “ and ” in the preceding paragraph using WordPress).

The answer lies in the interplay between the language and the development environment. EiffelStudio, like other modern IDEs, includes an automatic completion mechanism which lets you enter the beginning of a construct and will take care of filling in the rest. Already useful for complex structures (if you type “if” the tools will create the entire “if then else end” conditional structure for you to fill in), automatic completion will take care of inserting the appropriate Unicode symbols for you. Type for example “across”,  then CTRL-Space to trigger completion, and the choices will include the “∀” and “” forms. You can see below how this works:

across_all

Programming languages can be at the same time simple, easy to learn, consistent, and expressive. Start using quantifiers now!

Acknowledgments to the Ecma Technical Committee on Eiffel and the Eiffel Software team, particularly Alexander Kogtenkov (see his blog post here) and (for the completion mechanism and its animated illustration above) Jocelyn Fiat.

Questionnaire on deployed, formally verified systems

A group of us is preparing a survey on systems that have been both formally verified and deployed for actual use. To make sure we do not forget any important development, we have devised a questionnaire. If you have experience with such a system, please help by filling the questionnaire. It only includes a few questions and takes a few minutes to fill. You can find it here.

Please also bring it to the attention of others who might have relevant information.

Thanks in advance!

New video lecture: distances, invariants and recursion

I have started a new series of video lectures, which I call “Meyer’s Object-Oriented Classes” (MOOC). The goal is to share insights I have gained over the years on various aspects of programming and software engineering. Many presentations are focused on one area, such as coding, design, analysis, theoretical computer science (even there you find a division between “Theory A”, i.e. complexity, Turing machines and the like, and “Theory B”, i.e. semantics, type theory etc.), software project management, concurrency… I have an interest in all and try to explain connections.

 

The first lecture describes the edit distance (Levenshtein) algorithm, explains its correctness by introducing the loop invariant, expands on that notion, then shows a recursive version, explores the connection with the original version (it’s the invariant), and probes further into another view of recursive computations, leading to the concept of dynamic programming.

The videos are on YouTube and can be accessed from bertrandmeyer.com/levenshtein. (The general page for all lectures is at bertrandmeyer.com/mooc.)

The lecture is recorded in four segments of about 15 minutes each. In the future I will limit myself to 8-10 minutes. In fact I may record this lecture again; for example it would be better if I had a live audience rather than talking to my screen, and in general the recording is somewhat low-tech, but circumstances command. Also, I will correct a few hiccups (at some point in the recording I notice a typo on a slide and fix it on the fly), but the content will remain the same.

Feedback is of course welcome. I hope to record about a lecture a week from now on.

New master program at SIT: Webinar tomorrow

The Schaffhausen Institute of Technology (SIT) is holding a Webinar tomorrow with a set of three talks by: Serguei Beloussov, founder of Acronis and president of SIT; Michael Widenius, CTO of MariaDB and creator of MySQL Server; and Mauro Pezzè, my colleague at SIT, who will present the new master program that we have just announced, combining CS/SE topics with management and marketing courses to train future technology leaders.

The talks are in the form of a Webinar, starting at 9 AM this Tuesday (9 June). You can find all the details on the corresponding SIT page at here.

Getting a program right, in nine episodes

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 := 1 ; j := n + 1

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 := 1 ; j := n + 1

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:knuth

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 := 1 ; j := n + 1

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  log(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.

tool_stack

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.

loop_strategy

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.

Call for suggestions: beauty

On April 29 in the early evening at the Schaffhausen Institute of Technology I will give a talk on “The Beauty of Software”, exploring examples of what makes some concepts, algorithms, data structures etc. produce a sense of esthetics. (Full abstract below.) I gave a first version at TOOLS last year but am revising and expanding the talk extensively.

I obviously have my own examples but am interested in more. If you have some that you feel should be considered for inclusion, perhaps because you experienced a “Wow!” effect when you encountered them, please tell me. I am only asking for names or general pointers, not an in-depth analysis (that’s my job). To avoid having my thunder stolen I would prefer that you alert me by email. I will give credit for examples not previously considered.

Thanks!

Abstract of the talk as published:

Scientists often cite the search for beauty as one of their primary guiding forces. Programming and software engineering offer an inexhaustible source of astoundingly beautiful ideas, from strikingly elegant algorithms and data structures to powerful principles of methodology and language design.

Defining beauty is elusive, but true beauty imposes itself in such a way as to remove any doubt. Drawing comparisons from art, literature and other endeavours. He will show a sample of ideas from all walks of software, directly understandable to a wide audience of non-software-experts, offering practical applications in technology that we use daily, and awe-inspiring in their simplicity and elegance.

LASER 2020 in Elba Island: DevOps, Microservices and more, first week of June

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.