Bertrand Meyer's technology+ blog

Let me start, dear reader of this blog, by probing your view of equality, and also of assignment. Two questions:  

• Is a value x always equal to itself? (For the seasoned programmers in the audience: I am talking about a value, as in mathematics, not an expression, as in programming, which could have a side effect.)
• In programming, if we consider an assignment

       x := y

and v is the value of y before that assignment (again, this little detour is to avoid bothering with side effects), is the value of x always equal to v after the assignment?  

Maybe I should include here one of these Web polls that one finds on newspaper sites, so that you can vote and compare your answer to the Wisdom of Crowds. My own vote is clear: yes to both. Equality is reflexive (every value is equal to itself, at any longitude and temperature, no excuses and no exceptions); and the purpose of assignment is to make the value of the target equal to the value of the source. Such properties are some of the last ramparts of civilization. If they go away, what else is left?  

754 enters the picture

Now come floating-point numbers and the famous IEEE “754” floating-point standard [1]. Because not all floating point operations yield a result usable as a floating-point number, the standard introduces a notion of “NaN”, Not a Number; certain operations involving floating-point numbers may yield a NaN. The term NaN does not denote a single value but a set of values, distinguished by their “payloads”.  

Now assume that the value of x is a NaN. If you use a programming language that supports IEEE 754 (as they all do, I think, today) the test in  

        if x = x then …  

is supposed to yield False. Yes, that is specified in the standard: NaN is never equal to NaN (even with the same payload); nor is it equal to anything else; the result of an equality comparison involving NaN will always be False.  

Assignment behavior is consistent with this: if y (a variable, or an expression with no side effect) has a NaN value, then after  

        x := y  

the test x = y will yield False. 

Before commenting further let me note the obvious: I am by no means a numerics expert; I know that IEEE 754 was a tremendous advance, and that it was designed by some of the best minds in the field, headed by Velvel Kahan who received a Turing Award in part for that success. So it is quite possible that I am about to bury myself in piles of ridicule. On the other hand I have also learned that (1) ridicule does not kill, so I am game; and more importantly (2) experts are often right but not always, and it is always proper to question their reasoning. So without fear let me not stop at the arguments that “the committee must have voted on this point and they obviously knew what they were doing” and “it is the standard and implemented on zillions of machines, you cannot change it now”. Both are true enough, but not an excuse to censor questions.  

What are the criteria?

The question is: compatibility with an existing computer standard is great, but what about compatibility with a few hundred years of mathematics? Reflexivity of equality  is something that we expect for any data type, and it seems hard to justify that a value is not equal to itself. As to assignment, what good can it be if it does not make the target equal to the source value?  

The question of assignment is particularly vivid in Eiffel because we express the expected abstract properties of programs in the form of contracts. For example, the following “setter” procedure may have a postcondition (expressed by the ensure clause):  

        set_x (v: REAL)
                        — Set the value of x (an attribute, also of type REAL) the value of v.
                do
                        …
                        x := v  
                ensure
                        x = v
                end  

   
If you call this procedure with a NaN argument for a compiler that applies IEEE 754 semantics, and monitor contracts at run time for testing and debugging, the execution will report a contract violation. This is very difficult for a programmer to accept.  

A typical example arises when you have an assignment to an item of an array of REAL values. Assume you are executing a [i] := x. In an object-oriented view of the world (as in Eiffel), this is considered simplified syntax  for the routine call a.put (x, i). The postcondition is that a [i] = x. It will be violated!  

The experts’ view

I queried a number of experts on the topic. (This is the opportunity to express my gratitude to members of the IFIP working group 2.5 on numerical software [2], some of the world’s top experts in the field, for their willingness to respond quickly and with many insights.) A representative answer, from Stuart Feldman, was:  

If I remember the debate correctly (many moons ago), NaN represents an indefinite value, so there is no reason to believe that the result of one calculation with unclear value should match that of another calculation with unclear value. (Different orders of infinity, different asymptotic approaches toward 0/0, etc.)  

Absolutely correct! Only one little detail, though: this is an argument against using the value True as a result of the test; but it is not an argument for using the value False! The exact same argument can be used to assert that the result should not be False:  

… there is no reason to believe that the result of one calculation with unclear value should not match that of another calculation with unclear value.  

Just as convincing! Both arguments complement each other: there is no compelling reason for demanding that the values be equal; and there is no compelling argument either to demand that they be different. If you ignore one of the two sides, you are biased.  

What do we do then?

The conclusion is not that the result should be False. The rational conclusion is that True and False are both unsatisfactory solutions. The reason is very simple: in a proper theory (I will sketch it below) the result of such a comparison should be some special undefined below; the same way that IEEE 754 extends the set of floating-point numbers with NaN, a satisfactory solution would extend the set of booleans with some NaB (Not a Boolean). But there is no NaB, probably because no one (understandably) wanted to bother, and also because being able to represent a value of type BOOLEAN on a single bit is, if not itself a pillar of civilization, one of the secrets of a happy life.  

If both True and False are unsatisfactory solutions, we should use the one that is the “least” bad, according to some convincing criterion . That is not the attitude that 754 takes; it seems to consider (as illustrated by the justification cited above) that False is somehow less committing than True. But it is not! Stating that something is false is just as much of a commitment as stating that it is true. False is no closer to NaB than True is. A better criterion is: which of the two possibilities is going to be least damaging to time-honored assumptions embedded in mathematics? One of these assumptions is the reflexivity of equality:  come rain or shine, x is equal to itself. Its counterpart for programming is that after an assignment the target will be equal to the original value of the source. This applies to numbers, and it applies to a NaN as well. 

Note that this argument does not address equality between different NaNs. The standard as it is states that a specific NaN, with a specific payload, is not equal to itself.  

What do you think?

A few of us who had to examine the issue recently think that — whatever the standard says at the machine level — a programming language should support the venerable properties that equality is reflexive and that assignment yields equality.

Every programming language should decide this on its own; for Eiffel we think this should be the specification. Do you agree?  

Some theory

For readers who like theory, here is a mathematical restatement of the observations above. There is nothing fundamentally new in this section, so if you do not like strange symbols you can stop here.  

The math helps explain the earlier observation that neither True nor False is more“committing” than the other. A standard technique (coming from denotational semantics) for dealing with undefinedness in basic data types, is to extend every data type into a lattice, with a partial order relation meaning “less defined than”. The lattice includes a bottom element, traditionally written “⊥” (pronounced “Bottom”) and a top element written ⊤ (“Top”). ⊥ represents an unknown value (not enough information) and ⊤ an error value (too much information). Pictorially, the lattice for natural numbers would look like this:  

The lattice of integers

On basic types, we always use very simple lattices of this form, with three kinds of element: ⊥, less than every other element; ⊤, larger than all other elements; and in-between all the normal values of the type, which for the partial order of interest are all equal. (No, this is not a new math in which all integers are equal. The order in question simply means “is less defined than”. Every integer is as defined as all other integers, more defined than ⊥, and less defined than ⊤.) Such lattices are not very exciting, but they serve as a starting point; lattices with more interesting structures are those applying to functions on such spaces — including functions of functions —, which represent programs.  

The modeling of floating-point numbers with NaN involves such a lattice; introducing NaN means introducing a ⊥ value. (Actually, one might prefer to interpret NaN as ⊤, but the reasoning transposes immediately.)  More accurately, since there are many NaN values, the lattice will look more like this:

The lattice of floats in IEEE 754

For simplicity we can ignore the variety of NaNs and consider a single ⊤.

Functions on lattices — as implemented by programs — should satisfy a fundamental property: monotonicity. A function f  is monotone (as in ordinary analysis) if, whenever x ≤ y, then f (x) ≤ f (y). Monotonicity is a common-sense requirement: we cannot get more information from less information. If we know less about x than about y, we cannot expect that any function (with a computer, any program) f will, out of nowhere, restore the missing information.  

Demanding monotonicity of all floating-point operations reflects this exigency of monotonicity: indeed, in IEEE 754, any arithmetic operation — addition, multiplication … — that has a NaN as one of its arguments must yield a Nan as its result. Great. Now for soundness we should also have such a property for boolean-valued operations, such as equality. If we had a NaB as the ⊥ of booleans, just like NaN is the ⊥ of floating-point numbers,  then the result of NaN = NaN would be NaB. But the world is not perfect and the IEEE 754 standard does not govern booleans. Somehow (I think) the designers thought of False as somehow less defined than True. But it is not! False is just as defined as True in the very simple lattice of booleans; according to the partial order, True and False are equal for the relevant partial order:

The lattice of booleans

Because any solution that cannot use a NaB will violate monotonicity and hence will be imperfect, we must resort to heuristic criteria. A very strong criterion in favor of choosing True is reflexivity — remaining compatible with a fundamental property of equality. I do not know of any argument for choosing False. 

The Spartan approach

There is, by the way, a technique that accepts booleans as we love them (with two values and no NaB) and achieves mathematical rigor. If operations involving NaNs  truly give us pimples, we can make any such operation trigger an exception. In the absence of ⊥ values,  this is an acceptable programming technique for representing undefinedness. The downside, of course, is that just about everywhere the program must be ready to handle the exception in some way. 

It is unlikely that in practice many people would be comfortable with such a solution. 

Final observations

Let me point out two objections that I have seen. Van Snyder wrote: 

NaN is not part of the set of floating-point numbers, so it doesn’t behave as if “bottom” were added to the set. 

Interesting point, but, in my opinion not applicable: ⊥ is indeed not part of the mathematical set of floating point numbers, but in the form of NaN it is part of the corresponding type (float in C, REAL in Eiffel); and the operations of the type are applicable to all values, including NaN if, as noted, we have not taken the extreme step of triggering an exception every time an operation uses a NaN as one of its operands. So we cannot free ourselves from the monotonicity concern by just a sleight of hands. 

Another comment, also by Van Snyder (slightly abridged): 

Think of [the status of NaN] as a variety of dynamic run-time typing. With static typing, if  x is an integer variable and y

        x := y 

does not inevitably lead to 

        x = y

 True; and a compelling argument to require that conversions satisfy equality as a postcondition! Such  reasoning — reflexivity again — was essential in the design of the Eiffel conversion mechanism [3], which makes it possible to define conversions between various data types (not just integers and reals and the other classical examples, but also any other user types as long as the conversion does not conflict with inheritance). The same conversion rules apply to assignment and equality, so that yes, whenever the assignment x := y is permitted, including when it involves a conversion, the property x = y  is afterwards always guaranteed to hold.

It is rather dangerous indeed to depart from the fundamental laws of mathematics. 

References

[1] IEEE floating-point standard, 754-2008; see summary and references in the Wikipedia entry.  

[2] IFIP Working Group 2.5 on numerical software: home page. 

[3] Eiffel standard (ECMA and ISO), available on the ECMA site.

In a previous post I briefly mentioned some work that I am doing on aliasing. There is a draft paper [1], describing the theory, calculus, graphical notation (alias diagrams) and implementation. Here I will try to give an idea of what it’s about, with the hope that you will be intrigued enough to read the article. Even before you read it you might try out the implementation[2], a simple interactive interface with all the examples of the article .

What the article does not describe in detail — that will be for a companion paper— is how the calculus will be used as part of a general framework for developing object-oriented software proved correct from the start, the focus of our overall  “Trusted Components” project’ [3]. Let me simply state that the computation of aliases is the key missing step in the effort to make correctness proofs a standard part of software development. This is a strong claim which requires some elaboration, but not here.

The alias calculus asks a simple question: given two expressions representing references (pointers),  can their values, at a given point in the program, ever reference the same object during execution?

As an example  application, consider two linked lists x and y, which can be manipulated with operations such as extend, which creates a new cell and adds it at the end of the list:

 The calculus makes it possible to prove that if  x and y are not aliased to each other, then none of the pointers in any of the cells in either of the lists can point to  (be aliased to) any cell of  the other. If  x is initially aliased to y, the property no longer holds. You can run the proof (examples 18 and 19) in the downloadable implementation.

The calculus gives a set of rules, each applying to a particular construct of the language, and listed below.

The rule for a construct p is of the form

          a |= p      =   a’ 

where a and a’ are alias relations; this states that executing p  in a state where the alias relation is a will yield the alias relation a’ in the resulting state. An alias relation is a symmetric, irreflexive relation; it indicates which expressions and variables can be aliased to each other in a given state.

The constructs p considered in the discussion are those of a simplified programming language; a modern object-oriented language such as Eiffel can easily be translated into that language. Some precision will be lost in the process, meaning that the alias calculus (itself precise) can find aliases that would not exist in the original program; if this prevents proofs of desired properties, the cut instruction discussed below serves to correct the problem.

The first rule is for the forget instruction (Eiffel: x := Void):

          a |= forget x       =   a \- {x}

where the \- operator removes from a relation all the elements belonging to a given set A. In the case of object-oriented programming, with multidot expressions x.y.z, the application of this rule must remove all elements whose first component, here x, belongs to A.

The rule for creation is the same as for forget:

         a |= create x          =   a \- {x}

The two instructions have different semantics, but the same effect on aliasing.

The calculus has a rule for the cut instruction, which removes the connection between two expressions:

        a |= cut x, y       =   a — <x, y>

where — is set difference and <x, y> includes the pairs [x, y] and [y, x] (this is a special case of a general notation defined in the article, using the overline symbol). The cut   instruction corresponds, in Eiffel, to cut   x /= y end:  a hint given to the alias calculus (and proved through some other means, such as standard axiomatic semantics) that some references will not be aliased.

The rule for assignment is

      a |= (x := y)      =   given  b = a \- {x}   then   <b È {x} x (b / y)}> end

where b /y (“quotient”), similar to an equivalence class in an equivalence relation, is the set of elements aliased to y in b, plus y itself (remember that the relation is irreflexive). This rule works well for object-oriented programming thanks to the definition of the \- operator: in x := x.y, we must not alias x to x.y, although we must alias it to any z that was aliased to x.y.

The paper introduces a graphical notation, alias diagrams, which makes it possible to reason effectively about such situations. Here for example is a diagram illustrating the last comment:

Alias diagram for a multidot assignment

(The grayed elements are for explanation and not part of the final alias relation.)

For the compound instruction, the rule is:

           a |= (p ;  q)      =   (a |= p) |= q)

For the conditional instruction, we get:

           a |= (then p else  q end)      =   (a |= p) È  (a |= q)

Note the form of the instruction: the alias calculus ignores information from the then clause present in the source language. The union operator is the reason why  alias relations,  irreflexive and symmetric, are not  necessarily transitive.

The loop instruction, which also ignores the test (exit or continuation condition), is governed by the following rule:

           a |= (loop p end)       =   t_N

where span style=”color: #0000ff;”>N is the first value such that t_N = t_N+1 (in other words, t_N is the fixpoint) in the following sequence:

            t₀          =    a
           t_n+1       =   (t_n È (t_n |= p))     

The existence of a fixpoint and the correctness of this formula to compute it are the result of a theorem in the paper, the “loop aliasing theorem”; the proof is surprisingly elaborate (maybe I missed a simpler one).

For procedures, the rule is

         a |= call p        =   a |= p.body

where p.body is the body of the procedure. In the presence of recursion, this gives rise to a set of equations, whose solution is the fixpoint; again a theorem is needed to demonstrate that the fixpoint exists. The implementation directly applies fixpoint computation (see examples 11 to 13 in the paper and implementation).

The calculus does not directly consider routine arguments but treats them as attributes of the corresponding class; so a call is considered to start with assignments of the form f : = a for every pair of formal and actual arguments f and a. Like the omission of conditions in loops and conditionals, this is a source of possible imprecision in translating from an actual programming language into the calculus, since values passed to recursive activations of the same routine will be conflated.

Particularly interesting is the last rule, which generalizes the previous one to qualified calls of the form x. f (…)  as they exist in object-oriented programming. The rule involves the new notion of inverse variable, written x’ where x is a variable. Laws of the calculus (with Current denoting the current object, one of the fundamental notions of object-oriented programming) are

        Current.x            = x   
        x.Current            = x
        x.x’                      = Current
        x’.x                      = Current

In other words, Current plays the role of zero element for the dot operator, and variable inversion is the inverse operation. In a call x.f, where x denotes the supplier object (the target of the call), the inverse variable provides a back reference to the client object (the caller), indispensable to interpret references in the original context. This property is reflected by the qualified client rule, which uses  the auxiliary operator n (where x n a, for a relation a and a variable x, is the set of pairs [x.u, y.v] such that the pair [u, v] is in a). The rule is:

         a |= call x.r       =   x n ((x’ n a ) |= call r)

You need to read the article for the full explanation, but let me offer the following quote from the corresponding section (maybe you will note a familiar turn of phrase):

Thus we are permitted to prove that the unqualified call creates certain aliasings, on the assumption that it starts in its own alias environment but has access to the caller’s environment through the inverted variable, and then to assert categorically that the qualified call has the same aliasings transposed back to the original environment. This change of environment to prove the unqualified property, followed by a change back to the original environment to prove the qualified property, explains well the aura of magic which attends a programmer’s first introduction to object-oriented programming.

I hope you will enjoy the calculus and try the examples in the implementation. It is fun to apply, and there seems to be no end to the potential applications.

Reference

[1] Bertrand Meyer: The Theory and Calculus of Aliasing, draft paper, first published 12 January 2009 (revised 21 January 2010), available here and also at arxiv.org.
[2] Implementation (interactive version to try all the examples of the paper): downloadable Windows executable here.
[3] Bertrand Meyer: The Grand Challenge of Trusted Components, in 2003 International Conference on Software Engineering, available here.

In the past few weeks I wrote a program to compute the aliases of variables and expressions in an object-oriented program (based on a new theory [1]).

For one of the data structures, I needed a specific notion of equality, so I did the standard thing in Eiffel: redefine the is_equal function inherited from the top class ANY, to implement the desired variant.

The first time I ran the result, I got a postcondition violation. The violated postcondition clauses was not even any that I wrote: it was an original postcondition of is_equal (other: like Current)  in ANY, which my redefinition inherited as per the rules of Design by Contract; it reads

symmetric: Result implies other ~ Current

meaning: equality is symmetric, so if Result is true, i.e. the Current object is equal to other, then other must also be equal to Current. (~ is object equality, which applies the local version is is_equal).  What was I doing wrong? The structure is a list, so the code iterates on both the current list and the other list:

from
    start ; other.start ; Result := True
until (not Result) or after loop
        if other.after then Result := False else
              Result := (item ~ other.item)
              forth ; other.forth
        end
end

Simple enough: at each position check whether the item in the current list is equal to the item in the other list, and if so move forth in both the current list and the other one; stop whenever we find two unequal elements, or we exhaust either list as told by after list. (Since is_equal is a function and not produce any side effect, the actual code saves the cursors before the iteration and restores them afterwards. Thanks to Ian Warrington for asking about this point in a comment to this post. The new across loop variant described in  two later postings uses external cursors and manages them automatically, so this business of maintaining the cursor manually goes away.)

The problem is that with this algorithm it is possible to return True if the first list was exhausted but not the second, so that the first list is a subset of the other rather than identical. The correction is immediate: add

Result and other_list.after

after the loop; alternatively, enclose the loop in a conditional so that it is only executed if count = other.count (this solution is  better since it saves much computation in cases of lists of different sizes, which cannot be equal).

The lesson (other than that I need to be more careful) is that the error was caught immediately, thanks to a postcondition violation — and one that I did not even have to write. Just another day at the office; and let us shed a tear for the poor folks who still program without this kind of capability in their language and development environment.

Reference

[1] Bertrand Meyer: The Theory and Calculus of Aliasing, draft paper, available here.

The book page for Touch of Class (my introductory programming textbook), announced in the book, is finally available, courtesy Vladimir Tochilin:

touch.ethz.ch

It includes some book extracts (prefaces, table of contents, an entire sample chapter, for which I chose the Recursion chapter), a list of known errata and a wiki page to report new errata, a discussion forum, links to the full set of slides (PowerPoint, PDF) for the associated course, video recordings of that course at ETH, and a special “instructor’s corner” for those having adopted the textbook for their courses.

“For someone in search of a software development method, the problem is not to find answers; it’s to find out how good the proposed answers are. We have lots of methods — every year brings its new harvest — but the poor practitioner is left wondering why last year’s recipe is not good enough after all, and why he or she has to embrace this year’s buzz instead. Anyone looking for serious conceptual arguments has to break through the hype and find the precious few jewels of applicable wisdom.”

This is the start of an article that Ivar Jacobson and I just wrote for Dr. Dobb’s Journal;  it is available in the online edition [1] and will appear (as I understand) in the next paper edition. The article is a plea for a rational, science-based approach to software development methodology, and a call for others to join us in establishing a sound basis.

Reference

[1] Ivar Jacobson and Bertrand Meyer, Methods Need Theory, Dr. Dobb’s Journal, August 2009, available online.