Frama-C news and ideas

To content | To menu | To search | Frama-C Home

Tuesday, March 18 2014

Nginx buffer overflow

A buffer overflow has been discovered in recent versions of the HTTP server Nginx.

Hacker News user jmnicolas pondered out loud: “I wonder if this discovery is a result of OpenBSD switching its focus from Apache to Nginx?”

It took me one minute to understand what ey meant. I was actually wondering why OpenBSD would be putting buffer overflows in Nginx, now that they are done putting buffer overflows in Apache. I mean, surely there is room for one more buffer overflow in Apache?

The analysis of what this means about this industry and/or me is left as an exercise to the reader.

Tuesday, February 4 2014


Jesse Ruderman on assertions and fuzzing

Jesse Ruderman has published a blog post on assertions and how they complement fuzzing. Key quote: “Fuzzers make things go wrong. Assertions make sure we find out.”

Readers of this blog are accustomed to me talking about differential testing, where a reference result (say, obtained by compiling a random C program with a quality compiler) is used to detect bugs in the target program (say, a static analysis framework for C programs). Differential testing imposes constraints: the random input must have a definite meaning, and a reference implementation needs to be available to compute this meaning. Often one finds bugs in the reference implementation together with bugs in the target program.

Besides, there are deeply burrowed bugs that are difficult to reveal with such a black-box approach. Assertions simplify the problem: if the code being tested contains enough well-chosen assertions, bugs can be caught without a reference implementation.

Sometimes, an assertion is a reference implementation: some of the assertions in Frama-C are alternative computations of the same intermediate result, followed by a comparison of the normally computed result with the alternately computed result. These assertions initially caught bugs in either computation. Since then, they have caught many more bugs in hash-consing, where the two results are structurally identical but are not shared.

Even when an assertion happens to contain a reference implementation for an intermediate result, it saves a lot of time to write the assertion as opposed to producing a complete reference implementation for the whole problem (say, interpreting C programs). The alternative implementation in the assertion does not have to be efficient: in Frama-C's value analysis, it is considered acceptable to spend 20% of the analysis time executing inefficient reference implementations in assertions.

John Regehr on writing and using assertions

John Regehr just published a blog post too on the subject of assertions. What a coincidence! Key quote: “we have tools that are specifically designed to help us reason about assertions; my favorite one for C code is Frama-C”.

In most of his post, John describes executable assertions written in the same language as the program being debugged. In the section quoted above, he moves to specific annotation languages to write assertions in. The advantages of using the same language as the programming language for assertions require no elaboration: it's the same language! The programmer already knows it, and the reviewer of the code already knows it.

But there is a spectrum of annotation languages for assertions. Eiffel stands its ground somewhere on this spectrum, very close to the underlying programming language but with enough good intentions to be noted. I think that the JML annotation language for Java was initially mostly intended for run-time checking, but ended up being very popular too as the annotation language used by static analyzers (I would be happy to be corrected if I am getting this wrong). Nearby JML lies E-ACSL, an executable subset of ACSL and also a Frama-C plug-in to convert a C program annotated with /*@ ... */ assertions into an instrumented C program that detects at run-time violated assertions. SPARK 2014 aims at making both camps happy.

I should point out for the sake of scientific integrity that I am being slightly cheeky in the choice of the key quote to represent John's post. I recommend you read the whole thing, of which the Frama-C endorsement is only a tiny fraction.

Taking advantage of C assertions with Frama-C

One can also use Frama-C in conjunction with existing executable assertions, typically implemented with the standard function assert() in header assert.h. The function assert() takes a C expression representing a property that should hold.

One way to take advantage of assertions is to fail to establish that they always hold, and warn the user that perhaps they don't. It is easy for anyone who wants this behavior to obtain it. One simply needs to specify assert() thus:

/*@ requires x != 0 ; */
void assert(int x);

With this specification, any existing call to the C function assert(), intended for the programmer to be executed at run-time, is required to have an argument that demonstrably corresponds to a non-null expression. This specification creates a link of sorts between static verification and run-time checking (or expressions intended to be checked at run-time, anyway).

Here is an example:

/*@ requires x >= 0 ;*/
double sqrt(double x);

int get_int(void);

double dist(double x, double y)
  double s = x * x + y * y;
  assert(s >= 0.0);
  return sqrt(s);

int main(int c, char **v){
  dist(get_int(), get_int());

When this program is analyzed, the value analysis detects that the assertion s >= 0.0 helpfully inserted by the programmer as an argument to assert() may not hold. In this particular example, the warning is a false positive, but never mind that for the moment.

$ frama-c -val t.c
t.c:4:[value] Function assert: precondition got status unknown.
t.c:1:[value] Function sqrt: precondition got status unknown.

Even more irritating in the example above is that after producing a false positive for assert(), the analyzer produces a false positive for the pre-condition of sqrt(). This brings us to another way a static checker could use C assertions to its advantage: it could take them as hints, properties that can be assumed to hold in order to avoid warning for problems that would seem to arise later if they did not.

With the user-specified function assert() above, the analyzer computes the truth value of the argument s >= 0.0. Because the set of values computed for s is approximated, the set of values for the expression s >= 0.0 is {0; 1}. The analyzer can tell that the pre-condition for assert() may not hold, but the relationship with the possible values of s is too indirect for it to decide that the forbidden situation corresponds to s negative and that only s positive is allowed.

There exists a better modelization of assert(). It was implemented as a value analysis built-in in order to offer the very functionality we are describing here:

$ frama-c -val -val-builtin assert:Frama_C_assert t.c
t.c:11:[value] warning: Frama_C_assert: unknown
t.c:1:[value] Function sqrt: precondition got status valid.
[value] Values at end of function dist:
  s ∈ [-0. .. 9.22337203685e+18]

On the same example, the builtin version of assert detects that it may be passed a null expression and warns about that. These is no improvement there (annotating the example to convince the analyzer that s is always positive is left as an exercise to the reader). Subsequently, and this is an improvement with respect to the previous analysis, the builtin does its best to incorporate the information provided in the assertion, so that it can tell that s is henceforth positive and that the pre-condition of sqrt() is satisfied.

Friday, January 24 2014

Bear-joke security is dead

Likely, you have heard this one before:

Two campers are surprised by an angry bear. One of them starts putting on eir running shoes. Surprised, the other exclaims “What are you doing, Alex? You can't outrun a bear!”

To which Alex replies: “I don't have to outrun the bear. I only have to outrun you.”

This joke used to be popular with security experts, who employed it as a metaphor. I only ever heard it in that context, the first time as a student in the late nineties. The bear joke was still used for this purpose in 2008.

You may also have heard that there is a new bear in town, and the new bear can take both Alex and eir friend, with or without running shoes. Also, the new bear could literally(*) eat a horse. And the bear is part of an organized network of hungry bears with walkie-talkies and sniper guns.

If your approach to security was based on bear jokes with dubious morals, now must really suck. Andy Green's blog post “Cryptography may not be dead, but it is on life support” is representative of the change. One of the quote he takes from Schneier's talk is:

Most of how the NSA deals with cryptography is by getting around it … They exploit bad implementations—we have lots of those.

Yes, we do, but we don't have to use them.

Here is to 2014 being the year of reliable cryptography implementations that cannot be circumvented through defects.

(*) “literally” within the confines of this metaphor

Monday, October 21 2013

Bruce Dawson on compiler bugs

Bruce Dawson has written a superb blog post on a Visual C++ compiler bug (now fixed), covering every aspect an essay on compiler bugs should cover.

I really like one section, that I am going to quote in full:


In these paranoid days of the NSA subverting every computer system available every bug is a possible security flaw. This bug seems curiously commonplace – why wasn’t it discovered when building Windows or other 64-bit products? The most likely explanation is chance and happenstance, but compiler bugs can be used to make innocuous source code do surprising things. Maybe this bug is used as part of a back door to let code execution go in a direction that the source code says is impossible. This happened accidentally on the Xbox 360 where the low 32 bits of r0 were checked and then all 64 bits were used. This inconsistency allowed a hypervisor privilege escalation that led to arbitrary code execution.

This bug is probably not security related, but security is another reason to ensure that compiler bugs get fixed.

Bruce's post also illustrates the “most compiler bugs aren't” problem and provides a step-by-step tutorial on manual bug reduction. Regarding the latter, my reaction to Bruce's post was to bring up automatic testcase reduction with “delta debugging”, but Gravatar user jrrr beat me to it.

Frama-C contributes to making automatic testcase reduction work in practice for C compiler bugs by providing a reference C implementation that detects the undefined behaviors that could make the (original or reduced) bug reports invalid. We also use delta debugging internally to simplify our own reports of bugs in Frama-C. The next post in this blog will illustrate this use of delta debugging on a concrete example.

Monday, September 2 2013

The case for formal verification of existing software

Perry E. Metzger takes a look at formal verification. This is good stuff; there is a lot to agree with here.

However, agreeing with Perry's post alone would not make a very interesting counterpoint. If agreeing was the only thing I intended to do, I might even not have written this post. Instead I intended to add, and I apologize in advance for the predictability of my views, that while creating formally verified software from scratch is useful, verifying existing software is useful too.

Yes, formally verified software written from scratch can now be large enough in scope to be a compiler or a microkernel, but when verifying existing software, we can tackle the problem from the other end: we can pick any useful software component, and verify that. We can pick a software component so useful that it is already used by millions. If we succeed in verifying it, we have put formally verified software in the hands of millions of satisfied users. Transparently.

Take the example of the SSL implementation I am taking a couple of weeks to finish massaging through Frama-C. It is not as wide in scope as Quark, seL4 or CompCert. Neither am I aiming for the same kind of functional correctness as these projects are: I am only verifying the absence of undefined behaviors in the component, and verifying the functional dependencies of the cryptographic primitives(*).

But PolarSSL is useful. Plus, by its very nature, it is exposed to security attacks (SSL is the protocol behind HTTPS). And the former three examples are full-blown research projects, in contrast to my single person.month effort.

The bonus is that the formally verified PolarSSL can be compiled and embedded with all the same compilers and into all the same firmwares as the earlier non-verified version. It is the same as the non-verified version except maybe for a couple of bugfixes, and the confidence that for an identified usage pattern, no more undefined behavior bugs will ever need to be fixed.

All in all, the formal verification of existing code, despite its differences from its “from scratch” counterpart, has too become a force to be reckoned with.

(*) translation: I use Frama-C option -deps and I compare the result to what I expected

Acknowledgement: I got the link to Perry E. Metzger's post through Toby Murray.

Saturday, August 24 2013

Function pointers in C

This post contains a complete list of everything a C program can do with a function pointer, for a rather reasonable definition of “do”. Examples of things not to do with a function pointer are also provided. That list, in contrast, is in no way exhaustive.

What a C program can do with a function pointer

Convert it to a different function pointer type

A function pointer can be converted to a different function pointer type. The C99 standard's clause starts:

“A pointer to a function of one type may be converted to a pointer to a function of another type and back again; the result shall compare equal to the original pointer.”

Call the pointed function with the original type

Clause continues:

“If a converted pointer is used to call a function whose type is not compatible with the pointed-to type, the behavior is undefined.”

Alright, so the above title is slightly sensationalistic: the pointed function can be called with a compatible type. After typedef int t;, the types t and int are compatible, and so are t (*)(t) and int (*)(int), the types of functions taking a t and returning a t and of functions taking an int and returning an int, respectively.

There is no third thing a C program can do with a function pointer

Seriously. The C99 standard has uintptr_t, a recommended integer type to convert data pointers to, but there is not even an equivalent integer type to store function pointers.

What a C program cannot do with a function pointer

Convert it to an ordinary pointer

Function pointers should not be converted to char * or void *, both of which are intended for pointers to data (“objects” in the vocabulary of the C standard). Historically, there has been plenty of reasons why pointers to functions and pointers to data might not have the same representation. With 64-bit architectures, the same reasons continue to apply nowadays.

Call the pointed function with an incompatible type

Even if you know that type float is 32-bit, the same as int on your platform, the following is undefined:

void f(int x);

int main(){
  void (*p)(float) = f;

The line void (*p)(float) = f;, which defines a variable p of type “pointer to function that takes a float”, and initializes it with the conversion of f, is legal as per However, the following statement, (*p)(3); is actually equivalent to (*p)((float)3);, because the type of p is used to decide how to convert the argument prior to the call, and it is undefined because p points to a function that requires an int as argument.

Even if you know that the types int and long are both 32-bit and virtually indistinguishable on your platform (you may be using an ILP32 or an IL32P64 platform), the types int and long are not compatible. Josh Haberman has written a nice essay on this precise topic.

Consider the program:

void f(int x);

int main(){
  void (*p)(long) = f;

This time the statement is equivalent to (*p)((long)3);, and it is undefined, even if long and int are both 32-bit (substitute long and long long if you have a typical I32LP64 platform).

Lastly, the example that prompted this post was, in a bit of Open-Source code, the creation of a new execution thread. The example can be simplified into:

void apply(void (*f)(void*), void *arg)

void fun(int *x){
  // work work
  *x = 1;

int data;

int main(){
  apply(fun, &data);

The undefined behavior is not visible: it takes place inside function apply(), which is a standard library function (it was pthread_create() in the original example). But it is there: the function apply() expects a pointer to function that takes a void* and applies it as such. The types int * and void * are not compatible, and neither are the types of functions that take these arguments.

Note that gcc -Wall warns about the conversion when passing fun to apply():

t.c:11: warning: passing argument 1 of ‘apply’ from incompatible pointer type

Fixing this warning with a cast to void (*)(void*) is a programmer mistake. The bug indicated by the warning is that there is a risk that fun() will be applied with the wrong type, and this warning is justified here, since fun() will be applied with the wrong type inside function apply(). If we “fix” the program this way:

$ tail -3 t.c
int main(){
  apply((void (*)(void*))fun, &data);
$ gcc -std=c99 -Wall t.c

The explicit cast to (void (*)(void*) silences the compiler, but the bug is still in the same place, in function apply().

Fortunately gcc -std=c99 -Wall is not the only static analyzer we can rely on. Frama-C's value analysis warns where the problem really is, in function apply(), and it warns for both the version with implicit conversion and the version with explicit cast:

$ frama-c -val t.c
[value] computing for function apply <- main.
        Called from t.c:14.
t.c:3:[value] warning: Function pointer and pointed function 'fun'  have incompatible types:
        void (void *) vs. void (int *x). assert(function type matches)

The correct way to use function apply() without changing it is to make a function with the correct type for it, and to pass that function to apply():

void stub(void *x){
  apply(stub, &data);

Note that in the above, x is implicitly converted when passed to function fun(), the same way that &data is implicitly converted to void* when passed to apply().


There is almost nothing you can do in C with a function pointer. The feature is still very useful and instills a bit of genericity in an otherwise decidedly low-level language.

Function pointers are not often used in the standard library, considering: qsort() is, with pthread_create(), another of the few functions that requires a function pointer. Like it, it is often misused: it has its own entry in the C FAQ.

Jens Gustedt provided advice in the writing of this post.

Friday, July 19 2013

The word “binade” now has its Wikipedia page…

… but that's only because I created it.

If you are more familiar than me with Wikipedia etiquette, feel free to adjust, edit, or delete this page. Also, although a Wikipedia account is necessary to create a page, I think it is not required for editing, so you can add to the story too (but if you do not have an account you are perhaps no more familiar than me with Wikipedia etiquette).

Wednesday, June 19 2013

Microsoft's bug bounty program

I like Robert Graham's analysis on Microsoft's new bug bounty program.

I would never have thought of selling vulnerabilities to the NSA (but then, I am not American and not a security researcher). Does the NSA not employ qualified people to look for vulnerabilities as their day job? Is that not like trying to sell a loaf of bread to a company whose business is to make bread?

Sometimes, you have a really good loaf of bread, but still… Regardless of whether the NSA already owns your particular loaf of bread, and independently of the payment-by-carrot-or-stick discussion, you are a competitor, not a provider.

Monday, May 20 2013

Attack by Compiler

The title of this post, “Attack by Compiler”, has been at the back of my mind for several weeks. It started with a comment by jduck on a post earlier this year. The post's topic, the practical undefinedness of reading from uninitialized memory, and jduck's comment, awakened memories from a 2008 incident with the random number generator in OpenSSL.

As I am writing this, if I google “attack by compiler”, the first page of results include the classic essay Reflections on Trusting Trust by Ken Thompson, Wikipedia's definition of a backdoor in computing, an article by David A. Wheeler for countering the attack described by Thompson, a commentary by Bruce Schneier on Wheeler's article, and a Linux Journal article by David Maynor on the practicality of the attack described by Thompson on the widespread GNU/Linux platform.

This post is about a slightly different idea.

Initial conditions: trustworthy compiler, peer-reviewed changes

Suppose that we start with a trustworthy compiler, widely distributed in both source and binary form. Some people are in the position to make changes to the compiler's source code and could, like Thompson in his essay, attempt to insert a backdoor in the compiler itself.

But now, each change is scrutinized by innumerable witnesses from the Open-Source community. I say “witnesses”, in fact they are mostly students, but we will probably be forced to assume that they can't all be mischievous.

The attackers could try to obfuscate the backdoor as they insert it, but the problem is that some of the witnesses are bound to possess this character trait that the less they understand a change, the more they investigate it. Furthermore, once these witnesses have uncovered the truth, loss of credibility will ensue for the person who tried to sneak the backdoor in. This person will lose eir commit privilege to the compiler's sources, and people will recover untainted compilers in source and binary form from their archives. This kind of approach is risky and may only result in a temporary advantage—which may still be enough.

The underhanded approach

The 2013 edition of the Underhanded C Contest is under way. The contest defines underhanded code as:

code that is as readable, clear, innocent and straightforward as possible, and yet [fails] to perform at its apparent function

Underhandedness is exactly what an attacker with commit access to the source code of the widely used compiler should aim for. If the commit is underhanded enough, the committer may not only enjoy full deniability, but ey may obtain that the incriminating change stays in ulterior versions of the compiler, as a “good” change. This implies that all affected security-sensitive applications, like the “login” program in Thompson's essay, must be updated to work around the now official backdoor in the compiler. In this scenario, even after the attack has been discovered, anytime someone unknowingly compiles an old version of “login” with a recent compiler, it's another win for the attacker.

Fortunately, we agree with Scott Craver that the C programming language is a very good context to be underhanded in, and a C compiler is even better. How about the following ideas?

  1. making pseudo-random number generators that rely on uninitialized memory less random, in the hope that this will result in weaker cryptographic keys for those who do not know about the flaw;
  2. optimizing a NULL test out of kernel code when it is one of several defenses that need to be bypassed;
  3. optimizing buffer + len >= buffer_end || buffer + len < buffer overflow tests out from application code, so that buffer overflows do take place in code that is guarded thus;
  4. optimizing source code that was written to take constant-time into binary code that reveals secrets by terminating early.

I am not being very original. According to this post by Xi Wang, idea (1) is only waiting for someone to give the compiler one last well-calibrated shove. The NULL test optimization was already implemented in the compiler when it was needed for a famous Linux kernel exploit. The interesting scenario would have been if someone had found that the code in tun_chr_poll() was almost exploitable and had submitted the GCC optimization to activate the problem, but it did not happen in this order. Idea (3) really happened.

Idea (4) has not been exploited that I know of, but it is only only a matter of time. If I google for “constant-time memcmp”, I may find an implementation such as follows:

int memcmp_nta(const void *cs, const void *ct, size_t count)
  const unsigned char *su1, *su2;
  int res = 0;

  for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
  res |= (*su1 ^ *su2);

  return res;

Nothing in the C language definition forces a compiler to compile the above function into a function that does not return as soon as variable res contains (unsigned char)-1, not to mention the possibilities if the compiler first inlines the function in a site where its result is only compared to 0, and then optimizes the code for early termination. If I was trying to sneak in a compiler change that defeats the purpose of this memcmp_nta() function, I would bundle it with auto-vectorization improvements. It is a fashionable topic, and quite exciting if one does not care about non-functional properties such as execution time.


The impracticability of the described attack is counter-balanced by some unusual benefits: at the time of the attack, someone may already have audited the pseudo-random number generator or function memcmp_nta() we used as examples. The audit may have considered both source and generated assembly code, and involved actual tests, at a time when the code was “safely” compiled. But the auditor is not going to come back to the review again and again each time a new compiler comes out. Like Monty Python's Spanish Inquisition, nobody expects the compiler-generated backdoor.

Three of my four examples involve undefined behavior. More generally, my examples all involve unwarranted programmer expectations about C idioms. This is the key to plausibly deniable compiler changes that reveal latent security flaws. What other unwarranted expectations should we take advantage of for an “attack by compiler”?

Saturday, April 6 2013

Non-expert floating-point-using developers need accurate floating-point libraries the most

Quotes on the Internet

In 2012, Lukas Mathis took a quote out of the context of a blog post by Marco Arment and ran with it. The result was a though-provoking essay. Key quote:

This is a sentiment you often hear from people: casual users only need «entry-level» devices. Even casual users themselves perpetuate it: «Oh, I’m not doing much on my computer, so I always just go with the cheapest option.» And then they buy a horrid, underpowered netbook, find out that it has a tiny screen, is incredibly slow, the keyboard sucks, and they either never actually use it, or eventually come to the conclusion that they just hate computers.

In reality, it’s exactly backwards: proficient users can deal with a crappy computer, but casual users need as good a computer as possible.

Lukas fully develops the idea in his post, Crappy Computers. Go ahead and read it now if you haven't already. This blog will still be here when you come back.

Floating-point libraries

The idea expressed by Lukas Mathis applies identically in a much more specific setting: developing with floating-point computations. The developers who most need accurate floating-point libraries are those who least care about floating-point. These developers will themselves tell you that it is all the same to them. They do not know what an ULP (“unit in the last place”) is, so what difference is it to them if they get two of them as error where they could have had one or half of one?

In this, they are just as wrong as the casual computer users who pick horrid netbooks for themselves.

Floating-point-wise, programming environments are not born equal

All recent processors for desktop computers provide basic operations +, -, *, / and square root for IEEE 754 single- and double-precision floating-point numbers. Each operation has its assembly instruction, and since the assembly instruction is the fastest way to implement the operation, compilers have no opportunity to mess things up in a misguided attempt at optimizing for speed.

Who am I kidding? Of course compilers have plenty of opportunities to mess things up.

  1. It may seem to a compiler that a compile-time computation is even faster than the assembly instruction provided by the processor, so that if the program computes x / 10.0, the compiler may compute 1 / 10.0 at compile-time and generate assembly code that multiplies x by this constant instead. This transformation causes the result to be less accurate in some rare cases.
  2. Or a compiler may simplify source-code expressions as if floating-point operations were associative when they aren't. It may for instance optimize a carefully crafted floating-point expression such as a + b - a - b into 0.

Nevertheless, there has been much progress recently in standard compliance for compilers' implementations of floating-point. Overall, for programs that only use the basic operators, the situation has never been better.

The situation is not as bright-looking when it comes to mathematical libraries. These libraries provide conversion to and from decimal, and transcendental elementary functions implemented on top of the basic operations. They are typically part of the operating system. Implementations vary wildly in quality from an operating system to the next.

Expert developers know exactly what compromise between accuracy and speed they need, and they typically use their own functions instead of the operating system's. By way of illustration, a famous super-fast pretty-accurate implementation of the inverse square root function is used in Quake III and has been much analyzed.

The casual developer of floating-point programs, on the other hand, will certainly use the functions provided by the system. Some of eir expectations may be naive, or altogether impossible to reconcile with the constraints imposed by the IEEE 754 standard. Other expectations may be common sense, such as a sin() function that does not return -2.76.

For such a developer, the mathematical libraries should strive to be as accommodating and standard-compliant as possible, because ey needs it, regardless of what ey thinks.

An example

To illustrate, I have written a string-to-long conversion function. It could have been an entry in John Regehr's contest, but since the deadline is passed, I have allowed the function to expect only positive numbers and to fail miserably when the input is ill-formed.

The function looks like this:

long str2long(char *p)
  size_t l = strlen(p);
  long acc = 0;
  for (size_t i=0; i<l; i++)
      int digit = p[i] - '0';
      long pow10 = pow(10, l - 1U - i);
      acc += digit * pow10;;
  return acc;

Neat, huh?

I tested this function more than the last function. This time I compiled it and invoked it on a few strings:

  printf("%ld %ld %ld %ld\n",

You can download the entire C code for yourself. If you run it, you should get:

0 123 999 123456789123456789

I wrote my function to work for all well-formed inputs that fit in a long (but I only tested it for four values, so do not embed it in your space shuttle, please). Some of the reasons why I expect it to work are implicit: for one, powers of ten up to 10^22 are exactly representable as double-precision floating-point numbers. Also, I happen to know that on the system I use, the mathematical library is one of the best available.

I am not, in fact, a floating-point expert. I could be completely illiterate with respect to floating-point and have written the exact same function. In fact, this happened to StackOverflow user1257. (I am not saying that StackOverflow user1257 is illiterate with respect to floating-point, either. Ey wrote a function similar to mine, after all.)

User1257's function returned 122 when applied to the string "123" !!!

This was so troubling that user1257 suspected a compiler bug. The reality is more likely that on eir computer, the statement long pow10 = pow(10, 2); sets variable pow10 to 99. The function pow() only needs to be inaccurate by 1ULP for this result to come up because of C's truncation behavior (towards zero) when converting from floating-point to integer.


My str2long() function would fail just the same if it was run in user1257's compilation environment. I still think that my function is correct and that I should be able to expect results to the ULP from the math library's pow() function. A floating-point expert would never even encounter the issue at all. I might be able to diagnose it and to cope. But the floating-point beginner simply needs an environment in which long pow10 = pow(10, 2); sets pow10 to 100.

If you program, and if you use floating-point at all, beware of relying on the math library equivalent of a crappy netbook.

- page 1 of 4