Verifying pointer arithmetic

July 16, 2010 4 comments

Today I’ll look at whether code that uses pointer arithmetic is any harder to verify than equivalent code that does not use pointer arithmetic.

Consider this function for copying an array (or part of an array) into another array (or part of another array):

void arrayCopy(const int* src, int* dst, size_t num) {
  size_t i;
  for (i = 0; i < num; ++i) {
    dst[i] = src[i];
  }
}

Read more…

Run-time checks: Are they worth it?

July 7, 2010 4 comments

One of the criticisms levelled against the use of C in safety-critical software is that the C language does not provide run-time checks automatically. For example, when indexing into an array, there is no check that the index is in bounds. Likewise, when doing integer arithmetic in C, there is no check for arithmetic overflow. Read more…

Aliasing and how to control it

June 22, 2010 4 comments

Today I’ll start by writing a simple function that determines the maximum and minimum of two integers. We want to return two values, and C doesn’t make that easy unless we declare a struct to hold them. So I’ll pass two pointers to where I want the results stored instead. Here goes:

#include "arc.h"
 void minMax(int a, int b, out int *min, out int *max)
post(*min <= a; *min <= b; *min == a || *min == b) post(*max >= a; *max >= b; *max == a || *max == b)
{ *min = a < b ? a : b;
  *max = a > b ? a : b;
}

I’ve highlighted the ArC annotations in green. Read more…

Verifying absence of integer overflow

June 7, 2010 2 comments

One class of errors we need to guard against when writing critical software is arithmetic overflow. Before I go into detail, I invite you to consider the following program and decide what it prints:

#include <stdio.h>
int main(int argc, char *argv[]) {
unsigned int x = 42;
long y = -10;
printf("%s\n", (x > y ? "Hello, normal world!" : "Hello, strange world!"));
return 0;
}

Read more…

Specification with Ghost Functions

May 26, 2010 Comments off

In my previous post I showed that the C expression sublanguage extended with quantified expressions (forall and exists) is insufficient to allow some specifications to be expressed. I presented this function (annotated with an incomplete specification) to average an array of data:

int16_t average(const int16_t * array readings, size_t numReadings)
pre(readings.lwb == 0; readings.lim == numReadings)
pre(numReadings != 0)
post(result == ? /* sum of elements of readings */ /numReadings)
{
int sum = 0;
size_t i;
for (i = 0; i < numReadings; ++i)
keep(i <= numReadings)
keep(sum == ? /* sum of first i elements of readings */ )
decrease(numReadings - i)
{
sum += readings[i];
}
return (int16_t)(sum/numReadings);
}

Read more…

Expressing the Inexpressible

May 24, 2010 3 comments

When writing preconditions, postconditions and other specifications for C programs, sometimes we need to write expressions that can’t be expressed in plain C. That’s why formal verification systems based on annotated programming languages almost always augment the expression sublanguage with forall and exists expressions. In previous posts, I’ve introduced ArC’s implementations of these. For example, the following expression yields true if all elements of the array arr are between 0 and 100 inclusive:

forall ind in arr.indices :- arr[ind] >= 0 && arr[ind] <= 100

Here, ind is declared as a bound variable that ranges over the values in the expression that follows the keyword in, which in this case is all the indices into arr. Read more…

Verifying a binary search, part 2

May 6, 2010 2 comments

In my last entry I showed how to use a correct-by-construction approach to develop a binary search function. We got as far as specifying the function and the loop, but we left the loop body undefined. The function declaration looked like this: Read more…