cppcheck/man/cppcheck-design.docbook

<?xml version="1.0" encoding="UTF-8"?>
<article version="5.0" xmlns="http://docbook.org/ns/docbook"
         xmlns:xlink="http://www.w3.org/1999/xlink"
         xmlns:xi="http://www.w3.org/2001/XInclude"
         xmlns:svg="http://www.w3.org/2000/svg"
         xmlns:m="http://www.w3.org/1998/Math/MathML"
         xmlns:html="http://www.w3.org/1999/xhtml"
         xmlns:db="http://docbook.org/ns/docbook">
  <info>
    <title>Cppcheck Design</title>

    <author>
      <personname><firstname>Daniel</firstname><surname>Marjamäki</surname></personname>

      <affiliation>
        <orgname>Cppcheck</orgname>
      </affiliation>
    </author>

    <pubdate>2010</pubdate>
  </info>

  <section>
    <title>Introduction</title>

    <para>This article contains an overview of how Cppcheck works.</para>

    <para>The primary goal is that Cppcheck won't write any false warnings.
    This means that when an error is reported there must definitely be a bug
    in the code.</para>

    <para>The secondary goal is to detect as many bugs as possible.</para>
  </section>

  <section>
    <title>Limitations of static analysis</title>

    <para>There are many bugs in programs that are really hard to detect for
    tools. Here is an example:</para>

    <programlisting>// calculate the number of days
int days(int hours)
{
    return hours / 23;
}</programlisting>

    <para>A human programmer knows that there are 24 hours in a day and
    therefore he could see that "23" is wrong. A tool will probably not know
    that there are 24 hours in a day.</para>

    <para>A tool that tries to guarantee that all bugs are found could write a
    warning message for every "suspicious" calculation in the program. It
    might correctly report that "hours / 23" is wrong but incorrectly warn
    about "hours / 24".</para>

    <para>Cppcheck will only write a warning message if it can determine that
    the calculation is wrong. In this case, no error will be written.</para>
  </section>

  <section>
    <title>Buffer overflows</title>

    <para>This is a simple description of how buffer overflows are detected by
    Cppcheck.</para>

    <para>For simple cases, no control flow analysis is used. If an array is
    accessed out of bounds somewhere in its scope then an error message will
    be written.</para>

    <para>An example code:</para>

    <programlisting>void f()
{
    char a[10];
    if (x + y == 2) {
        a[20] = 0;
    }
}</programlisting>

    <para>Cppcheck will report this message:</para>

    <programlisting>Array 'a[10]' index 20 out of bounds</programlisting>

    <para>Cppcheck will not try to determine how execution can reach the
    "a[20] = 0;" statement. It is assumed that all statements are reachable.
    Cppcheck will detect the error even if it is really impossible that "x + y
    == 2" is true. I still claim that this is a correct warning because the
    statement is there and it has the error.</para>

    <para>Cppcheck will also investigate function calls. But then control flow
    analysis is needed to avoid false warnings. Here is an example that
    logically is the same as the previous example:</para>

    <para><programlisting>void f1(char *s)
{
    s[20] = 0;
}

void f2()
{
    char a[10];
    if (x + y == 2) {
        f1(a);
    }
}</programlisting>Cppcheck will report this message:</para>

    <programlisting>Array 'a[10]' index 20 out of bounds</programlisting>

    <para>If the execution reaches the function call then there will be an
    error.</para>

    <para>But if the condition is moved into "f1" then it will be necessary to
    prove that "x+y==2" can be true when the function is called from
    "f2".</para>

    <para>No error message is reported for this code:</para>

    <para><programlisting>void f1(char *s)
{
    if (x + y == 2) {
        s[20] = 0;
    }
}

void f2()
{
    char a[10];
    f1(a);
}</programlisting></para>
  </section>

  <section>
    <title>Memory leaks</title>

    <para>Simple control-flow analysis is made. The assumtion is that all
    conditions can always be either true or false. It is assumed that all
    statements are reachable.</para>

    <para>Here is an example:</para>

    <programlisting>void f()
{
    char *a = malloc(10);
    if (x + y == 2) {
        return;
    }
    free(a);
}</programlisting>

    <para>Cppcheck will determine that there is a leak at the "return;"
    statement:</para>

    <programlisting>Memory leak: a</programlisting>

    <para>Cppcheck doesn't try to determine how the execution reaches the
    "return;" statement. It will only see that if the execution reaches the
    "return;" then there will be a memory leak.</para>

    <para>Lack of advanced control-flow analysis means that many bugs are not
    detected:</para>

    <programlisting>void f(int x)
{
    char *a = 0;

    if (x == 10)
        a = malloc(10);

    if (x == 20)
        free(a);
}</programlisting>

    <para>Cppcheck doesn't detect any error. The "all conditions can be either
    true/false" means that cppcheck doesn't know that "if (x==20)" is false
    when "if (x==10)" is true. Many other static analysis tools will probably
    detect that there will be a leak if x is 10.</para>
  </section>

  <section>
    <title>Final thoughts</title>

    <para>You can not trust that Cppcheck will detect all bugs.</para>

    <para>Cppcheck will just find some bugs. It is likely that you won't find
    these bugs unless you use Cppcheck. Cppcheck has found bugs in production
    code that has been used for years.</para>

    <para></para>
  </section>
</article>