Add new check for nesting level

[Zetten]

Using a method inspired by Sonar-Java, check code for the depth of
nested blocks and raise issues accordingly.

[guwirth]

What is the difference to cyclomatic complexity (http://en.m.wikipedia.org/wiki/Cyclomatic_complexity)?

[jmecosta]

They measure different things based on same concept? For example you can say you should only 5 levels of nesting in a function. And the complexity can still not flag any violation because its 10. Complexity gives the idea as if the function is only with one nested level.

[Zetten]

Yeah, they are related, but cyclomatic complexity is more of a flat measure, e.g. lots of sequential unrelated if blocks = high complexity, while nesting level is a measure of a second dimension and is a good indicator for spaghetti code.

I think the nesting level is more indicative of a need for refactoring, but also easier refactoring. Pulling a deeply-nested block out to a separate function is easier and usually conceptually 'better' than splitting any general high-complexity method.

This PR does have one subtle point that may need consideration. In the AST, the way descendent nodes of ifStatement tokens are calculated mean that else-if blocks like:

if (cond1) {
  // do something
} else if (cond2) {
  // do something else
} else {
  // do a third thing
}

are actually equivalent to this:

if (cond1) {
  // do something
} else {
  if (cond2) {
    // do something else
  } else {
    // do a third thing
  }
}

So a series of else-ifs will increase the nesting level by one each time. But I think this is standard behaviour for most C++ static analysers (I recall one of my colleagues coded a rule for cppcheck which found the same thing).

However it would also be possible to catch the differing cases of an else if () {} and else { if () {} } and increase nesting level in the second case but not the first. It just takes an additional check on parent/preceding nodes in the AST.

Perhaps I should make this alternative behaviour a parameter for the check?

[guwirth]

If this metric is really measuring the nesting level (second case above) it makes sense. Would add no parameter because first case is equal to cyclomatic complexity. Especially with 'if/else if' and 'switch/case' the cyclomatic complexity metric is mostly wrong.

[Zetten]

Okay, the if-else behaviour has been rectified.

The test case is reasonably comprehensive, but please do let me know if there are any glaring errors or omissions in there.