DIR-4-15: Add support for fpclassify() functions.

[MichaelRFairhurst]

Affected rules

• DIR-4-15\

Description

The first implementation of the rule handles guards of the form isinf, isnan, isfinite, iszero, etc., however, it doesn't support guards relating to fpclassify().

Example

float may_be_inf() {
  ...
  return 1.0/0.0;
}

void fpclassify_guard() {
  float x = may_be_inf();
  int y;

  // error, casting possible infinity to int
  y = x; 

  // example 1: easiest to support:
  if (fpclassify(x) != FP_INFINITE) {
    // or "fpclassify(x) == FP_NORMAL"
    y = x;
  }

  // example 2: perhaps slightly harder to support, but guards library already has infra for this:
  switch(x) {
    case FP_NORMAL:
    case FP_SUBNORMAL:
    case FP_ZERO:
      // or, "case FP_INFINITE: break; default:"
      y = x;
  }

  // example 3: this is harder to support but a common pattern:
  int cls = fpclassify(x);
  if (cls != FP_INFINITE) {
    // or "cls == FP_NORMAL || cls == FP_ZERO"
    y = x;
  }   
}