Page History

Converting String representations of floating point values may lead to incorrect conclusions about the precision of the values. For example, consider converting a value of type float to the type double is , a widening primitive conversion. Refer to the guideline INT33-J. Do not cast numeric types to wider floating-point types without range checking for more details about such conversions. If the value of the float variable must be represented exactly using the double type, an explicit assignment is inappropriate. Note that this is a problem that results from the limited capacity of computers in representing floating point values preciselymore appropriate than first converting the floating point value to a String and then to a double.

Noncompliant Code Example

In this noncompliant code example, the programmer attempts to convert the fractional value {{1/3}} (represented by a variable of type {{float}}) to the {{double}} type. The correct representation of this fraction in floating point arithmetic is 0.33333334 (the last digit 4, is rounded up). When converted to {{double}}, instead of observing a value such as 0.3333333400000000 or 0.333333333333333, the programmer obtains an unexpected value because floating point arithmetic in Java _rounds to the nearest_, a mode defined by the IEEE 754 floating point standard. If the exact value lies midway between two valid floating point values, the lower one (with the least significant bit as 0) is chosen. If it is more than midway, a small positive error is introduced, else a small negative error. \[[Mak 02|AA. Java References#Mak 02]\]

This noncompliant code example converts a value of type float to a value of type double. The float variable contains the equivalent of 1/3 (0.3333333432674408). However, the default string representation can mislead a programmer into thinking that the correct value is 0.33333334. Consequently, the double variable may acquire an imprecise value. The imprecise value arises from the default precision used for converting floating point numbers to a string.

double d;
float f = 1/3f; // Contains the value 0.33333334
d = Double.valueOf(String.valueOf(f)); // Now contains the value 0.33333334

Compliant Solution

This compliant solution assigns the value of type float to the double variable directly.

double d;
float f = 1/3f; // Contains the value 0.333333343333333432674408 
d = f; // NowAssigned contains the value 0.3333333432674408

Compliant Solution

Noncompliant Code Example

This noncompliant code example obtains the result of 1/1000.0 and represents it as a String. However, even though the resulting value is precise, an extra zero digit is appended at the end. Any operations on the string, such as comparisons, may yield incorrect resultsIf the exact value must be preserved, the string representation of the floating point number must be obtained. This can be converted to the double type without losing the initial precision.

double dint i = 1;
floatString fs = Double.valueOf (i 1/3f/ 1000.0).toString(); // Contains the value 0.33333334
d = Double.valueOf(String.valueOf(f)); // Now contains the value 0.33333334

Risk Assessment

 s contains 0.0010
if(s.equals(0.001)) { // Fails
  // Do something
}

Compliant Solution

This compliant solution uses the BigDecimal class and strips the trailing zeros so that future operations do not fail.

int i = 1;
BigDecimal d = new BigDecimal(Double.valueOf (i / 1000.0).toString()).stripTrailingZeros();
if(s.equals(0.001)) { // Passes
  // Do something
}

Risk Assessment

Relying on the string representation of floating point types can result in imprecise valuesConverting from float to double type when exact values are required can affect the precision of the converted value.

...

TODO

Related Vulnerabilities

Search for vulnerabilities resulting from the violation of this rule on the CERT website.HV-192

References

\[[API 06|AA. Java References#API 06]\] 
\[[JLS 05|AA. Java References#JLS 05]\] 
\[[Mak 02|AA. Java References#Mak 02]\] Section 3.7 "Another Look at Roundoff Errors"

...

FLP04-J. Check floating point inputs for exceptional values      07. Floating Point (FLP)      FLP30-J. Do not use floating point variables as loop counters