False positive: Redundant nullcheck

[willhains]

I have a generic type as follows, which calls .toString() on an instance of the parameter type.

public final class Simple<T>
{
  private final @Nonnull T _raw;
  public Simple(@Nonnull T raw) { _raw = raw; }
  public @Nonnull String toString()
  {
    final String string = _raw.toString();
    return string == null ? "" : string;
  }
}

FindBugs gives a RCN_REDUNDANT_NULLCHECK_OF_NONNULL_VALUE warning on the null check of string, saying that it is "a known non-null value". However, this can be trivially show to be incorrect, as the following code demonstrates.

class Naughty
{
  public String toString() { return null; }
  public static void main(String... args)
  {
    final Simple<Naughty> simplyNaughty = new Simple<>(new Naughty());
    System.out.println(simplyNaughty.toString().length());
  }
}

If Simple.toString() did not have the null check on string, and return an alternate non-null value, then we would get an NPE thrown when calling .length(). Since _raw is of the unknown type T, FindBugs should not treat _raw.toString() as "a known non-null value".

[iloveeclipse]

toSting() should never return null.

[castocolina]

@iloveeclipse , you asume any third-party API class method dont return null

[iloveeclipse]

Yes, I assume that no one in the world should ever return null from toString(), because not doing so would violate the contract of Object.toString().

[castocolina]

But this rule is applicable for every variable, not only the return of toString(). This assumption fails my code every time when check variables depends on third-party API methods. My you what to add rules for null return in toString :-p

[iloveeclipse]

Please note, that this project is dead. If you want to contribute something, please do it via https://github.com/spotbugs/spotbugs.