Name: mgC56079 Date: 12/02/97



> From ###@###.### Tue Dec 2 06:47 MSK 1997
> To: "'JCK-comments'" <###@###.###>
>
> In this code:
> String s1 = ""; // empty string
> String s2 = ""; // empty string
> int n1 = s1.indexOf( s2 );
> int n2 = s1.indexOf( s2, 0 );
>
> the JDK sets both the variables "n1" and "n2" to -1.
> That is, the JDK reports that the empty string cannot
> occur within the empty string.
>
> This result both defies logic and is **NOT** in accord
> with the language specification.
>
> Consider: The JDK and the spec agree that the empty
> string is a subset of any non-empty string. That is,
> "xyz".indexOf( "" )
> is always zero. And if we put that in terms of sets, it
> makes sense: "The empty set is the subset of every
> set."
>
> But the JDK does not follow through on the full and proper
> definition: "The empty set is the subset of every set,
> INCLUDING the empty set."
>
> And, although it is obscure and difficult to follow, the JDK
> spec agrees with this.
>
> Specifically, in 20.12.26, the spec reads (in part):
>
> [The method indexOf returns] the smallest value
> "k" such that
> this.startsWith(str, k ) && ( k >= fromIndex )
> is true.
>
> In either of the cases shown above, "fromIndex" is zero,
> so "k" must simply be greater to or equal to zero, so
> the task is to find
>
> ...the smallest non-negative value "k" such that
> this.startsWith( str, k )
> is true.
>
> Well, the smallest non-negative value is, of course, zero.
> So let's test to see what
> this.startsWith( str, 0 )
> means.
>
> We now go to section 20.12.21, which says (after ignoring
> the subcases that do not apply in this situation):
>
> [the method startsWith returns] true if and only if
> this.subString( toffset ).startsWith( prefix )
> [is true].
>
> But here the value of "toffset" is zero, and 20.12.31 that,
> effectively, says that
> substring( 0 )
> returns a String object that is identical to the original.
>
> So we are finally effectively reduced to finding out if
> this.startsWith( prefix )
> is true, given that "prefix" is the empty string.
>
> And section 20.12.20 says SPECFICALLY
> "Note that the result will be true is the argument
> is an empty string..."
>
> In other words,
>
> "".startsWith( "" ) is TRUE
> so "".substring( 0 ).startsWith("") is TRUE
> so "".startsWith( "", 0 ) is TRUE
> so "".startsWith( "", k ) is TRUE for k==0
> and 0 is the minimum value of k such that k is >= 0
> so "".indexOf( "", 0 ) is 0.
>
> ***************
>
> Note that our in-house tests found this discrepancy between the
> JDK and the spec. So far as I know, the JCK String tests did
> NOT find this.
>
> Our version of class java.lang.String follows the spec, not the JDK,
> in this matter.


======================================================================