8230342: LineNumberReader.getLineNumber() returns inconsistent results after EOF

[Brian Burkhalter]

Hi Roger,

Thanks for the review.

> On Sep 9, 2019, at 7:35 AM, Roger Riggs <Roger.Riggs at oracle.com> wrote:
> 
> I would lean toward updating the spec to reflect the current implementation.

It seems strange however that if one read an entire stream using read() in one case and readLine() in another that the results would differ.

> A simple program that uses readline and prints the line number and line would show a duplicate line number.
> 
> try (LineNumberReader r =
>             new LineNumberReader(Files.newBufferedReader(Path.of(args[0])))) {
>    String s;
>    while ((s = r.readLine()) !=null) {
>        System.out.printf("%3d: %s%n", r.getLineNumber(), s);
>    }
> }
> 
>   1: 123
>   1: ABC

For a file with this contents (two lines)

123
ABC

I see the same result for the above code both with and without my patch:

  1: 123
  2: ABC

> Other questions and comments:
> 
> - Is the use of AtomicBoolean due to concurrency concerns?
>   If not, a new boolean[1] would be less overhead

No concurrency so that would be better.

> - BufferedReader also keeps track of CR/LF pairs and I've thought it would simpler
>   not to duplicate the line number counting logic in LineNumberReader.
>   If BufferedReader kept track of the line number, it could expose a package private field
>   to LineNumberReader to make it visible to the get/SetLineNumber methods of LineNumberReader.

That could be a good simplification.

Thanks,

Brian