RFR: 8306308: (ch) Writer created by Channels::newWriter may lose data

[Brian Burkhalter]

On Tue, 17 Oct 2023 05:57:37 GMT, Alan Bateman <alanb at openjdk.org> wrote:

> [...] I think you'll need to create to some benchmarks to demonstrate that the channel handling in StreamEncoder can be removed.

I don't know whether this was the intended sort of benchmark,

    @Param({"4096", "8192", "16384", "32768"})
    public int capacity;

    @Setup
    public void create() throws IOException {
        cbuf = new char[capacity];
        Arrays.fill(cbuf, (char)42);
        fc = FileChannel.open(path, CREATE, TRUNCATE_EXISTING,
                               WRITE, DSYNC);
        writer = Channels.newWriter(fc, "UTF-8");
    }

    @Benchmark
    public char[] testMethod() throws IOException {
        writer.write(cbuf);
        return cbuf;
    }

but I see little difference between the master code results:

Benchmark               (capacity)   Mode  Cnt      Score      Error  Units
MyBenchmark.testMethod        4096  thrpt    5  46120.101 ± 3514.975  ops/s
MyBenchmark.testMethod        8192  thrpt    5  23014.621 ±  703.382  ops/s
MyBenchmark.testMethod       16384  thrpt    5  11355.081 ±  357.569  ops/s
MyBenchmark.testMethod       32768  thrpt    5   5669.087 ±  401.619  ops/s

and those with `WritableByteChannel` support removed from `StreamEncoder`:

Benchmark               (capacity)   Mode  Cnt      Score      Error  Units
MyBenchmark.testMethod        4096  thrpt    5  46431.382 ± 1641.437  ops/s
MyBenchmark.testMethod        8192  thrpt    5  23079.163 ± 1333.578  ops/s
MyBenchmark.testMethod       16384  thrpt    5  11594.550 ±  793.716  ops/s
MyBenchmark.testMethod       32768  thrpt    5   5619.294 ± 1252.580  ops/s

-------------

PR Comment: https://git.openjdk.org/jdk/pull/16207#issuecomment-1769164290