RFR (XS) CR 8014509: @Contended: explicit default value behaves differently from the implicit value

[Vladimir Kozlov]

Aleksey,

Why you need to test 2 fields but not class annotation?

It would be nice to have a comment in parse_annotations() explaining what formats of Contended are expected.
And the code could be rearranged to simplify it a little:

      if (id == AnnotationCollector::_sun_misc_Contended) {
        u2 group_index = 0; // default contended group
        if (count == 1
            && s_size == (index - index0)  // match size
            && s_tag_val == *(abase + tag_off)
            && member == vmSymbols::value_name()) {
          u2 group_index = Bytes::get_Java_u2(abase + s_con_off);
          if (_cp->symbol_at(group_index)->utf8_length() == 0) {
            group_index = 0; // empty tag => default contended group
          }
        }
        coll->set_contended_group(group_index);
      }

Thanks,
Vladimir

On 5/15/13 8:36 AM, Aleksey Shipilev wrote:
> Hi,
>
> Please review the little parameter handling fix for @Contended:
>    http://cr.openjdk.java.net/~shade/8014509/webrev.01/
>
> Bottom-line, these guys should be equivalent:
>   public class Test {
>      private static class T0 {
>          @Contended("") private double double1;
>          @Contended("") private double double2;
>      }
>      private static class T1 {
>          @Contended()   private double double1;
>          @Contended()   private double double2;
>      }
>      private static class T2 {
>          @Contended     private double double1;
>          @Contended     private double double2;
>      }
> }
>
> They are not, because we handle the default value in a weird way.
>
> Testing:
>    - 8014509 regression test on Linux x86_64/fastdebug
>    - full JPRT cycle against hotspot-rt
>
> Thanks,
> Aleksey.
>
>
>