Re: IdentityHashMap & natStackTrace

[Bryce McKinlay <mckinlay at redhat dot com>]

Ranjit Mathew wrote:

> As Chris points, due to operator precedence weirdness this
> becomes a left shift by 3 or multiplication by 23, in line
> with what you observed.
>
> This was introduced almost two years ago in a merge from
> Classpath:
>
> http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libjava/java/util/IdentityHashMap.java.diff?r1=1.8&r2=1.9&only_with_tag=MAIN
>
> Relevant change:
>
>  // This isn't necessarily prime, but it is an odd number of key/value
>  // slots, which has a higher probability of fewer collisions.
> - table = new Object[old.length * 2 + 2];
> + table = new Object[old.length << 1 + 2];
>  

I'd like to make a plea to Classpath developers not to make this kind of 
change in the future. Not only does the new code introduce a bug, it is 
also more difficult to read, and it won't make any performance 
difference on any modern compiler. Its trivial for optimizers to 
substitute constant multiply's to left-shift operations, so please let 
the compiler take care of it!

Regards

Bryce