8058779: Faster implementation of String.replace(CharSequence, CharSequence) (original) (raw)

Xueming Shen xueming.shen at oracle.com
Tue May 26 17:17:16 UTC 2015

On 5/26/15 10:08 AM, Ivan Gerasimov wrote:

Thank you Roger for looking at this!

On 26.05.2015 19:40, Roger Riggs wrote: Hi Ivan,

Did you consider doing the optimization inside of Pattern.compile/replaceAll. That would have a broader impact and benefit. In Pattern.compile the flag LITERAL can be combined with several other flags, which makes things more complex. However, in String.replace we've got a special case, which can be optimized in a straight-forward (though a little bit verbose) way.

It probably makes big difference to access those characters via CharSequence.charAt() and the direct internal char[] access. Though avoiding allocating those objects created by Pattern/Matcher/SB definitely helps.

-Sherman

Sincerely yours, Ivan

Roger

On 5/26/2015 12:36 PM, Ivan Gerasimov wrote: Thank you Mark for looking at this! On 26.05.2015 18:39, mark.reinhold at oracle.com wrote: Your micro-benchmark improvements are significant, but do you have evidence to suggest that the performance of this method is actually critical to real applications?

In other words, is the added code complexity really worth it? The enhancement request contains a few links to the discussions of this method's performance at open forums. The most frequent suggestion is to use alternatives from 3rd party libraries. That should prove the benefits of this fix -- by improving performance we can keep some users from moving away from JDK :) grep shows that langtools would also benefit from making replace() faster. Sincerely yours, Ivan - Mark

More information about the core-libs-dev mailing list