RFR: 8215380: Backout accidental change to String::length
Stuart Marks
stuart.marks at oracle.com
Thu Dec 13 23:43:30 UTC 2018
Hi Claes,
Thank for catching this. Looks good.
Note that the original changeset (JDK-8215281) went into JDK 12 before the RDP1
fork, so this fix should also go into JDK 12. That's a different repo now. It
will then be auto-propagated to the JDK mainline ("JDK 13").
Since JDK-8215380 is a P3 bug, it's eligible to go into JDK 12 after RDP1
without any further approval.
s'marks
On 12/13/18 3:20 PM, Claes Redestad wrote:
> Hi,
>
> I need to revert an accidental change to String.length
>
> Bug: https://bugs.openjdk.java.net/browse/JDK-8215380
> Patch inlined below
>
> Running the accidentally pushed version in naive microbenchmarks showed
> that avoiding the shift operation can improve throughput of str.length()
> by a small measure (~1.06x) for latin1-only inputs, neutral for mixed
> or utf16-only inputs, but also adds a branch (visible in profiling)
> which could blow up in more real cases. Regardless, it should be
> reviewed and discussed on it's own merits. Sorry!
>
> /Claes
>
> diff -r 8bf9268df0e2 src/java.base/share/classes/java/lang/String.java
> --- a/src/java.base/share/classes/java/lang/String.java Thu Dec 13 15:31:05
> 2018 +0100
> +++ b/src/java.base/share/classes/java/lang/String.java Thu Dec 13 23:59:43
> 2018 +0100
> @@ -664,7 +664,7 @@
> * object.
> */
> public int length() {
> - return isLatin1() ? value.length : value.length >> UTF16;
> + return value.length >> coder();
> }
>
> /**
More information about the core-libs-dev
mailing list