RFR: JDK-8227831: Avoid using volatile for write-once, read-many class field

[Langer, Christoph]

Hi Ogata,

this seems to make sense. So, +1 from my end.

Can you please add a space after "if" in line "734                 if(langReflectAccess == null) {"?

Thanks
Christoph

> -----Original Message-----
> From: core-libs-dev <core-libs-dev-bounces at openjdk.java.net> On Behalf
> Of Kazunori Ogata
> Sent: Mittwoch, 17. Juli 2019 08:49
> To: core-libs-dev at openjdk.java.net
> Subject: RFR: JDK-8227831: Avoid using volatile for write-once, read-many
> class field
>
> Hi,
>
> May I have a review for "JDK-8227831: Avoid using volatile for write-once,
> read-many class field"?
>
> In jdk.internal.reflect.ReflectionFactory, there is a private class field
> named "langReflectAccess", which is referenced every time when the library
> handles various reflective operations.  This field is initialized on the
> first access to the ReflectionFactory class.  This field is declared as
> volatile to avoid (or reduce) race condition between initialization and
> references to the field.
>
> On the platforms with weak memory model (i.e, POWER and ARM), reading a
> volatile variable requires memory fence and incurs overhead.  So it is
> preferable to avoid use of volatile for such a write-once, read-many
> variable.
>
> langReflectAccess can be modified only in setLangReflectAccess() method.
> So we can avoid using volatile by modifying setLangReflectAccess() to use
> a synchronized block to avoid race condition.  This change reduced elapsed
> time of a micro benchmark by 9%, which repeatedly invoke
> Class.getMethods().
>
>
> Bug: https://bugs.openjdk.java.net/browse/JDK-8227831
>
> Webrev: http://cr.openjdk.java.net/~ogatak/8227831/webrev/
>
>
> Regards,
> Ogata