hg: hsx/hsx16/baseline: 6909480: Disable Escape Analysis in jdk6u18

Osvaldo Doederlein opinali at gmail.com
Fri Dec 11 07:50:29 PST 2009

I support Ben's request - this wouldn't be the first time that unstable,
beta-quality features are backported to production JDKs. I remember the G1
backport to JDK 6u14 - I've tested it, and found it very far from adequate
for production use; confirmed by the large number of G1 changes (not only
fixes) that followed in further JDK7 builds. Still it was available (jut not
enabled by default) in a production JDK, which is good so interested
developers like me could test it. (Being available in JDK7 builds is not
good enough, it's a new platform release with tons of changes and still
potentially much more unstable, and I wouldn't risk using it in production
even if I found that its G1 was kickin' for my app... but I could do that
with a fresh 6uXX release, btw I have high hopes that 6u18's G1 will be the
first I can put in production.)

If the idea was not destabilize JDK 6u18 with new EA code from JDK7, then OK
let's keep this dangerous code away from the production VM. But it seems
everything is already merged into 6u18, so I suppose (I'm not a HotSpot dev)
the EA stuff is well isolated and doesn't risk any regressions if not
enabled. So the restriction is completely artificial, you're just removing
the little red button I must press to enable EA. Why not just leaving it
there? If some user is irresponsible enough to enable a VM feature that's
documented as experimental, and uses it in production without sufficient
testing to assert that its existing bugs don't hit the app, it's the user's
fault not Sun's. Unless said feature is SO much buggy that there's no chance
that any nontrival app won't be screwed.


2009/12/11 Ben Evans <ben.evans at db.com>

> hotspot-dev-bounces at openjdk.java.net wrote on 11/12/2009 06:39:32:
> >
> > Yes, users will not be able to switch it on in jdk6u18.
> >
> > Currently it is still experimental optimization
> > and it could produce incorrect results.
> Can you be more specific, please. All our testing of this feature shows it
> as good-to-go.
> Exactly what aspect of it is still troubling?
> > Use early access jdk7 if you want to try it.
> Been there, done that.
> EA shows consistent benefit of at least a few %, and we want to switch it
> on in production, and can't do that with a non Sun-supported build.
> Please give us as much detail as you can (contact me offlist if necessary)
> - we'd really like to know what's caused this decision, and if possible see
> some code with corresponding PrintAssembly traces where errors could occur.
> Thanks,
> Ben
> --
> Ben Evans
> eFX Algorithmic Trading
> Deutsche Bank, London
> Office: +44 (0)20 7541 3953
> ---
> This e-mail may contain confidential and/or privileged information. If you
> are not the intended recipient (or have received this e-mail in error)
> please notify the sender immediately and delete this e-mail. Any
> unauthorized copying, disclosure or distribution of the material in this
> e-mail is strictly forbidden.
> Please refer to http://www.db.com/en/content/eu_disclosures.htm for
> additional EU corporate and regulatory disclosures.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-dev/attachments/20091211/cfb4aa4f/attachment.html 

More information about the hotspot-dev mailing list