jjs and the tiny heap that couldn't

[Claes Redestad]

On 2016-08-26 05:01, Sundararajan Athijegannathan wrote:
> I don't think jjs was ever GC tuned! Most likely default copied/adapted
> from somewhere else!
>
> Please file a bug with your findings.

Sure:

https://bugs.openjdk.java.net/browse/JDK-8164848

>
> PS. jjs of jdk9 is very different (in interactive mode) from that of
> jdk8u. jdk9 versions on jline+many other desktop stuff and so would load
> lot more code.

Right, I noticed that on jdk8 running the exit.js script loads fewer
classes (1827), while on jdk9 we load 2359 classes right now (down from
2450 OOTB thanks to the --generate-jli-classes plugin and other work on
improving indy bootstrap[1]).

When I link a custom image with a --generate-jli-classes configuration
created by running jjs exit.js I can currently get it down to 2114
classes loaded (meaning we're avoiding generating more than 300 LF
classes at bootstrap), which together with the small heap tuning puts
startup of jjs on 9 well on par with 8.

Thanks!

/Claes

[1] https://bugs.openjdk.java.net/browse/JDK-8086045