overhead of reflection info

[Adam Megacz]

Per Bothner <per@bothner.com> writes:
> I think such a flag will be worthwhile.  I think it would be easier
> and do more if you also disable classloading, interpretation, and JNI,
> since those are big uses of reflection info.
> I'd keep just enough to keep the "JDK 1.0" level of reflection:

Absolutely.  There are a large number of quasi-java environemnts
(jump2, etc) which support this exact level of funcitonality: jdk1.1
minus java.lang.reflect (but including Class.forName()), minus
ClassLoader.fromBytes() and minus JNI.  In fact, this is the subset of
Java which XWT is designed to run on.

> have to make sure all classes can be initialized at compile-time,
> and reject any non-trivial class initialization code.  I don't know
> how much of the library should have to be changed - a good chunk,
> I suspect, so this is not a trivial change.

Yeah, I think this is a pretty major change; most Java code
inadvertently relies on Java's ability to infer the order in which
<clinit>s run at runtime.

Still, it would be really great to eliminate all those
is-the-class-loaded checks.  I wonder if the MMU (write-protecting
pages) could be used to do this.  Put all of the methods/fields
related to a class on a contiguous set of pages and write protect them
until <clinit> runs.

  - a